Datasets:

bzantium

/

LongPerplexity

like 0

in time - reversal experiments a signal emitted by a localized source is recorded at an array of transducers . it is then re - emitted into the medium reversed in time , that is , the part of the signal that is recorded first is sent back last . because of the time - reversibility of the wave equation the back - propagated signal refocuses approximately at the location of the original source because the array is limited in size . a striking experimental observation is that the presence of inhomogeneities in the medium improves the refocusing resolution . the explanation for this super - resolution is multipathing : waves in complex media that are captured by the recording array have undergone multiple scattering making it effectively larger than its physical size . another important feature of super - resolution in time reversal is that the refocused signal does not depend on the realization of the random medium . that is , the refocused signal is deterministic . super - resolution and self - averaging of refocused signals in complicated media has been observed both in laboratory experiments ( see reviews @xcite and references therein ) and in underwater acoustic wave propagation over long distances ( tens of kilometers ) @xcite . time - reversal techniques have numerous applications ranging from medicine to communications and , more recently , imaging in random media @xcite . the first mathematical analysis of time - reversal in random media was given by clouet and fouque @xcite , who analyzed refocusing and self - averaging of time - reversed pulses in a one - dimensional layered random medium . their result was extended to a three - dimensional layered medium in @xcite . super - resolution in spatial refocusing and its statistical stability for multi - dimensional waves in random media was analyzed in @xcite , in a remote - sensing regime where the paraxial or parabolic wave equation can be used . the refocusing of the average signal in a full three - dimensional medium , in the regimes of random geometrical optics and radiative transfer ( transport ) , was studied in @xcite . we also mention that another source of multipathing is the mixing of waves by the boundaries in an ergodic cavity . this has been studied experimentally in @xcite and mathematically in @xcite . the purpose of this paper is to analyze time reversal in the radiative transfer regime using the parabolic wave equation , when the waves interact fully with the random inhomogeneities . we prove mathematically that the refocused signal is self - averaging , which means that it does not depend on the realization of the random medium . the mathematical quantitiy that we analyze is the wigner measure of a pair of oscillatory solutions of the random schrdinger equation . in the present setting , the random potential depends in a markovian way on the variable @xmath1 , the main direction of propagation of the waves . this allowed us to use in @xcite a martingale method to prove that the average of the wigner distribution converges to a solution of the radiative transfer equation . in this paper we use additional regularity of the wigner measure , available in time - reversal when there is some blurring at the recording array , to show that the whole wigner distribution , and not only its average , converges weakly , as a schwartz distribution and in probability , to the deterministic solution of the transport equation . the blurring at the recording array provides a priori bounds for the wigner transform in @xmath0 . these bounds and the markovianity of the random potential in the direction of propagation make the time - reversal problem more tractable mathematically and allow us to prove in a fairly simple and straightforward manner self - averaging of the time - reversed signal . we recall that the wigner transform is a convenient tool to analyze high frequency wave propagation in deterministic @xcite and random media @xcite . introduced by wigner in 1932 @xcite , it has been used extensively in the mathematical literature recently . convergence of the average wigner distribution to the solution of the radiative transfer equation was first proved by h. spohn in @xcite for time - independent potentials on small time intervals . this result was extended to global in time convergence by l. erds and h .- yau @xcite . these proofs involve infinite neumann ( diagrammatic ) expansions for the solution of the schrdinger equation and are quite involved technically . the corresponding problem with time - dependent potentials is much simpler mathematically . it was treated by us in @xcite in the markovian case , and by f. poupaud and a. vasseur @xcite in the case of finite - range time correlations . in this paper we use the fact that the wigner family arising in time - reversal problems is more regular than the usual one because blurring is added at the recording array . this provides some additional regularity usually obtained by considering mixtures of states as , for instance , in @xcite . the paper is organized as follows : we describe the scaling and obtain an expression for the back - propagated signal in terms of the wigner transform in section [ sec : signal ] . the main result and assumptions on the random medium are formulated in section [ sec : mainresult ] . the proofs are presented in section [ sec : wigner ] . this work was partially supported by onr grants n00014 - 02 - 1 - 0088 and n00014 - 02 - 1 - 0089 . g. bal was supported in part by nsf grant dms-0072008 , g. papanicolaou by grants afosr f49620 - 01 - 1 - 0465 and nsf dms-9971972 , and l. ryzhik in part by nsf grant dms-9971742 and the alfred p. sloan foundation . the pressure field @xmath2 satisfies the scalar wave equation @xmath3 here @xmath4 is the local wave speed that we will assume to be random , and the laplacian operator includes both direction of propagation , @xmath1 , and the transverse variable @xmath5 . in the physical setting , we have @xmath6 . we consider dimensions @xmath7 to stress that our analysis is independent of the dimension . if we assume that at time @xmath8 , the wave field has a `` beam - like '' structure in the @xmath1 direction , and if back - scattering may be neglected , we can replace the wave equation by its parabolic approximation @xcite . more precisely , the pressure @xmath9 may be approximated as @xmath10 where @xmath11 satisfies the schrdinger equation @xmath12 with @xmath13 the transverse laplacian in the variable @xmath14 . the refraction index @xmath15 , and @xmath16 in ( [ eq : ansatz ] ) is a reference speed . the rigorous passage to the parabolic approximation from the wave equation has been analyzed in @xcite in the deterministic case , and @xcite in a one dimensional random medium . a formal derivation of the paraxial approximation that leads to the radiative transfer regime is given below in section [ sec : scaling ] . we assume that the original source is located in the plane @xmath17 , and the time - reversal mirror is located in the plane @xmath18 as depicted in figure [ fig1 ] . during the first stage of a time reversal experiment the signal propagates for a time @xmath19 , or equivalently , over a distance @xmath20 , so that the signal arriving at the time - reversal mirror is given by @xmath21 where the green s function solves @xmath22 then the signal is time reversed . for three - dimensional acoustic pulses , this means that the pressure field is kept unchanged and that the sign of its time derivative is reversed . in the parabolic approximation , this is equivalent to phase conjugation , where @xmath11 is replaced by its complex conjugate @xmath23 . we assume that the recording array occupies a compact subset of the plane @xmath18 , and introduce a real - valued aperture function @xmath24 . it represents the restriction of the signal onto the array , and possible amplification by the array that may vary from one receiver to another . in the absence of amplification it is given by the characteristic function of an array set @xmath25 . we also allow for some blurring of the recorded signal , modeled by a convolution with kernel @xmath26 . the signal at @xmath27 after time reversal takes then the form @xmath28 the last step consists in letting the signal propagate back to the origin @xmath17 and time reversing it one more time @xmath29 the last step , phase - conjugating at the source , is not performed in real physical experiments but is convenient if we need to compare the back - propagated signal to the original signal . it affects neither the degree of refocusing nor the self - averaging effect . the back - propagated signal in time is given at the plane @xmath17 by @xmath30 we will be interested in the sequel in the refocusing and self - averaging properties of the back - propagated signal @xmath31 for each fixed frequency . self - averaging of the time signal @xmath32 then follows from ( [ eq : psib - time ] ) when @xmath33 has compact support in @xmath34 . an interesting conclusion of this paper is that the additional blurring by convolution with the kernel @xmath35 makes the signal @xmath31 self - averaging for every frequency @xmath34 . this should be contrasted with the situation studied in @xcite , where no blurring was introduced and self - averaging was observed only for the full time signal . we recall the formal passage from the reduced wave equation to the parabolic equation ( [ eq : sch ] ) described in the appendix to @xcite and explain how the radiative transfer scaling arises in this context . we start with the reduced wave equation @xmath36 and look for solutions of ( [ red - wave ] ) in the form @xmath37 then we obtain @xmath38 the refraction index @xmath39 is weakly fluctuating so that it has the form @xmath40 where @xmath41 models random fluctuations that will be described in detail in section [ sec : random - refr - ind ] . here , @xmath42 and @xmath43 are the correlation lengths of @xmath41 in the transverse and longitudinal directions , respectively , and the small parameter @xmath44 measures the strength of the fluctuation . the waves propagate over a distance @xmath45 in the @xmath14-plane and @xmath46 in the @xmath1-direction , and we rescale @xmath14 and @xmath1 accordingly . we also introduce a carrier wave number @xmath47 and replace @xmath48 , @xmath49 being the non - dimensional wave number ( we drop the prime below ) . the physical parameters determined by the medium are the length scales @xmath42 , @xmath43 and the non - dimensional parameter @xmath50 . we now explain the scaling of the parameters @xmath45 , @xmath46 and @xmath47 in the radiative transfer regime . equation ( [ red - unscaled ] ) in the non - dimensional variables @xmath51 , @xmath52 becomes after we drop the primes @xmath53 we introduce the following small parameters : @xmath54 and rewrite ( [ red - scaled ] ) as @xmath55 let us assume now that @xmath56 and denote @xmath57 . then ( [ red - param ] ) becomes after multiplication by @xmath58 @xmath59 observe that when @xmath60 and @xmath61 , the first term in ( [ red - param1 ] ) is small and may be neglected in the leading order since @xmath62 . then ( [ red - param1 ] ) becomes @xmath63 which is the parabolic wave equation ( [ eq : sch ] ) in the radiative transfer scaling . the second condition in ( [ parab - assump ] ) implies that the carrier wave number should be chosen as @xmath64 . then @xmath65 implies that @xmath66 . therefore the correlation length in the longitudinal direction @xmath1 should be much longer than in the transverse plane @xmath14 . furthermore , we should have @xmath67 which in turn implies that the fluctuation strength @xmath68 . given that these constraints are satisfied the last condition in ( [ parab - assump ] ) implies that the spatial scales @xmath45 and @xmath46 should be chosen according to @xmath69 note that the constraint @xmath66 implies that @xmath70 , which is the usual constraint for the validity of the parabolic approximation . the connection between the relations of the physical parameters and the resulting scaling of the parabolic approximation is further discussed in @xcite . the rigorous passage to the parabolic approximation in one dimension in a random medium in a similar scaling is discussed in @xcite . to define what we mean by the quality of the refocused signal , we assume that the initial pulse is centered around a point @xmath71 with support comparable to the transverse correlation length . then the initial data for the schrdinger equation ( [ eq : sch - rescaled ] ) becomes in the rescaled variables @xmath72 the transducers should obviously be capable of capturing signals of frequency @xmath73 so that the blurring occurs on the scale of the source . we therefore replace @xmath26 by @xmath74 . finally , we are interested in the refocusing properties of @xmath75 in the vicinity of @xmath71 and consequently introduce the scaling @xmath76 . we then recast ( [ eq : psib ] ) as @xmath77 where @xmath78 we observe that @xmath79 , so that @xmath80 let us now introduce the auxiliary function @xmath81 as in @xcite , @xmath82 which satisfies the initial value problem @xmath83 it physically describes the component with wave vector @xmath84 that is sent back to the plane @xmath17 by the array . we then define the wigner transform as @xmath85 where @xmath86 the wigner transform @xmath87 is more regular than @xmath88 because of additional averaging in the wave vector @xmath84 . more , precisely , while the family @xmath88 is not uniformly bounded in @xmath0 , the family @xmath87 has a uniform bound in @xmath0 . this regularizing effect is essentially the same as the one obtained by considering mixtures of states for the schrdinger equation as in @xcite . we can then recast ( [ eq : psibeps2 ] ) as @xmath89 the above formula shows that the asymptotic behavior of @xmath90 as @xmath91 is characterized by that of the wigner transform @xmath92 . this section presents our two main results . the first one describes the self - averaging of the back - propagated signal @xmath93 in ( [ eq : psibeps3 ] ) . its proof is based on the second theorem , of independent interest , which shows that the wigner transform converges in probability to a unique deterministic limit , solution of a transport equation . [ thm1 ] let the array function @xmath94 and the radially symmetric filter @xmath95 be in @xmath96 , while @xmath97 for a given @xmath98 . assume also that the refraction index @xmath39 satisfies assumptions in section [ sec : random - refr - ind ] below . then for each @xmath99 the back - propagated signal @xmath100 given by ( [ eq : psibeps3 ] ) converges in probability and weakly in @xmath101 as @xmath102 to the deterministic function @xmath103 the function @xmath104 satisfies the transport equation @xmath105 with initial data @xmath106 and where the operator @xmath107 is defined by @xmath108 here @xmath109 is the fourier transform of the correlation function of @xmath41 , defined by ( [ eq : tilder ] ) below . the convergence of the wigner transform is described by the following theorem . [ thm2 ] under the assumptions of theorem [ thm1 ] the wigner distribution @xmath87 converges in probability and weakly in @xmath110 to the solution @xmath104 of the transport equation ( [ eq : treqformal ] ) . more precisely , for any test function @xmath111 the process @xmath112 converges to @xmath113 in probability as @xmath102 , uniformly on finite intervals @xmath114 . here , @xmath115 is the usual scalar product in @xmath116 . theorem [ thm1 ] is then a corollary of theorem [ thm2 ] and lemma [ lemma2 ] below . theorem [ thm1 ] provides a mathematical explanation for the important properties of the time - reversal experiment . first , it ensures that the back - propagated signal is essentially independent of the realization of the random medium in the high - frequency limit since @xmath117 is deterministic . second , it provides a quantitative description of the back - propagated field , which may be written as @xmath118 the kernel @xmath119 is the fourier transform in @xmath120 of the solution of ( [ eq : treqformal ] ) . we deduce from this formula that the back - propagated signal has much better refocusing properties in random media than in homogeneous media . as is explained in @xcite , the reason is that the limiting wigner measure @xmath121 is more regular in its @xmath122 variable when the right - hand side in ( [ eq : treqformal ] ) is present . this leads to better spatial decay of @xmath123 and hence @xmath124 is localized tighter in a random medium than in a homogeneous medium . the improvement of the refocusing in random media has been carefully analyzed theoretically and numerically in @xcite , in the fokker - plank approximation of ( [ eq : treqformal ] ) . we summarize here the main steps of the derivation of the results stated in theorems [ thm1 ] and [ thm2 ] . we first obtain in lemma [ lemma1 ] uniform bounds in @xmath0 for the wigner transform @xmath87 independently of the realization of the random medium . lemma [ lemma2 ] then shows that theorem [ thm1 ] is a consequence of theorem [ thm2 ] . the main mathematical restriction is to assume that the random potential @xmath125 is markovian in its first variable . we refer to section [ sec : random - refr - ind ] for the details of its construction . following our previous work in @xcite , we show the convergence of the expectation @xmath126 for every test function @xmath127 . its proof is simpler ( and less general ) than in @xcite because of the available a priori @xmath0 bounds on the wigner transform , which satisfies the following cauchy problem @xmath128 with @xmath129.\ ] ] since @xmath34 is a fixed parameter , we set @xmath130 without loss of generality . the solution of ( [ eq : weps ] ) with initial data in @xmath0 is understood in the sense that for every smooth test function @xmath131 , we have @xmath132 here , we have used that @xmath133 is a self - adjoint operator for @xmath115 . therefore for a smooth function @xmath134 we obtain @xmath135 , where @xmath136 is the solution of the backward problem @xmath137 with the terminal condition @xmath138 . this defines the process @xmath139 in @xmath140 and generates a corresponding measure @xmath141 on the space @xmath142;l^2({\mathbb r}^{2d}))$ ] of continuous functions in time with values in @xmath0 . the measure @xmath141 is actually supported on paths inside a ball @xmath143 with the constant @xmath144 as in lemma [ lemma1 ] . the set @xmath145 is the state space for the random process we have proved in @xcite and will use in the sequel that the family @xmath141 is tight : [ lemma - tight ] the family of measures @xmath141 is weakly compact . the proof of convergence of @xmath87 to its deterministic limit is obtained in two steps . let us fix a deterministic test function @xmath131 . we use the markovian property of the random field @xmath125 in @xmath1 to construct a first functional @xmath146;x)\!\to c[0,l]$ ] by @xmath147(z)=\langle w,\lambda\rangle(z)-\int_0^z\langle w,{\dfrac{\partial{\lambda}}{\partial{z}}}+{\mathbf k}\cdot\nabla_{\mathbf x}\lambda+ { \cal l}\lambda\rangle(\zeta)d\zeta\ ] ] and show that it is an approximate @xmath141-martingale . more precisely , we show that @xmath148(z)|{\cal f}_s\right\ } -g_\lambda[w](s)\right|\le c_{\lambda , l}\sqrt{{\varepsilon}}\ ] ] uniformly for all @xmath149;x)$ ] and @xmath150 . lemma [ lemma - tight ] implies that there exists a subsequence @xmath151 so that @xmath152 converges weakly to a measure @xmath153 supported on @xmath142;x)$ ] . weak convergence of @xmath141 and the strong convergence ( [ eq : approxmart ] ) together imply that @xmath154(z)$ ] is a @xmath153-martingale so that @xmath155(z)|{\cal f}_s\right\}-g_\lambda[w](s)=0.\ ] ] taking @xmath156 above we obtain as in @xcite the transport equation ( [ eq : treqformal ] ) for @xmath157 in its weak formulation . the second step is to show that for every test function @xmath131 the new functional @xmath158(z)=\langle w,\lambda\rangle^2(z)- 2\int_0^z\langle w,\lambda\rangle(\zeta ) \langle w,{\dfrac{\partial{\lambda}}{\partial{z}}}+{\mathbf k}\cdot\nabla_{\mathbf x}\lambda+ { \cal l}\lambda\rangle(\zeta)d\zeta\end{aligned}\ ] ] is also an approximate @xmath141-martingale . we then obtain that @xmath159 , which implies convergence in probability . it follows that the limit measure @xmath153 is unique and deterministic , and that the whole sequence @xmath141 converges . that @xmath160(z)$ ] is an approximate @xmath141-martingale uses very explicitly the uniform a priori @xmath0 bound on the wigner distribution @xmath87 . when the a priori bound is available only in a much larger space ( the space @xmath161 of @xcite ) , we are not able to prove that the functional @xmath162 is an approximate @xmath141-martingale , and actually suspect that the result is not true . we expect to observe that some long - range correlations survive for sufficiently singular initial data , which would imply that the limiting measure @xmath153 is no longer deterministic . we describe here the construction of the random potential @xmath125 . the refraction index @xmath39 in non - dimensional variables is a random function of the form @xmath163 where the non - dimensional parameter @xmath44 measures the strength of fluctuations . the mathematical analysis of the wigner distribution in random media can be obtained by the method of diagrammatic expansions of the solution of the schrdinger equation as in @xcite . however , here we assume that the random field @xmath125 is a markov process in @xmath1 , as in @xcite , and we are able to analyze the wigner transform @xmath87 directly . the markovian hypothesis is crucial to simplify the mathematical analysis because it allows us to treat the process @xmath164 as jointly markov . in addition , @xmath125 is assumed to be stationary in @xmath14 and @xmath1 , mean zero , and is constructed in the fourier space as follows . let @xmath165 be the set of measures of bounded total variation with support inside a ball @xmath166 @xmath167 and let @xmath168 be a mean - zero markov process on @xmath169 with generator @xmath170 . the random potential @xmath125 is given by @xmath171 and is real and uniformly bounded : @xmath172 we assume that the process @xmath125 is stationary in @xmath1 and @xmath14 with correlation function @xmath173 @xmath174 in terms of the process @xmath175 this means that given any two bounded continuous functions @xmath176 and @xmath177 we have @xmath178 here @xmath179 is the usual duality product on @xmath180 , and the power spectrum @xmath181 is the fourier transform of @xmath173 in @xmath14 : @xmath182 we assume that @xmath183 for simplicity and define @xmath109 as @xmath184 which is the space - time fourier transform of @xmath185 . we assume that the generator @xmath170 is a bounded operator on @xmath186 with a unique invariant measure @xmath187 @xmath188 and that there exists @xmath189 such that if @xmath190 then @xmath191 the simplest example of a generator with gap in the spectrum and invariant measure @xmath192 is a jump process on @xmath165 where @xmath193 given ( [ eq : expdecaynew ] ) , the fredholm alternative holds for the poisson equation @xmath194 provided that @xmath195 satisfies @xmath196 . it has a unique solution @xmath35 with @xmath197 and @xmath198 . the solution @xmath35 is given explicitly by @xmath199 and the integral converges absolutely because of ( [ eq : expdecaynew ] ) . the particular markovian model adopted in this paper is somewhat restrictive . however , the only information about the process that enters into the main result , the transport equation for the wigner distribution in the limit @xmath102 , is the two - point correlation function of @xmath41 . therefore , we expect that self - averaging of the wigner transform and convergence to the solution of the transport equation hold for much more general classes of random perturbations of the refraction index . * proof . * it is shown in @xcite that equation ( [ eq : weps ] ) preserves the @xmath0-norm of @xmath87 : @xmath203 . the initial conditions for @xmath87 are given by @xmath204 so that @xmath205 and ( [ weps - bd ] ) follows . the above calculation also shows that the initial wigner distribution @xmath206 converges to @xmath207 notice however that one can not expect strong convergence in @xmath0 of @xmath139 to @xmath208 , the solution of the transport equation ( [ eq : treqformal ] ) , because the latter has an @xmath0-norm that decreases as @xmath1 increases while the @xmath0-norm of @xmath87 is preserved . to obtain the approximate martingale property ( [ eq : approxmart ] ) , one has to consider the conditional expectation of functionals @xmath215 with respect to the probability measure @xmath216 on the space @xmath142;{\cal v}\times x)$ ] generated by @xmath217 and the cauchy problem ( [ eq : weps ] ) . the only functions we need to consider are actually of the form @xmath218 with @xmath219;{\cal s}({\mathbb r}^{2d})))$ ] . given a function @xmath215 let us define the conditional expectation @xmath220 the weak form of the infinitesimal generator of the markov process generated by @xmath216 is given by @xmath221\right)\lambda\right\rangle,\ ] ] hence @xmath222\right)\lambda\right\rangle(s)ds\ ] ] is a @xmath216-martingale . the operator @xmath223 is defined by @xmath224\psi({\mathbf x},{{\boldsymbol{\eta}}},{\mathbf k},\hat v)= \frac 1i\int_{\rm^d } \frac{d\hat v({\mathbf p})}{(2\pi)^d } e^{i{\mathbf p}\cdot { { \boldsymbol{\eta } } } } \left[\psi({\mathbf x},{{\boldsymbol{\eta}}},{\mathbf k}-\frac { \mathbf p}2)-\psi({\mathbf x},{{\boldsymbol{\eta}}},{\mathbf k}+\frac { \mathbf p}2)\right].\ ] ] the generator ( [ generator ] ) comes from equation ( [ eq : weps ] ) written in the form @xmath225 w_{\varepsilon}.\ ] ] given a test function @xmath226;{\cal s})$ ] we construct a function @xmath227 with @xmath228 bounded in @xmath229 uniformly in @xmath230 $ ] . the functions @xmath231 will be chosen so that @xmath232 for all @xmath230 $ ] . here @xmath233 is defined by ( [ g - eps ] ) with @xmath127 replaced by @xmath234 , and @xmath235 is defined by ( [ g - lambda ] ) . the approximate martingale property ( [ eq : approxmart ] ) follows from this . the functions @xmath236 and @xmath237 are as follows . let @xmath238 be the mean - zero solution of the poisson equation @xmath239 it is given explicitly by @xmath240.\ ] ] then we let @xmath241 . furthermore , the second order corrector is given by @xmath242 where @xmath243 is the mean - zero solution of @xmath244 which exists because @xmath245 , and is given by @xmath246(z,{\mathbf x},{{\boldsymbol{\eta}}}+r{\mathbf k},{\mathbf k},\hat v)\right].\end{aligned}\ ] ] using ( [ lambda1eq ] ) and ( [ lambda2eq ] ) we have @xmath247 + \frac 1{\varepsilon}q\right ) \left(\lambda+\sqrt{{\varepsilon}}\lambda_1^{\varepsilon}+{\varepsilon}\lambda_2^{\varepsilon}\right)\right\rangle \\ & & = \left\langle w , \left({\dfrac{\partial{}}{\partial{z}}}+{\mathbf k}\cdot\nabla_{\mathbf x}\right ) \lambda+{\cal l}\lambda\right\rangle + \left\langle w , \left({\dfrac{\partial{}}{\partial{z}}}+{\mathbf k}\cdot\nabla_{\mathbf x}\right)\left(\sqrt{{\varepsilon}}\lambda_1^{\varepsilon}+ { \varepsilon}\lambda_2^{\varepsilon}\right ) + \sqrt{{\varepsilon}}{\cal k}[\hat v,\frac { \mathbf x}{\varepsilon}]\lambda_2^{\varepsilon}\right\rangle\\ & & = \left\langle w,\left({\dfrac{\partial{}}{\partial{z}}}+{\mathbf k}\cdot\nabla_{\mathbf x}\right)\lambda+{\cal l}\lambda\right\rangle+\sqrt{{\varepsilon}}\langle w,\zeta_{\varepsilon}^\lambda\rangle\end{aligned}\ ] ] with @xmath248\lambda_2^{\varepsilon}.\ ] ] the terms @xmath249 above are understood as differentiation with respect to the slow variable @xmath14 only , and not with respect to @xmath250 . it follows that @xmath233 is given by @xmath251 and is a martingale with respect to the measure @xmath216 defined on @xmath252;x\times{\cal v})$ ] , the space of right - continuous paths with left - side limits @xcite . the estimate ( [ eq : approxmart ] ) follows from the following two lemmas . [ lemma1new ] let @xmath253;{\cal s}({\mathbb r}^{2d}))$ ] . then there exists a constant @xmath254 independent of @xmath230 $ ] so that the correctors @xmath255 and @xmath256 satisfy the uniform bounds @xmath257 and @xmath258 indeed , ( [ eq : lambda1bdnew ] ) implies that @xmath263 for all @xmath264 and @xmath261 , while ( [ eq : lambda2bdnew ] ) and lemma [ lemma3 ] imply that for all @xmath230 $ ] @xmath265 for all @xmath261 so that ( [ eq : approxmart ] ) follows . * proof of lemma [ lemma1new ] . * we only prove ( [ eq : lambda1bdnew ] ) . since @xmath266 , there exists a constant @xmath259 so that @xmath267 then we obtain using ( [ defcalv ] ) and ( [ eq : expdecaynew ] ) @xmath268\right| \\ & & \le { c}\int_0^\infty dr e^{-\alpha r}\sup_{\hat v } \int_{\rm^d } |d\hat v({\mathbf p})| \left[|\lambda(z,{\mathbf x},{\mathbf k}-\frac{{\mathbf p}}{2})|+ we show next that @xmath237 is uniformly bounded . we have @xmath270\right].\end{aligned}\ ] ] the second term above may be written as @xmath271\\ & & = -\int_{\rm^d } \frac{d\hat v({\mathbf p})}{(2\pi)^{d}}e^{i{\mathbf p}\cdot({\mathbf x}/{\varepsilon}+r{\mathbf k } ) } \int_0^\infty ds e^{sq}\int_{\rm^d } \frac{d\hat v({\mathbf q})}{(2\pi)^d } e^{is({\mathbf k}-{\mathbf p}/2)\cdot { \mathbf q}+i({\mathbf x}/{\varepsilon}+r{\mathbf k})\cdot { \mathbf q}}\\ & & \times \left[\lambda({\mathbf x},{\mathbf k}-\frac { \mathbf p}2-\frac { \mathbf q}2)- \lambda({\mathbf x},{\mathbf k}-\frac { \mathbf p}2+\frac { \mathbf q}2)\right ] \\ & & + \int_{\rm^d } \frac{d\hat v({\mathbf p})}{(2\pi)^d } e^{i{\mathbf p}\cdot({\mathbf x}/{\varepsilon}+r{\mathbf k } ) } \int_0^\infty ds e^{sq}\int_{\rm^d } \frac{d\hat v({\mathbf q})}{(2\pi)^d } e^{is({\mathbf k}+{\mathbf p}/2)\cdot { \mathbf q}+i({\mathbf x}/{\varepsilon}+r{\mathbf k})\cdot { \mathbf q}}\\ & & \times\left [ \lambda({\mathbf x},{\mathbf k}+\frac { \mathbf p}2-\frac { \mathbf q}2)- \lambda({\mathbf x},{\mathbf k}+\frac { \mathbf p}2+\frac { \mathbf q}2)\right].\end{aligned}\ ] ] therefore we obtain @xmath272\\ & & \le c\left[|{\cal l}\lambda({\mathbf x},{\mathbf k})|+ \frac{1}{(1+|{\mathbf x}|^{5d})(1+(|{\mathbf k}|-l)^{5d}\chi_{|{\mathbf k}|\ge 5l}({\mathbf k}))}\right]\end{aligned}\ ] ] and the @xmath0-bound on @xmath237 in ( [ eq : lambda1bdnew ] ) follows because the operator @xmath273 is bounded . the proof of ( [ eq : lambda2bdnew ] ) is very similar and is omitted . lemma [ lemma1new ] and lemma [ lemma3 ] together with ( [ geps2 ] ) imply the bound ( [ glambda - glambdaeps ] ) . the tightness of measures @xmath141 given by lemma [ lemma - tight ] implies then that the expectation @xmath274 converges weakly in @xmath140 to the solution @xmath275 of the transport equation for each @xmath230 $ ] . we now prove that for any test function @xmath127 the second moment @xmath276 converges to @xmath277 . this will imply the convergence in probability claimed in theorem [ thm2 ] . the proof is similar to that for @xmath278 and is based on constructing an appropriate approximate martingale for the functional @xmath279 , where @xmath280 is a test function , and @xmath281 . we need to consider the action of the infinitesimal generator on functions of @xmath282 and @xmath283 of the form @xmath284 where @xmath285 is a given function . the infinitesimal generator acts on such functions as @xmath286 where @xmath287\mu+ { \mathbf k}^j\cdot\nabla_{{\mathbf x}^j}\mu,\ ] ] with @xmath288\mu=\frac 1i \int_{\rm^d } d\hat v({\mathbf p})e^{i({\mathbf p}\cdot { { \boldsymbol{\eta}}}_1 ) } \left[\mu({\mathbf k}_1-\frac { \mathbf p}2,{\mathbf k}_2)-\mu({\mathbf k}_1+\frac { \mathbf p}2,{\mathbf k}_2)\right]\ ] ] and @xmath289\mu=\frac 1i \int_{\rm^d } d\hat v({\mathbf p})e^{i({\mathbf p}\cdot { { \boldsymbol{\eta}}}_2 ) } \left[\mu({\mathbf k}_1,{\mathbf k}_2-\frac { \mathbf p}2)-\mu({\mathbf k}_1,{\mathbf k}_2+\frac { \mathbf p}2)\right].\ ] ] therefore the functional @xmath290 + { \cal k}_2[\hat v,\dfrac{{\mathbf x}_2}{{\varepsilon } } ] ) \big ) \mu\right\rangle(s)ds \nonumber\end{aligned}\ ] ] is a @xmath291 martingale . we let @xmath292 be a test function independent of @xmath283 , where @xmath293 , and @xmath294 . we define an approximation @xmath295 we will use the notation @xmath296 and @xmath297 . the functions @xmath298 and @xmath299 are to be determined . we now use ( [ generator2 ] ) to get @xmath300\mu \right\rangle \nonumber\\ & & + \left\langle w\otimes w , \left(q+{\displaystyle\sum}_{j=1}^2 { \mathbf k}^j\cdot\nabla_{{{\boldsymbol{\eta}}}^j } \right ) \mu_2 + { \displaystyle\sum}_{j=1}^2 { \cal k}_j\left[\hat v,{{\boldsymbol{\eta}}}^j\right ] \mu_1 + { \dfrac{\partial{\mu}}{\partial{z}}}+ { \displaystyle\sum}_{j=1}^2{\mathbf k}^j\cdot\nabla_{{\mathbf x}^j } \mu \right\rangle \nonumber\\ & & + \sqrt{\varepsilon}\left\langle w\otimes w , { \displaystyle\sum}_{j=1}^2 { \cal k}_j\left[\hat v,{{\boldsymbol{\eta}}}^j\right ] \mu_2 + \left({\dfrac{\partial{}}{\partial{z}}}+{\displaystyle\sum}_{j=1}^2{\mathbf k}^j\cdot\nabla_{{\mathbf x}^j}\right ) ( \mu_1+\sqrt{\varepsilon}\mu_2)\right\rangle\nonumber . \end{aligned}\ ] ] the above expression is evaluated at @xmath301 . the term of order @xmath73 in @xmath302 vanishes since @xmath285 is independent of @xmath41 and the fast variable @xmath303 . we cancel the term of order @xmath304 in the same way as before by defining @xmath298 as the unique mean - zero ( in the variables @xmath283 and @xmath305 ) solution of @xmath306\mu=0.\ ] ] it is given explicitly by @xmath307\\ & & ~~~~~~~~~~~~~~~~~~+ \frac 1i\int_0^\infty dr e^{rq}\int_{\rm^d } d\hat v({\mathbf p } ) e^{ir({\mathbf k}_2\cdot { \mathbf p})+i({{\boldsymbol{\eta}}}_2\cdot { \mathbf p})}\left[\mu({\mathbf k}_1,{\mathbf k}_2-\frac { \mathbf p}2)- \mu({\mathbf k}_1,{\mathbf k}_2+\frac { \mathbf p}2)\right].\end{aligned}\ ] ] when @xmath285 has the form @xmath308 , then @xmath298 has the form @xmath309 with the corrector @xmath269 given by ( [ lambda1-eq ] ) . let us also define @xmath299 as the mean zero with respect to @xmath310 solution of @xmath311 \mu_1 = \overline{{\displaystyle\sum}_{j=1}^2 { \cal k}_j\big[\hat v,{{\boldsymbol{\eta}}}^j\big ] \mu_1},\ ] ] where @xmath312 . the function @xmath299 is given by @xmath313 \mu_1({\mathbf x},{{\boldsymbol{\eta}}}+r{\mathbf k},{\mathbf k})}\\ & & ~~~~~~~~~~~~~~~~~~~~-[{\cal k}_1[\hat v,{{\boldsymbol{\eta}}}_1+r{\mathbf k}_1 ] \mu_1]({\mathbf x},{{\boldsymbol{\eta}}}+r{\mathbf k},{\mathbf k},\hat v)\big]\nonumber\\ & & ~~~~~~~~~~~~~~~~~~~~- \int_0^\infty dr e^{rq } \big[\overline{{\cal k}_2[\hat v,{\mathbf k}_2+r{{\boldsymbol{\eta}}}_2 ] \mu_1({\mathbf x},{{\boldsymbol{\eta}}}+r{\mathbf k},{\mathbf k } ) } \nonumber \\ & & ~~~~~~~~~~~~~~~~~~~~ -[{\cal k}_2[\hat v,{{\boldsymbol{\eta}}}_2+r{\mathbf k}_2]\mu_1 ] ( { \mathbf x},{{\boldsymbol{\eta}}}+r{\mathbf k},{\mathbf k},\hat v ) \big].\nonumber\end{aligned}\ ] ] unlike the first corrector @xmath298 , the second corrector @xmath299 may not be written as an explicit sum of tensor products even if @xmath285 has the form @xmath308 because @xmath298 depends on @xmath283 . the @xmath291-martingale @xmath314 is given by @xmath315 where @xmath316 is given by @xmath317\mu_2^{\varepsilon}+ \left({\dfrac{\partial{}}{\partial{z}}}+{\displaystyle\sum}_{j=1}^2{\mathbf k}^j\cdot\nabla_{{\mathbf x}^j}\right ) ( \mu_1^{\varepsilon}+\sqrt{\varepsilon}\mu_2^{\varepsilon})\end{aligned}\ ] ] and the operator @xmath318 is defined by @xmath319 \\ + \big[e^{i{\mathbf p}\cdot\frac{{\mathbf x}_2-{\mathbf x}_1}{{\varepsilon}}+ir{\mathbf k}_2\cdot { \mathbf p } } ( \mu(z,{\mathbf x}_1,{\mathbf k}_1+\frac{{\mathbf p}}{2},{\mathbf x}_2,{\mathbf k}_2-\frac{{\mathbf p}}{2})- \mu(z,{\mathbf x}_1,{\mathbf k}_1+\frac{{\mathbf p}}{2},{\mathbf x}_2,{\mathbf k}_2+\frac{{\mathbf p}}{2}))\\ -e^{i{\mathbf p}\cdot \frac{{\mathbf x}_2-{\mathbf x}_1}{{\varepsilon}}+ir{\mathbf k}_2\cdot { \mathbf p } } ( \mu(z,{\mathbf x}_1,{\mathbf k}_1-\frac{{\mathbf p}}{2},{\mathbf x}_2,{\mathbf k}_2-\frac{{\mathbf p}}{2})- \mu(z,{\mathbf x}_1,{\mathbf k}_1-\frac{{\mathbf p}}{2},{\mathbf x}_2,{\mathbf k}_2+\frac{{\mathbf p}}{2}))\big]\\ + \big[e^{ir{\mathbf k}_1\cdot { \mathbf p}+ i\frac{{\mathbf x}_1-{\mathbf x}_2}{{\varepsilon}}\cdot { \mathbf p } } ( \mu(z,{\mathbf x}_1,{\mathbf k}_1-\frac{{\mathbf p}}{2},{\mathbf x}_2,{\mathbf k}_2+\frac{{\mathbf p}}{2})- \mu(z,{\mathbf x}_1,{\mathbf k}_1-\frac{{\mathbf p}}{2},{\mathbf x}_2,{\mathbf k}_2-\frac{{\mathbf p}}{2}))\\ -e^{ir{\mathbf k}_1\cdot { \mathbf p}+ i\frac{{\mathbf x}_1-{\mathbf x}_2}{{\varepsilon}}\cdot { \mathbf p } } ( \mu(z,{\mathbf x}_1,{\mathbf k}_1+\frac{{\mathbf p}}{2},{\mathbf x}_2,{\mathbf k}_2+\frac{{\mathbf p}}{2})- \mu(z,{\mathbf x}_1,{\mathbf k}_1+\frac{{\mathbf p}}{2},{\mathbf x}_2,{\mathbf k}_2-\frac{{\mathbf p}}{2}))\big ] \\ + \big[e^{ir({\mathbf k}_2+\frac{{\mathbf p}}{2})\cdot { \mathbf p } } ( \mu(z,{\mathbf x}_1,{\mathbf k}_1,{\mathbf x}_2,{\mathbf k}_2)-\mu(z,{\mathbf x}_1,{\mathbf k}_1,{\mathbf x}_2,{\mathbf k}_2+{\mathbf p}))\\ -e^{ir({\mathbf k}_2-\frac{{\mathbf p}}{2})\cdot { \mathbf p } } ( \mu(z,{\mathbf x}_1,{\mathbf k}_1,{\mathbf x}_2,{\mathbf k}_2-{\mathbf p})-\mu(z,{\mathbf x}_1,{\mathbf k}_1,{\mathbf x}_2,{\mathbf k}_2))\big ] . \end{array}\ ] ] we have used in the calculation of @xmath318 that for a sufficiently regular function @xmath35 , we have @xmath320 = { \displaystyle\int}_0^\infty dr { \displaystyle\int}_{\rm^d } \tilde r(r,{\mathbf p } ) f(r,{\mathbf p},-{\mathbf p } ) d{\mathbf p}.\ ] ] the bound on @xmath316 is similar to that on @xmath321 obtained previously as the correctors @xmath322 satisfy the same kind of estimates as the correctors @xmath323 : unlike the first moment case , the averaged operator @xmath318 still depends on @xmath200 . we therefore do not have strong convergence of the @xmath291-martingale @xmath314 to its limit yet . however , the a priori bound on @xmath87 in @xmath0 allows us to characterize the limit of @xmath314 and show strong convergence . this is shown as follows . the first and last terms in ( [ eq : l-2eps ] ) that are independent of @xmath200 give the contribution : @xmath329\\ & & = { \displaystyle\int}_{\rm^d } \dfrac{d{\mathbf p}}{(2\pi)^{d } } \hat r(\frac{p^2-k_1 ^ 2}{2},{\mathbf p}-{\mathbf k}_1 ) ( \mu(z,{\mathbf x}_1,{\mathbf p},{\mathbf x}_2,{\mathbf k}_2)-\mu(z,{\mathbf x}_1,{\mathbf k}_1,{\mathbf x}_2,{\mathbf k}_2))\\ & & + \hat r(\dfrac{p^2-k_2 ^ 2}{2},{\mathbf p}-{\mathbf k}_2 ) ( \mu(z,{\mathbf x}_1,{\mathbf k}_1,{\mathbf x}_2,{\mathbf p})-\mu(z,{\mathbf x}_1,{\mathbf k}_1,{\mathbf x}_2,{\mathbf k}_2 ) ) . \end{aligned}\ ] ] the two remaining terms give a contribution that tends to @xmath330 as @xmath91 for sufficiently smooth test functions . they are given by @xmath331 we have @xmath332 by bochner s theorem . since @xmath333 and @xmath127 are real quantities , we can take the real part of the above term and , after the change of variables @xmath334 and @xmath335 , obtain @xmath336 we have ( since @xmath285 is real - valued ) @xmath337 using density arguments we may assume that @xmath285 has the form @xmath338 then we have @xmath339 where @xmath340 . we introduce @xmath341 and use the cauchy - schwartz inequality in @xmath342 and @xmath343 : @xmath344 we use again the cauchy - schwartz inequality , now in @xmath345 , to get @xmath346 this proves that @xmath347 as @xmath102 . notice that oscillatory integrals of the form @xmath348 are not small in the bigger space @xmath161 , which is natural in the context of wigner transforms and was used in @xcite . in this bigger space , we can not control @xmath349 and actually suspect that the limit measure @xmath153 may no longer be deterministic . we therefore deduce that @xmath350 is an approximate @xmath216 martingale . the limit of the second moment @xmath351 thus satisfies ( weakly ) the transport equation @xmath352 with initial data @xmath353 . moreover , the operator @xmath354 acting on a tensor product @xmath355 has the form @xmath356={\cal l}\lambda\otimes\lambda+ \lambda\otimes{\cal l}\lambda.\ ] ] this implies that @xmath357 by uniqueness of the solution to the above transport equation with initial conditions given by @xmath358 . this proves that the limiting measure @xmath359 is deterministic and unique ( because characterized by the transport equation ) and that the sequence @xmath360 converges in probability to @xmath361 . w. kuperman , w. hodgkiss , h. song , t. akal , c. ferla and d. jackson , phase conjugation in the ocean : experimental demonstration of an acoustic time reversal mirror , j. acoust . * 103 * , 1998 , 25 - 40 .

diluted magnetic semiconductors ( dms ) based on iii - v semiconductors doped with mn have attracted a lot of interest recently , after critical temperatures for the onset of ferromagnetism of the order of 110 k have been found in ga@xmath0mn@xmath1as , for @xmath2.@xcite more recently , critical temperatures larger than room temperatures have been reported in mn - doped gan , enhancing the hope for extensive technological applications of these materials.@xcite while there is general agreement that the magnetism is due to charge - carrier mediated , effectively ferromagnetic interactions between the mn spins , there are various theoretical models attempting to understand their detailed behavior . the dms are alloy systems , with inherent positional disorder of the mn atoms ; further , spin - orbit coupling may play a significant role for hole doping . a theoretical treatment which takes into account all these factors , and their effects on the magnetic and transport properties of dms , is not yet available . instead , theoretical models have tended to concentrate on different aspects of the problem . one class of models , where much work has been done , neglects both the disorder and spin - orbit coupling effects . at large carrier densities , where coulomb potentials of the impurity ( mn ) atoms are effectively screened out , and other disorder effects ( such as those coming from as antisite defects which are believed to be the cause of the large compensation observed experimentally ) can be neglected , the former at least can be justified . in such a case , holes occupy a fermi sea at the top of the valence band.@xcite studies of the spin - wave spectra@xcite , monte - carlo studies@xcite as well as dynamical mean - field theory studies@xcite have been performed . overall , their results are rather similar to the physics one would expect to hold in a conventional ferromagnet . more recently , it has been proposed that spin - orbit coupling of the valence band , in the high carrier density ( strongly metallic ) regime leads to an rkky coupling between mn moments that is _ anisotropic _ in spin space , and where the positional disorder effectively leads to random anisotropy.@xcite this would lead to frustration effects on the magnetic ordering , and therefore unusual ferromagnetism . a number of very useful ab - initio studies have also been published.@xcite in parallel , we have studied effects of positional disorder on the magnetic ordering ( in the absence of spin - orbit coupling ) of an impurity band model,@xcite developed from previous work done in the context of insulating ii - vi dms compounds.@xcite such an approach should be of relevance at low doping concentrations x , at and below the metal - insulator transition , and possibly even above . evidence for the existence of impurity - like states is provided by ab - initio studies,@xcite which found that occupied hole states near the fermi level have wave - functions mostly concentrated on and near the mn impurity . more recently , angle resolved photoemission spectroscopy has revealed the existence of the impurity band in ga@xmath0mn@xmath1as with @xmath3 ( very close to the metal - insulator transition).@xcite a scanning tunneling microscope study @xcite demonstrates the existence of an impurity band in ( ga , mn)as samples with x=0.005 - 0.06 . optical spectroscopy@xcite also identifies the impurity band for x=0.0001 and x=0.05 samples . electron densities in localized states , as well as states close to the mobility edge , can be far from homogeneous , unlike bloch waves . at low mn concentrations and high degree of compensation ( which comes from charged centers ) seen in dms , short length - scale density fluctuations could be significant . this can lead to inhomogeneities in the local charge densities at different mn sites , which in turn leads to ( microscopically ) inhomogeneous magnetizations of the magnetic ions in the ordered phase . such inhomogeneity would be expected to alter the nature of the collective magnetic excitations ( spin waves ) of the magnetically ordered system , which in turn would affect charge transport through the magnetic excitation processes which give rise to spin - flip scattering . in this paper , we develop a numerical scheme to calculate the spin wave spectrum for a finite but large size model of coupled fermions and spins , in the presence of quenched disorder . we show that the method accurately reproduces the results for a lattice model of small size obtained from a standard treatment of spin waves via the random phase approximation.@xcite we then apply the method to the model of ref . , and study in detail its collective magnetic excitations . the plan of the paper is as follows . the model hamiltonian for dms is described in section ii . section iii is devoted to developing the numerical scheme . the accuracy of the scheme is tested by comparing it with the standard rpa method in section iv . in particular , we demonstrate that we can use our scheme to study large size systems , beyond the capability of the standard rpa approach . section v presents the results obtained by applying the scheme to the model of section ii , including the density of states and nature of the eigenvectors of the magnetic excitations of the coupled fermion - spin system . we summarize our conclusions in section vi . in the following we develop the formalism for an impurity band hamiltonian that we believe to be appropriate for describing ( at least qualitatively ) the diluted magnetic semiconductor ga@xmath0mn@xmath4as for low doping @xmath5 ( in the insulating phase , or not too far from the metal - insulator transition ) . however , generalizations of this method for other types of hamiltonians , crystal structures , parameters etc . can be done in a straightforward manner . the iii - v host semiconductor is assumed to have the zinc - blende structure appropriate for gaas . experimental measurements@xcite suggest that valence - ii mn substitutes for valence - iii ga . as a result , the @xmath6 mn dopants are randomly distributed at positions @xmath7 , @xmath8 on the @xmath9 fcc ga sublattice , corresponding to @xmath10 . all throughout this paper , we assume periodic boundary conditions . due to the valence mismatch , each mn introduces a charge carrier ( hole ) in the system . however , experimentally it is found that the hole concentration is considerably suppressed through compensation processes . as a result , the number of holes is only a fraction @xmath11 of the number of mn atoms @xmath12 . each mn atom also has a @xmath13spin @xmath14 from its half - filled @xmath15 level . the magnetic properties of the doped semiconductor are related to the exchange interaction between the mn spins and the charge carrier spins . this interaction is known to be antiferromagnetic ( afm ) , and proportional to the probability of finding the charge carrier at the corresponding mn site . recently , we proposed a simple impurity - band model to describe the behavior of the charge carriers for the low - doping regime.@xcite the main justification is that near and below the metal - insulator transition ( @xmath16 ) the density of charge carriers is not large enough to effectively screen the attractive coulomb potential of the mn dopants . as a result , bound , hydrogen - like impurity states are created about each mn site , at an energy @xmath17 ry above the top of the valence band . due to interactions , these impurity states lead to the appearance of an impurity band , and the holes first occupy states in this band . only if the concentration of holes ( or the temperature ) is large enough , are states in the valence band itself occupied by holes . thus , it is reasonable to attempt a description of the charge carriers behavior only in terms of this impurity band . as in previous work,@xcite in the following we will use the electron formalism to treat the problem . this is equivalent to performing a particle - hole transformation which leads to an inversion @xmath18 of the charge carrier spectrum . thus , instead of emptying the top of a valence - like impurity band ( i.e. , introducing holes in the system ) we instead occupy the bottom of a conduction - like impurity band . we also make the simplifying assumption that the isolated impurity wave - function for a charge carrier trapped near the mn at site @xmath7 is the @xmath19 hydrogen orbital @xmath20 , where @xmath21 is the appropriate bohr radius related to the effective mass of the heavy hole ( @xmath22 for gaas ) and the binding energy ( @xmath23 mev for mn in gaas).@xcite our approach neglects both the complicated orbital form of the acceptor wave - function ( ref . ) and spin - orbit coupling . the former is not expected to lead to any qualitative changes . it has recently been proposed that spin - orbit coupling leads to frustration in the magnetic ordering.@xcite these effects are left out in the present study , which concentrates on the effect of disorder . a study combining both effects is indicated for future work . we consider the hamiltonian @xmath24 here , @xmath25 creates a charge carrier with spin @xmath26 in the impurity state centered at site @xmath7 . the first term describes the hopping of charge carriers between impurity states . we use the parameterization @xmath27 ry , where @xmath28 , of magnitude and form appropriate for hopping between two isolated 1@xmath29 impurities which are not too close to one another . @xcite this hopping term has been shown to lead to the appearance of an impurity band which has a mobility edge , as well as a characteristic energy for the occupied states in agreement with physical expectation.@xcite the second term of the hamiltonian ( [ 1.1 ] ) describes the afm exchange between the mn spin @xmath30 and the charge carrier spin @xmath31 [ @xmath32 are the pauli spin matrices ] . this afm exchange is proportional to the probability of finding the charge carrier trapped at @xmath33 near the mn spin at @xmath34 , and therefore @xmath35 . based on calculations of the isolated mn impurity in gaas , we estimate the exchange coupling between a hole and the trapping mn ( @xmath36 ) to be @xmath37 mev.@xcite as already stated , the number of mn atoms is @xmath6 , and therefore there are a total of @xmath38 states in the impurity band . the number of charge carriers is fixed to @xmath39 , where we take @xmath40 . in ref . we studied the relevance of various other terms , such as an on - site potential ( associated with the coulomb potential of the compensation centers ) , hubbard - like on - site repulsion ( describing interactions between charge - carriers ) , external magnetic field etc . while they lead to various _ quantitative _ changes , we believe that the _ qualitative _ picture we present in the following sections is not changed by their absence . the random phase approximation ( rpa ) describes the collective excitations of a system about its self - consistent @xmath41 mean - field ground state . in order to clarify the notation used , we begin this section with a very short review of the derivation of the relevant equations for the mean - field ground state . we use the equation - of - motion approach of ref . . at the mean - field level , we are trying to find non - interacting fermionic quasiparticles through a unitary transformation of the on - site charge carrier operators : @xmath42 such that @xmath43 \approx e_{n\sigma } a^{\dagger}_{n\sigma}\ ] ] and @xmath44 \approx - \hbar h_i s_i^{+}\ ] ] for a non - interacting system , eqs . ( [ 22 ] ) and ( [ 23 ] ) would be exact . for an interacting system at the mean - field level , the exact hamiltonian is approximated by the diagonalized , non - interacting hartree - fock hamiltonian @xmath45 for which eqs . ( [ 22 ] ) and ( [ 23 ] ) become exact . it is straightforward to show that for hamiltonian ( [ 1.1 ] ) @xmath46 = \sum_{ij}^ { } t_{ij } \psi_{n\sigma}(j ) c^{\dagger}_{i\sigma } + \sum_{ij}^ { } j_{ij } \vec{s}_j { \vec{\sigma}_{\alpha\sigma } \over 2 } \psi_{n\sigma}(i ) c^{\dagger}_{i\alpha}\ ] ] the right - hand - side of eq . ( [ 25 ] ) contains spin operators , so it can not be put in the form of eq . ( [ 22 ] ) unless we linearize it , by replacing the spin operators with their mean - field ground state expectation value @xmath47 . while the choice of a collinear ground - state , with all mn spins aligned along the @xmath48-axis , is not the most general possibility , we have actually shown in ref . that the ground - state of this model , for the range of parameters we are interested in , is indeed collinear . this is why we choose to apriori make this assumption in this case . after the linearization , we can now equate the right - hand - side of eq . ( [ 25 ] ) with @xmath49 [ see eq . ( [ 22 ] ) ] to obtain the hartree - fock equation for the one - particle orbitals : @xmath50 the mean - field ground - state of the charge carriers is obtained by occupying the @xmath51 lowest - energy orbitals , @xmath52 the effective magnetic fields @xmath53 [ see eq . [ 24 ] ) ] are obtained in a similar way . we evaluate the commutator of the exact hamiltonian with the spin raising operator @xmath54 @xmath55 = -\hbar \sum_{j}^{}j_{ij } s_i^z s_i^{+ } + \hbar \sum_{j}^ { } j_{ij } s_i^{+ } s_i^{z}\ ] ] the right - hand - side now contains charge carrier operators ( in the charge carrier spin operator ) , beside the spin operators . we again use linearization @xmath56 [ see eq . ( [ 27 ] ) ] , @xmath57 , and by comparison with eq . ( [ 23 ] ) we find @xmath58 from eq . ( [ 24 ] ) we see that if @xmath59 , then in the mean - field ground state the spin @xmath14 is in the `` up '' state @xmath60 and therefore @xmath61 , while if @xmath62 the spin @xmath14 is in the `` down '' state @xmath63 and @xmath64 ( @xmath65 ) . as a result , the spin - part of the mean - field ground - state can also be easily found if the expectation value of the charge carrier spins @xmath66 are known from eqs . ( [ 26 ] ) , ( [ 27 ] ) . thus , we obtain the usual self - consistent hartree - fock equations . diluted magnetic semiconductors exhibit ferromagnetism at low temperatures . as a result , we assume that in the mean - field ground state , all mn spins are fully polarized @xmath61 . then , from eq . ( [ 26 ] ) we find the lowest - lying @xmath51 states . if the charge carriers are also fully polarized at @xmath41 , all the occupied @xmath51 state have @xmath67 , and therefore the mean - field ground - state is of the form : @xmath68 of course , one must check that self - consistency is obeyed by verifying that indeed the first @xmath51 lowest energy one - electron states are all spin - down . we find that this always holds true for the parameters we study in this paper . as discussed later , the spin - wave spectrum of the collective excitations also confirms that this ground - state is indeed stable . however , for higher charge carrier concentrations ( or hamiltonians with other types of interactions ) it is likely that the hf ground - state is only partially spin polarized . in that case , one must do the full iterational search for the self - consistent hf ground state . the computation for the spin - wave spectrum in the partially - polarized case ( for instance due to spin - orbit coupling ) , can be derived in a similar way to the one we present in the following for the fully polarized case . we will use the same equation - of - motion approach to derive the rpa equations . we are interested in spin - waves , which are related to spin - flip ( spin - lowering ) processes . therefore , the rpa operators for spin - wave collective modes must be constructed in terms of spin - flip operators @xmath69{c } b^{\dagger}_{hp } = a^{\dagger}_{h\uparrow } a_{p\downarrow } \\ b^{\dagger}_i = { 1 \over \sqrt{2s } } s^-_i \\ \end{array } \right.\ ] ] the index @xmath70 runs over the empty ( `` hole '' ) states of the hf ground state , while the index @xmath71 runs over the occupied ( `` particle '' ) states of the hf ground state . the index @xmath72 runs over all the mn spin positions . the hf ground - state @xmath73 is fully polarized , and given by eq . ( [ 30 ] ) . the spin - flip operators have `` bosonic '' nature , as expected for collective modes operators . indeed , it is straightforward to verify that @xmath74\rangle = \delta_{pp ' } \delta_{hh'}$ ] , @xmath75\rangle = \delta_{ij}$ ] , with all the other commutators identically zero . here , the notation @xmath76 signifies an average over the hf ground state @xmath77 . the hamiltonian can be rewritten in terms of the @xmath78 operators as [ see eqs . ( [ 1.1 ] ) , ( [ 21 ] ) ] @xmath79 where @xmath80 and @xmath81 one can easily derive the equations of motion for the spin - flip operators . for instance @xmath82= { 2 \over \sqrt{2s } } \sum_{nm}^ { } j_{n\downarrow , m\uparrow}(i ) a^{\dagger}_{n\downarrow } a_{m\uparrow } s^z_i\ ] ] @xmath83 since rpa is an approximation , not an exact solution , we again must linearize this commutator about the hf ground - state , and keep only the meaningful terms . in other words , we replace @xmath84 here , the first transformation is the linearization about the hf ground state @xmath73 . the second transformation results if one uses the hf expectation values @xmath85 , @xmath86 . the restriction to relevant terms means that in any sum over @xmath87 , we restrict @xmath88 to being an occupied orbital in the hf ground - state @xmath89 , because otherwise @xmath90 , i.e. a spin - flip process from the hf ground - state is impossible [ see eq . ( [ a.1 ] ) ] . performing a similar linearization for @xmath91 and collecting the various terms , we find the linearized equation of motion @xmath92\approx h_i b_i^{\dagger } + \sqrt{2s } \sum_{p=1}^{n_h } \sum_{h=1}^{n_d } j_{p\downarrow , h\uparrow}(i ) b_{hp}\ ] ] after similar steps , the linearized equation of motion for the other spin - flip operator @xmath93 is found to be @xmath94 \approx -\left ( e_{h\uparrow } - e_{p\downarrow}\right ) b_{hp } - \sqrt{2s } \sum_{i}^ { } j^*_{p\downarrow , h\uparrow}(i ) b_i^{\dagger}\ ] ] eqs . ( [ a4 ] ) and ( [ a5 ] ) allow us to `` guess '' the rpa part of the hamiltonian , for which eqs . ( [ a4 ] ) and ( [ a5 ] ) become exact , to be given by @xmath95 @xmath96 thus , the rpa hamiltonian is quartic in electron operators and quadratic in spin operators , showing that it is the next order term after the hartree - fock hamiltonian ( which is quadratic in electron operators and linear in spin operators ) @xmath97 higher order approximation terms describe interactions between the collective modes , which are neglected at the rpa level . we want to diagonalize the rpa hamiltonian to the canonical form @xmath98 where @xmath99 is the creation operator for a spin - wave with energy @xmath100 , and @xmath101 is an index ( in homogeneous systems , it would be a wave - vector ) . since they create collective mode excitations , these operators must have bosonic nature @xmath102=0 $ ] . however , since they do not describe exact , but only approximative solutions of the exact hamiltonian , in fact only the weaker condition @xmath103\rangle=\delta_{\alpha\beta}$ ] is obeyed . the most general form for these operators [ see eq . ( [ a.6 ] ) ] is @xmath104 using the equation of motion @xmath105= \hbar \omega_{\alpha } q^{\dagger}_{\alpha } $ ] and computing the commutator using eqs . ( [ a.6 ] ) , ( [ a.8 ] ) , we find the diagonalization condition to be @xmath106 x^{(\alpha)}_{hp } = \sqrt{2s } \sum_{i}^{}j_{p\downarrow , h\uparrow}(i ) y^{(\alpha)}_i \\ \label{a.9b } & & \left [ \hbar \omega_{\alpha } - h_i\right ] y^{(\alpha)}_i = - \sqrt{2s } \sum_{p=1}^{n_h}\sum_{h=1}^{n_d } j_{p\downarrow , h\uparrow}^*(i ) x^{(\alpha)}_{hp}\end{aligned}\ ] ] these are the rpa equations and they can be recast in the customary standard eigenvalue rpa equation form@xcite @xmath107{c } { \bf x } \\ { - \bf y } \\ \end{array } \right ) = \left ( \begin{array}[c]{cc } { \bf e } & { \bf j}\\ { \bf j}^ * & { \bf h } \\ \end{array } \right ) \cdot \left ( \begin{array}[c]{c } { \bf x } \\ { \bf y } \\ \end{array } \right)\ ] ] where the vectors @xmath108 and @xmath109 contain all the unknowns @xmath110 and @xmath111 , and the matrices @xmath112 and @xmath113 have the elements @xmath114 , @xmath115 and @xmath116 . the dimension of the rpa matrix ( [ 5.2 ] ) , and therefore the number of normal modes , is @xmath117 . of these solutions , @xmath6 are proper spin - wave collective modes , while the rest of @xmath118 are spin - flip processes associated with particle - hole excitations . if there were no interactions ( j=0 ) , the eigenenergies of these spin - flip processes would be @xmath53 for the lowering of the mn spin @xmath119 , respectively @xmath120 for a hole spin - flip [ see the left - hand side of eqs . ( [ a.9]),([a.9b ] ) ] . interactions renormalize these values [ as described by the right - hand - side of eqs . ( [ a.9]),([a.9b ] ) ] , but one still expects @xmath6 spin - wave collective solutions at low energies , coming from the mn spin lowering , and @xmath121 spin - flip particle - hole excitations at energies comparable or larger than the spin - flip gap @xmath122 . we now comment on the sign of the rpa spectrum frequencies @xmath100 . since @xmath59 ( [ 29 ] ) ] and @xmath123 ( by definition of the hf ground - state ) , it is apparent from eqs . ( [ a.9]),([a.9b ] ) that @xmath6 spin - wave solutions will have positive energies @xmath100 , while the @xmath121 spin - flip particle - hole excitations will have negative energies . this _ does not _ imply that the system is unstable , but simply that we have not chosen the proper spin - flip creation operators . instead of the choice of eq . ( [ a.8 ] ) , we can also denote @xmath124 in terms of these new operators , we have now @xmath125= \hbar \omega_{\alpha } p_{\alpha}$ ] , i.e. @xmath126 in other words , a negative solution for @xmath127 simply means that we chose the corresponding annihilation operator instead of the creation operator when we wrote eq . ( [ a.8 ] ) . for spin - flip processes , it is obvious that this is the case from the dependence of @xmath99 on @xmath93 , instead of @xmath128 . an unstable mean - field ground - state is signaled by _ complex _ values of the spectrum frequencies @xmath129.@xcite excited states about the mean - field ground - state @xmath73 are of the general form @xmath130 @xmath131 , leading to a time dependence : @xmath132 |\psi\rangle,\ ] ] where @xmath133 is the hartree - fock energy of the system . if any of the frequencies @xmath134 has a non - trivial imaginary part , it follows that the expectation value of any operator @xmath135 will move in time exponentially away from its mean - field ground - state expectation value @xmath136 , i.e. the mean - field is unstable to small perturbations . the advantage of the standard rpa approach is that by solving the eigenvalue problem ( [ 5.2 ] ) , one finds the normal mode spectrum @xmath127 and the spatial distribution of the spin - waves ( related to the @xmath111 coefficients ) at once . the obvious disadvantage is of a numerical nature : for a disordered system the rpa matrix must be diagonalized numerically . the size of the matrix is @xmath137 . the typical sizes we consider are systems with around @xmath138 mn spins , and @xmath139 holes ( although concentrations up to @xmath140 might have to be considered , depending on the doping @xmath5 ) , leading to rpa matrices of typical size @xmath141 . as a result , one has to either consider much smaller systems ( in which case , finite size effects may be overwhelmingly important ) , or to try sparse matrix techniques to obtain the first few low - energy modes of the rpa matrix ( although the off - diagonal @xmath113 matrix is not necessarily sparse ) . there is , however , an alternative way of finding the collective mode spectrum and spatial distributions . we can directly solve for the @xmath110 coefficients from eq . ( [ a.9 ] ) and rewrite eq . ( [ a.9b ] ) as @xmath142 where @xmath143 the advantage of this formulation is that one has to deal with much smaller matrices ( the dimension of the @xmath144 matrix is @xmath145 ) . ( [ 5.3 ] ) tells us that for frequencies @xmath146 belonging to the spin - wave spectrum , the matrix @xmath147 has at least one eigenvector corresponding to an eigenvalue @xmath148 ( if the mode is degenerate , there are several such eigenvectors ) . the corresponding eigenvectors give us the desired spatial mode distribution @xmath149 . thus , the problem is reduced to sweeping the range of frequencies of interest ( for low - energy collective excitations , this is usually a fairly small range of frequencies near @xmath150 ) , and monitoring the eigenvalues of the @xmath151 matrix . the dependence on @xmath152 of the matrix elements @xmath153 , and therefore of the eigenvalues @xmath154 is monotonic for small @xmath155 [ see eq . ( [ 5.4 ] ) ] . as a result , each equation @xmath156 has a single solution , and we expect to find @xmath6 collective mode eigenenergies , one for each eigenvalue . the monotonic dependence of @xmath154 on @xmath152 also simplifies the search for the collective spectrum , since it is enough to evaluate the eigenvalues @xmath154 in the range of interest on a grid of step @xmath157 , and use linear interpolation to find the solutions of the equations @xmath156 . as we show in the following by comparison with standard rpa ( for small system sizes ) we can very easily achieve relative errors less than @xmath158 . the rpa spectrum has a second type of solutions , the spin - flip particle - hole continuum . these solutions are at frequencies of the order of the gap between the last occupied and first empty hf states @xmath159 , and one expects @xmath121 such solutions . in principle , one can use the same method to find these energies as well , although the fact that they extend over a larger range of frequencies and that the @xmath152 dependence is very strong makes the search more difficult . typically , one is only interested in the lower and upper limits of the spin - flip continuum , which can be found easily with the method just described . finally , we would like to mention that it is always possible to reformulate the standard rpa equation ( [ 5.2 ] ) in a `` self - consistent '' form of the type shown in eq . ( [ 5.3 ] ) , with a matrix @xmath151 of much smaller dimension than the rpa matrix . for instance , for hamiltonians describing interacting electron systems , the rpa matrix has a dimension @xmath160 , where @xmath161 is the number of occupied , and @xmath162 is the number of empty orbitals in the mean - field ground - state.@xcite for a system with a finite concentration of electrons @xmath5 on a lattice of linear dimension @xmath163 in a @xmath164-dimensional space , we have @xmath165 and @xmath166 , and the rpa matrix scales as @xmath167 . for such systems , one can always find an equivalent reformulation of the rpa equation in terms of a @xmath151 matrix of size @xmath168.@xcite this allows numerical handling of considerably larger systems , without having to resort to sparse matrix techniques , which often have issues related to instability . in this section we show , by direct comparison against known calculations , as well as by direct comparison against solving the rpa matrix equation , that the formulation of eqs . ( [ 5.3 ] ) , ( [ 5.4 ] ) gives correct and numerically very accurate results . in the second part of the section we describe in more detail the numerical implementation as well as efficiency of our method . we first apply our approach by solving eqs . ( [ 5.3 ] ) , ( [ 5.4 ] ) analytically for a simplified case . we assume that the mn spins are arranged on an ordered superlattice , instead of having random positions . then , the charge - carrier part of the mean - field hamiltonian is easily diagonalized in the @xmath169-space : @xmath170 where @xmath171 here , @xmath172 is the kinetic energy of the non - interacting electrons , where @xmath173 indexes all the neighboring sites and @xmath174 for which @xmath175 . also , @xmath176 , where again @xmath177 for which @xmath175 . in the hf ground state , the states @xmath178 given by the condition @xmath179 are occupied . since the @xmath51 electrons are equally distributed among the @xmath6 mn sites , and are fully polarized , the z - component of the total spin created by electrons at each site , within hfa , is @xmath180 , and therefore we find @xmath181 [ see eq . ( [ 29 ] ) ] . using the fact that the solutions of eq . ( [ 5.3 ] ) must also have plane - wave structure @xmath182 , it follows that the index @xmath183 becomes a wave - vector , as expected , and we can reduce eqs . ( [ 5.3 ] ) and ( [ 5.4 ] ) to a self - consistent equation for each collective mode @xmath184 : @xmath185 here , @xmath186 . the occupation number @xmath187 obviously comes from the sum over occupied states we had in eq . ( [ 5.4 ] ) . for a finite - size ordered lattice , one expects considerable degeneracy for each mode @xmath188 , due to invariance to various symmetry transformations of the @xmath169-vector . in most cases , the number of charge carriers is such that the last orbital ( of degeneracy @xmath189 ) is only partially occupied by @xmath190 charge carriers . in this case , one must obviously choose @xmath191 for all these orbitals , and @xmath192 for lower , fully occupied orbitals , and @xmath193 for higher , empty orbitals . otherwise , the translational invariance is broken and the collective spectrum is not indexed by a wave - vector . the sum in eq . ( [ 55 ] ) can be performed if we assume that the dispersion at the bottom of the band is of quadratic form @xmath194 ( this is a reasonable approximation since we are interested in low filling fraction @xmath195 ) . then , the sum over occupied states can be transformed to an integral which is straightforward to evaluate , leading to the solution : @xmath196 here , @xmath197 is the fermi velocity , where the fermi vector is given by @xmath198 the function @xmath199 is given by @xmath200 @xmath201 \ln { { \hbar\omega + \epsilon_{q } + j_0s- v_f q } \over { \hbar\omega + \epsilon_{q } + j_0s+ v_f q}}\ ] ] where @xmath202 . for comparison purposes , we will assume a local afm interaction @xmath203 , leading to @xmath204 . in this case , the hamiltonian becomes identical to the hamiltonian used by konig . al _ in ref . , provided that the effective mass @xmath205 in the dispersion relation is assumed to be the band effective mass . we can easily solve eq . ( [ 56 ] ) in the asymptotic limit @xmath206 , to find @xmath207\ ] ] this is indeed identical with the asymptotic long - wavelength spin - wave spectrum obtained in ref . , and has the typical quadratic dependence of spin - waves in conventional ferromagnetic systems . in this section we briefly illustrate the accuracy and speed of our formulation of the rpa problem [ eqs . ( [ 5.3 ] ) and ( [ 5.4 ] ) ] , in comparison with the standard rpa approach [ eq . ( [ 5.2 ] ) ] . we use a rather small system , with @xmath208 mn spins randomly distributed on a fcc lattice with linear size @xmath209 , corresponding to a mn concentration @xmath210 . the number of charge carriers is @xmath211 , and the other parameters are as defined in section [ sec2 ] . thus , the standard rpa involves the diagonalization of a non - hermitian matrix of dimension @xmath212 . the largest eight eigenvalues @xmath154 of the @xmath151 matrix ( circles ) , evaluated on a grid with a step @xmath213 mev . the full line is @xmath214 . the spin - wave spectrum is given by the condition @xmath215 . ] accuracy @xmath216 for all @xmath217 spin - wave frequencies @xmath218 , between the exact rpa ( standard ) values and values obtained in our formulation using a grid with @xmath213 [ see fig . [ fig1 ] ] . ] in fig . [ fig1 ] we show the @xmath152-dependence of the largest eight eigenvalues @xmath154 of the @xmath144 matrix ( circles ) , evaluated on a grid with a step @xmath213 mev . [ for real @xmath152 and @xmath219 , the matrix @xmath151 is hermitian and all eigenvalues @xmath154 are real , see eq . ( [ 5.4 ] ) ] . the full line is @xmath214 , and the spin - wave spectrum is given by the condition @xmath156 . one can see that the eigenvalues @xmath154 have monotonically increasing dependence on @xmath152 for small @xmath220 mev , and therefore the equation @xmath221 can have at most one solution for each eigenvalue , for small @xmath152 . if each eigenvalue yields a solution , we have found all @xmath6 spin - wave frequencies . if one or more eigenvalues do not intersect the @xmath222 line , this means that the mean - field ground - state is unstable ( there are _ complex _ spin - wave frequencies ) . for all the cases and parameters we investigate , we found that the fully - polarized ground - state is stable for our model . largest relative error @xmath216 among @xmath223 spin - wave frequencies @xmath218 , between the exact rpa ( standard ) values and values obtained in our formulation using a grid with @xmath224 and 1 mev . the line is drawn to guide the eye . even for a very large grid , the relative error is still reasonably small . ] to find the spectrum frequencies @xmath100 , we use linear interpolation over the interval @xmath157 where each @xmath154 curve intersects the @xmath214 curve . this method avoids the need of evaluating the eigenvalues at too many @xmath152 points ( the most time - consuming part of the computation is the evaluation of the matrix elements of @xmath144 ) . the high degree of accuracy obtained with this method is demonstrated in fig . [ fig2 ] , where we compare the exact rpa spectrum @xmath218 obtained through direct diagonalization of the rpa matrix [ see eq . ( [ 5.2 ] ) ] with the the values @xmath225 obtained through our formulation . in fact , we plot the relative error @xmath226 for each spectrum frequency @xmath218 . as one can see from fig . [ fig2 ] , the largest error is of the order @xmath227 , for the grid @xmath213 . this suggests that one could use a much larger grid @xmath157 and still have a good accuracy . indeed , we have computed the spectrum through both methods for five different realization of disorder , using various grid values @xmath157 in our formulation , and selected the largest relative error for each case ( from a total of @xmath223 values ) . the results are shown in fig . the error is found to scale as the square of the grid size . even with a very large grid @xmath228 mev , ( which implies the evaluation of the eigenvalues @xmath154 in only very few points ) , we still get a relative error smaller than @xmath229 . thus , one can use the grid step @xmath157 to optimize and considerably speed up the computation . however , as the number of spins ( and eigenvalues ) @xmath6 increases , one must take into consideration other complications , as discussed below . cpu time for the two rpa methods , for systems with @xmath230 and 216 spins . the standard rpa ( full line ) is significantly more time consuming than our formulation ( dashed line ) . ] in fig . [ fig4 ] we compare cpu times for finding the rpa spin - wave spectrum using the standard vs. our rpa formulations . all simulations were done on the same processor . we used a small grid @xmath213 mev for our approach , and verified that all relative errors were less than @xmath158 . we use systems with @xmath231 and 216 spins , randomly distributed on fcc lattices of linear size @xmath232 and 18 ( @xmath233 ) . as expected , the computational time depends in a power - law manner on the size @xmath6 , with exponents 6.4 and 3.2 respectively for the standard and current approach . clearly , our formulation can be successfully used for much larger sizes than those afforded by the standard approach . eigenvalues @xmath154 for a system with @xmath234 random spins , evaluated for a grid @xmath157=0.05mev . crossing and anti - crossings of neighboring eigenvalues is apparent . the dotted line is @xmath214 . since we are interested in the solutions of the equations @xmath235 , it is clear that too large a grid @xmath157 may lead to considerable errors . ] one important aspect to keep in mind is that as the system size @xmath6 increases , neighboring eigenvalues become more closely spaced , and accidental crossings and anti - crossings appear . some typical examples are shown in fig . [ fig5 ] , for a disordered system with @xmath234 spins . tracking the crossing of the eigenvalues is essential , since such events lead to a change in the indexing of the eigenvalues from one grid point to the next . we found that requiring continuity in first and second derivatives allows a unique identification of each continuous eigenvalue from one grid point to the next ( 3 steps are shown for each eigenvalue in fig . [ fig4 ] ) . in fact , the only relevant crossings are the ones that take place within the step @xmath157 where the eigenvalues intersects the @xmath214 line ( shown as a dotted line in fig . [ fig4 ] ) . such an example is shown in the left panel of fig . [ fig4 ] . for the largest system size investigated , of @xmath234 randomly distributed spins , we find that only 4.5% of the eigenvalues have such relevant crossings within @xmath157 of @xmath156 . the percentage decreases with decreasing @xmath6 , to 1.1% and 0.3% for @xmath236 216 and 125 , respectively . this percentage also depends on the grid step @xmath157 ; for increasing @xmath157 the identification of crossings and anti - crossings becomes more difficult , leading to possibly large errors . one can optimize the choice of the grid step @xmath157 by starting with a larger value . the well separated eigenvalues ( such as the ones depicted in fig . [ fig1 ] ) will provide unique identification and very accurate values for their corresponding spin - wave spectra values . where considerable mixing and therefore non - linear variation of the eigenvalues @xmath154 is apparent , a finer mesh is necessary in order to correctly characterize their variation near @xmath214 . in all our simulations , we use the grid @xmath213 mev , which allows for comfortable tracking of each eigenvalue and also is sufficiently small to allow us to approximate the variation of @xmath154 as being linear within each @xmath157 step . these comparisons clearly demonstrate the accuracy and speed of our formulation of rpa , as compared to the standard rpa . the biggest advantage , though , is that it can easily be applied to systems with large sizes , for which standard rpa is numerically cumbersome . in this section , we use the rpa method described above to study the density of states ( dos ) , as well as nature ( extended or localized ) of the spin - wave spectrum of ( iii , mn)v dms described by the model introduced in section [ sec2 ] . spin - wave dispersion @xmath237 for an ordered , simple cubic superlattice arrangement of mn spins inside the host semiconductor . lattices of linear size @xmath238 and 150 , with a total of respectively @xmath239 and @xmath240 mn spins are considered . this corresponds to @xmath241 and @xmath40 . dispersion is plotted along three linear cuts in the brillouin zone : @xmath242 , @xmath243 and @xmath244 . ] we study the spin - wave density of states for three types of arrangements of the mn spins inside the host semiconductor . all samples studied correspond to @xmath241 and @xmath40 . other parameters are as specified in section [ sec2 ] . first , we consider fully ordered systems , in which the mn spins are arranged on a simple cubic superlattice . for @xmath241 , the superlattice constant is @xmath245 . in this case , we can study the spin - wave dispersion and density of states for very large sizes , since we can use directly eq . ( [ 56 ] ) to compute the spin - wave frequency @xmath246 for each wave - vector @xmath247 inside the brillouin zone . spin - wave dispersion obtained along three cuts in the brillouin zone is shown in fig . the dispersion has quadratic behavior near the center of the brillouin zone , as expected from the discussion for an ordered case provided above . the finite - size effects are reasonably small . for the small sizes we used both eq . ( [ 56 ] ) and our method to compute the dispersion . the results of the two agree with a relative error of less than @xmath248 . one important aspect to notice is the small range of the spin - wave spectrum , as compared to the afm exchange @xmath37 mev . this is a consequence of the fact that the mn spins do not interact directly with one another . instead , their interaction is mediated by the rather small concentration of charge carriers present . we compute the density of states ( dos ) @xmath249 associated with the spin - wave dispersion for the superlattice case employing the standard method of dividing the brillouin zone into tetrahedra and linearizing the dispersion ( ref . ) . the dos obtained for the lattice with @xmath250 , @xmath251 is shown as a dotted line in fig . we use a logarithmic scale for the energy , and normalize the density of states such that @xmath252 where @xmath253 . this convention will be maintained throughout the rest of the paper . upper panel : average density of states @xmath254 _ on a logarithmic scale _ for systems with @xmath255 and 512 mn spins in moderately disordered configurations ( full lines ) . lower panel : average density of states for systems with @xmath255 and 512 mn spins in strongly disordered configurations ( full lines ) . the dotted line is the spin - wave density of states of a dms with fully ordered ( superlattice ) configuration of mn spins . all samples correspond to @xmath256 , @xmath40 . ] in the upper panel of fig . [ fig7 ] we also plot the densities of states obtained for moderately disordered configurations with @xmath257 and @xmath258 mn spins ( full lines ) . these are configurations in which we place the mn spins randomly on the fcc ga sublattice of the host semiconductor , subject to the constraint that the distance between any two spins is _ larger than @xmath259_. [ in the ordered cubic superlattice , nearest neighbor spin separation is @xmath245 ] . this moderate amount of disorder breaks translational invariance , and the wave - vectors are no longer good quantum numbers . also , the large degeneracies of the superlattice spectrum are lifted . we computed the spin - wave spectrum for @xmath261 realizations of the disorder with @xmath262 spins , 100 realizations with @xmath263 spins and 50 realizations with @xmath234 spins . thus , we have a total of over 21,000 spin - wave energies for each size , and statistics can be comfortably carried out . in particular , from fig . [ fig7 ] we see that the dos histograms corresponding to the three sizes are smooth functions , i.e. the average over disorder is well accounted for . also , the curves are practically indistinguishable from one another , implying that finite size effects are negligible . in the lower panel of fig . [ fig7 ] we plot the dos corresponding to fully disordered configurations with @xmath257 and @xmath258 mn spins ( full lines ) . these are configurations in which we place the mn spins randomly on the fcc ga sublattice of the host semiconductor , with no restrictions ( except that each mn occupies a different site ) . the number of samples analyzed is the same as in the previous case . again , the curves show smooth behavior and no finite - size effects . the origin of the peak appearing near @xmath264mev will be discussed later . before continuing the analysis , we must also point out that the goldstone modes have been left out in the dos shown in fig . the rpa system _ always _ has one solution of energy @xmath150 ( numerically , we find its magnitude to be less that @xmath265 ) , corresponding to @xmath266 . this is the goldstone mode , describing the same overall rotation of all the spins . these goldstone modes would appear in the dos as a @xmath267 function at @xmath150 ( @xmath268 ) . the effect of disorder in the positions of the mn is reflected in the considerable widening of the dos ( on a logarithmic scale ) , and rounding of the van hove singularities of the superlattice dos , as the amount of disorder increases . more significantly , however , is the _ substantial _ enhancement of the dos at _ low energies_. this behavior is in agreement with the general expectation of the effects of disorder on an energy - band dispersion . a question of considerable interest concerns the _ nature _ of the spin - wave excitations . we know that the ordered superlattice has extended , plane - wave type spin - waves . the general expectation is that disorder will lead to localization , and thus one expects the appearance of localized spin - waves as the disorder increases . one way to characterize the nature of the spin - waves is to compute their inverse participation ratio ( ipr ) , defined as : @xmath269 for an extended mode @xmath101 ( plane - wave , for instance ) , we expect that all @xmath270 coefficients are roughly of equal magnitude , since all spins are expected to participate equally in the spin - wave . then , it follows that for an _ extended mode @xmath101 _ , ipr@xmath271 , i.e. it is inversely proportional to the size of the system . on the other hand , in a localized mode @xmath101 , only the spins within the localization volume have non - vanishing values for @xmath270 . thus , it follows that for a _ localized mode @xmath101 _ , ipr@xmath272 is independent of @xmath6 . instead , its value is given by the inverse number of mn spins participating in the localized mode . upper panel : average ipr(e ) for systems with @xmath255 and 512 mn spins in moderately disordered configurations ( circles , square and triangles , respectively ) . lower panel : same for strongly disordered configurations . all samples correspond to @xmath256 , @xmath40 . ] we compute the ipr of the spin - waves in the following manner . as we sweep the frequencies @xmath152 of interest and compute the eigenvalues @xmath154 of the @xmath151 matrix on the grid @xmath157 , we also compute the corresponding ipr of the eigenvectors @xmath273 ( which are normalized to unity ) . the ipr is a continuous function of @xmath152 , and we use linear interpolation to find its value at the spin - wave frequencies @xmath134 of interest . whenever eigenvalues @xmath154 cross , the iprs are no longer well defined : one can choose any linear combination of the eigenvectors of the degenerate modes , which would lead to different values for the corresponding iprs . we have checked that these accidental crossings are rather rare ( below a few percent ) and equally distributed over the entire energy scale . if we exclude all these degenerate cases , we get a dos which is virtually indistinguishable from the one obtained when these modes are kept . more importantly , we find that modes which cross predominantly have the same nature ( either localized or extended ) . as a result , the ensemble averaged ipr is not sensitive to the precise treatment of these cases . in fig . [ fig8 ] we plot the ( geometric ) average ipr(e ) for the moderate ( upper panel ) and strong ( lower panel ) disorder configurations . again , the goldstone modes are not shown . we use the geometric mean ( i.e. , arithmetic mean of the @xmath274 values ) in order to insure proper weight for the extended modes , with low ipr . for moderate disorder configurations , we see that the spin - wave modes at high energies @xmath275 mev are localized : the values corresponding to the three different sizes collapse on top of one another . this is expected , given the fact that the upper band - edge of the superlattice spectra is just below @xmath276 mev ( see fig . [ fig7 ] ) . thus , states of higher energy have been split off the band by disorder , and are expected to be localized . the low - energy part of the band also appears to be localized : while the three curves do not quite meet , the value of the ipr is just below unity , showing that these spin - waves have considerable weight on only one spin ( i.e. , they correspond to a individual spin - flip ) . however , for the central part of the spectrum , the spin - waves are extended , with the iprs for the three sizes clearly distinct and decreasing with increasing @xmath6 . histogram of the ipr values for all spin - waves ( spin flips ) of energy @xmath277 , for @xmath278 and 512 . results are shown only for strong disorder configurations . except for the low - lying goldstone modes , the histograms are identical , suggesting only localized modes at these energies . ] for the strong disorder configurations , this tendency is even more apparent . the high- and low - energy regions , @xmath279 1 mev , respectively @xmath280 mev , contain localized spin - waves : in both limits , the ipr curves collapse on top of one another . the low - energy localized modes are individual spin - flips . the associated @xmath111 are vanishingly small at all sites except one , leading to ipr @xmath281 . these sites are always situated far apart from all other mn spins , and the probability to be visited by charge carriers is exponentially small ( tailing from far occupied regions ) . as a result , the mn spins at these sites are virtually isolated , and their spin excitations are individual spin - flips . the energy for such a spin - flip is equal to @xmath53 , if one neglects small corrections due to the extremely weak interactions [ see eq . ( [ a.9b ] ) and following discussion ] . histograms of the ipr of all the modes with energies below @xmath282 are shown in fig . [ fig9 ] . here , we also show the goldstone modes , which have zero energy , and @xmath283 . the histograms have been scaled by the total number of modes for all disorder realizations considered for each particular size . since the number of goldstone modes exactly equals the number of different realizations of disorder considered , their peaks are in a ratio of roughly 200/100/50 with respect to one another . the main observation is that the histograms for the three sizes are very similar , with a huge peak just below ipr=1 . the absence of a dependence on @xmath6 confirms that all these modes are localized ( except the goldstone modes , of course ) . histogram of the ipr values for all spin - waves ( spin flips ) of energy @xmath284 , for @xmath278 and 512 . results are shown only for strong disorder configurations . the histograms are almost identical , suggesting only localized modes at these energies . ] the high - energy localized states are of very different nature : their ipr is close to @xmath285 , suggesting that it has large contributions from only two sites . analysis of the @xmath111 values for such modes shows that most of them are associated with nearest - neighbor ( n.n . ) mn spins on the fcc ga sublattice . the characteristic energy for the spin - waves centered on such n.n . sites is just below 5 mev , and this is the feature responsible for the peak appearing in the dos at these energies ( see fig . [ fig7 ] , lower panel ) . careful inspection of the dos reveals the appearance of smaller peaks at somewhat lower energies ( even the dos for moderate disorder has a small peak at around 2 mev , and its ipr for these modes is close to 0.5 ) . these peaks are associated with excitations of spin - pairs ( or larger clusters ) with varying separations , but they are not as well defined as the one corresponding to nearest - neighbors spins . the small clusters of mn spins giving rise to such modes are always found to be in regions densely populated with charge carriers . due to their closeness in space , these spins are much more strongly coupled to one another than they are to other spins with which they share charge carriers . this leads to the `` resonance - like '' character of these modes : either of the spins can be flipped with equal probability . as a result of the strong - coupling , the cost of flipping either spin is high , since it frustrates their fm arrangement . indeed , the characteristic energy of roughly @xmath286 reflects much stronger coupling than the average one present in the superlattice case . the histogram of the ipr of all modes of energy @xmath284 shown in fig . [ fig10 ] confirms all these conclusions . we caution that these energies may be substantially modified due to direct antiferromagnetic mn - mn exchange ( left out in our model ) in real systems . histogram of the ipr values for all spin - waves ( spin flips ) of energy @xmath287 , for @xmath288 ( upper ) , 216 ( middle ) and 512 ( lower panel ) . results are shown only for strong disorder configurations . with increasing @xmath6 the histograms shift to lower values , suggesting that these spin - wave modes are extended . ] finally , the spin - waves at intermediate energies @xmath289 correspond to extended modes . it is rather difficult to establish exactly the `` mobility edge '' corresponding to the transitions from localized to extended modes at either end , given the rather small system sizes considered here . thus , we use the values quoted above only as plausible estimates of these boundaries . clearly , for these energies the average ipr decreases with increasing @xmath6 ( see fig . [ fig8 ] ) . in fig . [ fig11 ] we plot the histograms of the ipr for all the modes with energy @xmath290 . in this case , size dependence is clearly visible , both in the position of the broad peak maximum as well as in the lower cut - off in ipr value . the position of the maxima scale roughly like @xmath291 . we have attempted both gaussian and lorentzian fits , but neither seems to capture the low - ipr tail properly . in the smaller samples , a second peak appears just below ipr=1 . at first sight , this might seem to indicate the existence of a finite density of localized states at these energies as well . however , we believe that this peak is a finite - size effect . clearly , its amplitude decreases with increasing @xmath6 and would vanish in the thermodynamic limit . this is very different from the behavior observed in either of the two localized energy ranges , where the distributions for all sizes are virtually identical ( see figs . [ fig9 ] and [ fig10 ] ) . the reason for the appearance of this second peak is simply the restriction ipr @xmath292 1 . as the broad peak moves towards higher values with decreasing @xmath6 , all the values in the upper tail `` bunch '' at ipr=1 . one final important observation relates to the absolute values of the iprs in the extended spin - wave regime . although scaling with system size is present , the corresponding ipr values are much higher than the ones expected for modes extended over _ all _ mn sites ( these values are shown by the positions of the goldstone modes in fig . [ fig9 ] , and are well below the cut - offs observed in fig . [ fig11 ] ) . this means that in the disordered system , the extended spin - waves are delocalized only over a fraction of the total number of sites . as the amount of disorder increases , this fraction decreases , since ipr averages for the moderate disorder are smaller than ipr averages for full disorder . these results are consistent with the picture provided by the temperature dependent mean - field study of this model ( ref . ) . there , we concluded that in the disordered system , the ( small fraction @xmath195 ) of charge carriers are concentrated in a small volume of the sample , where the local mn concentration is larger than average . as a result , the concentration of charge carriers is strongly enhanced in these regions , and the exchange with the mn spins inside these regions ( which is proportional to the probability of finding charge carriers nearby ) is greatly increased , preserving magnetization of these regions up to high temperatures . on the other hand , the mn spins in the regions devoid of charge carriers are very weakly coupled , and behave as free spins down to very low temperatures . due to the very inhomogeneous assembly of weakly- and strongly- interacting spins , the magnetization curves have very unusual , concave shapes . the present study of the spin - waves corroborates the same picture . we find very low - energy , spin - flip like excitations , which are obviously due to the weakly interacting spins , and which are responsible for a very sharp decrease of the magnetization at exponentially small temperatures.@xcite this is is to be contrasted with the behavior of ordered conventional ferromagnets , where at low temperatures only low - energy long - wavelength spin - waves can be excited . since their phase space is vanishing in the long - wavelength limit ( @xmath293 in @xmath294 dimensions ) , the magnetization of conventional ferromagnets decreases very slowly from its saturation value with increasing temperature , leading to convex upwards magnetization curves.@xcite in our model , the extended spin - waves are concentrated around the high - density mn regions , where the charge carriers mediating the interactions are to be found . this is consistent with the appearance of modes whose ipr , while scaling with the system size , shows modes extended only over a fraction of all system sites . as the amount of disorder decreases all the way to a fully ordered superlattice , the charge carriers are more and more homogeneously spread throughout the sample and the ipr of the extended modes has lower and lower values , as observed from fig . however , the more homogeneous a sample is , the less the average afm coupling between the mn spins and the charge carriers , since the charge carriers have now equal probability of being found anywhere in the sample , instead of being concentrated in a small fraction of the space . this leads to a decrease of the critical temperature @xmath295 , as observed in both mean - field @xcite and monte - carlo@xcite analysis . the enhancement of the mean - field @xmath295 with increasing disorder is suggested also by the existence of the high - energy localized modes , which show that ferromagnetic alignment will persist in high - density clusters up to very high temperatures . it also suggests that local ferromagnetic fluctuations might be observed well above @xmath295 . the aim of this paper is two - fold . first , we demonstrate the accuracy and speed of a new scheme for computing rpa spectra , which allows tackling of systems of much larger sizes than the ones that can be analyzed with the standard rpa formulation . investigation of large system sizes and averages over many disorder realizations facilitate a clear picture for the problem we are interested in , namely the spectrum and nature of spin - waves of a disordered dms . we then demonstrate that disorder can significantly change the spectrum and the nature of the spin - waves . this is likely to lead to important consequences not only as far as magnetic properties are concerned ( we have already commented on the fast de - magnetization with increasing temperature , due to low - energy spin - flip modes ) . more importantly , this may have significant consequences for transport properties as well . for instance , charge carrier spin scattering is likely to be very different in various regions of the sample . large anomalous hall effect has been observed in ( ga , mn)as,@xcite but the theory used to interpret it is borrowed from phenomenology relevant to homogeneous ferromagnetic metals . in an inhomogeneous system , some of the accepted ideas might have to be modified or at least verified to still hold true . this research was supported by nsf dmr-9809483 . m.b . was supported in part by a postdoctoral fellowship from the natural sciences and engineering research council of canada . t. dietl , a. haury and y. m. daubign , phys . b * 55 * , r3347 ( 1997 ) ; m. takahashi , phys . rev . b * 56 * , 7389 ( 1997 ) ; t. jungwirth , w. a. atkinson , b. h. lee and a. h. macdonald , phys . rev . b * 59 * , 9818 ( 1999 ) ; t. dietl , h. ohno , f. matsukura , j. cibert and d. ferrand , science * 287 * , 1019 ( 2000 ) . uniform magnets as well as metallic alloys are characterized by weak deviations from a universal curve on a @xmath296 versus @xmath297 and @xmath298 plot . [ see t. kaneyoshi , _ introduction to amorphous magnets _ , p 56 - 61 ( world scientific , 1992 ) ] .

in a couple of experiments , the measurement of the frequencies of collective modes in trapped fermi gases revealed a lot of interesting information on the equation of state , the validity of superfluid hydrodynamics , and the superfluid - normal phase transition @xcite . more recently , the transition from the hydrodynamic to the collisionless regime in the normal phase was also studied @xcite . however , since the frequencies of the collective modes depend on the trap frequencies , a precise knowledge of the latter is required for a meaningful interpretation of the collective - mode data . a possibility to determine the trap frequency with high precision is the measurement of the frequency of the sloshing mode @xcite , which is an oscillation of the center of mass of the trapped atom cloud . in a harmonic trap , this oscillation ( also called kohn mode ) is undamped and its frequency coincides exactly with the trap frequency in the corresponding direction , independently of the number of atoms @xmath0 , or their interaction , their temperature @xmath1 etc . this is a consequence of the kohn theorem @xcite and follows from the fact that the center - of - mass oscillation decouples completely from the internal dynamics of the gas if the interaction is translationally invariant and the external potential is harmonic @xcite . however , in practice the trap potential is never exactly harmonic . in optical dipole traps @xcite , the potential is typically gaussian ( corresponding to the intensity profile of the laser beam ) . because of this anharmonicity , the frequency of the sloshing mode is shifted , and the shift depends on the system parameters such as @xmath0 , @xmath1 , the scattering length @xmath2 characterizing the interaction strength , etc . furthermore , the sloshing mode is no longer undamped . the anharmonicity of the trap complicates considerably the analysis of collective - mode experiments . in ref . @xcite the measured frequencies were corrected for the anharmonicity effects by giving them in units of the measured frequency of the sloshing mode . however , it is clear that the damping rates of the modes can not be corrected in this way . in ref . @xcite , the damping rate of the sloshing mode was used to estimate the increase of the damping rate of other modes due to the anharmonicity . both corrections are ad - hoc prescriptions without rigorous justification . it is therefore strongly desirable to get a better understanding of the anharmonicity effects on the sloshing mode . the aim of the present paper is to describe theoretically the frequency shift and the damping rate of the sloshing mode in an anharmonic trap . we will compare our results with the experimental data available from the innsbruck group @xcite and with the numerical results by wu and zhang @xcite . we will also discuss in detail the damping mechanism of the sloshing mode . in an anharmonic potential , the center - of - mass motion is no longer decoupled from the internal degrees of freedom of the cloud . we will see that the damping of the transverse sloshing mode is a consequence of its coupling to other damped collective modes , in particular to the radial dipole mode and the bending mode . the framework of our study is the boltzmann equation , including mean field @xcite and in - medium cross - section @xcite . especially the mean field is expected to be important in the present context , because it can have a sizable effect on the density profile , i.e. , on how far the cloud extends into the anharmonic region of the trap potential . the boltzmann equation is solved approximately with the help of the phase - space moments method . this method , when extended beyond the lowest order , has proven to be in very good quantitative agreement with the results of a full numerical simulation @xcite . it has also been quite successful for the description of the experimental results for the frequency and damping rate of the radial quadrupole mode @xcite . in the present case of the sloshing mode , we include phase - space moments of first and third order . the paper is organized as follows . the general formalism is briefly presented in sec . [ sec : formalism ] . then we specify our model for the experimental trap potential in sec . [ sec : potential ] . in sec . [ sec : firstorder ] , we give a formula for the frequency shift of the sloshing mode within the first - order moments method . then we extend the ansatz to third order in sec . [ sec : thirdorder ] . the physical contents of the extended ansatz and the numerical results are discussed . in sec . [ sec : cubic ] the third - order ansatz is used to describe also the radial dipole , radial octupole , and bending modes . finally , in sec . [ sec : cl ] , we will conclude . throughout the paper , we use units with @xmath3 . in this section , we give a short summary of the formalisms of refs . more details can be found there . we consider a balanced two - component ( @xmath4 ) fermi gas of atoms with mass @xmath5 and interspecies attractive interaction ( scattering length @xmath6 ) , trapped in a potential @xmath7 . the framework we use to describe the collective dynamics of the system in the normal - fluid phase is the boltzmann equation . we assume that the two components move in phase , so that only one distribution function @xmath8 is needed . it is normalized to @xmath9 , and expectation values of one - body operators are given by @xmath10 as in ref . @xcite , we include in - medium effects in both the transport and the collision parts of the boltzmann equation : a mean field like potential @xmath11 and the in - medium modified cross - section @xmath12 , respectively . within the thomas - fermi or local - density approximation ( lda ) , the equilibrium distribution function reads @xmath13 } + 1},\ ] ] where @xmath14 is the mean field in equilibrium , @xmath15 is the chemical potential , and @xmath16 is the inverse temperature . as in ref . @xcite , we obtain @xmath14 from the single - particle self - energy in ladder approximation , evaluated at the fermi level . in the weak coupling limit , this reduces to the hartree term @xmath17 , where @xmath18 is the density per spin state , and it remains finite for all interaction strengths up to the unitary limit , @xmath19 . in equilibrium , the main effect of the mean field is to enhance the density in the center of the trap , as shown in fig . 3 of ref . @xcite . for the study of collective oscillations , it is sufficient to consider small deviations from equilibrium and to linearize the boltzmann equation with respect to @xmath20 . if we write the variation of the distribution function in the form @xcite @xmath21 with @xmath22 , the linearized boltzmann equation can be written as @xmath23\ , , \label{eq : boltzlin}\end{gathered}\ ] ] where @xmath24 denotes the poisson bracket , @xmath25 is the perturbation of the trap potential that is used to excite the collective mode , @xmath26 is the variation of the mean field due to the variation of the density , and @xmath27 $ ] is the linearized collision term . since we want to calculate the so - called response function , we take the perturbation to be a pulse , @xmath28 as in ref . @xcite , we approximate the variation of the mean field by @xmath29 the linearized collision integral reads @xmath30 = \int \frac{d^3p_1}{(2\pi)^3 } \int d \omega\ , \frac{d\sigma}{d\omega } \frac{|{{\bm{\mathrm{p}}}}-{{\bm{\mathrm{p}}}}_1|}{m } { f_{\mathit{eq}}}f_{{\mathit{eq}}1 } { \bar{f}_{\mathit{eq}}}^\prime \bar{f}_{{\mathit{eq}}1}^\prime\\ \times ( \phi+\phi_1-\phi^\prime-\phi_1^\prime)\,,\end{gathered}\ ] ] where an obvious notation for the different @xmath31 and @xmath32 at the momenta before ( @xmath33 , @xmath34 ) and after the collision ( @xmath35 , @xmath36 ) has been used . note that , especially near the critical temperature , the in - medium cross - section @xmath12 can differ strongly from the free one @xcite . as in ref . @xcite , we are looking for a semi - analytical solution of the boltzmann equation ( [ eq : boltzlin ] ) by using the method of phase - space moments . in this section , we will generalize the formalism of that paper to the case of an arbitrary trap potential @xmath37 and with mean field @xmath11 . the basic idea is to approximate the function @xmath32 by a polynomial in the components of @xmath38 and @xmath33 with time - dependent coefficients @xmath39 , @xmath40 where the @xmath41 are suitable basis functions , e.g. , monomials in the components of @xmath38 and @xmath33 . multiplying the linearized boltzmann equation ( [ eq : boltzlin ] ) by @xmath42 and integrating over phase space , one obtains , after a fourier transform with respect to @xmath43 , a set of @xmath44 coupled linear algebraic equations for the @xmath44 coefficients @xmath45 . in matrix form , they read @xmath46 where @xmath47\ , , \label{eq : acollij}\end{gathered}\ ] ] and @xmath48 the variation of the mean field in eq . ( [ eq : adeltauij ] ) is defined as @xmath49 equations ( [ eq : atransij ] ) and ( [ eq : ai ] ) correspond to eqs . ( 14 ) and ( 16 ) in ref . @xcite which have been simplified by integration by parts . comparing eqs . ( [ eq : ac = a])-([eq : ai ] ) with the analogous ones of ref . @xcite , eqs . ( 11)-(16 ) , one sees that the mean field gives rise to a new term , @xmath50 . however , one should keep in mind that implicitly all terms depend on the mean field since it modifies the equilibrium distribution @xmath51 . one can see that the matrices @xmath52 and @xmath53 are symmetric , @xmath54 is antisymmetric , whereas @xmath50 has no defined symmetry . in practice , the calculation of the matrices is straight - forward , but tedious , and we made use of the mathematica software to express the numerous matrix elements in terms of a smaller number of integrals over equilibrium quantities . without an external perturbation @xmath55 , i.e. , for @xmath56 , eq . ( [ eq : ac = a ] ) has a solution with non - vanishing coefficients @xmath39 only if @xmath57 . the frequencies @xmath58 for which this happens are obviously given by the eigenvalues of the matrix @xmath59 . if they are well separated , it is possible to interpret them as the frequencies of the eigenmodes of the system . in general , they are complex , and their imaginary part describes the damping rate of the corresponding mode @xcite . however , as discussed in ref . @xcite , when the moments method is extended to higher order , there can be many eigenvalues belonging to a single collective mode . in this case , the scattering of the eigenvalues , which goes over into a continuous spectrum in the limit of an infinite number of moments @xcite , corresponds to a new contribution to the damping ( landau damping ) in addition to the imaginary parts coming from the collision term . in order to obtain the mode frequency and damping rate in this case , it is useful to look at the response function which contains the contributions of all eigenvalues . we denote by @xmath60 the fourier transform of the expectation value @xmath61 of some operator @xmath62 after the perturbation . the so - called response function is equal to @xmath63 in the special case that the excitation operator is @xmath64 . in all excitations considered in this paper , @xmath65 will be zero . the strength function is proportional to the imaginary part of the response function . in order to calculate the response function , we need the coefficients @xmath45 of the ansatz ( [ generalansatz ] ) . by diagonalizing the matrix @xmath66 with @xmath67 , we can write them as @xmath68 then , the response function can be easily obtained as @xmath69 , where @xmath70 the frequency and damping of the mode can be extracted from the response @xmath60 , e.g. , by fitting the peak in the strength function @xmath71 corresponding to the mode under consideration with a lorentzian . using the solution for the coefficients @xmath45 at the peak , we can also obtain the velocity field of the corresponding collective mode : @xmath72 in this section we recall the shape of a typical optical dipole trap . we concentrate on the configuration used by the innsbruck group , consisting of a focused - beam trap @xcite with additional magnetic confinement in the axial direction . following ref . @xcite , we parametrize the experimental trap potential as @xmath73\\ + \frac{1}{2 } m { \omega_{z,\mathit{mag}}}^2 z^2 \ , , \label{eq : vtrap}\end{gathered}\ ] ] where @xmath74 , @xmath75 is the trap depth , @xmath76 is the minimal waist of the laser beam , @xmath77 is the rayleigh length , @xmath78 the laser wavelength and @xmath79 defines the magnetic trapping in the @xmath80 direction . in order to compare our results with the data from refs . @xcite , we focus on the setup of that experiment . the trap was of the type ( [ eq : vtrap ] ) , with strong anharmonicity in the @xmath81-@xmath82 plane due to the gaussian shape of the laser beams . the confinement along the @xmath80 direction was practically harmonic . in our calculations , we will expand eq . ( [ eq : vtrap ] ) up to second order in @xmath80 and up to sixth order in @xmath83 . the result can be written as @xmath84 where @xmath85 and @xmath86 are the radial and axial trap frequencies , respectively . we use eq . ( [ eq : vsix ] ) instead of eq . ( [ eq : vtrap ] ) because it simplifies the calculations in the sense that it leads to a density that tends to zero for @xmath87 . the necessity to go beyond fourth order is also clear from eq . ( [ eq : vsix ] ) : otherwise the potential would be unbound from below . for illustration , fig . [ fig : vtrap ] for @xmath88 . the fermi energy for the parameters of ref . @xcite is indicated by the thin dashes . the potentials are in units of @xmath89 and @xmath81 is in units @xmath90 . ] shows the real trap potential ( [ eq : vtrap ] ) , its harmonic approximation , and the one we will use , eq . ( [ eq : vsix ] ) . in the present paper , we concentrate on axially symmetric traps . nevertheless , we write @xmath91 and @xmath92 instead of @xmath93 for the trap frequencies in @xmath81 and @xmath82 direction if the formulas can be generalized to the triaxial case . in our numerical calculations , we use as an example the parameters of the innsbruck experiment @xcite . as mentioned in ref . @xcite , the sloshing mode was studied with the same parameters as the compression mode , i.e. , @xmath94 hz , @xmath95 hz , and @xmath96 @xmath97k . the trap was loaded with @xmath98 atoms of @xmath99li in the unitary limit , @xmath100 , and the temperature was varied between @xmath101 and @xmath102 . since we can not describe the superfluid phase , we limit ourselves to temperatures above @xmath103 @xcite . moreover , we approximate the unitary limit numerically by setting @xmath104 . as pointed out in ref . @xcite , the anharmonicity effects depend mainly upon the ratio @xmath105 . defining the fermi energy as usual by @xmath106 , one obtains @xmath107 , i.e. , at low temperature the anharmonicity effects are relatively weak . for illustration , @xmath108 is indicated in fig . [ fig : vtrap ] by the thin dashed line . we see that , for the present choice of parameters , the atoms start to feel the anharmonicity of the potential when their energy exceeds @xmath108 , while the sixth - order approximation to the potential stays very precise up to about five times the fermi energy . in order to describe the sloshing motion , say , along the @xmath81 direction , the ansatz ( [ generalansatz ] ) has to contain at least two basis functions : @xmath109 the second term describes the collective velocity of the cloud in @xmath81 direction , while the first one corresponds approximately to a displacement of the center of mass . the latter statement becomes exact in the special case of a purely harmonic trap without mean field ( @xmath110 ) . in this case , the ansatz ( [ ansatzfirst ] ) is closed with respect to the operator @xmath111 on the left - hand side of the boltzmann equation ( [ eq : boltzlin ] ) , and since the collision term does not contribute ( @xmath112=i[p_x]=0 $ ] because of momentum conservation in a collision ) , the ansatz ( [ ansatzfirst ] ) becomes exact . however , if the trap is not purely harmonic , this is no longer true , since the gradient of the trap potential generates new terms . and even if the trap is harmonic , it is not true for interacting atoms , which feel a mean field @xmath113 in addition to the trap potential @xmath37 . in the next section , we will therefore extend the ansatz ( [ ansatzfirst ] ) , but let us first look what happens at first order . we now apply the formalism described in sec . [ sec : formalism ] to @xmath114 . the calculation of the matrix @xmath115 is very easy in this case . after some algebra , we obtain a quadratic equation for the frequency of the sloshing mode by imposing @xmath116 . its solution reads @xmath117 the parameters @xmath118 and @xmath119 characterize , respectively , the strength of the mean field and of the anharmonicity effects . equation ( [ eq : wk1st ] ) has two important features : ( a ) if @xmath37 is harmonic ( @xmath120 ) , then @xmath121 , independently of the interaction , in accordance with the kohn theorem @xcite . this point was already discussed in ref . @xcite and shows the consistency of our approach , in particular of eq . ( [ deltau ] ) . ( b ) if @xmath37 is not purely harmonic ( @xmath122 ) , then @xmath123 depends on both the anharmonicity of the trap and on the interaction . another property of eq . ( [ eq : wk1st ] ) is that the sloshing mode is undamped ( i.e. , @xmath123 is real ) even in the case of an anharmonic potential . this is , however , only a consequence of the first - order approximation . a similar calculation was performed in the appendix b of ref . @xcite . in that reference , only the leading anharmonicity correction was kept . if we combine eqs . ( [ eq : vsix ] ) , ( [ eq : wk1st ] ) , and ( [ eq : cse ] ) and keep only the leading order in @xmath124 , we find @xmath125 this result differs from eq . ( b2 ) of ref . @xcite in two respects . first , there is an additional factor of three in front of the @xmath126 term ( misprint in ref . second , the anharmonicity correction is enhanced by a factor @xmath127 . this factor is missing in ref . @xcite because there the mean field was not considered , although the expectation values @xmath126 and @xmath128 were calculated with the density profiles of an interacting gas . as in ref . @xcite , we will extend the ansatz ( [ ansatzfirst ] ) by including higher - order terms . at the next higher order , the ansatz contains 18 terms @xmath129 where : @xmath130 the first two terms of @xmath131 coincide with @xmath114 , and the subsequent ones are all possible terms of third order in the components of @xmath38 and @xmath33 satisfying the symmetries of the mode : odd under @xmath132 and even under @xmath133 and @xmath134 . note that some of the third - order terms , e.g. , @xmath135 , generate a velocity field that is quadratic in the coordinates , whereas the velocity field corresponding to the first - order ansatz is spatially constant . other terms , e.g. , @xmath136 or @xmath137 , describe momentum - sphere distortions . they give rise to a non - vanishing collision integral so that they generate a damping . the momentum - independent third - order terms such as @xmath138 describe deformations of the cloud in coordinate space that will be discussed in more detail in the next subsection . as explained in sec . [ sec : response ] , the frequency and damping rate should now be extracted from the response function . the suitable excitation operator is @xmath139 because it gives all atoms a constant kick in @xmath81 direction , @xmath140 , at the moment of the excitation @xmath141 . the response function is defined as the expectation value @xmath142 . we checked numerically that , in a purely harmonic potential @xmath143 , the strength function has only a single sharp peak ( no damping ) at @xmath144 within the precision of our calculation ( better than @xmath145 ) , independently of the temperature and of the interaction strength . in fig . [ fig : resp_sloshing ] for the sloshing mode ( excitation @xmath146 ) in the anharmonic trap for various temperatures . the system parameters are those of ref . [ fig : resp_sloshing ] ] we show the strength functions for the sloshing mode in the anharmonic trap for various temperatures . in all cases the strength is concentrated in a single peak and the width of the peak corresponds to the damping rate . the position of the peak is always below @xmath144 , and with increasing temperature , this shift gets stronger . this is easily understood : with increasing temperature , more and more atoms reach the peripheral region where the anharmonic potential is flatter than the harmonic one ( cf . [ fig : vtrap ] ) . since in all the cases shown in fig . [ fig : resp_sloshing ] the strength is concentrated in a single peak , this allows us to extract the frequency and damping rate by fitting it with a single lorentzian . ( this is not clear a priori : for instance , if we do the calculation without mean field , the strength function for low temperature has two separate peaks , and it is not evident how one has to define the average frequency and total width . ) the results for the frequency and damping rate as functions of temperature are shown in fig . [ fig : sloshing](a ) . solid line : third - order result ; dashed line : first - order result ; dash - dotted line : first - order result obtained without mean field . the experimental data points are taken from ref . ( b ) temperature dependence of the damping rate in units of @xmath91 . solid line : full third - order result ; dashed line : third - order result obtained with the free instead of the in - medium cross section . the filled and empty data points are measured damping rates of the sloshing modes in @xmath81 and @xmath82 direction , respectively , from ref . [ fig : sloshing],title="fig : " ] . solid line : third - order result ; dashed line : first - order result ; dash - dotted line : first - order result obtained without mean field . the experimental data points are taken from ref . ( b ) temperature dependence of the damping rate in units of @xmath91 . solid line : full third - order result ; dashed line : third - order result obtained with the free instead of the in - medium cross section . the filled and empty data points are measured damping rates of the sloshing modes in @xmath81 and @xmath82 direction , respectively , from ref . [ fig : sloshing],title="fig : " ] and ( b ) as the solid lines . for comparison , the dashed line in fig . [ fig : sloshing](a ) shows the first - order result for the frequency . the dash - dotted line corresponds to the first - order result obtained without mean field . as expected , the mean field reduces the shift of the frequency , but its effect is moderate and most pronounced at lower temperature . the experimental results from fig . 2 of ref . @xcite are also shown in fig . [ fig : sloshing](a ) . in that paper , the measured empirical temperatures @xmath147 were already converted into real temperatures @xmath148 as explained in @xcite , so that the results can immediately be compared with our calculation . the trend is correctly reproduced , and the third - order result is clearly in better agreement with the data than the first - order one , but the theoretical frequency shift is still stronger than the experimental shifts reported in ref . however , one should keep in mind that while @xmath149 was precisely determined , the ratio @xmath150 depends also on @xmath89 , which was not very well known in that experiment . actually , @xmath89 was deduced from @xmath149 under the assumption that the anharmonicity effects become negligible at @xmath151 @xcite . if we suppose that the true trap frequency @xmath89 was just @xmath152% higher , the data points for @xmath150 are slightly shifted downwards and the agreement between theory and data becomes much better . the damping rate , shown in fig . [ fig : sloshing](b ) , increases strongly with the temperature : at @xmath153 the sloshing mode survives for several hundreds of oscillation periods , whereas at @xmath154 its amplitude decreases by a factor of @xmath155 after @xmath156 oscillations . the in - medium modification of the cross - section has only a small effect on the damping of the sloshing mode . the experimental data from ref . @xcite are in quite good agreement with our theoretical result , especially at the high temperature . the strongly scattered damping rates at low temperature are probably an artefact due to a beat caused by the small residual ellipticity of the trap potential in the experiment . we also compared our results with those shown in fig . 5(a ) of the recent paper @xcite by wu and zhang , where they solve the boltzmann equation numerically within the relaxation - time approximation with a local relaxation time @xmath157 taken from ref . @xcite . as we discussed in our previous work @xcite , the inclusion of higher - order moments accounts in an approximate way for the spatial dependence of the relaxation time . for the sake of comparison , we use the same trap parameters as wu and zhang , namely , @xmath158 hz , @xmath159 hz , and @xmath160 @xmath97k , and neglect , as it is done in ref . @xcite , the mean field and in - medium effects on the cross section . because of the larger trap depth , the frequency shifts are somewhat weaker than those shown in fig . [ fig : sloshing](a ) . since in the calculation by wu and zhang the frequency depends on the oscillation amplitude , we have to compare with the results obtained for the smallest amplitude [ red squares in fig . 5(a ) of ref . @xcite ] . if we extrapolate these results , shown only up to @xmath153 in ref . @xcite , to higher temperatures , they are in quite good agreement with our results above @xmath161 . at lower temperatures , however , our frequency shift is slightly stronger than that by wu and zhang , probably because they use boltzmann instead of fermi distributions so that their density is more concentrated in the trap center at low temperatures . in a harmonic trap , a characteristic feature of the kohn mode @xcite is that , as long as the interaction is translationally invariant , the cloud moves as a whole without any change in size or shape @xcite , as illustrated in figs . [ fig : sloshskew](a ) direction ( @xmath162 being the density integrated over @xmath82 and @xmath80 ) for ( a ) the kohn mode , in which the cloud oscillates as a whole , and ( b ) the radial compressional dipole mode , where the cloud shape oscillates , while the center of mass of the cloud stays at rest . all density profiles have the same normalization ( number of atoms ) and the same width . [ fig : sloshskew],title="fig:",width=151 ] direction ( @xmath162 being the density integrated over @xmath82 and @xmath80 ) for ( a ) the kohn mode , in which the cloud oscillates as a whole , and ( b ) the radial compressional dipole mode , where the cloud shape oscillates , while the center of mass of the cloud stays at rest . all density profiles have the same normalization ( number of atoms ) and the same width . [ fig : sloshskew],title="fig:",width=151 ] and [ fig : sloshbend](a ) . , title="fig:",width=151 ] , title="fig:",width=151 ] in an anharmonic trap , this is no longer the case and the cloud will be slightly distorted during the oscillation . in other words , because of the anharmonicity , the pure center - of - mass motion gets coupled to other collective modes . these are damped as usual by collisions , which results in a damping of the sloshing mode . what kind of distortions of the cloud shape can one expect ? let us first note that , to first order in the perturbation , the cloud width can not be changed ( neither in the transverse nor in the axial direction ) , because the excitation operator is odd with respect to @xmath163 . however , the density profile can become skewed in the direction of the oscillation ( i.e. , in the @xmath81 direction in our case ) . for illustration , fig . [ fig : sloshskew](b ) shows how the density profile changes during a radial compressional dipole oscillation . to derive the corresponding operator , let us start from the usual definition of the skewness , which is proportional to @xmath164 . keeping only terms linear in the variations @xmath165 and @xmath166 , and using @xmath167 , we find that for small oscillations , this is equal to @xmath168 . if the trap is axially symmetric , collective modes of different multipolarity in the radial direction do not mix . it is therefore useful to decompose the operator into dipolar and octupolar parts : @xmath169 with @xmath170 only @xmath171 can get a non - vanishing expectation value during the sloshing oscillation . another possible distortion involves the axial ( @xmath80 ) direction . let us consider again the sloshing mode in @xmath81 direction . then one could imagine that the cloud near @xmath172 , owing to its larger radial size , is more sensitive to the anharmonicity of the trap and therefore oscillates slightly more slowly than the parts of the cloud at large @xmath173 . hence , after a few oscillations , the oscillations near @xmath172 and at large @xmath173 become out of phase , which results in a bending of the cloud in the @xmath81-@xmath80 plane , as illustrated in fig . [ fig : sloshbend](b ) . a suitable measure for such a deformation is the expectation value of the operator @xmath174 like @xmath171 and @xmath175 , this operator is defined in such a way that it is insensitive to a translation of the cloud as a whole in @xmath81 direction . within the first - order ansatz ( [ ansatzfirst ] ) , it is easy to see that in a harmonic trap without mean field and to linear order in the perturbation , the density oscillates as @xmath176 , with @xmath177 , i.e. , the cloud shape remains unchanged . however , in the presence of a mean field @xmath11 , the cloud starts to deform during the oscillation . for illustration , the measures discussed above for the cloud skewness , @xmath178 , and for the bending @xmath179 , obtained with the first - order ansatz in the case of a harmonic trap ( @xmath180 ) , are displayed in fig . [ fig : skewbend](a ) by the dash - dotted and dotted lines . and bending @xmath181 of the cloud during the sloshing motion excited by @xmath182 at @xmath153 . in order to have comparable orders of magnitude , we multiplied the bending strengths by a factor @xmath183 . ( a ) in a harmonic trap , the unphysical skewness and bending of the first - order approximation ( dash - dotted and dotted lines ) are strongly suppressed when the third order moments are included ( solid and dashed lines ) . since in a harmonic trap there is no damping , we replaced @xmath58 by @xmath184 to generate this plot . ( b ) skewness and bending during the sloshing motion in the anharmonic trap within the third - order approximation . [ fig : skewbend ] ] the existence of these distortions is in contradiction to the kohn theorem and an unphysical consequence of the crude ansatz . at third order , owing to the additional degrees of freedom of the ansatz ( [ ansatzthird ] ) , the unphysical change of the cloud shape in the harmonic potential is strongly suppressed , as shown by the solid and dashed lines in fig . [ fig : skewbend](a ) . while unphysical in a harmonic trap , a change of the cloud shape during the sloshing motion is expected in an anharmonic trap . from the preceding discussion it is clear that at least the third - order ansatz is required to get a meaningful description of this effect . in fig . [ fig : skewbend](b ) , third - order results for the expectation values of @xmath171 and @xmath185 during the sloshing motion are displayed . we note that , compared to the third - order results for a harmonic trap , both responses are now significantly increased , so that the peaks correspond to a real physical effect . moreover , the observed shapes are typical of what one can obtain in a schematic model by coupling two damped modes ( dipole and bending ) to an undamped one ( sloshing ) . finally one can also see that the peaks are centered around the sloshing frequency , which is lower than in the harmonic case . in the previous section we have discussed coupling of the sloshing mode to other collective modes due to the trap anharmonicity . in particular , the ansatz ( [ ansatzthird ] ) contains the necessary terms to describe the transverse compressional dipole and bending modes corresponding to the operators ( [ eq : qdip ] ) and ( [ eq : qbend ] ) . in this section , we will study these modes in more detail . however , one should keep in mind that , since the excitation operators are already third - order ones , the ansatz ( [ ansatzthird ] ) can only give the leading - order result for these modes . let us start with the radial dipole oscillation excited by the operator ( [ eq : qdip ] ) . in an ideal fermi gas in an harmonic trap , this operator would excite two modes with frequencies @xmath186 and @xmath187 . in fig . [ fig : resp_dip ] , for the radial dipole mode ( excitation @xmath188 ) in the anharmonic trap for various temperatures . the parameters are the same as in fig . [ fig : sloshing ] . [ fig : resp_dip ] ] the corresponding response function is shown for various temperatures . we see that in the collisionless regime , i.e. , at high temperature , both modes exist but their frequencies are significantly lowered by the anharmonicity of the trap . even at @xmath154 both modes are still very strongly damped ( damping rate @xmath189 ) . at lower temperature , the damping gets so strong that both modes disappear completely : the hydrodynamic regime is not reached at any temperature . by artificially increasing the collision cross section to reach the hydrodynamic regime at @xmath190 , we found that there would be a mode at @xmath191 ( @xmath192 in the case of a harmonic trap ) , but in order to reach this regime in an experiment one would have to use a trap with a much lower radial frequency ( i.e. , a more spherical or even pancake - shaped trap ) . at the frequency of the sloshing mode , the response exhibits a characteristic dip due to the coupling between the dipole mode and the almost undamped sloshing mode ( the sloshing is damped mainly because of this coupling ) . in order to get a better understanding of the character of the two modes in the collisionless regime , we display their velocity fields in fig . [ fig : v_dip ] . -@xmath82 plane , averaged over @xmath80 , corresponding to the dipole modes at @xmath193 ( a ) and @xmath194 ( b ) at @xmath195 ( collisionless regime ) . the background color indicates the density per spin state integrated over @xmath80 in units of @xmath196 , with @xmath197 . [ fig : v_dip],title="fig : " ] -@xmath82 plane , averaged over @xmath80 , corresponding to the dipole modes at @xmath193 ( a ) and @xmath194 ( b ) at @xmath195 ( collisionless regime ) . the background color indicates the density per spin state integrated over @xmath80 in units of @xmath196 , with @xmath197 . [ fig : v_dip],title="fig : " ] although both modes are excited by the same operator , their velocity fields are quite different . the velocity field of the low - lying mode ( @xmath198 at @xmath195 ) , cf . fig . [ fig : v_dip](a ) , confirms that this mode comprises both center - of - mass and skewness oscillations . the velocity field of the high - lying mode ( @xmath199 at @xmath195 ) , shown in fig . [ fig : v_dip](b ) , resembles that of the dipole compression modes in atomic nuclei , shown e.g. , in fig . . @xcite . although in an axially symmetric trap the octupole mode does not couple to the sloshing mode , let us for the sake of completeness briefly discuss this mode . like the dipole operator ( [ eq : qdip ] ) , the octupole operator ( [ eq : qoct ] ) would excite two modes at @xmath200 and @xmath187 if the gas was collisionless and in a harmonic trap . the octupole response function for the realistic case is displayed in fig . [ fig : octupole](a ) . for the radial octupole mode ( excitation @xmath201 ) in the anharmonic trap for various temperatures . the parameters are the same as in fig . [ fig : sloshing ] . ( b ) velocity field corresponding to the high - lying mode at @xmath202 ( see caption of fig . [ fig : v_dip ] for details ) . [ fig : octupole],title="fig : " ] for the radial octupole mode ( excitation @xmath201 ) in the anharmonic trap for various temperatures . the parameters are the same as in fig . [ fig : sloshing ] . ( b ) velocity field corresponding to the high - lying mode at @xmath202 ( see caption of fig . [ fig : v_dip ] for details ) . [ fig : octupole],title="fig : " ] as in the dipole response , both modes are considerably shifted downwards because of the trap anharmonicity . the damping is even stronger than in the case of the dipole modes . both octupole modes have a similar velocity field , as an example fig . [ fig : octupole](b ) shows that of the higher - lying mode . in a nuclear physics context it was found that their difference lies mainly in the non - diagonal pressure tensor ( quadrupole moments in momentum space ) @xcite . at @xmath190 , there is only a very broad peak around the hydrodynamic frequency @xmath203 predicted in ref . @xcite for a harmonic trap . however the strong damping shows that one is still far from the hydrodynamic regime . by increasing again artificially the cross - section , the hydrodynamic frequency is found to be lowered to @xmath204 due to anharmonicity . finally , in fig . [ fig : bending ] for the bending mode ( excitation @xmath205 ) in the anharmonic trap for various temperatures . the parameters are the same as in fig . [ fig : sloshing ] . ( b ) velocity field in the @xmath81-@xmath80 plane , averaged over @xmath82 , corresponding to the bending mode at @xmath206 . the background color indicates the density integrated over @xmath82 in units of @xmath207 with @xmath208 . [ fig : bending],title="fig : " ] for the bending mode ( excitation @xmath205 ) in the anharmonic trap for various temperatures . the parameters are the same as in fig . [ fig : sloshing ] . ( b ) velocity field in the @xmath81-@xmath80 plane , averaged over @xmath82 , corresponding to the bending mode at @xmath206 . the background color indicates the density integrated over @xmath82 in units of @xmath207 with @xmath208 . [ fig : bending],title="fig : " ] we display the strength function and the velocity field of the bending mode . in a non - interacting gas in a harmonic trap , there would be three modes at @xmath209 and @xmath210 . in the realistic case , these peaks can not be resolved and one sees a damped oscillation with a frequency close to @xmath89 . in fact , the behavior of its frequency and damping rate is qualitatively similar to that of the sloshing mode , but the temperature dependence of the frequency is stronger . above @xmath211 , the damping is weaker than that of the sloshing mode , but this is probably an artefact of our approximation : we expect that the bending mode will receive an additional damping ( analogous to that of the sloshing mode ) from its coupling to a radial dipole mode modulated in @xmath80 [ operator @xmath212 , which is not included in the third - order ansatz . except at the lowest temperature , the mixing between sloshing and bending mode is weaker than that between sloshing and dipole mode , and the bending response does not show the characteristic dip that was found in the dipole response . the velocity field , shown in fig . [ fig : bending](b ) confirms that this mode corresponds to a bending of the elongated cloud in the @xmath81-@xmath80 plane . in this paper , we studied the frequency shift and damping of the sloshing mode of a normal - fluid fermi gas in a realistic anharmonic trap potential . we used the moments method in order to find approximate solutions of the boltzmann equation with mean field and in - medium cross - section . already at first order , the moments method predicts a downwards shift of the frequency of the sloshing mode . however , the first order ansatz is insufficient to describe the coupling between the center - of - mass and the internal degrees of freedom of the gas , which is responsible for the damping of the sloshing mode in an anharmonic trap . we therefore extended the ansatz to include also third - order phase - space moments . the third - order ansatz contains not only the sloshing mode , but also more complicated radial dipole and bending modes . the important mechanism for the damping of the sloshing mode is its coupling to these modes ( especially to the dipole mode ) due to the trap anharmonicity . in addition , the third - order ansatz allowed us to discuss the radial octupole mode . the comparison with the available experimental data for the sloshing mode shows that the essential features of the frequency shift can be reproduced . quantitatively , the calculated ratios @xmath150 are somewhat lower than those given in ref . @xcite . since both the mean field and the third - order moments give only small corrections to the frequency shift predicted by the first - order moments method , further improvements of the theory will probably give even smaller corrections . an effect which has not been considered in our linear - response study is the dependence of the sloshing frequency on the amplitude of the mode . however , this effect goes into the wrong direction , since the restoring force gets weaker with increasing amplitude ( see fig . [ fig : vtrap ] ) and therefore the frequency of the sloshing mode will be further reduced ( cf . also @xcite ) . as discussed in sec . [ sec : extendedansatz ] , a possible explanation of the discrepancy is that the experimental trap frequency @xmath89 may have been slightly underestimated in the analysis of ref . actually , our results could be used to determine the true value of @xmath89 . the agreement between the calculated and measured damping rates of the sloshing mode is satisfactory and makes us confident that our approach contains the essential physics to describe the coupling between the sloshing and internal degrees of freedom . this may also help to better understand other effects where the dissipation of center - of - mass kinetic energy plays a role , such as the heating of the cloud due to laser - beam - pointing noise discussed in ref . @xcite . in addition to the sloshing mode , we studied the radial dipole and octupole modes and the bending mode in the @xmath81-@xmath80 plane . however , one should keep in mind that for these modes the third - order ansatz contains only the leading order , and higher - order corrections can be important . we found that , with the present trap parameters , the dipole and octupole modes do not behave hydrodynamically at any temperature . the frequencies of the dipole , octupole , and bending modes are more strongly affected by the anharmonicity than the frequency of the sloshing mode . for the dipole and octupole modes , this can be intuitively understood , since these higher - order modes are more sensitive to what happens at larger @xmath213 , where the atoms feel the anharmonicity of the trap . the same argument should also apply to the radial quadrupole and scissors modes not studied in the present paper . one should therefore be careful when comparing theoretical results obtained for a harmonic trap with experimental ones ( as done in @xcite ) , since the anharmonicity effects can not be completely absorbed in a renormalized radial trap frequency . as a final remark we note that a similar third - order ansatz as the one used here [ eq . ( [ ansatzthird ] ) with @xmath214 can be used to describe one of the higher - order axial modes studied recently by the innsbruck group @xcite ( the @xmath215 mode in the notation of @xcite , corresponding to a motion as shown in fig . [ fig : sloshskew](b ) but in @xmath80 instead of @xmath81 direction ) . having more nodes than the usually considered collective modes , these kinds of modes are very interesting since they are closer to sound waves in uniform systems . we thank l. sidorenkov and r. grimm for useful explanations concerning the experimental data of @xcite and for the permission to use the unpublished data of @xcite . is supported by the _ fundao para a cincia e a tecnologia _ ( fct , portugal ) and the _ european social fund _ ( esf ) via the post - doctoral grant sfrh / bpd/64405/2009 . s.c . also acknowledges partial support by qren / feder , the compete programme , and fct under the project no . ptdc / fis/113292/2009 . * j. kinast , s.l . hemmer , m.e . gehm , a. turlapov , and j.e . thomas , phys . lett . * 92 * , 150402 ( 2004 ) . m. bartenstein , a. altmeyer , s. riedl , s. jochim , c. chin , j. hecker denschlag , and r. grimm , phys . 92 , 203201 ( 2004 ) ; a. altmeyer , s. riedl , c. kohstall , m.j . wright , j. hecker denschlag , and r. grimm , e - print cond - mat/0611285 ( 2006 ) . j. kinast , a. turlapov , and j.e . thomas , phys . a * 70 * 051401(r ) ( 2004 ) ; j. kinast , a. turlapov , and j.e . thomas , phys . a * 71 * , 029901(e ) ( 2005 ) . a. altmeyer , s. riedl , c. kohstall , m.j . wright , r. geursen , m. bartenstein , c. chin , j. hecker denschlag , and r. grimm , phys . lett . * 98 * , 040401 ( 2007 ) . a. altmeyer , s. riedl , m.j . wright , c. kohstall , j. hecker denschlag , and r. grimm , phys . a * 76 * , 033610 ( 2007 ) . wright , s. riedl , a. altmeyer , c. kohstall , e.r . snchez guajardo , j. hecker denschlag , and r. grimm phys . 99 , 150403 ( 2007 ) . s. riedl , e. r. snchez guajardo , c. kohstall , a. altmeyer , m. j. wright , j. hecker denschlag , and r. grimm , g. m. bruun and h. smith , phys . a * 78 * , 053609 ( 2008 ) . w. kohn , phys . rev . * 123 * , 1242 ( 1961 ) . l. brey , n.f . johnson , and b.i . halperin , phys . b * 40 * , 10647 ( 1989 ) . dobson , phys . . lett . * 73 * , 2244 ( 1994 ) . r. grimm , m. weidemller , and y.b . ovchinnikov , adv . at . . phys . * 42 * , 95 ( 2000 ) . l. wu and y. zhang , phys . a * 85 * , 045601 ( 2012 ) . l. sidorenkov and r. grimm , unpublished ; the present damping data have been obtained by reanalyzing the raw data of ref . l. wu and y. zhang , europhys . lett . * 97 * , 16003 ( 2012 ) . s. chiacchiera , t. lepers , d. davesne , and m. urban , phys . rev . a * 79 * , 033613 ( 2009 ) . bruun and h. smith , phys . a * 75 * , 043612 ( 2007 ) . t. lepers , d. davesne , s. chiacchiera , and m. urban , phys . rev . a * 82 * , 023609 ( 2010 ) . s. chiacchiera , t. lepers , d. davesne , and m. urban , phys . rev . a * 84 * , 043634 ( 2011 ) . e.m . lifshitz and l.p . pitaevskii , _ physical kinetics _ , landau - lifshitz course of theoretical physics , vol . 10 ( pergamon , oxford , 1980 ) . da providncia , nucl . a * 489 * , 111 ( 1988 ) . a. altmeyer , phd thesis ( innsbruck , 2007 ) . j. kinast , a. turlapov , j.e . thomas , q. chen , j. stajic , and k. levin , science * 307 * , 1296 ( 2005 ) . q. chen , j. stajic , and k. levin , phys . rev . lett . * 95 * , 260405 ( 2005 ) . p. massignan , g.m . bruun , and h. smith , phys . a * 71 * , 033607 ( 2005 ) . serr , t.s . dumitrescu , t. suzuki , and c.h . dasso , nucl . phys . a * 404 * , 359 ( 1983 ) . h. kohl , p. schuck , and s. stringari , nucl . a * 459 * , 265 ( 1986 ) . a. griffin , w .- c . wu , and s. stringari , phys . * 78 * , 1838 ( 1997 ) . savard , k.m . ohara , and j.e . thomas , phys . rev . a * 56 * , r1095 ( 1997 ) . snchez guajardo , phd thesis ( innsbruck , 2011 ) .

the thermodynamic analogue of the casimir effect originating from the confinement of vacuum fluctuations @xcite is the critical casimir effect due to the presence of long - ranged thermal fluctuations in a fluid close to its critical point at @xmath0 . the corresponding critical casimir forces have been predicted theoretically by fisher and de gennes in 1978 @xcite . experimentally , critical casimir forces have been studied only during the last decade when first measurements were performed indirectly via monitoring the thickness of wetting films upon approaching a critical end point @xcite . later on , the critical casimir force has been measured directly by using colloidal particles suspended in a binary liquid mixture @xcite . generically , the surfaces , which confine a binary liquid mixture , preferentially attract one of its two components leading to either positive [ @xmath1 or negative [ @xmath2 values of the scalar order parameter @xmath3 which describes the difference between the local concentration of one of the two components and its critical value . this generic preference of the surfaces confining the liquid can be described by effective surface fields . upon approaching @xmath4 , the critical adsorption profiles , which describe the concentration enhancement near the surface , become long ranged due to the concomitant divergence of the bulk correlation length @xcite . in semi - infinite systems , the transition from the phase in which only the region near the single surface is ordered to the one in which also the bulk is ordered is known as the extraordinary or normal transition @xcite . for two surfaces opposing each other , depending on the mutual combinations of boundary conditions ( bcs ) critical casimir forces are either attractive or repulsive . their range is set by the bulk correlation length @xmath5 . whereas in fluids @xmath5 is typically of molecular size , it attains values of the order of micrometers upon approaching the critical point @xcite . thus , critical fluctuations may induce effective interaction potentials with a strength of several @xmath6 at the nanometer and micrometer scale . moreover , critical casimir forces are universal in character : due to the divergence of the correlation length molecular details of the confined binary liquid mixture become irrelevant and only a few gross features of the system determine the main characteristics of the critical casimir forces @xcite . from a lower , parallel , crenellated wall which exhibits a periodic pattern with period @xmath7 consisting of rectangular crenels of width @xmath8 and depth @xmath9 and merlons of width @xmath10 and height @xmath9 . the system is spatially invariant in the @xmath11-direction . for later reference , the box at the bottom summarizes the definitions of the various scaling variables which the scaling function of the critical casimir force depends on ; @xmath12 is the reduced temperature @xmath13 and @xmath14 is the nonuniversal amplitude of the bulk correlation length @xmath15 in the disordered phase , which defines the universal critical exponent @xmath16 . , title="fig:",width=302 ] from a lower , parallel , crenellated wall which exhibits a periodic pattern with period @xmath7 consisting of rectangular crenels of width @xmath8 and depth @xmath9 and merlons of width @xmath10 and height @xmath9 . the system is spatially invariant in the @xmath11-direction . for later reference , the box at the bottom summarizes the definitions of the various scaling variables which the scaling function of the critical casimir force depends on ; @xmath12 is the reduced temperature @xmath13 and @xmath14 is the nonuniversal amplitude of the bulk correlation length @xmath15 in the disordered phase , which defines the universal critical exponent @xmath16 . , title="fig:",width=321 ] [ fig : sketch ] in view of nano- and micro - electromechanical devices , nowadays various experimental techniques are available to endow solid surfaces with precisely defined geometrical or chemical structures ( see , e.g. , refs . critical casimir forces for _ chemically _ structured confinements have been studied theoretically @xcite as well as in experiments with colloidal particles @xcite . it has been demonstrated that such patterns induce _ lateral _ critical casimir forces which can be used to trap particles reversibly along the lateral direction in a designated way @xcite . moreover , a suitable combination of attractive and repulsive critical casimir forces may even lead to stable levitation @xcite . here , we study critical casimir forces for _ geometrically _ structured confinements . the quantum - electrodynamic casimir effect in the presence of geometrically structured surfaces has been studied theoretically and experimentally for various surface topographies @xcite . two opposing surface gratings are subject to lateral quantum electrodynamic casimir forces @xcite . in this context , experimental studies focus on spherical particles near crenellated surfaces , i.e. , forming grooves with rectangular cross - sections @xcite . typically , the radius of the spherical particles is much larger than the period of the pattern of the crenellated surface such that , effectively , in the region of closest approach the system mimics the geometrical setup of a planar wall near a parallel , crenellated surface . critical adsorption and critical casimir forces for geometrically structured confinements have been studied for structures shaped as wedges and ridges with triangular cross - section @xcite , as well as for truncated wedges @xcite . it was found that , for large distances between a sawtooth - shaped wall and a planar wall , the critical casimir force effectively corresponds to the one between two planar walls ; on the other hand , for short distances between the two surfaces , the tips of the ridges dominate the order parameter profile and the characteristic power law behavior of the critical casimir force differs from that for planar surfaces @xcite . the present study extends these previous investigations @xcite into various directions . we consider a crenellated substrate close to a planar substrate at minimal surface - to - surface distance @xmath17 , as shown in fig . [ fig : sketch ] . the details of the geometry as well as the finite - size scaling of the critical casimir phenomena are described in section [ sec : theory ] . we calculate universal scaling functions for the critical casimir forces and for the order parameter profiles within mean - field theory and for identical chemical bcs at both walls ( @xmath18 configuration ) . in section [ sec : profiles ] we first study the universal features of the order parameter profiles close to @xmath4 for the geometry under consideration . second , in section [ sec : force ] we study the critical casimir forces acting on such geometrically structured substrates . finally , in section [ sec : summary ] we summarize our main findings . according to the theory of finite - size scaling , the singular contribution to the critical casimir force is described by a universal scaling function , which is independent of the molecular details of the binary liquid mixture and depends only on the bulk universality class of the associated critical point @xcite . here , we focus on the ising universality class characterized by a scalar order parameter @xmath3 , which encompasses the experimentally relevant binary liquid mixtures and simple fluids . upon approaching the critical point of the fluid , the bulk correlation length diverges as @xmath19 , where @xmath20 in spatial dimension @xmath21 and @xmath22 in spatial dimension @xmath23 @xcite ; @xmath24 are nonuniversal amplitudes characterized by the universal ratio @xmath25 in @xmath21 and @xmath26 in @xmath23 . the sign of the reduced temperature @xmath27 is chosen such that @xmath28 corresponds to the mixed ( disordered ) phase of the fluid , whereas @xmath29 corresponds to the ordered phase , corresponding to spontaneous phase separation . for an upper critical point the homogeneous phase is found at high temperatures , and one has @xmath13 . however , many experimentally relevant binary liquid mixtures exhibit a lower critical point ; in this case @xmath30 . in general , the sign and the amplitude of the critical casimir force depend on the types of effective chemical bcs at the walls and on the geometry of the confining surfaces . here we focus on the case of equal , symmetry - breaking @xmath31 bcs , which corresponds to the generic case of preferential adsorption of one of the two species of a binary liquid mixture . this leads to an attractive critical casimir force @xcite . inspired by experiments encompassing binary liquid mixtures of water and lutidine ( with a lower critical point at @xmath32 ) @xcite , here a binary liquid mixture with a lower critical point is considered at fixed pressure and at its critical composition . first , we briefly review the film geometry . in this case the liquid is confined between two parallel , macroscopically extended walls at a distance @xmath33 . according to renormalization group theory the critical casimir force @xmath34 per area of one wall , which is acting on the parallel walls ( @xmath35 ) , scales as @xcite @xmath36 the scaling function @xmath37 depends only on a single scaling variable given by the film thickness @xmath33 in units of @xmath38 , raised to the power @xmath39 . for equal chemical bcs , as discussed here , @xmath40 is negative , so that the critical casimir force is attractive . for @xmath41 the scaling function of the critical casimir force reduces to a universal constant value , the so - called critical casimir amplitude @xcite : @xmath42 accordingly , at @xmath4 the critical casimir force decays algebraicly @xmath43 . away from criticality , the critical casimir force decays exponentially as a function of @xmath44 . the scaling function @xmath37 has been calculated exactly in @xmath45 @xcite , for @xmath46 using a perturbative field - theoretical method @xcite or a local - functional method @xcite , and in @xmath21 numerically via monte carlo simulations @xcite . in the following we consider a crenellated wall located at a minimal surface - to - surface distance @xmath17 from a planar wall as sketched in fig . [ fig : sketch ] . the width and depth of the crenels are given by @xmath8 and @xmath9 , respectively , and the structure is periodic along the lateral @xmath47-direction with period @xmath7 , so that the width of the merlons , i.e. , the surface - to - surface separation between two neighboring crenels , is given by @xmath10 . accordingly , the corresponding universal contribution to the critical casimir force @xmath48 per area of the planar wall scales as @xmath49 where the geometrical parameters form the following scaling variables : @xmath50 the critical casimir force between a crenellated and a planar wall attains the value of the corresponding force between two planar walls in various limits . for very shallow crenels , i.e. , in the limit @xmath51 or @xmath52 , the following relation holds : @xmath53 as will be discussed in more detail below , we find that for narrow crenels with @xmath54 , the order parameter profile attains the form of the planar wall geometry , _ independent _ of the value of @xmath55 : @xmath56 similarly , for @xmath57 , one has @xmath58 for very broad crenels compared with the film thickness , i.e. , @xmath59 , the limiting behavior of the scaling function of the critical casimir force is given by the average of the scaling functions for two planar walls at distances @xmath17 and @xmath60 , respectively . this corresponds to the so - called proximity force approximation ( pfa ) , which we discuss in the following . within the pfa the surfaces are subdivided into infinitely small elements parallel to each other , and the resulting force is obtained by pairwise adding the individual contributions to the force . the derjaguin approximation is a special case of the pfa for smoothly curved surfaces . detailed comparisons of the derjaguin approximation with experimental and theoretical results revealed a good agreement for a large range of parameters @xcite . for the present geometry the resulting critical casimir force @xmath61 per area [ eq . ] acts along the @xmath62-direction and is the sum of two contributions : ( i ) the force between the fraction @xmath63 of the upper flat wall and the top surfaces of the merlons separated by the distance @xmath17 , and ( ii ) the force between the fraction @xmath64 of the upper flat wall and the bottom surfaces of the crenels separated by the distance @xmath60 : @xmath65 accordingly , the scaling function of the critical casimir force within the pfa is given by @xmath66 the scaling function @xmath67 is independent of @xmath68 because within pfa the effective interactions between the steps and the upper wall are ignored . that is , as long the values of @xmath64 and @xmath55 are fixed , steps may be arbitrarily added to or removed from the structured substrate ( e.g. , via a transformation @xmath69 and @xmath70 , where @xmath71 ) without changing the force within pfa , independent of the value of @xmath68 . analogous to the case of chemically striped surfaces , discussed in detail in refs . @xcite , every isolated geometrical step gives rise to a contribution to the scaling function of the critical casimir force per area which is proportional to @xmath72 . the asymptotic behavior for @xmath59 of the universal scaling function for the critical casimir force between a planar wall and a crenellated surface is therefore given by @xmath73 where we define @xmath74 as the universal contribution of a pair of geometrical steps as sketched in fig . [ fig : sketch_step](a ) . for a fixed value @xmath75 , in eq . @xmath76 actually does not depend on @xmath64 because in the limit @xmath59 the distance @xmath17 between the steps and the upper wall is much smaller than the step - step distances @xmath8 and @xmath10 , respectively , so that step - step interactions are negligible in this limit . in the following we therefore define @xmath77 and consider this as the generic case . for very deep crenels @xmath78 attains a @xmath79-dependent value which corresponds to the contribution to the critical casimir force between a pair of top corners of the right - angled edges of the merlons opposite to a planar wall ( see the sketch in fig . [ fig : sketch_step](b ) ) . ( eq . ) for @xmath80 at fixed values of @xmath81 and @xmath82 . ( a ) the surfaces of the lower , geometrically structured substrate which are parallel to the film ( indicated by a green line ) give rise to the pfa contribution @xmath67 to the universal scaling function @xmath83 of the critical casimir force . in the limit @xmath59 , the steps consisting of the upper and lower edges as well as the sidewalls ( indicated by the red lines and dots ) generate the step contribution @xmath84 to the critical casimir force ( see eq . ) . for fixed @xmath75 , the step - step interaction vanishes and @xmath76 becomes independent of @xmath64 . ( b ) for @xmath85 the step contribution @xmath86 to the critical casimir force effectively corresponds to the contribution of two right - angled corners opposite to a planar substrate and attains a limiting value independent of @xmath55 . , title="fig:",width=302 ] ( eq . ) for @xmath80 at fixed values of @xmath81 and @xmath82 . ( a ) the surfaces of the lower , geometrically structured substrate which are parallel to the film ( indicated by a green line ) give rise to the pfa contribution @xmath67 to the universal scaling function @xmath83 of the critical casimir force . in the limit @xmath59 , the steps consisting of the upper and lower edges as well as the sidewalls ( indicated by the red lines and dots ) generate the step contribution @xmath84 to the critical casimir force ( see eq . ) . for fixed @xmath75 , the step - step interaction vanishes and @xmath76 becomes independent of @xmath64 . ( b ) for @xmath85 the step contribution @xmath86 to the critical casimir force effectively corresponds to the contribution of two right - angled corners opposite to a planar substrate and attains a limiting value independent of @xmath55 . , title="fig:",width=321 ] [ fig : sketch_step ] the standard landau - ginzburg - wilson fixed - point effective hamiltonian describing critical phenomena of the ising bulk universality class is given by @xcite @xmath87=\int_v\,{{\ensuremath{\textrm{d}}}}^d{r}\,\left\ { \frac{1}{2}(\nabla\phi)^2 + \frac{{\hat{\tau}}}{2}\phi^2 + \frac{u}{4!}\phi^4 -h\phi \right\},\ ] ] where @xmath88 is proportional to the order parameter describing the fluid , which completely fills the accessible volume @xmath89 in @xmath90-dimensional space . the statistical weight of a configuration @xmath88 is proportional to @xmath91 . the parameter @xmath92 in eq . is proportional to @xmath12 , and @xmath93 is a coupling constant . the last term in eq . vanishes for the case considered here ( i.e. , @xmath94 ) , which corresponds to the situation that the concentrations of the species forming the binary liquid mixture are fixed to their critical values . in a finite - size system the bulk hamiltonian @xmath95 $ ] is supplemented by appropriate surface and curvature ( edge ) contributions @xcite . this surface contribution , which adds to eq . , is given by @xcite @xmath96=\int_{\partial v}\,{{\ensuremath{\textrm{d}}}}^{(d-1)}{r}\,\left\ { \frac{c}{2}\phi^2 -h_1\phi \right\},\ ] ] where @xmath97 is the so - called surface enhancement and @xmath98 is a surface field ; @xmath99 is the surface of the volume @xmath89 . in the strong adsorption limit @xcite , as discussed in the present study , these contributions generate boundary conditions for the order parameter such that @xmath100 corresponding to @xmath31 bcs . thus , the use of eq . together with additional surface contributions can be replaced by applying the appropriate bc to @xmath3 and by using eq . throughout the bulk . the mean - field order parameter profile minimizes the hamiltonian , i.e. , @xmath101/\updelta\phi|_{\phi=\langle\phi\rangle}=0 $ ] . in the _ bulk _ , the mean - field order parameter is spatially constant and attains the values @xmath102 for @xmath103 and @xmath104 for @xmath105 , where , besides @xmath14 , @xmath106 is the only additional independent nonuniversal amplitude appearing in the description of bulk critical phenomena ; @xmath107 is a standard critical exponent . within mean - field theory ( mft ) the following relations hold : @xmath108 and @xmath109 . for the film geometry , the mft scaling function for the critical casimir force can be determined analytically @xcite . one finds [ see eq . ] for the case of the same strong adsorption at both surfaces the critical casimir amplitude @xmath110 ^ 4/u\simeq-283.61\times u^{-1}$ ] , where @xmath111 is the complete elliptic integral of the first kind @xcite . renormalization group arguments tell that mft provides the correct universal properties of critical phenomena for spatial dimensions above the upper critical dimension , i.e. , @xmath112 , up to logarithmic corrections in @xmath113 @xcite . moreover , mft provides the lowest - order contribution to universal quantities within an expansion in terms of @xmath114 . ( see eqs . and ) at @xmath115 ( i.e. , @xmath116 ) for various values of @xmath82 and @xmath117 with @xmath118 fixed ( see fig . [ fig : sketch ] ) : ( a ) @xmath119 ( deep crenels and widely spaced merlons ) , ( b ) @xmath120 , @xmath121 ( shallow crenels and widely spaced merlons ) , and ( c ) and @xmath122 ( shallow crenels and closely spaced merlons ) . the local values of @xmath123 are indicated by the color code given by the side bars . large values of @xmath124 indicate strong adsorption of the fluid close to the surfaces . from the shape of the contour lines for @xmath125 near the structured wall ( lower white lines ) the dependence of the critical adsorption profile on the scaling variables @xmath55 and @xmath68 is clearly visible . whereas for widely spaced merlons with @xmath121 the adsorption profile follows the shape of the crenelated surface , for closely spaced merlons with @xmath126 the crenels are filled with the adsorbed fluid and for increasing normal distances @xmath127 the order parameter profile rapidly adopts an effective planar wall behavior corresponding to rather straight contour lines . , title="fig:",width=321 ] ( see eqs . and ) at @xmath115 ( i.e. , @xmath116 ) for various values of @xmath82 and @xmath117 with @xmath118 fixed ( see fig . [ fig : sketch ] ) : ( a ) @xmath119 ( deep crenels and widely spaced merlons ) , ( b ) @xmath120 , @xmath121 ( shallow crenels and widely spaced merlons ) , and ( c ) and @xmath122 ( shallow crenels and closely spaced merlons ) . the local values of @xmath123 are indicated by the color code given by the side bars . large values of @xmath124 indicate strong adsorption of the fluid close to the surfaces . from the shape of the contour lines for @xmath125 near the structured wall ( lower white lines ) the dependence of the critical adsorption profile on the scaling variables @xmath55 and @xmath68 is clearly visible . whereas for widely spaced merlons with @xmath121 the adsorption profile follows the shape of the crenelated surface , for closely spaced merlons with @xmath126 the crenels are filled with the adsorbed fluid and for increasing normal distances @xmath127 the order parameter profile rapidly adopts an effective planar wall behavior corresponding to rather straight contour lines . , title="fig:",width=321 ] ( see eqs . and ) at @xmath115 ( i.e. , @xmath116 ) for various values of @xmath82 and @xmath117 with @xmath118 fixed ( see fig . [ fig : sketch ] ) : ( a ) @xmath119 ( deep crenels and widely spaced merlons ) , ( b ) @xmath120 , @xmath121 ( shallow crenels and widely spaced merlons ) , and ( c ) and @xmath122 ( shallow crenels and closely spaced merlons ) . the local values of @xmath123 are indicated by the color code given by the side bars . large values of @xmath124 indicate strong adsorption of the fluid close to the surfaces . from the shape of the contour lines for @xmath125 near the structured wall ( lower white lines ) the dependence of the critical adsorption profile on the scaling variables @xmath55 and @xmath68 is clearly visible . whereas for widely spaced merlons with @xmath121 the adsorption profile follows the shape of the crenelated surface , for closely spaced merlons with @xmath126 the crenels are filled with the adsorbed fluid and for increasing normal distances @xmath127 the order parameter profile rapidly adopts an effective planar wall behavior corresponding to rather straight contour lines . , title="fig:",width=321 ] ( see eqs . and ) at @xmath115 ( i.e. , @xmath116 ) for various values of @xmath82 and @xmath117 with @xmath118 fixed ( see fig . [ fig : sketch ] ) : ( a ) @xmath119 ( deep crenels and widely spaced merlons ) , ( b ) @xmath120 , @xmath121 ( shallow crenels and widely spaced merlons ) , and ( c ) and @xmath122 ( shallow crenels and closely spaced merlons ) . the local values of @xmath123 are indicated by the color code given by the side bars . large values of @xmath124 indicate strong adsorption of the fluid close to the surfaces . from the shape of the contour lines for @xmath125 near the structured wall ( lower white lines ) the dependence of the critical adsorption profile on the scaling variables @xmath55 and @xmath68 is clearly visible . whereas for widely spaced merlons with @xmath121 the adsorption profile follows the shape of the crenelated surface , for closely spaced merlons with @xmath126 the crenels are filled with the adsorbed fluid and for increasing normal distances @xmath127 the order parameter profile rapidly adopts an effective planar wall behavior corresponding to rather straight contour lines . , title="fig:",width=321 ] ( see eqs . and ) at @xmath115 ( i.e. , @xmath116 ) for various values of @xmath82 and @xmath117 with @xmath118 fixed ( see fig . [ fig : sketch ] ) : ( a ) @xmath119 ( deep crenels and widely spaced merlons ) , ( b ) @xmath120 , @xmath121 ( shallow crenels and widely spaced merlons ) , and ( c ) and @xmath122 ( shallow crenels and closely spaced merlons ) . the local values of @xmath123 are indicated by the color code given by the side bars . large values of @xmath124 indicate strong adsorption of the fluid close to the surfaces . from the shape of the contour lines for @xmath125 near the structured wall ( lower white lines ) the dependence of the critical adsorption profile on the scaling variables @xmath55 and @xmath68 is clearly visible . whereas for widely spaced merlons with @xmath121 the adsorption profile follows the shape of the crenelated surface , for closely spaced merlons with @xmath126 the crenels are filled with the adsorbed fluid and for increasing normal distances @xmath127 the order parameter profile rapidly adopts an effective planar wall behavior corresponding to rather straight contour lines . , title="fig:",width=321 ] ( see eqs . and ) at @xmath115 ( i.e. , @xmath116 ) for various values of @xmath82 and @xmath117 with @xmath118 fixed ( see fig . [ fig : sketch ] ) : ( a ) @xmath119 ( deep crenels and widely spaced merlons ) , ( b ) @xmath120 , @xmath121 ( shallow crenels and widely spaced merlons ) , and ( c ) and @xmath122 ( shallow crenels and closely spaced merlons ) . the local values of @xmath123 are indicated by the color code given by the side bars . large values of @xmath124 indicate strong adsorption of the fluid close to the surfaces . from the shape of the contour lines for @xmath125 near the structured wall ( lower white lines ) the dependence of the critical adsorption profile on the scaling variables @xmath55 and @xmath68 is clearly visible . whereas for widely spaced merlons with @xmath121 the adsorption profile follows the shape of the crenelated surface , for closely spaced merlons with @xmath126 the crenels are filled with the adsorbed fluid and for increasing normal distances @xmath127 the order parameter profile rapidly adopts an effective planar wall behavior corresponding to rather straight contour lines . , title="fig:",width=321 ] [ fig : op ] the order parameter @xmath3 exhibits the following scaling properties ( see subsec . 2.5 in ref . @xcite ) : @xmath128 or equivalently @xmath129 with universal scaling functions @xmath130 and @xmath131 . the bulk order parameter varies as @xmath132 . this implies @xmath133 within mft one has @xmath134 so that with the definition @xmath135 eq . renders the following mft approximation for the scaling function @xmath136 : @xmath137 in the following we present these mft results which we have obtained by minimizing numerically @xmath95 $ ] using a finite element method in order to obtain the ( spatially inhomogeneous ) profile @xmath138 for the geometries under consideration . here , we focus on the case of strong adsorption and the same chemical bcs at the two surfaces . for distances from the surface of a substrate which are small compared to @xmath38 , or for @xmath139 , the order parameter varies algebraicly . in order to obtain a bc for the numerical calculations we use a short distance expansion ( see , e.g. , ref . @xcite and reference therein ) with @xmath140 at the surfaces of the two walls shown in fig . [ fig : sketch ] . in fig . [ fig : op ] the order parameter profile of a fluid confined between a planar and a crenellated wall at the bulk critical point @xmath115 is shown for @xmath141 ; @xmath138 depends on @xmath47 and @xmath62 and is invariant along the @xmath11-direction . from fig . [ fig : op ] we can infer that for @xmath142 the order parameter profile follows the shape of the crenellated surface [ figs . [ fig : op ] ( a ) and ( b ) ] , whereas for @xmath126 the space between the merlons corresponds to high values of the order parameter , i.e. , the crenels are `` filled '' with the adsorbed fluid [ fig . [ fig : op ] ( c ) ] until an almost straight contour line has formed separating the fluid with high order parameter , which fills the crenels , from the fluid with lower order parameter in the middle of the slit . hence for narrow crenels , for increasing @xmath127 the order parameter profiles rapidly approach the ones of a corresponding film of thickness @xmath17 . the critical casimir forces are calculated directly from the numerically obtained mean - field order parameter profiles using the stress tensor @xcite . as in sec . [ sec : profiles ] , here throughout we focus on the case @xmath143 . we estimate the numerical error of the present method to be less than 1% . of the critical casimir force as a function of @xmath144 for a fixed ratio @xmath143 and various values of @xmath82 and @xmath117 . in addition to the numerically obtained full mft scaling functions ( full curves ) , the corresponding scaling functions @xmath67 obtained within the pfa [ eq . ] ( dashed curves ) and the scaling function for two planar walls ( pw ) at distance @xmath17 ( dashed - dotted curves ) are shown . for @xmath119 , in ( a ) the mft scaling function does neither agree with the one obtained within pfa nor with the one for the pw case , but lies roughly in between these two curves . however , for shallow crenels and widely spaced merlons with @xmath120 and @xmath121 , in ( b ) the pfa is closer to the full mft scaling function . for shallow crenels and closely spaced merlons with @xmath122 , in ( c ) the full mft scaling function is similar to the one for two planar walls ( pw ) . figures [ fig : op ] and [ fig : force ] allow a direct comparison between the order parameter distribution and the critical casimir force at @xmath4 , i.e. , for @xmath145 . here the scaling function @xmath83 is normalized with @xmath146 so that the ratio does not depend on the coupling constant @xmath147 , which within mft is undetermined . , title="fig:",width=302 ] of the critical casimir force as a function of @xmath144 for a fixed ratio @xmath143 and various values of @xmath82 and @xmath117 . in addition to the numerically obtained full mft scaling functions ( full curves ) , the corresponding scaling functions @xmath67 obtained within the pfa [ eq . ] ( dashed curves ) and the scaling function for two planar walls ( pw ) at distance @xmath17 ( dashed - dotted curves ) are shown . for @xmath119 , in ( a ) the mft scaling function does neither agree with the one obtained within pfa nor with the one for the pw case , but lies roughly in between these two curves . however , for shallow crenels and widely spaced merlons with @xmath120 and @xmath121 , in ( b ) the pfa is closer to the full mft scaling function . for shallow crenels and closely spaced merlons with @xmath122 , in ( c ) the full mft scaling function is similar to the one for two planar walls ( pw ) . figures [ fig : op ] and [ fig : force ] allow a direct comparison between the order parameter distribution and the critical casimir force at @xmath4 , i.e. , for @xmath145 . here the scaling function @xmath83 is normalized with @xmath146 so that the ratio does not depend on the coupling constant @xmath147 , which within mft is undetermined . , title="fig:",width=321 ] [ fig : force ] in fig . [ fig : force ] the scaling function @xmath148 of the critical casimir force is shown as a function of @xmath144 for various values of @xmath82 and @xmath117 . in addition to the numerically obtained full mft scaling functions ( solid curves ) , the corresponding scaling functions @xmath149 obtained within the pfa [ eq . ] ( dashed curves ) , and the scaling function @xmath150 for two planar walls ( pw ) at distance @xmath17 ( dashed - dotted curves ) are shown . for @xmath119 , in fig . [ fig : force](a ) the mft scaling function does neither agree with the one obtained within the pfa nor with the one for the pw case , but lies roughly in between these two curves . the corresponding mft order parameter profile at @xmath4 is shown in fig . [ fig : op](a ) . for @xmath120 and @xmath121 , in fig . [ fig : force](b ) the scaling function obtained within the pfa is closer to the full mft scaling function which corresponds to the critical mft order parameter profile shown in fig . [ fig : op](b ) . for these shallow crenels and widely spaced merlons the relative contribution from the the right - angled steps of the merlons to the total critical casimir force is smaller than for the case shown in fig . [ fig : force](a ) . for shallow crenels and closely spaced merlons with @xmath122 , in fig . [ fig : force](c ) the mft scaling function is similar to the one for two planar walls ( pw ) . this corresponds to the case in which the crenels are filled by a fluid with a high value of the order parameter ( see fig . [ fig : op](a ) for @xmath115 ) . as can be inferred from fig . [ fig : force ] , in general the pfa deviates from the full mft results because critical phenomena do not allow for linear superposition . in order to study the deviations of the results for the force from the corresponding ones following from the assumption of pairwise additivity , we study the ratio @xmath151 of the scaling function @xmath83 obtained within full mft and the one ( @xmath152 ) obtained within pfa . from eqs . ( [ eq : limit - omega-0 ] ) and ( [ eq : derjaguin - scaling ] ) we find for @xmath116 , i.e. , @xmath115 : @xmath153 of the scaling function @xmath83 of the critical casimir force obtained within full mft and the one ( @xmath152 ) obtained within pfa , for @xmath116 ( i.e. , @xmath115 ) and @xmath154 ( eq . ) . the ratio varies between @xmath155 and @xmath156 corresponding to deviations of the actual force from the force calculated within pfa between 0 in ( a ) @xmath151 is shown as a function of @xmath117 for several values of @xmath82 . for widely spaced merlons ( @xmath157 ) , according to eq . the ratio @xmath151 approaches @xmath155 . the corresponding decay @xmath158 is shown as dashed lines . for very shallow crenels ( @xmath52 ) , the ratio @xmath151 approaches @xmath155 , so that pfa is valid . in ( b ) @xmath151 is shown as a function of @xmath82 for various values of @xmath68 . we find that , for deep crenels ( @xmath85 ) , @xmath151 attains a plateau . for large values of @xmath68 this corresponds to the limiting behavior given in eq . ; ultimately , for @xmath59 pfa is valid . on the other hand for closely spaced merlons ( @xmath159 ) the critical casimir force approaches its value for an effective film geometry with the two parallel walls at distance @xmath17 . in this limit the ratio @xmath151 is given by eq . with @xmath154 and @xmath23 ( dashed line , `` pw '' ) . , title="fig:",width=302 ] of the scaling function @xmath83 of the critical casimir force obtained within full mft and the one ( @xmath152 ) obtained within pfa , for @xmath116 ( i.e. , @xmath115 ) and @xmath154 ( eq . ) . the ratio varies between @xmath155 and @xmath156 corresponding to deviations of the actual force from the force calculated within pfa between 0 in ( a ) @xmath151 is shown as a function of @xmath117 for several values of @xmath82 . for widely spaced merlons ( @xmath157 ) , according to eq . the ratio @xmath151 approaches @xmath155 . the corresponding decay @xmath158 is shown as dashed lines . for very shallow crenels ( @xmath52 ) , the ratio @xmath151 approaches @xmath155 , so that pfa is valid . in ( b ) @xmath151 is shown as a function of @xmath82 for various values of @xmath68 . we find that , for deep crenels ( @xmath85 ) , @xmath151 attains a plateau . for large values of @xmath68 this corresponds to the limiting behavior given in eq . ; ultimately , for @xmath59 pfa is valid . on the other hand for closely spaced merlons ( @xmath159 ) the critical casimir force approaches its value for an effective film geometry with the two parallel walls at distance @xmath17 . in this limit the ratio @xmath151 is given by eq . with @xmath154 and @xmath23 ( dashed line , `` pw '' ) . , title="fig:",width=321 ] [ fig : compare ] figure [ fig : compare ] shows the ratio @xmath151 as function of @xmath55 and @xmath68 for fixed values @xmath116 and @xmath154 ( eq . ) . the ratio @xmath151 varies between @xmath155 and @xmath156 which corresponds to deviations of the actual force from the one calculated within pfa between 0% and 100% , respectively . in fig . [ fig : compare](a ) @xmath151 is shown as a function of @xmath68 for various values of @xmath55 . this graph tells that the limiting behavior for @xmath59 given in eq . already holds for @xmath160 . the amplitude function @xmath161 [ eq . ] has been determined via a least square fit to the numerical data . as expected on physical grounds , for very shallow crenels the critical casimir force can be approximated by the corresponding pfa expression and hence @xmath162 for @xmath52 . in fig . [ fig : compare](b ) we show @xmath151 as a function of @xmath55 for various values of @xmath68 . for deep crenels with @xmath163 this ratio reaches a plateau , i.e. , for @xmath85 the strength of the critical casimir force is not affected by the depth of the crenels so that for @xmath59 pfa becomes valid . for @xmath159 the critical casimir force approaches its value for an effective film geometry with the two parallel walls ( dashed line , `` pw '' ) at distance @xmath17 ( see eq . for @xmath154 and @xmath23 ) . this means that for closely spaced merlons the critical casimir force reduces to that between two parallel flat surfaces at separation @xmath17 . [ see eq . ] of the step contribution to the critical casimir force for @xmath116 as a function of the reduced depth @xmath82 of the crenels . for @xmath164 a plateau value @xmath165 is reached and the step contribution to the critical casimir force becomes independent of the depth of the crenels . in ( b ) @xmath166 is shown as a function of @xmath79 as obtained for @xmath167 . according to our analysis , for @xmath167 the limiting behavior for @xmath168 is almost reached ( see ( a ) ) . ( according to eq . @xmath76 does not depend on @xmath64 as long as the value of the latter is not @xmath169 or @xmath155 . ) , title="fig:",width=302 ] [ see eq . ] of the step contribution to the critical casimir force for @xmath116 as a function of the reduced depth @xmath82 of the crenels . for @xmath164 a plateau value @xmath165 is reached and the step contribution to the critical casimir force becomes independent of the depth of the crenels . in ( b ) @xmath166 is shown as a function of @xmath79 as obtained for @xmath167 . according to our analysis , for @xmath167 the limiting behavior for @xmath168 is almost reached ( see ( a ) ) . ( according to eq . @xmath76 does not depend on @xmath64 as long as the value of the latter is not @xmath169 or @xmath155 . ) , title="fig:",width=321 ] [ fig : a ] as can be inferred from eq . and fig . [ fig : compare](a ) for large values of @xmath68 the critical casimir force acting between a flat and a crenellated wall can be described reliably by the sum of the pfa - contribution and a contribution @xmath170 . the latter contribution stems from effects due to the presence of geometrical steps ( see fig . [ fig : sketch_step ] ) . in fig . [ fig : a](a ) we show @xmath171 as a function of @xmath55 as obtained numerically within full mft via least square fits to the data for @xmath116 . as expected , for @xmath52 , i.e. , @xmath172 , @xmath171 vanishes for the case of chemically homogeneous boundary conditions as considered here . ( for chemically inhomogeneous boundary conditions in lateral direction a line contribution arises due to the presence of a _ chemical _ step even for a planar substrate @xcite . ) for @xmath85 , i.e. , deep crenels , @xmath171 attains negative values and depends on @xmath79 only . @xmath86 corresponds to the contribution to the critical casimir force of two infinitely extended right - angled corners opposite to a planar substrate ( see fig . [ fig : sketch_step](b ) ) . we find that this plateau value is reached for @xmath164 . in this limit the lower parts of the crenels including the right - angled wedges at the bottom do not contribute to the critical casimir force . figure [ fig : a](b ) shows the amplitude @xmath171 of the step contribution as a function of @xmath79 for deep crenels with @xmath167 , normalized by the absolute value @xmath146 of the critical casimir amplitude . for the chosen value @xmath167 we find that @xmath173 holds [ see eq . ] . we find that the functional shapes of this amplitude of the step contribution and of @xmath174 are similar but the minima of the two scaling functions are displaced such that the minimum of @xmath76 as a function of @xmath79 is located closer to the critical point at @xmath116 . for comparison we note that the functional shape of the amplitude of the step contribution is very close to the numerically obtained mft scaling function of the attractive critical casimir force acting between a thin cylinder and a planar substrate , where the symmetry axis of the cylinder is parallel to the substrate @xcite . indeed , the lower white contour line in fig . [ fig : op](a ) indicates that one may approximate each cross - section of the two right - angled edges of the merlons by a small inscribed quarter circle , keeping in mind that in the limit @xmath175 the critical adsorption profile near a thin cylinder is independent of the radius @xmath136 of the cylinder @xcite . within mean field theory ( mft ) we have calculated the critical casimir force between a planar and a crenellated surface , i.e. , a periodic structure of geometric steps forming crenels and merlons ( fig . [ fig : sketch ] ) . to this end we have numerically calculated the critical order parameter profiles of a fluid confined by such a structure . we have focused on the experimentally relevant case of binary liquid mixtures or simple liquids , which belong to the ising bulk universality class , and on @xmath31 boundary conditions ( bcs ) which correspond to the generic case of strong critical adsorption . using the stress tensor we have calculated the universal scaling function of the singular contribution to the critical casimir force , acting on the confining walls along the normal direction upon approaching the critical point . we have obtained the following main results : 1 . according to finite size scaling , the universal scaling function of the critical casimir force depends on only a few scaling variables which describe the geometry under consideration [ eqs . , , and [ fig : sketch ] ] . in the limits , that the width @xmath8 or the depth @xmath9 of the crenels are small compared to the film width or the period of the geometric structure , the force reduces to the one acting between two parallel flat substrates at a distance @xmath17 [ eqs . ] . on the other hand , for fixed reduced crenel depth @xmath55 and for reduced crenel width @xmath176 , the expression for the critical casimir force approaches the one obtained via the proximity force approximation ( pfa ) [ eq . ] as @xmath177 , which can be interpreted as a step contribution [ see eq . and fig . [ fig : sketch_step ] ] . 2 . in sec . [ sec : profiles ] we have shown generic examples of order parameter profiles ( i.e. , the universal contribution to the profiles of the deviations of the local concentration or density from its critical value in the bulk ) as obtained within mft at the critical point . whereas for crenel sizes comparable with the film thickness the order parameter profile follows the geometrical structure [ figs . [ fig : op](a ) and ( b ) ] , for shallow crenels and closely spaced merlons the order parameter profile resembles the one in between two planar walls [ fig . [ fig : op](c ) ] . in this latter case strong critical adsorption effectively suppresses the influence of the merlons . these properties of the order parameter profiles agree with our results for the universal scaling function @xmath83 [ eq . ] of the critical casimir force presented in section [ sec : force ] : for deep crenels and widely spaced merlons the full mft scaling function of the critical casimir force as a function of the scaling variable @xmath79 does neither agree with the planar wall limit ( pw ) nor with the pfa [ fig . [ fig : force](a ) ] . on the other hand , for shallow crenels and widely spaced merlons the pfa describes the actual full mft behavior better [ fig . [ fig : force](b ) ] , whereas for shallow crenels and closely spaced merlons the pw limit is approached [ fig . [ fig : force](c ) ] . thus , depending on the values of the scaling variables @xmath82 and @xmath117 , the critical casimir force interpolates between its limiting behaviors described by the pfa , which always underestimates the strength of the force , and the pw limit , which overestimates the critical casimir force this can also be seen in fig . [ fig : compare](b ) which shows the ratio of the universal scaling function of the critical casimir force and its corresponding pfa value as a function of @xmath55 . 4 . upon increasing @xmath68 the scaling function @xmath67 as obtained within pfa is approached as @xmath177 [ fig . [ fig : compare]a ) ] , which can be interpreted as a step contribution with @xmath77 de facto independent of @xmath64 [ eq . ] . thus , for widely spaced merlons the critical casimir force can be described as the linear superposition of the planar wall contributions at distances @xmath17 and @xmath60 , respectively , plus additional step contributions stemming from the geometrical steps . for large values of @xmath55 this step contribution approaches a specific negative value , independent of the crenel depth @xmath9 [ fig . [ fig : a](a ) ] . to conclude , our numerical calculations within mft extend previous investigations of the critical casimir force caused by the confinement due to structured substrates . in particular , we have analyzed in detail the crossover of the universal scaling function of the critical casimir force from an effective planar wall limit for small roughness to the pfa limit . this may not only be useful for the analysis of critical casimir forces in such experimentally relevant , designed geometries , but could also help in understanding unavoidable roughness corrections in the case of planar geometries . moreover , our results show that the first - order correction to the pfa can be expressed in terms of a contribution stemming from the individual geometrical steps . this is analogous to similar situations involving chemically structured surfaces @xcite , which are discussed in ref . @xcite . for comparison we note that such step contributions arising from crenellated surfaces are also of importance for non - critical fluids consisting of rod - like particles close to the isotropic to nematic phase transition of the bulk fluid @xcite . the authors thank nikolas brunner for very helpful contributions . 77 ifxundefined [ 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 * * , ( ) link:\doibase 10.1103/physrevlett.83.1187 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevlett.94.135702 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.97.075301 [ * * , ( ) ] link:\doibase 10.1016/j.physa.2007.07.072 [ * * , ( ) ] \doibase http://dx.doi.org/10.1038/nature06443 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1209/0295 - 5075/88/26001 [ * * , ( ) ] in @noop _ _ , vol . , ( , ) p. in @noop _ _ , vol . , ( , ) p. @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1209/0295 - 5075/88/40004 [ * * , ( ) ] link:\doibase 10.1063/1.3464770 [ * * , ( ) ] link:\doibase 10.1039/c0sm00635a [ * * , ( ) ] http://stacks.iop.org/1742-5468/2010/i=11/a=p11003 [ ( ) ] @noop * * , ( ) @noop ( ) link:\doibase 10.1039/c3sm52858h [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.208301 [ * * , ( ) ] http://www.informaworld.com/10.1080/00268976.2011.553639 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * ( ) link:\doibase 10.1103/physrevlett.101.160403 [ * * , ( ) ] http://dx.doi.org/10.1038/nphoton.2011.39 [ * * , ( ) ] link:\doibase 10.1103/physreva.86.062502 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.88.101801 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * ( ) @noop * * ( ) link:\doibase 10.1103/physrevb.81.115417 [ * * , ( ) ] @noop * * , ( ) ; _ ibid . _ * 107 * , 019901(e ) ( 2011 ) link:\doibase 10.1103/physrevlett.105.250402 [ * * , ( ) ] link:\doibase 10.1103/physreva.87.052514 [ * * , ( ) ] http://dx.doi.org/10.1038/ncomms3515 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1063/1.2977999 [ * * , ( ) ] , preprint arxiv:1406.0962 ( 2014 ) in @noop _ _ , vol . , ( , ) p. in @noop _ _ , ( , ) p. link:\doibase 10.1016/s0370 - 1573(02)00219 - 3 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevb.49.8842 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.125702 [ * * , ( ) ] link:\doibase 10.1209/0295 - 5075/80/60009 [ * * , ( ) ] link:\doibase 10.1103/physreve.79.041142 [ * * , ( ) ] ; _ ibid . _ * 80 * , 039902(e ) ( 2009 ) link:\doibase 10.1103/physrevb.82.104425 [ * * , ( ) ] link:\doibase 10.1103/physrevb.82.174434 [ * * , ( ) ] link:\doibase 10.1103/physrevb.82.174433 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevb.47.5841 [ * * , ( ) ] ; _ ibid . _ 6740 ( 1993 ) @noop _ _ , ed . ( , ) link:\doibase 10.1063/1.3259188 [ * * , ( ) ] link:\doibase 10.1103/physreve.59.5081 [ * * , ( ) ] @noop * * , ( )

perturbative methods are indispensable in order to establish consistent theories of particle interactions , and to predict quantitatively their experimental manifestations . the anticipation of new phenomena in modern experiments and theoretical extensions of the standard model , requires cross - sections for complicated processes . this has driven a remarkable progress in the development of new computational methods . at the one - loop level , the calculation of cross - sections with five external legs is gradually becoming a routine activity ( e.g. @xcite ) . at two - loops , there have been recent successful computations of amplitudes with four external legs and up to three kinematic scales ( e.g. @xcite ) . at three - loops and beyond , amplitudes with up to one parametric variable have also been computed ( e.g. @xcite ) . many new processes with higher final - state multiplicity , number of loops , and kinematic scales , have been identified to be important at the tev energy frontier . the aim of our paper is to provide a new method which can be used to compute loop amplitudes for such , more complicated , processes . loop integrations are cumbersome due to the presence of infrared singularities . loop integrals with many kinematic scales have , in addition , a complicated analytic structure with respect to their kinematic parameters . the tensor structure in gauge theories is also an issue , since it proliferates the number of terms . a general method for computing arbitrary loop integrals should extract their infrared and ultraviolet singularities , and treat simply kinematic discontinuities and threshold singularities . for practical applications , it should also be able to handle tensor integrals efficiently . there is no method which addresses satisfactorily all these issues ; known techniques can compute a limited number of amplitudes where some simplifications occur in special cases . following a traditional approach to calculate loop amplitudes , one reduces the number of terms to a few master integrals and computes the latter analytically . one - loop integrals can be reduced using the classical method of passarino and veltman @xcite . for generic multi - loop computations one can derive reduction identities from integration by parts @xcite and the invariance of scalar integrals under lorentz transformations @xcite . there is a large variety of approaches for evaluating the master integrals analytically . for example , one can compute integrals with a simple singularity structure and a small number of kinematic scales by integrating directly their feynman parameter representation . for more complicated cases one can use advanced techniques , such as the method of differential equations @xcite . this approach can fail , however , if we apply it to complicated processes . the reduction algebra is hard , and the expressions of amplitudes in terms of master integrals may have spurious singularities which hamper their numerical evaluation . the extraction of @xmath1 poles in the master integrals , the evaluation of the coefficients of the @xmath1-expansion in terms of known analytic functions , and the analytic continuation of the latter to physically interesting kinematic regions are also involved . it is , thus , very well motivated to improve or replace the `` traditional scheme '' and to develop new automated methods . one - loop amplitudes can be entirely determined in four dimensions in terms of basic functions such as logarithms and polylogarithms that appear in the one - loop scalar box . new methods introduce sophisticated algorithms for the reduction to the basic functions ; by either numerical or analytical techniques , they control the appearance of spurious singular terms and minimize the size of the intermediate expressions @xcite . recently , cross - sections for @xmath2 processes at nlo were computed @xcite using such techniques . a method , which is inspired by techniques for real radiation at nlo , renders one - loop graphs numerically integrable @xcite with the introduction of universal subtraction terms . a different approach uses unitarity , dualities , and analyticity properties for the determination of one - loop amplitudes @xcite . beyond one - loop , there has been a significant progress in the automation of the reduction methods to master integrals @xcite . the infrared structure of multi - loop integrals is substantially more complex than at one - loop ; the development of methods for their numerical evaluation is more difficult . nevertheless , there is significant progress in this direction @xcite . a powerful numerical method for multi - loop calculations is the method of sector decomposition @xcite ; it simplifies recursively the singularities of feynman parameterizations and allows a straightforward expansion in @xmath1 . the method of sector decomposition has been very succesfully employed in the computation of several multi - leg integrals at one , two and three loops @xcite . this method has been introduced , recently , for the purely numerical evaluation of multi - loop amplitudes @xcite . however , it is perplexing how to apply it for loop - integrals in non - euclidean regions . in 1999 , smirnov @xcite and , soon later , tausk @xcite , introduced a new method for the evaluation of loop integrals . in their pioneering papers , smirnov and tausk @xcite computed analytically the first infrared divergent double box integrals . in @xcite new two - loop integrals for @xmath3 massless processes were computed . in @xcite the method was applied to double - box integrals with one additional mass - scale . the method was spectacularly applied in the computation of three - loop amplitudes @xcite in @xmath4 supersymmetric yang - mills theory , shedding light to novel cross - order perturbative relations of the theory @xcite . the smirnov - tausk method is based on a few simple ideas . starting from the feynman parameterization of a loop integral we can derive a new representation in terms of a multiple complex contour integral . such , mellin - barnes ( mb ) , parameterizations have enabled compicated loop calculations by using powerful methods for complex integration @xcite . smirnov and tausk exploited a novel property of these representations . infrared divergences localize on simple poles inside the complex integration volume . we can isolate the divergent pieces of the integral at @xmath5 , by using the cauchy theorem . after subtracting the divergent residues , we can perform a taylor expansion in @xmath1 , and sum up the remaining infinite series of residues . finally , we can work to derive analytic expressions for the infinite sums in the coefficients of the expansion in terms of logarithms , generalized polylogarithms @xcite , and more complicated functions . the smirnov - tausk method is very powerful ; however , it is laborious and intricate . the isolation of the divergent residues in multiple mellin - barnes integrals is convoluted . in addition , it is difficult to identify infinite sums in terms of polylogarithm functions with known analyticity properties . as a consequence , the analytic continuation in physical kinematic regions is also involved . due to these complications , the method has been applied to a few master integrals with a small number of kinematic scales . in this paper , we generalize the method to a broader spectrum of applications . as a first task , we automate the procedure for the isolation of the divergent residues at @xmath0 . smirnov @xcite and tausk @xcite use different techniques for finding these residues . the approach of smirnov is very intuitive , but daedal . we have found that the technique described by tausk in ref . @xcite resembles closely to a programmable algorithm . we have used it as a guide and we have written computer programs which subtract the @xmath6 poles in arbitrary mellin - barnes integrals . in our method , we avoid entirely the painstaking tasks of finding analytic expressions for infinite sums in terms of polylogarithms , and performing the analytic continuation in the arguments of polylogarithms . mellin - barnes representations are valid in kinematic regions where loop integrals may be complex - valued . we have found that , in a broad spectrum of applications , it is simple to calculate the representations numerically . the only analytic continuation that is ever required is that of logarithms with a single kinematic scale as an argument . an important goal of our method is to calculate loop amplitudes in realistic gauge theories . we have found that tensor integrals and , furthermore , diagrams which belong to the same topology can be calculated collectively . as we will show , the integrand of a representation for a scalar integral will be a product of gamma functions and powers of kinematic invariants , while that of a generic tensor will be the same integrand as in the scalar integral multiplied by a polynomial in the integration variables . the evaluation of polynomials is fast in a numerical program ; the computational cost for evaluating tensor integrals or loop diagrams is not significantly larger than evaluating scalar integrals . the only practical issue that we need to address , is the book - keeping of the various terms that contribute to the polynomial ; we present an efficient solution of this problem here . in this paper , we apply our method to a number of examples . we , first , test our method in scalar and tensor integrals of the one - loop massless hexagon topology . the purpose of this computation is to introduce our method for tensor integrals and to demonstrate that we can tackle problems which are relevant in computations of physical amplitudes . we present here results for tensors through rank six , in both the euclidean and the physical region for @xmath7 processes . the numerical programs that we have constructed are suitable for the evaluation of the qcd amplitudes in four - jet production at hadron colliders . in the second set of examples , we compute scalar two and three - loop integrals which are known analytically in all kinematic regions : the massless planar @xcite and cross @xcite double - box , the massless double - box with one off - shell leg @xcite , and the massless planar triple - box @xcite . they serve to cross - check our algorithms and to demonstrate that we can easily reproduce state - of - the - art computations . our numerical results are in excellent agreement with the analytic expressions . we also present a number of new results that would require significant efforts for their computation with traditional approaches . we present , for the first time , double - box integrals with up to four kinematic parameters , and triple - box integrals with up to three kinematic parameters computed in all physical regions . in section [ sec : mbrepresentations ] , we explain our technique for deriving mellin - barnes representations for loop integrals . in section [ sec : analytic ] , we describe our routines for performing an @xmath1-expansion of mellin - barnes representations for scalar integrals . we extend these results to loop integrals with tensor numerators in section [ sec : tensors ] and , in section [ sec : numerical ] , we present our methods for the numerical integration . section [ sec : results ] is devoted to present our results for several integrals at the one , two and three loop level . finally we present our conclusions . we start with a brief discussion on the derivation of mellin - barnes representations for loop integrals from their feynman parameterization . the construction of parameterizations is not unique , and various representations of the same integral may have quite different features . for example , they could have a different integral dimensionality , or they could be better or worse suited for numerical evaluation . a valid parameterization , however , can always be found . a pedagogical introduction to the topic of mellin - barnes representations is presented in ref . @xcite . as a concrete example we consider the one - loop box integral with two adjacent external legs off - shell and massless propagators , @xmath8 with @xmath9 and @xmath10 , @xmath11 , @xmath12 , @xmath13 , @xmath14 . the powers of the propagators @xmath15 and the dimension @xmath16 are kept arbitrary and we will derive a mellin - barnes parameterization of the box - integral in this general case . in this section , we will seek values for @xmath17 where the representation is well defined and the integral is finite . in the next section , we will explain the technique for the analytic continuation to the values of the parameters , such as @xmath18 , where the integral develops divergences . our starting point is the feynman parameterization of the box integral : @xmath19 with @xmath20 . the main tool for obtaining the mellin - barnes representation of the above integral is the formula : @xmath21 where the following conditions are satisfied : 1 . @xmath22 are complex numbers with @xmath23 , 2 . the contour of integration is a straight line parallel to the imaginary axis separating the poles of @xmath24 and @xmath25 , i.e. @xmath26 . we can use the identity of eq . [ eq : mbrep ] to simplify the entangled denominator of eq . [ eq : box_2m_fp ] , with the cost of introducing new integrations in the complex plane . before we do so , we would like to point out an attractive feature of eq . [ eq : mbrep ] . the above two conditions guarantee that the integrand in the right hand side of eq . [ eq : mbrep ] vanishes at infinity . we can then close the contour of integration at infinity and calculate the integral using the cauchy theorem . if we close the contour to the side of the positive real - axis we obtain @xmath27 this is the taylor expansion of the left hand side of eq . [ eq : mbrep ] in the region @xmath28 . equivalently , by closing the contour of integration to the side of the negative real axis , we obtain a taylor expansion in the complementary region @xmath29 . this is a particularly useful property : representations of feynman integrals which are derived by using the mellin - barnes decomposition are valid in all kinematic regions . in addition , @xmath22 can be complex ; therefore , we can account for the infinitesimal imaginary part @xmath30 that is assigned to invariant masses and kinematic parameters . 3 may be recursively applied to denominators with more than two terms , yielding : @xmath31 we are now ready to apply the decomposition of eq . [ eq : mbrep_multi ] to the denominator of eq . [ eq : box_2m_fp ] , introducing three contour integrations : @xmath32 where the contours must satisfy condition 2 above , i.e. they must be chosen in such a way that the arguments of the gamma functions are all positive . it is now straightforward to integrate out the feynman parameters , using @xmath33 to exchange the order of integrations , we must assume that the exponents of the feynman parameters satisfy @xmath34 with these conditions the gamma functions on the right hand side of eq.[eq : int_feynman ] have arguments with positive real part as well . the final result is : @xmath35 where @xmath36 and we have adopted the shorthand notation @xmath37 in general , mellin - barnes representations for a loop - integral are well defined for number of dimensions and powers of propagators which render the real parts of the arguments in all gamma functions positive . for example , it is impossible to satisfy positivity in all gamma function arguments for @xmath38 . this is in accordance with the expectation that the scalar box integral is infrared divergent in four dimensions . but , for example , the representation of eq . [ eq : box_2m_mb ] is well defined if we choose the contour @xmath39 and the parameters @xmath40 . in the next section we will detail the analytic continuation from values of parameters for which the mellin - barnes representation is well - defined to values for which the integral develops divergences . that is , a procedure which makes the divergences appear explicitly in the form of poles in the parameter space . from eq . [ eq : box_2m_mb ] we observe that it is simple to implement the analytic continuation of kinematic parameters in a numerical evaluation , since they always appear in a factorized form , e.g. @xmath41 we need the analytic continuation of simple logarithms ; these are evaluated trivially in all kinematic regions : @xmath42 it is clear from the above example that we can derive a mellin - barnes representation for any integral starting from its feynman parameterization . the form of the parameterization depends on the choice of feynman parameters , the order in which the @xmath43 terms appear in the right hand side of eq . [ eq : mbrep_multi ] , and the implementation of the delta function constraint for the feynman parameters . this arbitrariness is more pronounced beyond one - loop . in section [ sec : tensors ] we will derive representations for tensor one - loop integrals which simplify their numerical evaluation . we aim to find similarly simple representations for multi - loop tensor integrals as well . for this purpose , it is convenient to use the one - loop representations as building blocks , by employing a re - insertion technique @xcite . we explain this technique in our second example . we now derive a mellin - barnes representation for the massless planar double - box integral with two adjacent legs off - shell . the integral is defined : @xmath44 with @xmath45 and @xmath10 , @xmath11 , @xmath12 , @xmath13 , @xmath14 . for simplicity , we have set the powers of propagators to one . we perform , first , the @xmath46-loop integral and derive the representation for this integral reading it off from the one - loop result of eq . [ eq : box_2m_mb ] , @xmath47 the off - shell legs of the @xmath46-loop box correspond to the @xmath48 propagators while the @xmath49 invariant mass gives rise to a new propagator for the second integration : @xmath50 the propagators @xmath51 form a new one - loop box with two adjacent legs off - shell , and we can read off , once again , the representation for this second integration using eq . [ eq : box_2m_mb ] . the final result for the two - loop box with two adjacent legs off - shell is : @xmath52 we have produced a representation for a two - loop integral by embedding representations for one - loop integrals into other representations . it is obvious that we can use the re - insertion method for writing representations for arbitrary multi - loop integrals . representations for integrals with given kinematic scales can be used to derive , easily , results for simpler integrals where some of the scales are taken to zero . let us , for illustration , derive the representation for the double - box with one - leg off - shell and the on - shell double - box . we must first take the limit @xmath53 in eq . [ eq : dbox_2m_mb ] . the term @xmath54 is vanishing in this limit unless @xmath55 at the same time . if we use the cauchy theorem , we find that the wanted limit is given by taking the residue of the integrand at @xmath56 . therefore , the double - box with one - leg off - shell is : @xmath57 similarly , the on - shell double - box is given by : @xmath58 note that , in order to have a finite limit for a vanishing kinematic scale , the mellin - barnes representation should have terms in the integrand which behave as @xmath59 . these emerge naturally in non - trivial representations due to eq . [ eq : mbrep ] . in the previous section we have noted that the mellin - barnes representation of a loop integral is valid if appropriate poles of the gamma functions lay separated on the right and left of the integration contours . this condition guarantees the equivalence between the mellin - barnes integral and the original denominator in the loop integral . it often occurs that this condition can not be satisfied for values of the dimension parameter and the powers of the propagators which are relevant for a realistic application . let us consider , as an example , the one - loop box representation of eq . [ eq : box_2m_mb ] in the case of powers of propagators set to unity and the dimension in @xmath16 , @xmath60 this representation can only be valid for values of @xmath1 different from zero . for example , if we choose the contour @xmath61 we find that @xmath1 should be in the interval @xmath62 . we can use the cauchy theorem to obtain a representation in terms of a sum of contour integrals , which is valid at @xmath63 and the @xmath6 poles appear explicitly . the key point is that if the value of @xmath1 is chosen outside the allowed region , for example @xmath63 , some of the poles of the gamma functions will be on the wrong side of the contours ( see fig . [ fig : pole_pos ] ) . to recover the original result , we must compare the position of the poles relative to the contours for values of @xmath1 inside the allowed region and the point outside . depending whether the poles crossed the contours from left or right , we should correct the representation by adding or subtracting the residue of the integrand on those poles . this simple observation can be easily cast into general purpose algorithms which extract the poles in @xmath1 of arbitrary mellin - barnes representations . , and some of the poles cross to the other side of the contour . to recover a valid representation for the loop - integral , these poles sould be isolated using the cauchy theorem . , height=151 ] let us choose a value of @xmath1 inside the allowed region and observe the position of the poles with respect to the contours for some @xmath1-dependent gamma functions in our example . the poles of @xmath64 and @xmath65 , at @xmath66 and @xmath67 , are situated to the right of the @xmath68 and @xmath69 contours , respectively , for all non - negative integers @xmath70 . let us now take @xmath0 . we observe that the first poles , for @xmath71 , moved to new positions , @xmath72 and @xmath73 , which are on the left side of the contours . the poles for @xmath74 remain on the right side . therefore , the representation is not valid at @xmath75 , since the contours separate poles which originate from the same gamma function . to obtain a valid expression for the integral at @xmath76 , we need to use the cauchy theorem to isolate the crossing poles ( @xmath71 ) . smirnov , in ref . @xcite , provides a number of pedagogical examples where this is done . the approach of smirnov is to deform the multiple contours , using the cauchy theorem , so that the `` offending '' poles never cross the contour of integration . tausk , in ref . @xcite , describes in detail this task as well , using as an explicit example the mellin - barnes representation for the two - loop cross box integral . the technique of tausk is different from the one of smirnov ; the main idea is to account for the poles that end up on the `` wrong '' side of the contour in a progressive way , as they cross the contours when changing continuously the value of @xmath1 . in this paper , we have followed the approach of tausk . we would like to recommend that the interested reader studies the pedagogical example of ref . @xcite . here we discuss our implementation of the algorithm which is depicted in the flow diagram of figure [ fig : flow1 ] . consider the mellin - barnes representation @xmath77 of a loop integral . the integrand of @xmath77 depends initially on gamma functions . the application of the analytic continuation algorithm requires the iterative evaluation of residues . these may also give rise to psi functions , which have poles in the same positions the gamma functions do , and are treated identically for the purposes of analytic continuation . the contours @xmath78 of the representation are straight lines parallel to the imaginary axis , and are chosen so that the real parts of the arguments of all gamma functions are positive . the same condition determines an allowed region for @xmath1 , which we will assume that includes the point @xmath79 . for the sake of clarity , we will consider the case @xmath80 , but the method proceeds analogously when @xmath81 . the first step is to find the maximum value @xmath82 , with @xmath83 , for which the representation holds . at this value , one or some of the positions of the poles run into a contour . if the set of these poles , @xmath84 , is empty , i.e. no poles cross the contours when moving @xmath1 from @xmath85 to 0 , then the integral can be safely expanded around @xmath63 . so , let us focus on the more interesting case in which @xmath84 is non empty . here we have to distinguish between @xmath86 and @xmath87 . if @xmath86 , at least one pole lies on the contours for the terminating value of @xmath1 . then , the representation has an unregulated singularity , and it is not feasible to perform an @xmath1 expansion and evaluate the contour integrals numerically . to remedy the situation we must change the contour of integration ; we use the cauchy theorem and displace one or more of the contours parallel to the imaginary axis . then , we attempt afresh the analytic continuation ; the poles should now hit the contour at @xmath87 . if this does not happen , due to an unfortunate choice of the contour we displaced , we simply iterate the procedure until we succeed . when @xmath88 , we have to determine the residues in the poles that cross the contour . if there is only one pole in @xmath84 the situation is simple . let us look back at the specific example of the one - loop box . by applying the previous steps of the algorithm , we find that the first pole to cross a contour is at @xmath89 . we then choose one of the multiple integrals , e.g. the @xmath69 integral , and subtract the residue of the integrand on this pole , at @xmath90 , to the representation : @xmath91 the residue term in the above expression is a new multiple mellin - barnes integral with one integration variable less than in @xmath77 . in our example , the pole crosses the @xmath69 contour from the right side and we subtract its residue from the original representation . if a pole crosses the contour from the left , we should add its residue . at this point , we have achieved the analytic continuation in @xmath1 from @xmath85 to @xmath92 . we now repeat the algorithm setting as initial value @xmath93 , and perform the analytic continuation to the next value of @xmath1 where a new crossing occurs . we proceed iteratively , until we find all poles which cross the contours of integration in the terms of eq . [ eq : ancont ] and its descendants . the algorithm is more complicated if many poles cross simultaneously the contour at @xmath82 . in our one - loop box example , if we had chosen a symmetric contour @xmath94 then we would have two poles crossing at the same time : @xmath95 . one way to deal with this situation is to displace the contours in such a way that the degeneracy is removed , i.e. introduce asymmetric contours from the beginning . another alternative is to take combined residues in all the poles . to correctly account for all the terms in this case , we can imagine that the contours are only slightly displaced from their current position . then the poles will hit them one at the time and we can take residues successively . finally , we can restore the original position of the contours without affecting the position of the poles . there is , however , a caveat in taking multiple residues . let us consider the case of the one - loop box , where the poles @xmath96 and @xmath97 hit the contour at the same time . we take residues in sequence , considering first @xmath98 . we make the `` unfortunate '' choice to take the residue on the @xmath99 integration variable , and not on @xmath68 . the analytic continuation yields : @xmath100 then we consider the crossing of the @xmath101 pole on each of the two terms on the right hand side of the above expression . we find no problems in taking the residue of the first term . however , the position of the @xmath101 pole for the second term is now written : @xmath102 the pole @xmath101 is now independent of @xmath1 and gets nailed to the contour . this situation is similar to the one we encountered for @xmath86 , and we can deal with it in exactly the same way by displacing the contours . we apply the algorithm recursively , to all the terms which are generated by adding the residues of crossing poles to the original parameterization . at the end of this procedure , we obtain a sum of contour integrals plus some terms that do not contain any remaining integral due to taking residues on all integration variables . the method assures that the sum of all these contributions equals the original loop integral when @xmath103 . so we can expand in a power series of @xmath1 . in this section , we have presented the algorithm assuming that only the dimension parameter @xmath1 is required to regulate the mellin - barnes parameterization . however , the method proceeds identically for integrals where multiple analytic regulators are required . this could be the case in non - covariant gauges or in integrals with irreducible numerators @xcite or additional positive powers of propagators @xcite . the actual implementation of the algorithm in a set of routines in both maple and mathematica is very efficient , and allows to perform the analytic continuation in the regulator parameters in a few seconds , for all the mellin - barnes representations that we have studied in this paper . mellin - barnes representations , with the exception of the massless two - loop diagonal box topology @xcite , have been traditionally employed for the evaluation of master integrals . in this section , we introduce an efficient generalization of the method to tensor integrals ; our method does not require any reduction techniques to master integrals . common reduction methods produce extremely large algebraic expressions which may also have spurious singular denominators . here we derive efficient representations for one - loop tensor integrals , which are amenable to numerical integration . we can derive analogous representations for multi - loop integrals by using the re - insertion method of section [ sec : mbrepresentations ] . we consider a generic one - loop tensor integral of rank @xmath104 with @xmath70 external legs : @xmath105 \left [ ( k+q_2)^2-m_2 ^ 2 \right ] \cdots \left [ ( k+q_n)^2-m_n^2 \right ] } \,\ ] ] where , @xmath106 and @xmath107 are external incoming momenta . we first introduce feynman parameters for the denominator . we obtain , @xmath108^n}\,,\ ] ] where @xmath109 and @xmath110\ ] ] now we perform the usual shift in the loop momentum , @xmath111 , and we obtain , @xmath112 } } { ( k^2-\delta)^n}\ , , \ ] ] where , we denote @xmath113 } = \sum_{\{j_1,\ldots , j_m\ } } k^{\mu_{j_1}}\cdots k^{\mu_{j_r } } p^{\mu_{j_{r+1}}}\cdots p^{\mu_{j_m } } , \\ & & \hspace{2 cm } \{j_1,\ldots , j_m\ } \in \mbox{permutations}(1,\ldots , m ) \nonumber . \end{aligned}\ ] ] we find the standard feynman representation for the generic tensor one - loop @xmath70-point function , by integrating the loop momentum : @xmath114 } \delta^{\frac{r}{2 } } , \ , \end{aligned}\ ] ] where @xmath115 for @xmath116 odd , and @xmath117}$ ] for @xmath116 even . we observe that the sum , in the second line of the equation , is a polynomial in the feynman parameters , with tensor coefficients : @xmath118 } \delta^{\frac{r}{2 } } = \sum_{\{\nu_i > 0 \ } } c(\{\nu_i\ } ; \{p_i^{\mu_j}\ } ) x_1^{\nu_1 - 1}\cdots x_n^{\nu_n-1}\ ] ] for the scalar integral with unit powers of propagators , this sum is @xmath119 . we will use eq . [ eq : tensor - feynf ] to derive mellin - barnes representations for the tensor loop - integral . we would like to exploit the fact that the feynman representation of the scalar and tensor integrals have the same denominator : @xmath120 in section [ sec : mbrepresentations ] , we have seen that we can obtain a mellin - barnes representation for a loop integral by decomposing the denominator of the feynman parameterization with eq . [ eq : mbrep_multi ] . we should anticipate that the mellin - barnes representations of tensors and scalars are , therefore , very similar . we can verify this intuition , if we consider one generic term in the polynomial of eq . [ eq : fp_polynomial ] , and follow the steps of section [ sec : mbrepresentations ] . decomposing the denominator with eq . [ eq : mbrep_multi ] , we obtain @xmath121 where @xmath122 is the number of kinematic scales @xmath123 in the integral . the factors @xmath124 are universal for all terms in eq . [ eq : fp_polynomial ] , and originate from the dependence of @xmath125 on the feynman parameters . we now proceed to integrate out the feynman parameters using eq . [ eq : int_feynman ] . this integration yields gamma functions @xmath126 rewriting all gamma functions as above in eq . [ eq : fp_polynomial ] , we obtain , for the tensor integral , the gamma functions of the scalar integral multiplied by simple factors with mellin - barnes integration variables . the result for the one - loop tensor integrals depends on the topology and the rank of the tensors in a rather complicated manner . however , we find that the representation is of the following form @xmath127 the first line of the above representation contains all the terms which already appear in the analogous representation of the scalar integral with unit powers of propagators . the function @xmath128 is a _ polynomial in the mellin - barnes variables _ , with tensor coefficients in terms of the external momenta . this is very important for our strategy to evaluate tensor integrals and feynman diagrams . the polynomial in the numerator is smooth , and it does not affect the analytic continuation in @xmath1 . we can therefore perform the continuation collectively for all tensor integrals and diagrams of the same topology . we organize the evaluation of tensor integrals and diagrams of a topology as follows . we first find the mellin - barnes parameterization of the scalar integral of the topology , and multiply the integrand with a template function @xmath129 which we assume depends on all mellin - barnes variables . then , we apply the analytic continuation algorithm of section [ sec : analytic ] keeping @xmath130 general . we expand in @xmath1 , and create numerical programs for the evaluation of the expansion for any smooth function @xmath130 . this part of the evaluation needs to be performed only once for a given topology . then we must identify the polynomial for the evaluation of the specific tensors or diagrams that we are interested in . we have written form @xcite and mathematica programs which perform the steps that we described in this section , and derive the explicit functional form of the polynomials in eq . [ eq : tensor_mbrep ] . we then write numerical routines for their evaluation , and link them to the general routines for the @xmath1 expansion of the integrals of the topology . this approach is efficient for the application of our technique to tensor integrals . it turns out , however , that the expressions for the polynomials are quite long for high rank tensors . we have observed that we can reduce the size of the expressions substantially with a simple modification . the terms @xmath131 in eq . [ eq : tensor - feynf ] are lengthy , and when we integrate out the feynman parameters they give rise to large expressions in @xmath128 of eq . [ eq : tensor_mbrep ] . we rewrite eq . [ eq : tensor - feynf ] in a different way , @xmath132 } \,\end{aligned}\ ] ] then , we introduce a mellin - barnes decomposition for each denominator : @xmath133 and integrate out the feynman parameters . at the end of this procedure , we obtain a sum of @xmath134 + 1 $ ] mellin - barnes integrals for a tensor of rank @xmath104 . all integrals correspond to the mellin - barnes representation of the scalar integral with shifted dimension @xmath135 . each of them has a different polynomial @xmath136 in the numerator . however , these are substantially shorter than the polynomial @xmath128 in eq . [ eq : tensor_mbrep ] . it is worth to note that , with our method , we never introduce spurious singularities . the evaluation of the higher rank tensors and diagrams in gauge theories is very efficient , since we never require the manipulation of lengthy expressions . the polynomials @xmath136 can be large , however , they require a minimum amount of handling before generating numerical codes for their evaluation . here we have presented the derivation of efficient representations for one - loop integrals . the method may be applied to multi - loop integrals using the one - loop tensor integral representations as building blocks . the analogous functions of @xmath136 in multi - loop tensors contain gamma functions in the denominator . however , they are smooth and maintain the properties which are required for a collective evaluation of all tensors of a topology . explicit applications with multi - loop tensor integrals will be examined in future work . after analytic continuation , the contour integrals can be safely expanded in power series of @xmath1 . the coefficients of the expansion are also contour integrals which , in general , contain in their integrand gamma functions , gamma function derivatives , simple logarithms , and powers of the kinematic parameters . in earlier works , these integrals were computed analytically . the standard approach is to evaluate them by means of the residue theorem , i.e. summing an infinite number of residues and expressing the power series as logarithms , polylogarithms and harmonic polylogarithms . this method is cumbersome for the evaluation of integrals with more than one mellin - barnes variable . in all cases presented in the literature , integrals with more than one dimension are reduced to one - dimensional integrals with a variety of clever , however not general , tricks . typical procedures include direct integration in simple situations , the application of barnes lemmas @xcite , and the expansion in laurent series around an integer value of one of the variables . in a few cases , a careful inspection of the multi - dimensional integrals reveals exact cancellations between different components which yield significant simplifications . many integrals which are relevant for physical applications , such as the massless double box with two external legs off - shell and one - loop tensor integrals with six external legs , after the analytic continuation in @xmath1 , are expressed in terms of hundreds of mellin - barnes components ; some of them have a very high dimensionality . the analytic methods are not suitable for such problems . in our method , we choose to evaluate the mb integrals numerically by direct integration over the contours . the procedure we follow is , conceptually , straightforward . in practice , the implementation of the numerical evaluation requires a significant work in automating the book - keeping of the various terms . tensor integrals require significantly more complex book - keeping due to large expressions that multiply the gamma functions in the integrand of their mellin - barnes representation . our computer programs are organized in the following way . we first use maple , form @xcite , and mathematica routines to derive mellin - barnes representations for the integrals that we need to compute . we then perform the analytic continuation of section [ sec : analytic ] , and we expand the integrands in powers of @xmath1 using maple and mathematica . as the next step , we translate the integrands into fortran routines which evaluate them as complex quantities . there are many options to calculate numerically the sum of the mellin - barnes integrals which emerge from the routines for the @xmath1 expansion . for example , we can combine all contributions in a single common integrand , so that large numerical cancellations take place before integration . however , the finding of the peaks and the adaptation of the integration routines become less efficient . following a diametric approach , we could integrate all components separately and compute their sum at the end . then , the adaptation of the numerical integration is optimal , however , the numerical precision is sensitive to cancellations . in practice , we follow a hybrid approach and group together integrals with equivalent contours . we have achieved a reasonable efficiency and speed for all the computations that we present in this paper . since we aim to keep our integration method general , we do not attempt any special simplifications by , for example , applying barnes lemmas or exploiting symmetries of particular contours . for the numerical integration , we maintain the contours as straight lines parallel to the imaginary axis . we map them onto a real interval with a simple transformation , @xmath137 in all cases that we have considered , the integrands vanish rapidly when the integration variables take values away from the real line @xmath138 . the mellin - barnes integrals do not present problematic numerical instabilities , and converge rather fast . we perform the multidimensional integrations using the cuhre and vegas routines of the cuba library @xcite . to gain some speed , we integrate the real and imaginary parts of the integrals concurrently , using the same grid . the grid is adapted to the peaks of the real part , however , the quality of the integration for the imaginary part is only mildly affected . on the other hand , there is another subtle point concerning the convergence of the imaginary parts . as they always start an order later in the expansion in @xmath1 , the analytic expressions for them would be simpler than the corresponding real ones . when integrating numerically this relative simplicity is reflected by a faster convergence when compared to the real pieces at the same order . the numerical results depend on the values for the kinematic scales of the integrals . with our approach , we use the same expressions for the evaluation of the integrals in all kinematic regions . as we explained earlier , the only analytic continuations in kinematic variables that we must perform , is for simple functions of the type @xmath139 and @xmath140 . these are implemented trivially , in a branched format , in numerical programs . in this paper , we present multi - loop integrals for fixed kinematic parameters . however , we have arranged our routines so that a numerical integration over the phase - space can be performed simultaneously with the mellin - barnes integrations . a problematic case for our numerical algorithms is the evaluation of loop integrals with massive internal propagators in non - euclidean kinematic regions . typically , mellin - barnes represenations for such integrals exhibit a very slow damping at @xmath141 in physical regions . the mapping of the integration region to a finite interval that we have considered in eq . [ eq : mapping ] is not adequate , and contour deformations together with more sophisticated algorithms for the evaluation of oscillatory integrals are required . we defer the study of such situations for a future project . we have achieved a full automatization of the code generation for the numerical evaluation . this has allowed us to apply our method to diverse problems . our approach can be improved and refined further in the future by considering , for example , a clever analysis of the terms before integration in order to avoid numerical cancellations or more sophisticated mappings which can smoothen the peak structure of the integrands . however , as we will see in the rest of the paper , our first , naive , implementation is already sufficient to obtain accurate results for complicated integrals in a reasonable time . in what follows we present results for one , two , and three loop integrals . some of the integrals are known analytically , and we use them to verify our numerical programs . we also present new loop integrals , which would be extremely tedious to calculate with traditional methods . in this section we present results for one , two , and three - loop integrals that we obtain with our method . for each integral , we present the coefficients of its expansion in @xmath142 through the finite term : @xmath143 where @xmath116 is the depth of the leading pole in @xmath1 . we estimate the errors associated with our results for @xmath144 , by adding , in quadrature , the integration errors of all contributions from the residue decomposition . the first integral we consider is the massless one - loop hexagon , depicted in figure [ fig : hexagon ] . for massless external legs , we derive an eight - dimensional mellin - barnes representation for this integral . due to the high dimensionality , the analytic continuation in @xmath1 involves the crossing of the contours of integration from many poles , and the mellin - barnes expression after the continuation consists of about two hundred terms . this example demonstrates the advantage of the automatic algorithm we described in section [ sec : analytic ] , since the book - keeping is done automatically and our routines perform the @xmath1 expansion in fractions of a minute . after the analytic continuation and the expansion in series of @xmath1 , only integrals with low dimensionality contribute to the poles and finite pieces of the series . the integration over feynman parameters produces a denominator factor @xmath145 , which contributes factors of @xmath1 in the series expansion . the eight - dimensional component , in which no residues have been taken , drops out from the finite part . in addition , most of the residues in the analytic continuation do not give rise to poles in @xmath1 ; a few components only develop a singularity that compensates the @xmath1 in the overall factor . for the scalar hexagon , we find integrals with up to three dimensions which contribute to the poles and the finite term . this mechanism , which reduces the dimensionality of the representation in the leading coefficients of the @xmath1 expansion , is not specific to the hexagon topology , but is typical of mellin - barnes representations of scalar multi - leg integrals . in table [ tb : hex - scalar.physical ] we present explicit results for the scalar hexagon in phase space points corresponding to the physical region for the @xmath146 process . to check the efficiency of our integration routines in all the phase - space , we have sampled hundreds of points . here , we only show representative results in the phase - space points of table [ tb : hexagon - phasepoints ] . these points were generated starting from explicit values for the momenta in four dimensions , so they have a vanishing gram determinant . .explicit values of the invariants for some of the phase space points used for the evaluation of the hexagon topology . the points correspond to the physical region of the @xmath146 scattering of massless particles . [ cols="^,^,^,^,^,^,^,^,^,^",options="header " , ] we also present our results in the kinematical region corresponding to the process @xmath147 shown in figure [ fig:2b2mass ] , relevant , for instance , for the process @xmath148 . in figure [ fig:2off2bem ] we plot the values of the coefficients , for this configuration , up to the finite terms as a function of @xmath49 in the case of two equal masses , @xmath149 , where the energy scale is fixed by @xmath150 . figure [ fig:2off2bm10 ] corresponds to the same kinematical region but for @xmath150 , @xmath151 and @xmath152 . the error bars again lie within the points in the plot and are , for the finite pieces , typically less that 1% after a run of a couple of minutes per point on a 2.8ghz cpu . process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath149.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath149.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath149.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath149.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath149.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath149.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath149.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath149.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath149.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath149.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath151 and @xmath152.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath151 and @xmath152.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath151 and @xmath152.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath151 and @xmath152.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath151 and @xmath152.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath151 and @xmath152.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath151 and @xmath152.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath151 and @xmath152.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath151 and @xmath152.,width=226 ] process where the two particles in the final state are massive . we plot the values of the coefficients as function of @xmath49 for @xmath150 , @xmath151 and @xmath152.,width=226 ] at the three loop level , we start by considering the on - shell triple box ( figure [ fig:3bonshell ] ) , impressively computed in an analytic way in @xcite using the mb technique . using the re - insertion method described in section ii , we get a 7-fold mb representation . the analytic continuation leaves us with contributions with up to 5 contour integrals for the constant pieces in the @xmath1 expansion , and poles up to @xmath153 . although the numerical integration is certainly more involved in this case , due to the depth of the poles in @xmath1 which generates very complicated expressions for the finite terms , it is relatively easy to achieve a precision of @xmath154 , as shown in figure [ fig : on3bem ] . as in the two loop case we show results for the physical region corresponding to @xmath155 , @xmath156 with @xmath157 and @xmath158 . we compare our results with the analytic computation of @xcite . we used the mathematica package hpl @xcite for the evaluation of the one - dimensional harmonic polylogarithms . process . on the upper and lower left panel we plot the real and imaginary parts , respectively , of @xmath159 as a function of @xmath49 for fixed value of @xmath150 . on the right panels we show the corresponding ratios of the numerical calculation to the analytic result of @xcite for the same kinematics , the bands are given by the error in the numerical integrations.,width=302 ] process . on the upper and lower left panel we plot the real and imaginary parts , respectively , of @xmath159 as a function of @xmath49 for fixed value of @xmath150 . on the right panels we show the corresponding ratios of the numerical calculation to the analytic result of @xcite for the same kinematics , the bands are given by the error in the numerical integrations.,width=302 ] process . on the upper and lower left panel we plot the real and imaginary parts , respectively , of @xmath159 as a function of @xmath49 for fixed value of @xmath150 . on the right panels we show the corresponding ratios of the numerical calculation to the analytic result of @xcite for the same kinematics , the bands are given by the error in the numerical integrations.,width=302 ] process . on the upper and lower left panel we plot the real and imaginary parts , respectively , of @xmath159 as a function of @xmath49 for fixed value of @xmath150 . on the right panels we show the corresponding ratios of the numerical calculation to the analytic result of @xcite for the same kinematics , the bands are given by the error in the numerical integrations.,width=302 ] as we did for the double box topology , we consider now a further step in complication for the triple box , adding a mass in the external state . this is , as far as we know , the first evaluation of a three - loop box with three mass scales . we obtained a mb representation with 8 variables . as usual , the dimensionality of the problem is reduced after the analytic continuation and we found that , up to order @xmath160 , only up to six - fold integrals contribute . we present results for this novel integral in the physical region of the decay of a heavy particle , @xmath161 ( figure [ fig:3bdecay ] ) as we did in the two loops case . in figure [ fig:1off3bm ] we show the the corresponding coefficients of the expansion in @xmath1 as a function of the invariant @xmath162 for fixed values of the other two variables : @xmath163 and @xmath164 . with the notable exception of the left - most point in the real part of the constant term , the error bars are all contained in the plotted points . they are typically of the order of @xmath165 or better for the finite parts after 50 minutes , in average , per point on a 2.8ghz cpu . , width=151 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] ( figure [ fig:3bdecay ] ) . we show the results for fixed values of @xmath163 and @xmath164 as a function of the remaining invariant @xmath162.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] process with the massive leg on the final state . we plot the real and imaginary parts as a function of the invariant @xmath158 for fixed value of @xmath166 and @xmath167.,width=226 ] in this paper , we have introduced a new method for the numerical evaluation of arbitrary loop integrals . we first derive mellin - barnes representations by either decomposing directly the feynman representation , or by using one - loop representations as building blocks for representations of multi - loop integrals . for a given representation , we find values of the space - time dimension and the powers of the propagators which yield a well defined representation . then we use automated programs to perform the analytic continuation to values of the parameters for which the integral may develop divergences . finally , we expand in @xmath1 and compute the integral coefficients of the expansion numerically . the method is based on the earlier work of smirnov @xcite and tausk @xcite . the novel features in our publication are : ( i ) the automation of the procedure for the @xmath1 expansion , ( ii ) the numerical evaluation of the coefficients of the expansion , avoiding cumbersome summations of multiple infinite series and the analytic continuation in the arguments of polylogarithms , ( iii ) and the efficient generalization of the method to tensor integrals . we believe that these new features render our method suitable for the practical evaluation of multi - loop amplitudes in gauge theories . we have performed a number of explicit calculations to verify our algorithms and demonstrate the power of our method . we first recalculated complicated loop integrals which are only recently known in the literature . namely , we evaluated the planar and cross on - shell double boxes , the planar double - box with one leg off - shell , and the triple planar box with on - shell legs . in all cases , we have found an excellent agreement between our numerical results and the known analytic results . the analytic results require complicated analytic continuations of polylogarithms to non - euclidean kinematic regions . with our method , we were able to compute the integrals in all kinematic regions effortlessly . in this paper , we have presented results for loop - integrals which were previously unknown . at two - loops , we have computed the double - box integral with two adjacent legs off - shell for independent external mass - parameters . this is one of the most complicated two - loop box integrals that enters the evaluation of two - loop amplitudes for heavy boson pair production at colliders . at three - loops , we have computed the planar triple - box with one off - shell leg ; this integral emerges in @xmath168 , or the production of a single heavy boson in association with a jet at hadron colliders at nnnlo in qcd . to the best of our knowledge , it is the first time that a two - loop box with four kinematic scales or a three - loop box with three scales are ever computed in all physical regions . many processes with six external legs are particularly important at the lhc . at present , there has been no nlo calculation of a cross - section for a hadron collider process with six external states , due to the lack of efficient methods for evaluating loop amplitudes . in this paper , we applied our method to the evaluation of tensors through rank six for the hexagon topology . the purpose of this application was two - fold ; first , we wanted to demonstrate that we were able to extend the method to tensor integrals economically , and second , to setup programs that are efficient for the evaluation of multi - scale one - loop amplitudes ( e.g. six parton qcd amplitudes ) . we have found that the evaluation of the tensor integrals is not significantly more complicated than the scalar integral . the programs that we developed can be immediately used for the evaluation of the six parton one - loop qcd amplitudes . our technical goals for the current study were to build the necessary programs , implement an efficient book - keeping platform , and to examine the scaling of computing intensity in applications with diverse features . we anticipate that our method can be improved and generalized even further , when we consider issues that were only briefly examined in this first study . for example , in preliminary investigations , we have found that it is possible to exploit further the cauchy theorem . if we specify a kinematic region , it turns out that some of the numerical integrals can be approximated very accurately by the sum of a finite number of residues . we expect that we can improve substantially our efficiency if we replace some of the integrations which involve kinematic scales with appropriate finite sums of such residues . in addition , we are investigating hybrid approaches , combining our method with ideas in ref . @xcite . the applications of our method are numerous . we believe that it will be particularly useful for precision calculations of observables in collider physics . we plan to apply our techniques to key nlo and nnlo calculations for the lhc and to answer more formal questions on the perturbative behavior of gauge theories in the near future . we are indebted to vladimir smirnov , bas tausk , simon schwarz and jeppe andersen for extensive discussions on the topic . we would like to thank aude gehrmann - de ridder , thomas gehrmann , gudrun heinrich , zoltan kunszt , zoltan nagy , and bas tausk for their suggestions and for reading the manuscript . we are grateful to thomas gehrmann for providing us results and numerical routines from refs . @xcite , and daniel matre for discussions about his program to evaluate harmonic polylogarithms @xcite . this work was supported in part by the swiss national science foundation ( snf ) under contract number 200020 - 109162 and by the forschungskredit der universitt zrich . j. campbell , r. k. ellis , f. maltoni and s. willenbrock , arxiv : hep - ph/0510362 . l. w. garland , t. gehrmann , e. w. n. glover , a. koukoutsakis and e. remiddi , nucl . b * 642 * , 227 ( 2002 ) [ arxiv : hep - ph/0206067 ] . g. passarino and m. j. g. veltman , nucl . b * 160 * , 151 ( 1979 ) . f. v. tkachov , phys . b * 100 * , 65 ( 1981 ) ; k. g. chetyrkin and f. v. tkachov , nucl . b * 192 * , 159 ( 1981 ) . t. gehrmann and e. remiddi , nucl . b * 580 * , 485 ( 2000 ) [ arxiv : hep - ph/9912329 ] . z. bern , l. j. dixon and d. a. kosower , nucl . b * 412 * , 751 ( 1994 ) [ arxiv : hep - ph/9306240 ] . a. v. kotikov , phys . b * 254 * , 158 ( 1991 ) . w. t. giele and e. w. n. glover , jhep * 0404 * , 029 ( 2004 ) [ arxiv : hep - ph/0402152 ] ; r. k. ellis , w. t. giele and g. zanderighi , arxiv : hep - ph/0508308 ; r. k. ellis , w. t. giele and g. zanderighi , phys . d * 72 * , 054018 ( 2005 ) [ arxiv : hep - ph/0506196 ] . a. denner and s. dittmaier , arxiv : hep - ph/0509141 ; a. denner and s. dittmaier , nucl . b * 658 * , 175 ( 2003 ) [ arxiv : hep - ph/0212259 ] ; t. binoth , j. p. guillet , g. heinrich , e. pilon and c. schubert , arxiv : hep - ph/0504267 ; t. binoth , g. heinrich and n. kauer , nucl . phys . b * 654 * , 277 ( 2003 ) [ arxiv : hep - ph/0210023 ] ; t. binoth , j. p. guillet , g. heinrich and c. schubert , nucl . b * 615 * , 385 ( 2001 ) [ arxiv : hep - ph/0106243 ] ; t. binoth , j. p. guillet and g. heinrich , nucl . b * 572 * , 361 ( 2000 ) [ arxiv : hep - ph/9911342 ] . a. van hameren , j. vollinga and s. weinzierl , eur . j. c * 41 * , 361 ( 2005 ) [ arxiv : hep - ph/0502165 ] . f. del aguila and r. pittau , jhep * 0407 * , 017 ( 2004 ) [ arxiv : hep - ph/0404120 ] . a. denner , s. dittmaier , m. roth and l. h. wieders , arxiv : hep - ph/0505042 ; a. denner , s. dittmaier , m. roth and l. h. wieders , phys . b * 612 * , 223 ( 2005 ) [ arxiv : hep - ph/0502063 ] . z. nagy and d. e. soper , jhep * 0309 * , 055 ( 2003 ) [ arxiv : hep - ph/0308127 ] . z. bern , l. j. dixon and d. a. kosower , arxiv : hep - ph/0507005 . s. laporta , int . j. mod . a * 15 * , 5087 ( 2000 ) [ arxiv : hep - ph/0102033 ] . c. anastasiou and a. lazopoulos , jhep * 0407 * , 046 ( 2004 ) [ arxiv : hep - ph/0404258 ] . a. v. smirnov and v. a. smirnov , arxiv : hep - lat/0509187 ; o. v. tarasov , nucl . instrum . a * 534 * , 293 ( 2004 ) [ arxiv : hep - ph/0403253 ] ; o. v. tarasov , acta phys . b * 29 * , 2655 ( 1998 ) [ arxiv : hep - ph/9812250 ] ; p. a. baikov , arxiv : hep - ph/0507053 ; p. a. baikov , nucl . phys . proc . * 116 * ( 2003 ) 378 . v. a. smirnov and m. steinhauser , nucl . b * 672 * , 199 ( 2003 ) [ arxiv : hep - ph/0307088 ] . s. actis , a. ferroglia , g. passarino , m. passera and s. uccirati , nucl . b * 703 * , 3 ( 2004 ) [ arxiv : hep - ph/0402132 ] ; a. ferroglia , m. passera , g. passarino and s. uccirati , nucl . b * 680 * , 199 ( 2004 ) [ arxiv : hep - ph/0311186 ] ; a. ferroglia , m. passera , g. passarino and s. uccirati , nucl . b * 650 * , 162 ( 2003 ) [ arxiv : hep - ph/0209219 ] ; g. passarino and s. uccirati , nucl . b * 629 * , 97 ( 2002 ) [ arxiv : hep - ph/0112004 ] . s. p. martin and d. g. robertson , arxiv : hep - ph/0501132 . m. caffo , h. czyz and e. remiddi , nucl . suppl . * 116 * , 422 ( 2003 ) [ arxiv : hep - ph/0211178 ] ; s. laporta , phys . b * 504 * , 188 ( 2001 ) [ arxiv : hep - ph/0102032 ] . t. binoth and g. heinrich , nucl . b * 585 * , 741 ( 2000 ) [ arxiv : hep - ph/0004013 ] ; t. binoth and g. heinrich , nucl . b * 680 * , 375 ( 2004 ) [ arxiv : hep - ph/0305234 ] . k. hepp , commun . * 2 * , 301 ( 1966 ) . m. roth and a. denner , `` high - energy approximation of one - loop feynman integrals , '' nucl . b * 479 * , 495 ( 1996 ) [ arxiv : hep - ph/9605420 ] . t. binoth , g. heinrich and n. kauer , nucl . b * 654 * , 277 ( 2003 ) [ arxiv : hep - ph/0210023 ] . c. anastasiou , k. melnikov and f. petriello , arxiv : hep - ph/0505069 . v. a. smirnov , phys . b * 460 * , 397 ( 1999 ) [ arxiv : hep - ph/9905323 ] . j. b. tausk , phys . b * 469 * , 225 ( 1999 ) [ arxiv : hep - ph/9909506 ] . v. a. smirnov and o. l. veretin , nucl . b * 566 * , 469 ( 2000 ) [ arxiv : hep - ph/9907385 ] . c. anastasiou , j. b. tausk and m. e. tejeda - yeomans , nucl . suppl . * 89 * , 262 ( 2000 ) [ arxiv : hep - ph/0005328 ] . v. a. smirnov , phys . b * 491 * , 130 ( 2000 ) [ arxiv : hep - ph/0007032 ] ; v. a. smirnov , phys . b * 500 * , 330 ( 2001 ) [ arxiv : hep - ph/0011056 ] ; v. a. smirnov , phys . b * 524 * , 129 ( 2002 ) [ arxiv : hep - ph/0111160 ] ; g. heinrich and v. a. smirnov , phys . b * 598 * , 55 ( 2004 ) [ arxiv : hep - ph/0406053 ] . v. a. smirnov , phys . b * 567 * , 193 ( 2003 ) [ arxiv : hep - ph/0305142 ] . z. bern , l. j. dixon and v. a. smirnov , phys . d * 72 * , 085001 ( 2005 ) [ arxiv : hep - th/0505205 ] . c. anastasiou , z. bern , l. j. dixon and d. a. kosower , phys . * 91 * , 251602 ( 2003 ) [ arxiv : hep - th/0309040 ] . e. remiddi and j. a. m. vermaseren , int . j. mod . a * 15 * , 725 ( 2000 ) [ arxiv : hep - ph/9905237 ] . ussyukina , teor.mat.fiz . 22 ( 1975 ) 300 ; e. e. boos and a. i. davydychev , theor . * 89 * , 1052 ( 1991 ) [ teor . mat . fiz . * 89 * , 56 ( 1991 ) ] ; n. i. ussyukina and a. i. davydychev , phys . b * 305 * , 136 ( 1993 ) ; n. i. ussyukina and a. i. davydychev , phys . b * 298 * , 363 ( 1993 ) . t. gehrmann and e. remiddi , nucl . b * 601 * , 248 ( 2001 ) [ arxiv : hep - ph/0008287 ] . t. gehrmann and e. remiddi , comput . commun . * 141 * , 296 ( 2001 ) [ arxiv : hep - ph/0107173 ] . t. gehrmann and e. remiddi , comput . phys . commun . * 144 * , 200 ( 2002 ) [ arxiv : hep - ph/0111255 ] . t. gehrmann and e. remiddi , nucl . b * 640 * , 379 ( 2002 ) [ arxiv : hep - ph/0207020 ] .

currently , experiment suggests that there are more scalar mesons than can be accommodated in a single quark - model nonet . many qcd motivated models predict the existence of non-@xmath8 mesons such as glueballs @xcite , hybrids @xcite , @xmath9 @xcite and @xmath10 @xcite states . it is tempting to say that these ` spare ' experimental states are the unconventional mesons allowed by qcd but , with scalar glueball and multiquark states expected to have masses comparable to the conventional @xmath11 states , which _ are _ the extra states ? in the work that follows we hope to shed some light on this question by attempting to determine which of the 5 scalar - isoscalar mesons currently listed in the particle data group tables @xcite is the lightest @xmath12 meson in this channel . to address this question we will use the qcd sum rule technique . the answer has importance for chiral symmetry breaking in qcd , see for instance @xcite . qcd sum rules are integral expressions that relate the hadronic and partonic regimes , _ i.e. _ the low energy world of resonances with the high energy world of ( tractable ) qcd . since their conception over twenty years ago @xcite , qcd sum rules have become an established technique both for calculating hadronic properties in channels , where the qcd expressions are under control , and , conversely , estimating qcd parameters ( such as the masses of the quarks ) in channels where there is good experimental information . using qcd sum rules in the scalar channel is not a new idea . they were first applied to the scalar mesons in @xcite where laplace sum rules were used ( for a recent review of laplace sum rules see @xcite ) . here the phenomenological side was represented using the , now standard , resonance + continuum ansatz , with resonances being represented as @xmath13-functions and the continuum being calculated entirely within perturbative qcd . as only the operator product expansion ( ope ) terms , discussed in sect . [ sec : ope ] , were taken into account on the theory side , _ exact _ mass degeneracy between the lightest isoscalar and isovector was found , with . the @xmath14 state was predicted to have a mass of around 1.35 gev . these findings were supported by bramon and narison @xcite , who used qcd sum rules to calculate the couplings of the @xmath4 and @xmath15 to two photons . again modelling the resonances with @xmath13-functions the authors concluded that a @xmath11 interpretation of these two mesons could not be ruled out with the data then available . the effects of going beyond the ope were studied in @xcite , where instantons ( see sect . [ sec : instantons ] ) were included on the theoretical side , but not in the qcd continuum on the phenomenological side . once again the resonances were treated as @xmath13-functions and only the isospin-1 channel was considered . it was concluded that the lightest @xmath11 state in this channel had a mass @xmath16 1 gev . for the light scalar mesons the zero - width approximation is not a good one and the first attempt to go beyond it in a sum rule investigation was by elias _ _ @xcite . here resonances were represented by breit - wigner formulae , which , whilst better than a @xmath13-function , still do not adequately describe the complex structure in the scalar sector . a further abstraction was introduced by replacing these breit - wigner shapes with a riemann sum of rectangular pulses . laplace sum rules were used , but in an integral form which requires the calculation of perturbative expressions at rather low energy scales . again instantons were included , but not in the continuum . in the isovector channel , the @xmath15 was found to decouple from the sum rule and the mass of the lightest quarkonium isovector was consistent with the @xmath17 . in the isoscalar channel the conclusions were not so clear cut , it was predicted that the lightest quarkonium here should have a width less than half of its mass , and although the @xmath4 could not be ruled out as the main contributor a lighter state was preferred . similar conclusions were reached in @xcite . this study used both laplace and finite energy sum rules , again requiring perturbative expressions to be evaluated at low energies . the authors noted , that for a consistent treatment of the borel sum rules , the instanton effects should also be included in the qcd continuum contribution . the sum rules were dominated by an isoscalar with mass around 1 gev and an isovector with mass around 1.5 gev , thus suggesting the @xmath4 and @xmath17 as the lightest @xmath11 states in their respective channels . however , in the same work , a comparison to a more realistic resonance shape predicted resonance parameters for the lightest quarkonium of @xmath18 mev and @xmath19 mev . these parameters are not consistent with the @xmath4 , but could describe a breit - wigner fit to the @xmath3 . in @xcite , maltman used pinched weight finite energy sum rules to calculate the decay constants of the @xmath15 and @xmath17 , which were then compared with the decay constant of the , presumably @xmath11 , @xmath20 . instantons were included and all qcd integrals were carried out in the complex plane , which improves the convergence of perturbative expressions . the isovector decay constants were found to be of comparable size , suggesting a similar structure for both . the author concluded that this favoured a unitarised quark model @xcite scenario , where two physical hadrons can arise from one ` seed ' state . in the work that follows we will apply qcd finite energy sum rules introduced in sect . 2 to the scalar - isoscalar channel . on the theoretical side we will include in sects . 3 and 4 instanton effects as well as the ope terms up to dimension six and use the contour improvement prescription to increase the convergence of perturbative expressions . on the the phenomenological side in sect . 5 we will incorporate experimental data directly by relating the hadronic vacuum polarisation to the @xmath21 scattering amplitude via the pion s scalar form - factor . the sum rules are evaluated in sect . 6 and we present our conclusions in sect . 7 . the connection between quarks and hadrons is through the vacuum polarisation , @xmath22 . this is defined by the correlation function of two currents , _ i.e. _ @xmath23 where @xmath24 and the currents are chosen so as to select the desired channel . in this work we are considering the scalar - isoscalar mesons and so we choose the renormalisation group invariant current @xmath25 where @xmath26 and we use the symbol @xmath27 to denote an effective light - quark . the second equality follows because we ignore the two - gluon intermediate state , whose contribution is @xmath28 and so can be safely neglected . -plane showing the so called ` pacman ' contour which is used to derive the fesrs . [ fig : pacman ] ] ' '' '' @xmath22 is analytic everywhere in the complex @xmath29plane except along the time - like axis , so for any weight function , @xmath30 , which is real for all real @xmath31 and analytic within the ` pacman ' contour shown in fig . [ fig : pacman ] we can write @xmath32 on the left - hand side of eq . ( [ eq : fesr ] ) we have an integral in the physical region , and here we choose to describe the integrand using hadronic physics ( either with experimental data or a phenomenological parameterisation ) . on the right - hand side we have an integral in the non - physical region ( complex @xmath31 ) , and here we use field theory ( _ i.e_. qcd ) to calculate the integrand . the assumption that eq . ( [ eq : fesr ] ) holds for a range of values of @xmath1 is a statement of global quark - hadron duality , _ i.e. _ the integral of an hadronic expression is equal to the integral of an expression written in terms of partonic degrees of freedom . early sum rule work used positive integer powers of @xmath31 as weight functions . in this work we wish to investigate the lightest meson in the channel , and so we choose a decaying exponential , . on the phenomenological side , this weight function will suppress the region of the integral @xmath33 , and so by varying @xmath34 we can control which regions of the integrand contribute most significantly to the sum rule . if we also include integer powers of @xmath31 in the weight function then we can build up a family of sum rules . almost by definition , at energies where resonances dominate the spectral density , the implications of qcd are not straightforwardly calculable . if qcd _ were _ tractable then the technology of qcd sum rules would be redundant . this means that , for moderate values of @xmath1 , our straightforward qcd description of @xmath22 must fail , at the very least , in the region of the positive real axis ( see fig . [ fig : pacman ] ) . however , it is believed @xcite that it is _ only _ in the region of the positive , real axis that the straightforward qcd description is particularly unreliable and that in the rest of the @xmath31-plane calculable qcd provides a good description of the correlator . consequently we introduce a zero into the weight function at the point where the circle joins the real axis , _ i.e. _ @xmath1 . thus our sum rules have the weight function and we denote them by the symbol @xmath35 . elsewhere in the literature they are often referred to as pinched weight sum rules . then the left hand side of eq . ( 3 ) becomes @xmath36 introducing this zero into the weight function has a number of advantages . firstly , on the theoretical side we avoid the problem , just described , of the failure of qcd on the part of the circle near the positive real axis . secondly , on the phenomenological side , we have introduced a second factor with a tendency to suppress the higher energy portion of the integral and consequently reduce the dependence on the exact value of @xmath1 @xcite . the parameter @xmath1 is not entirely free : it must be large enough that the qcd expressions we are able to calculate are expected to be a good approximation to the full correlator , yet @xmath1 should be small enough that we have experimental information . the art of sum - ruling is to find ( sensible ) values of @xmath1 for which over a wide range of the unphysical parameter @xmath34 . when this occurs we say that the sum rules are _ although the current eq . ( [ eq : current ] ) is rg - invariant , the correlator it gives rise to is not . as well as making the calculation of the correlator within qcd more difficult , this would also lead to an unwanted scale dependence in our final results . the second derivative of this correlator , @xmath37 , _ is _ rg - invariant . as @xmath37 is also analytic within the contour shown in fig . [ fig : pacman ] , we could choose to write the sum rules completely in terms of this quantity , but @xmath38 is not directly related to any physical quantity and so we would lose our link to experiment . instead we use partial integration to re - express the theoretical side of the sum rules in terms of @xmath37 . to do this we write the integral on the theoretical side as @xmath39 we now integrate this expression by parts twice to give , @xmath40 and arrange the constants of integration which appear in @xmath41 and @xmath42 so that these functions disappear at @xmath43 . then our sum rule is @xmath44_{had } \ , \dd s = -\frac{1}{2 \pi i } \oint_{|s|=s_0 } \!\!\!\ ! v(s ) \ , \left[\pi^{\prime \prime}(s)\right]_{qcd } \ , \dd s \ . \label{eq : rgifesr}\ ] ] = 6.0 mm in order to calculate the vacuum polarisation within qcd we make use of the operator product expansion ( ope ) @xcite . this replaces the matrix element of the product of currents in eq . ( [ eq : corr ] ) by a linear combination of local operators ordered in terms of increasing dimensionality , @xmath45 the first term in this expansion is the purely perturbative contribution , with higher dimension terms giving the contribution due to various condensates . we will consider terms up to dimension six , higher order terms are expected to be negligible . expressions are available in the literature for the dimension zero term to four - loop order , the dimension four term to two loops and the dimension six term at leading order . we quote these expressions with refernces to the original works for the full information . all expressions are given in the @xmath46 scheme . the dimension zero expression can be found in @xcite and is given by @xmath47 \right . \nonumber\\ & & \left . { } + a_s^2 \left [ \left ( \frac{5071}{144 } - \frac{35}{2}\zeta(3 ) \right ) - \frac{139}{6 } \l + \frac{17}{4 } \ll \right ] \right.\nonumber \\ & & \left . { } + a_s^3 \left [ \left ( -\frac{4781}{9 } + \frac{b_1}{6 } + \frac{475}{4}\zeta(3)\right ) + \left ( -\frac{2720}{9 } + \frac{475}{4}\zeta(3 ) \right ) \l \right . \nonumber \\ & & \left . \qquad \quad { } + \frac{695}{8}\ll - \frac{221}{24}\lll \right ] \right\ } \ , \label{eq : dim0}\end{aligned}\ ] ] where @xmath48 and @xmath49 is the riemann zeta function . as @xmath50 is renormalisation group invariant , we are free to make any convenient choice of the renormalisation scale . we use the ` contour improvement ' prescription , whereby the logs in eq . ( [ eq : dim0 ] ) are ` re - summed ' by making the choice @xmath51 . this improves the convergence of the perturbative series by removing the dependence on large powers of possibly large logs which would appear in the ( unknown ) higher order terms . the contour improved perturbative contribution to our correlator is then , @xmath52 navely , the only gauge - invariant operator of dimension two that can be constructed within qcd is the so - called mass insertion term , @xmath53 . for the light quarks that we are considering this term can safely be ignored . some recent works have suggested that there may be mechanisms outside of the normal ope , which could lead to dimension two contributions to correlation functions . these mechanisms include renormalons @xcite and a tachyonic ( _ i.e. _ imaginary ) gluon mass @xcite . it has been argued that , if they do exist , the effects of dimension two operators would not be significant in most sum rule analyses and that even in calculations where they would be expected to be most noticeable , these effects are consistent with zero @xcite . hence , in this work , we choose to ignore them . the contour - improved contribution at dimension four can be inferred from @xcite and is given by @xmath54 - \frac{i_g}{9 } \left[1 + \frac{121 a_s(q^2)}{18 } \right ] \right . \nonumber \\ & & \qquad \qquad \left . { } + i_s \frac{4 a_s(q^2)}{9 } - \frac{3 m_s^4(q^2)}{28 \pi^2 } \right\ } \ , \label{eq : dim4inv}\end{aligned}\ ] ] where @xmath55 and @xmath56 the leading order dimension six contribution can be taken from @xcite and we follow this reference in taking this term at the fixed scale of 1 gev . the dimension six contribution is then @xmath57 \right ) \ , \label{eq : dim6}\end{aligned}\ ] ] where @xmath58 are the dirac matrices , and @xmath59 are the usual gell - mann matrices normalised such that . currently there is no reliable method to calculate the vacuum expectation values appearing in eq . ( [ eq : dim6 ] ) , so we follow standard practice and relate these dimension six condensates to the ( rg variant ) dimension three light quark condensate @xmath60 . for the mixed condensate , @xmath61 , this is done via the parameterisation @xmath62 and for the four - quark condensates , @xmath63 and @xmath64 , via the _ vacuum saturation hypothesis _ @xcite . to take into account the possible violation of the vacuum saturation hypothesis we include a multiplicative factor @xmath65 . thus our final expression for the dimension six contribution is @xmath66 in order to evaluate all these expressions we need to calculate the running coupling and quark masses at complex values . this is done by solving the four - loop renormalisation group equations that define the qcd @xmath67 and @xmath68 functions ( our conventions for these functions are given in appendix [ app : rg ] ) . for the coupling , we choose to solve the implicit equation @xmath69 iteratively using newton s method . as input , we take the experimental value @xmath70 , as measured by the aleph collaboration @xcite . for reference , had we used the standard expansion for @xmath71 , then at four - loops and with three active flavours , this central input value would correspond to @xmath72 mev . for higher values of @xmath73 , or correspondingly @xmath74 , the radiative corrections to eq . ( [ eq : dim0q ] ) become more important and start to swamp the higher dimension terms , _ i.e _ it becomes more important to include higher loop corrections to the perturbative contribution than to include the condensates . it has been argued @xcite that the sum rule methodology could become invalid for @xmath75 mev . this is especially problematic for attempts to determine condensate values via the sum rule technique . as this is not our aim and considering the success of previous sum rule studies with similar and higher values of @xmath73 ( and @xmath74 ) , we feel justified in applying sum rules to this problem . the momentum dependence of the quark mass is given by @xmath76 \ . \label{eq : mass}\ ] ] the opes for the scalar - isoscalar and scalar - isovector channels are identical . this was originally interpreted as an explanation for the almost exact mass degeneracy of the @xmath4 and the @xmath15 @xcite . however , we know that the ope is not enough to describe the qcd vacuum completely . other effects are important , particularly in the scalar and pseudoscalar channels @xcite . the complex nature of the qcd vacuum means that it is not currently possible to solve the equations of motion of the full theory directly . we must make use of approximations , which in this work we choose to be the instanton liquid model @xcite . here the qcd vacuum is modelled as a four - dimensional ` liquid ' of instantons , which is assumed to be dilute enough to make the idea of individual instantons meaningful . the liquid is characterised by the average instanton size @xmath77 and four - volume density @xmath78 ( or alternatively the average instanton separation , @xmath79 ) . the instanton liquid model is then thought to be a good approximation , if the ratio @xmath80 is small . within this model the instanton contribution to the scalar - isoscalar light - quark correlator was found in @xcite to be @xmath81 ^ 2 \ , \label{eq : inscorr}\ ] ] where @xmath82 is a modified bessel function of the second kind ( or macdonald function ) . taking two derivatives we obtain @xmath83 \ . \label{eq : inssecdir}\ ] ] as stated earlier @xmath37 is an rg - invariant quantity . the only quantity in eq . ( [ eq : inssecdir ] ) that could have a renormalisation scale dependence is the quark mass . this renormalisation scale dependence can not be cancelled out anywhere else , so we conclude that the quark mass appearing in eq . ( [ eq : inssecdir ] ) must be the quark mass at some fixed scale , @xmath84 . _ a priori _ we do not know what this scale is . it is reasonable to expect that it is related to the instanton scale given by @xmath85 or the hadronic scale of @xmath86 gev , but we have no physical reason for choosing @xmath85 rather than _ e.g. _ @xmath87 . thus we take this scale as an input parameter in the calculation . if this scale is found to be considerably different from the hadronic and instanton scales , then it would call into doubt our treatment of the instanton contribution . at low energies the only hadronic process possible in the scalar - isoscalar channel is @xmath21 scattering . as pions are made of non - strange quarks , it is reasonable to expect that , in the elastic region , the spectral density for the correlator is related to the absorptive part of the @xmath88 scattering amplitude , @xmath89 . this link can be made via the scalar form - factor of the pion , @xmath90 , which is defined by @xmath91 inserting a complete set of hadronic states into the correlator eq . ( [ eq : corr ] ) and keeping only the lowest state , _ i.e. _ two pions , gives the spectral density in the elastic region , @xmath92 where @xmath93 . watson s final state interaction theorem @xcite implies that the form factor and the scattering amplitude have the same phase . thus , in the region where elastic unitarity holds , we can write @xmath94 where the coupling function , @xmath95 , is real for real @xmath96 . substituting eq . ( [ eq : watson ] ) into eq . ( [ eq : spec ] ) and making use of the elastic unitarity relationship , we obtain @xmath97 it is well known that @xmath21 scattering contains an adler zero @xcite , and that this zero does not appear in the scalar pion form factor . we now make the simplifying assumption that this is the only @xmath31-dependence appearing in @xmath95 , _ i.e. _ @xmath98 where @xmath99 is the position of the adler zero and @xmath100 is a constant . substituting eq . ( [ eq : adler ] ) into eq . ( [ eq : spec2 ] ) gives our first ansatz @xmath101 where @xmath102 is an unknown constant of proportionality . although the @xmath103 threshold is at 560 mev , it is well known that multi - pion channels are not significant below around 1400 mev , when quasi two - body channels , _ @xmath104 , @xmath105 , become important @xcite . the first important inelastic channel is @xmath106 . even after this has opened the @xmath21 final state will continue to pick out the non - strange contribution to the sum over states . thus we might expect eq . ( [ eq : ansatz1 ] ) to give a reasonable approximation to the true spectral density even above 1 gev . we call this coupling scheme i. the multi - pion final states mentioned above will increase the @xmath107 spectral density , causing eq . ( [ eq : ansatz1 ] ) to be an underestimate at higher energies . phenomenologically , we might expect the true spectral density to be enhanced by the ratio of the total to elastic @xmath21 cross - sections , thus giving @xmath108 which we call coupling scheme ii . in the elastic region , the two coupling schemes are , of course , identical . above the @xmath10 threshold , whereas coupling scheme i ignores all inelastic channels and so will underestimate the spectral density , coupling scheme ii takes into account all inelastic channels , including those with hidden strangeness , and so may be an overestimate . @xmath21 scattering amplitude as calculated from the data set described in the text ( line ) . superimposed are experimental points from @xcite ( diamonds ) , @xcite ( boxes ) , @xcite ( triangles ) and @xcite ( circles ) . [ fig : sketch ] ] ' '' '' as input we take the parameterisation of the @xmath109 scattering data carried out by bugg _ et al _ @xcite . however , when continued down to threshold this fails to reproduce the scattering length predicted by chiral perturbation theory ( @xmath110pt ) and so in the region below 450 mev we use the roy equation extrapolation of the classic ochs - wagner phase - shifts @xcite , as carried out by pennington and used in the discussion of @xmath111 @xcite . [ fig : sketch ] shows a sketch of the absorptive part of the isospin zero , @xmath112-wave amplitude for @xmath21 scattering as calculated from our data set . from this we can clearly see that the @xmath3 and the @xmath113 appear as peaks , while the @xmath4 and the @xmath114 show up as sharp dips . thus these latter two states , and in particular the @xmath4 , will largely decouple from the spectral density and if saturation of the sum rules is achieved with either of our two coupling schemes , it will not be due in any great part to the @xmath4 or @xmath114 . by varying both @xmath34 and @xmath1 it may be possible to determine which of the @xmath3 and the @xmath113 plays the dominant role in saturation , but for the evaluation of the wilson coefficients in the ope to be trustworthy , we would expect @xmath1 to be too high to exclude the @xmath113 . nevertheless , as the decaying exponential in the weight functions of our sum rules suppresses the integrand for @xmath33 , it is still possible to determine which resonance is most important for saturation . a value for the normalisation factor @xmath102 , can be determined by making a comparison with the spectral density obtained using an omns representation @xcite of the form factor . this representation is given by @xmath115\ , \label{eq : omnes}\ ] ] where @xmath116 is the phase of @xmath117 scattering . the form factor at zero momentum transfer , which normalises the omns function , can be determined from the feynman - hellman theorem @xcite and is given by @xmath118 to leading order in @xmath110pt , the physical pion mass is given by the gell - mann - oakes - renner relation , @xcite , and thus we see that @xmath119 going to next - to - leading order in @xmath110pt would introduce a correction to eq . ( [ eq : dzero ] ) of the order of 1% ( see for example @xcite ) , which is safe to neglect . to evaluate the integral in eq . ( [ eq : omnes ] ) exactly requires knowledge of the scattering phase in the high energy region where it is unknown . however , the integrand is dominated by the region around @xmath31 . so if we restrict our comparison to the low energy region then the behaviour of the phase above some point @xmath120 , where @xmath121 , will not greatly affect the value of the integral . we take @xmath120 to correspond to the last data point and approximate the phase above this point as a constant , @xmath122 . thus , the spectral density , as calculated from the omns representation , is written @xmath123 \ . \label{eq : specomnes}\ ] ] where the bar on the integral sign denotes the cauchy principal value . ] ' '' '' the value of the constant , @xmath102 , can then be determined by fitting eq . ( [ eq : ansatz1 ] ) to eq . ( [ eq : specomnes ] ) . the position of the adler zero for isospin-0 , s - wave @xmath21 scattering is fixed to be close to @xmath124 , which is its position at lowest order in @xmath110pt and so we essentially only have one parameter , @xmath102 , to determine . the fit , shown in fig . [ fig : lowrho ] , was carried out up to a maximum energy of 400 mev . over this energy range it does not matter which coupling scheme is used . each gives the value . however , this value is quite sensitive to the maximum energy of the fit and should not be trusted to an accuracy better than 20% . in any sum rule calculation , we hope to find regions of the parameter space where the theoretical and phenomenological sides are equal . we measure this _ saturation _ of the sum rules by introducing the double ratios @xmath125_{had}}{\left[t_{k+1 } / t_k \right]_{qcd } } \ . \label{eq : drat}\ ] ] this ratio is independent of the overall normalising constants on either side , _ i.e _ @xmath102 and @xmath126 , and obviously is equal to one when the sum rules are saturated . as sum rule analyses are generally accurate to around 10 - 20% we set an agreement region of @xmath127 . in practice we shall use only the lowest such double ratio for each type of sum rule , _ i.e _ @xmath128 , as the integer powers of @xmath31 in the higher sum rules increase the importance of the upper end of the integrand on the phenomenological side . = 5.9 mm .standard values of all parameters used to evaluate the sum rules . [ tab : parameters ] [ cols="^,^,^",options="header " , ] ' '' '' we have performed a number of tests to determine how stable this estimate is to the variation of input parameters within the ranges found to give good saturation curves . the parameters @xmath77 , @xmath129 and @xmath65 each yield an uncertainty in the quark mass of 0.1 mev , see fig . [ fig : mass ] . varying the remaining condensate values from zero to twice the value listed in table [ tab : parameters ] produces no noticeable effect on the quark mass obtained . the largest source of uncertainty in our calculation is the value of the phenomenological constant of proportionality , @xmath102 , defined in eqs . ( 27,28 ) . this could only be determined to an accuracy of 20% , and so gives the dominant uncertainty in the final result of 0.5 mev . our final estimate is then @xmath130 . in table [ tab : masses ] we list some other recent determinations of this quantity . we see that our estimate is consistent , though lower , than earlier sum rule determinations . though not agreeing with the results of quenched lattice qcd , our value is reassuringly consistent with the early results of unquenched lattice calculations which are just beginning to appear in the literature . the application of qcd sum rules to the isoscalar - scalar current requires a realistic representation of the contributing resonances . none in this channel remotely approximates a @xmath131function , fig . 2 . indeed , states like the @xmath132 ( or @xmath3 ) are far too broad to be sensibly described by simple breit - wigner forms . moreover they overlap with other resonances like the @xmath4 and strongly coupled thresholds , like @xmath133 . it is quite natural that the @xmath134 correlator is expressly related to the @xmath135 scattering amplitude and so the vacuum polarization can be described directly in terms of experimental information without the need for further ( breit - wigner - like ) approximations . we have shown that such coupling schemes , defined in sect . [ subsec : ansatz ] , are perfectly capable of saturating qcd finite energy sum rules . in @xmath21 scattering the @xmath4 appears as a sharp dip . consequently , in our coupling schemes this state makes relatively little contribution to the correlator , in total contrast to the conclusions of shakin and wang @xcite . that good saturation is possible strongly suggests then that the @xmath4 does indeed make only a small contribution to the scalar - isoscalar sum rules , which in turn implies that it does not contain a large @xmath136 component in its wave function . this result is also at odds with the conclusion reached in @xcite . however , as explained above , when the same authors @xcite considered a more realistic approximation for the resonance shape , the parameters found were more consistent with the @xmath3 than the @xmath4 . from the present analysis , beyond saying that it is not strongly @xmath136 , we can make no comment on the structure of the @xmath4 . the saturation curves we obtain are remarkably flat and close to one from low values of @xmath34 , see figs . as the exponential in the weight function tends to suppress the integrand in the region @xmath33 , saturation at these values of @xmath34 means it is the lower energy portion of the experimental data that is most important in saturating the sum rules . furthermore , the @xmath113 ( with mass @xmath137 and width @xmath138 ) will not contribute until . that the sum rules are fulfilled from @xmath139 implies it is the @xmath3 that plays the most important role in saturating the @xmath136 scalar - isoscalar sum rules . our results again show that the operator product expansion is not enough to describe the physics of the scalar mesons completely . we find saturation only when instantons are included . one potentially worrying point of this analysis is the need to tune the average instanton size away from its accepted value of . to or 0.40 0.42 fm , which is about 25% larger . the validity of the instanton liquid model relies on the mean separation of instantons being significantly larger than their average size . if this is not true then neighbouring instantons will overlap significantly and it becomes meaningless to talk about individual instantons . the assumed average instanton separation of 1 fm is only about 2.5 times the value of @xmath77 required by our sum rule , and so we are probably approaching this limit . reassuringly though , lattice simulations of the topological features of the qcd vacuum seem to find average instanton sizes similar to the one we are forced to use ( for a summary of lattice determinations of instanton parameters see @xcite ) . the above results follow from ratios of the finite energy sum rules defined in eq . ( 33 ) and displayed in figs . 4 - 8 . moreover , the absolute value of each is determined by the quark mass curves of sect . [ subsec : mass ] . these are also found to be flat and this constancy sets in already below @xmath140 , fig . this is again evidence that our coupling schemes are capable of satisfying the sum rules and that the state most important for this is the @xmath3 . the value we find for the average @xmath141 and @xmath142 quark mass is @xmath143 the most conservative way to interpret this result is merely as a consistency check . a number of assumptions are made in this analysis , _ e.g. _ the validity of the coupling schemes , the truncation of the ope at dimension-6 and the perturbative contribution at @xmath144 , the form of the instanton contribution _ etc_. that our final result is of the correct size reassures us that none of our assumptions are too far off the mark . going further and taking it foregranted that the isoscalar - scalar correlator is indeed directly related to @xmath135 scattering , then the sum rules provide an independent determination of the average light quark mass , as given in eq . our result is quite compatible with sum rule determinations in the pseudoscalar sector and with the early unquenched lattice estimates that are starting to appear in the literature . we can then be certain that the @xmath3 is the dominant contribution to the @xmath145 current . m.r.p . is most grateful for the warm hospitality as well as support at the special research centre for the subatomic structure of matter at the university of adelaide during the course of this work . the authors would like to thank kim maltman for most helpful discussions in adelaide and subsequent e - mail correspondence . is also grateful to the theory group of the institut de physique nuclaire , universit paris - sud , orsay for its kind hospitality . this work was supported in part under the eu - tmr programme , contract no . ct98 - 0169 , eurodaphne . the qcd @xmath67 and @xmath68-functions are defined by the the renormalisation group equations @xmath146 and @xmath147 where @xmath148 . currently , the qcd @xmath67 and @xmath68-functions are known to four - loop order . for three active flavours , the coefficients of the @xmath67 function within the @xmath46 scheme are @xcite @xmath149 the coefficients of the @xmath68-function , also within the @xmath46 scheme , are @xcite @xmath150 @xmath151 99 h. fritzsch and m. gell - mann , proc . of _ xvi international conference on high energy physics _ , chicago , 1972 , ( ed . j. jackson , a. roberts and r. donaldson ) vol . 2 p. 135 ; + h. fritzsch and p. minkowski , nuovo cim . a * 30 * ( 1975 ) 393 . d. horn and j. mandula , phys . d * 17 * ( 1978 ) 898 . r. l. jaffe , phys . rev . * d15 * ( 1977 ) 267 ; + r. l. jaffe , phys . rev . * d15 * ( 1977 ) 281 . j. weinstein and n. isgur , phys . rev . * 48 * ( 1982 ) 659 . d. e. groom _ et al . _ , eur . j. * c15 * ( 2000 ) 1 . r. delbourgo and m.d . scadron , phys . * 48 * ( 1982 ) 379 . m. a. shifman , a. i. vainshtein and v. i. zakharov , nucl . b * 147 * ( 1979 ) 385 . l. j. reinders , s. yazaki and h. r. rubinstein , nucl . b * 196 * ( 1982 ) 125 . p. colangelo and a. khodjamirian , hep - ph/0010175 . a. bramon and s. narison , mod . lett . a * 4 * ( 1989 ) 1113 . e. v. shuryak , nucl . phys . b * 319 * ( 1989 ) 541 . v. elias , a. h. fariborz , f. shi and t. g. steele , nucl . a * 633 * ( 1998 ) 279 [ hep - ph/9801415 ] . f. shi , t. g. steele , v. elias , k. b. sprague , y. xue and a. h. fariborz , nucl . a * 671 * ( 2000 ) 416 [ hep - ph/9909475 ] . k. maltman , phys . b * 462 * ( 1999 ) 14 [ hep - ph/9906267 ] . e. van beveren , t. a. rijken , k. metzger , c. dullemond , g. rupp and j. e. ribeiro , z. phys . * c30 * ( 1986 ) 615 ; + n. a. tornqvist , z. phys . c * 68 * ( 1995 ) 647 [ hep - ph/9504372 ] . e. c. poggio , h. r. quinn and s. weinberg , phys . d * 13 * ( 1976 ) 1958 . f. le diberder and a. pich , phys . lett . b * 289 * ( 1992 ) 165 . c. a. dominguez and k. schilcher , phys . b * 448 * ( 1999 ) 93 [ hep - ph/9811261 ] . k. g. wilson , phys . * 179 * ( 1969 ) 1499 . k. g. chetyrkin , d. pirjol and k. schilcher , phys . b * 404 * ( 1997 ) 337 [ hep - ph/9612394 ] . m. beneke , phys . * 317 * ( 1999 ) 1 [ hep - ph/9807443 ] . k. g. chetyrkin , s. narison and v. i. zakharov , nucl . b * 550 * ( 1999 ) 353 [ hep - ph/9811275 ] . c. a. dominguez and k. schilcher , phys . d * 61 * ( 2000 ) 114020 [ hep - ph/9903483 ] . m. jamin and m. munz , z. phys . c * 66 * ( 1995 ) 633 [ hep - ph/9409335 ] . l. r. surguladze and f. v. tkachov , nucl . b * 331 * ( 1990 ) 35 . k. g. chetyrkin , c. a. dominguez , d. pirjol and k. schilcher , phys . d * 51 * ( 1995 ) 5090 [ hep - ph/9409371 ] . r. barate _ [ aleph collaboration ] , eur . j. c * 4 * ( 1998 ) 409 . c. a. dominguez , a. ramlakan and k. schilcher , hep - ph/0104262 . v. a. novikov , m. a. shifman , a. i. vainshtein and v. i. zakharov , nucl . b * 191 * ( 1981 ) 301 . e. m. ilgenfritz and m. muller - preussker , nucl . b * 184 * ( 1981 ) 443 . e. v. shuryak , nucl . b * 203 * ( 1982 ) 93 . e. v. shuryak , nucl . phys . b * 203 * ( 1982 ) 116 . v. elias , f. shi and t. g. steele , j. phys . g * g24 * ( 1998 ) 267 . k. m. watson , phys . * 88 * ( 1952 ) 1163 . s. l. adler , phys . rev . * 137 * ( 1965 ) b1022 . d. v. bugg , i. scott , b. s. zou , v. v. anisovich , a. v. sarantsev , t. h. burnett and s. sutlief , phys . b * 353 * ( 1995 ) 378 . a. abele _ et al . _ [ crystal barrel collaboration ] , phys . lett . b * 380 * ( 1996 ) 453 . d. v. bugg , b. s. zou and a. v. sarantsev , nucl . phys . b * 471 * ( 1996 ) 59 . b. hyams _ et al . _ , nucl . b * 64 * ( 1973 ) 134 [ aip conf . * 13 * ( 1973 ) 206 ] ; + w. ochs , ph.d . thesis , ludwig - maximilians - universitt , mnchen ( 1974 ) . m. r. pennington , _ the da@xmath152ne physics handbook , _ eds . l. maiani , g. pancheri and n. paver ( infn , frascati , 1992 ) vol . 379 - 418 ; + m. r. pennington , _ the second da@xmath152ne physics handbook , _ eds . l. maiani , g. pancheri and n. paver ( infn , frascati , 1995 ) vol . 531 - 558 . p. estabrooks and a. d. martin , phys . lett . b * 41 * ( 1972 ) 350 ; + p. estabrooks and a. d. martin , nucl . b * 79 * ( 1974 ) 301 . r. kamiski , l. leniak and k. rybicki , z. phys . c * 74 * ( 1997 ) 79 [ hep - ph/9606362 ] . r. omns , nuovo cim . * 8 * ( 1958 ) 316 . h. hellmann , _ einfrung in die quantenchemie _ , deuticke ( 1937 ) + r. p. feynman , phys . * 56 * , 340 ( 1939 ) . m. gell - mann , r. j. oakes and b. renner , phys . * 175 * ( 1968 ) 2195 . g. colangelo , j. gasser and h. leutwyler , hep - ph/0103088 . a. a. ovchinnikov and a. a. pivovarov , sov . j. nucl . phys . * 48 * ( 1988 ) 721 . j. prades , nucl . * 64 * ( 1998 ) 253 [ hep - ph/9708395 ] . k. maltman and j. kambor , [ hep - ph/0107060 ] . h. g. dosch and s. narison , phys . b * 417 * ( 1998 ) 173 [ hep - ph/9709215 ] . d. becirevic , p. boucaud , j. p. leroy , v. lubicz , g. martinelli and f. mescia , phys . b * 444 * ( 1998 ) 401 [ hep - lat/9807046 ] . s. aoki _ et al . _ [ jlqcd collaboration ] , phys . rev . lett . * 82 * ( 1999 ) 4392 [ hep - lat/9901019 ] . m. gockeler _ et al . _ , phys . d * 62 * ( 2000 ) 054504 [ hep - lat/9908005 ] . d. becirevic , v. gimenez , v. lubicz and g. martinelli , phys . d * 61 * ( 2000 ) 114507 [ hep - lat/9909082 ] . a. ali khan _ et al . _ [ cp - pacs collaboration ] , phys . * 85 * ( 2000 ) 4674 [ hep - lat/0004010 ] . n. eicker _ et al . _ [ sesam collaboration ] , phys . rev . d * 59 * ( 1999 ) 014509 [ hep - lat/9806027 ] . j. w. negele , nucl . * 73 * ( 1999 ) 92 [ hep - lat/9810053 ] . j. a. vermaseren , s. a. larin and t. van ritbergen , phys . b * 405 * ( 1997 ) 327 [ hep - ph/9703284 ] . t. van ritbergen , j. a. vermaseren and s. a. larin , phys . b * 400 * ( 1997 ) 379 [ hep - ph/9701390 ] . k. g. chetyrkin , phys . b * 404 * ( 1997 ) 161 [ hep - ph/9703278 ] .

liquid state theory experienced remarkable successes in the understanding of the properties of dense , simple fluids @xcite and more recently , of complex liquids , like colloidal suspensions or globular polymers @xcite . however a fully quantitative description of phase transitions can not be obtained by standard integral equations or diagrammatic expansions and requires specially devised approaches . the freezing transition , often dominated by geometrical constraints , is usually understood on the basis of density functional theory @xcite , while fluctuation induced phenomena , like the liquid - vapour transition , need the completely different theoretical framework provided by the renormalization group method @xcite . the hierarchical reference theory of fluids ( hrt ) was formulated more than 25 years ago as an attempt to reconcile the renormalization group ( rg ) approach , an extremely general and successful theory of second order phase transitions formulated in field - theoretical language , and the microscopic point of view adopted within liquid state theory @xcite . the main goal of hrt was to develop a unified theory able to reproduce the universal properties at the critical point , as obtained by rg techniques , together with the non universal features of realistic models of fluids , present both in the critical region ( e.g. critical density and temperature , short range correlations ) as well as far from the critical point . a comprehensive presentation of the hrt formalism , including a detailed derivation of the basic equations and an extensive discussion of its critical properties , can be found in ref . the earliest applications of hrt to simple fluids and to the ising model are briefly discussed in the same reference . however , most of the more interesting results and extensions were developed later . by now it is therefore appropriate to summarize the advances in hrt which took place in the last decade , providing a guidance to the most significant recent applications . such a critical discussion of the hrt formalism also gives the opportunity to emphasize its relationship with other theoretical tools which have been independently developed in different frameworks . since the early nineties , sophisticated functional , non perturbative rg techniques ( nprg ) were devised within field theory , leading to a boost of applications to many and diverse physical situations : from particle physics , to quantum many body theory , to classical statistical mechanics @xcite . only recently it was recognized that , despite the different language and physical context , a deep relation ties hrt and nprg @xcite , whose developments have proceeded quite independently for 20 years in distinct scientific communities . to introduce the hierarchical reference theory of fluids it is convenient to refer to a system of @xmath0 classical particles confined in a volume @xmath1 , in spatial dimension @xmath2 , interacting via a pairwise potential @xmath3 . generalizations to classical spin systems , quantum fluids and magnets do not present additional difficulties and will be examined in the next sections . the grand canonical partition function @xmath4 of the model is written as : @xmath5= \sum_n \frac{z^{n}}{n!}\int { \rm d } { \bf r_1} ... {\rm d } { \bf r_n } \ , e^ { -\beta\sum_{i < j } v({\bf r}_i - { \bf r}_j)-\beta\sum_i u({\bf r}_i ) } \label{gran}\ ] ] where the fugacity @xmath6 is related to the chemical potential @xmath7 and to the thermal wavelength @xmath8 by @xmath9 with @xmath10 . an external field @xmath11 is also included for convenience and @xmath4 is considered as a functional of @xmath11 . a detailed derivation of the hrt equations and a thorough discussion of their properties can be found in ref . @xcite via diagrammatic methods @xcite . here we provide a brief introduction of the physical content of the method together with a simpler , less technical derivation . since the pioneering works by van der waals , a large part of liquid state theory is based on the idea that the strongly repulsive short range interaction , distinctive of the interatomic potentials , provides a constraint to the available phase space , determining an important contribution to the entropy of the fluid , while the mainly attractive , regular part of @xmath3 dominates the internal energy . this is certainly true for the special , although prototypical , case in which @xmath3 has hard core at short distances , but holds also in the general case of soft core potentials . the different role played by the singular and regular contributions to the interparticle potential is manifest in lattice gas models , where particles live on the sites of a lattice : hard core repulsion between particles gives rise to the single occupancy constraint , which allows to map the lattice gas onto an ising model with a spin - spin interaction proportional to the regular part of the interparticle potential . mean field approaches developed for specific models , like van der waals equation in simple fluids or weiss theory in magnets , suggest that only the latter interaction triggers the order parameter fluctuations , responsible for the order - disorder transitions . in fluid models , both the singular and the regular part contributions are included in the interaction @xmath3 which should then be split in a reference , singular , part @xmath12 and in the regular remainder @xmath13 . the partition function ( [ gran ] ) can be written in a suggestive form by introducing the configurational density " @xmath14 whose thermal average is just the density profile @xmath15 , which reduces to the ordinary average density @xmath16 for @xmath17 . in the following , the properties of the reference system will be considered as known . the fluctuating order parameter of the liquid - vapour phase transition in fluids is then identified as the configurational density and @xmath4 is written as @xmath5 = \sum_n\frac{(z\,e^{-\frac{1}{2}\phi(0)})^{n}}{n!}\ , \int { \rm d } { \bf r_1} ... {\rm d } { \bf r_n } \ , e^{-\beta h_r}\ , e^{\frac{1}{2}\int { \rm d}{{\bf x}}_1 \,{\rm d}{{\bf x}}_2 \phi({{\bf x}}_1-{{\bf x}}_2 ) \ , \hat\rho({{\bf x}}_1 ) \,\hat\rho({{\bf x}}_2 ) } \label{xi1}\ ] ] where the regular dimensionless interaction has been defined as @xmath18 and the reference hamiltonian is @xmath19 according to the previous remarks , @xmath20 defines a non - trivial integration measure , limiting the phase space available to the particles . by factorizing the reference @xmath0-particle canonical partition function @xmath21 , ( [ xi1 ] ) can be also expressed as : @xmath5 = \sum_n\frac{(z\,e^{-\frac{1}{2}\phi(0)})^{n}}{n!}\,e^{-\beta\,a_r(n ) } \,\left \langle \exp\left ( \frac{1}{2}\int \frac{{\rm d}{{\bf q}}}{(2\pi)^d } \,\tilde\phi({{\bf q}})\ , \hat\rho({{\bf q}})\,\hat\rho(-{{\bf q}})\right ) \right \rangle_r \label{xi2}\ ] ] where the average @xmath22 refers to the gibbs measure of the reference fluid , and the fourier transforms of the potential @xmath23 and of the density fluctuations @xmath24 have been defined . such a form is particularly suitable as a starting point for a reformulation of wilson s renormalization group picture in hamiltonian systems . in the momentum space rg formulation @xcite , a sequence of systems , labeled by a wave - vector @xmath25 , is defined by allowing density fluctuations only at wave - vectors @xmath26 larger than the cut - off @xmath25 . the evolution of the physical properties of the model is then followed starting from the limit @xmath27 , where the system reduces to the reference one with a shift @xmath28 in the chemical potential , to the final value @xmath29 , where the cut - off is removed and the original partition function @xmath4 is retrieved . expression ( [ xi2 ] ) can easily accommodate the presence of a cut - off @xmath25 : @xmath30 = \sum_n\frac{(z\,e^{-\frac{1}{2}\phi(0)})^{n}}{n!}\,e^{-\beta\,a_r(n ) } \,\left \langle \exp\left ( \frac{1}{2}\int \frac{{\rm d}{{\bf q}}}{(2\pi)^d } \,\tilde\phi({{\bf q}})\,\theta ( q - q)\ , \hat\rho({{\bf q}})\,\hat\rho(-{{\bf q}})\right ) \right \rangle_r \label{xiq}\ ] ] where @xmath31 is the heaviside step function . this definition embodies basic principle of hrt : a cut - off in density fluctuations is equivalent to a cut - off in the regular part of the two body interaction for the evaluation of the fluctuation corrections to the free energy , leading to the definition of a cut - off dependent potential @xmath32 . it is customary to perform a legendre transform defining a cut - off dependent helmholtz free energy @xmath33 as a functional of the density profile @xmath15 : @xmath34 = -kt\,\log \xi_q[u({{\bf r } } ) ] - \int { \rm d}{{\bf r}}\ , \left ( u({{\bf r}})-kt\,\log z\right ) \,\rho({{\bf r } } ) \label{leg0}\ ] ] finally , it is also convenient to include in the definition of the free energy functional the mean field contribution due to the residual interaction @xmath35 : @xmath36 = a_q[\rho({{\bf r } } ) ] + \frac{1}{2}\int { \rm d}{{\bf r}}\,{\rm d}{{\bf r}}^\prime \left [ w({{\bf r}}-{{\bf r}}^\prime ) - w_q({{\bf r}}-{{\bf r}}^\prime)\right ] \ , \rho({{\bf r } } ) \,\rho({{\bf r}}^\prime ) \label{leg}\ ] ] in this way , for @xmath37 , @xmath38 reduces to the mean field approximation of the full free energy , which is recovered as @xmath39 . this way to artificially eliminate the long wave - length fluctuations from the thermodynamics is referred to as sharp cut - off " . however it is often more convenient to deal with a smooth cut - off " procedure , where the @xmath40-function in the definition of @xmath41 is substituted by a smoothed step function , denoted in the following as @xmath42 which is continuous and differentiable for every @xmath43 and tends to the step function in the @xmath44 limit : @xmath45 after having defined the cut - off dependent free energy , @xmath38 via eq . ( [ leg ] ) , the next step is to write the exact evolution ( or flow ) equation governing the change in the physical properties of the model when the cut - off @xmath25 is varied . this goal is most easily achieved in the smooth cut - off formulation , where an infinitesimal increase of the cut - off @xmath25 leads to an infinitesimal change in the potential @xmath41 . according to standard first order perturbation theory @xcite we get , in a uniform fluid , @xmath46 where @xmath47 is the total correlation function of a fluid interacting through a potential @xmath48 and @xmath49 . this smooth cut - off hrt equation has been obtained for the first time in ref . @xcite as an attempt to apply the renormalization group ideas to microscopic models . note that the total correlation function @xmath50 rather than the radial distribution function @xmath51 appears in the evolution equation ( [ evol1 ] ) because the mean field contribution has been subtracted in the definition of the @xmath25-dependent free energy ( [ leg ] ) . the total correlation function can be conveniently expressed in terms of the direct correlation function @xmath52 by use of the ornstein - zernike relation : @xmath53 the limit @xmath54 in eq . ( [ wqsmooth ] ) requires some care , as already recognized in the description of the sharp cut - off hrt equation and also discussed in field theoretical approaches @xcite , because of the emergence of a discontinuity in @xmath55 precisely on the same surface @xmath56 where the @xmath57-function singularity develops in @xmath58 . such a discontinuity originates from the random phase contribution to the exact direct correlation function , being the single term where the regular part of the potential appears outside any integration . if @xmath59 denotes the direct correlation function of the reference fluid , the first terms of the perturbative expansion read @xmath60 all the remaining contributions , being integrated , do not display discontinuities in the whole @xmath26-space . this suggests to introduce a modified correlation function @xmath61 which remains continuous also in the sharp cut - off limit @xmath54 : @xmath62 interestingly , this newly defined function is just the second functional derivative of the free energy functional @xmath63 $ ] with respect to density variations , evaluated at uniform density : @xmath64 where the last term is due to the ideal gas contribution , included in our definition of @xmath38 . we first write the right hand side of eq . ( [ evol1 ] ) by introducing the fourier transforms : @xmath65 } \ , \frac{\partial \tilde\phi_q({{\bf q}})}{\partial q } \label{smooth}\ ] ] the easiest way to take the sharp cut - off limit of this flow equation is to introduce the differential operator @xmath66 acting only on the q - dependence of @xmath67 ( while @xmath61 is treated as @xmath25-independent ) : @xmath68 \label{evol2}\ ] ] taking the @xmath54 limit and writing @xmath69 as a finite difference , eq . ( [ evol2 ] ) is then written in the form @xcite : @xmath70 \label{sharp}\ ] ] where the integration is limited to the spherical surface ( in dimension @xmath2 ) defined by @xmath56 and the function @xmath71 is defined by the ornstein - zernike relation as : @xmath72 equations ( [ smooth ] ) and ( [ sharp ] ) govern the hrt flow of the free energy in the smooth and sharp formulations respectively . as previously mentioned , the initial condition at @xmath27 coincides in both cases with the known mean field approximation , while for @xmath39 all density fluctuations are included and the exact free energy is in principle recovered if @xmath71 is known . hrt then provides the exact evolution of the free energy in terms of the two body correlation function @xmath73 ( or equivalently @xmath71 ) whose flow is determined by similar equations involving the three and four particle correlations @xcite . these equations , which can be derived along the same lines previously sketched , are most easily obtained by functional differentiation of ( [ smooth ] ) and ( [ sharp ] ) with respect to @xmath15 leading to a hierarchy of differential equations describing the rg flow of the free energy and of the @xmath74-particle direct correlation functions @xmath75 , formally defined as the @xmath76 functional derivative of @xmath77 with respect to the density profile @xmath15 evaluated at uniform density @xmath78 . the ideal gas contribution is included in our definition of @xmath38 leading to an additional term @xmath79 with respect to the standard definition of the many particle direct correlation functions @xcite . as an example , we explicitly report the flow equations for the previously defined two point function in smooth cut - off formulation @xcite @xmath80 and in sharp cut - off formulation @xcite : @xmath81 where , in the latter equation , @xmath82 is defined as : @xmath83^{-1 } & \text{for $ \vert { { \bf k}}+{{\bf q}}\vert < q$ } \\ { \cal f}_q({{\bf k}}+{{\bf q } } ) & \text{for $ \vert { { \bf k}}+{{\bf q}}\vert > q$ } \end{cases}\ ] ] note that , in the limit @xmath84 , these flow equations reproduce exactly the second density derivative of the corresponding free energy evolution ( [ smooth],[sharp ] ) , showing that the density dependence of the free energy flow encodes some ( partial ) information contained in the full infinite hierarchy . since the first derivation of the hrt equations in the sharp cut - off formulation @xcite a close similarity to some previous differential rg formulation was noticed . in particular it was shown that close to a critical point , in the long wave - length limit the full hrt hierarchy precisely reproduces the flow equations for the one particle irreducible rg generator @xcite . the analogy between the hierarchical reference theory and functional rg approaches is however even deeper , as recently shown by caillol @xcite . to highlight the connection between the two approaches it is convenient to start from the definition of the partition function @xmath85 ( [ xiq ] ) . as pointed out in an early attempt to apply rg techniques to microscopic models @xcite , by performing a hubbard - stratonovich transformation , the partition function can be expressed as a functional integral over an auxiliary scalar field @xmath86 governed by an effective action @xmath87 $ ] . this commonly adopted procedure , however , is rather formal and does not lend a well defined physical meaning for the hubbard stratonovich field @xmath86 . it is possible to circumvent this problem by first introducing a reference effective action @xmath88 $ ] by : @xmath89 = -\log \left [ \sum_n\frac{z^{n}}{n!}\ , \int { \rm d } { \bf r_1} ... {\rm d } { \bf r_n } \ , e^{-\beta\sum_{i < j } v_r({\bf r}_i - { \bf r}_j)}\ , \prod_{{\bf q}}\delta(\hat\rho({{\bf q}})-\psi_{{\bf q } } ) \right ] \label{sr}\ ] ] this definition clearly forces the effective field @xmath86 to faithfully represent density fluctuations . the exact partition function is then written as @xmath90 & = & \int { \cal d } \psi \ , e^{-s_q[\psi_{{\bf q } } ] } \\ s_q[\psi_{{\bf q } } ] & = & s_r[\psi_{{\bf q } } ] - \frac{1}{2v}\,\sum_{{\bf q}}\tilde\phi_q({{\bf q}})\ , \psi_{{\bf q}}\,\psi_{-{{\bf q } } } + \frac{1}{v } \sum_{{\bf q}}\left ( \frac{\phi(0)}{2 } + \beta \tilde u({{\bf q}})\right ) \psi_{-{{\bf q } } } \label{action}\end{aligned}\ ] ] this expression shows that @xmath87 $ ] can be written as the sum of the effective action in the absence of any cut - off @xmath91 $ ] ( i.e. the value attained at @xmath29 ) plus a momentum dependent mass term " @xcite : @xmath92 = s[\psi_{{\bf q } } ] + \frac{1}{2v}\,\sum_{{\bf q}}r_q({{\bf q } } ) \ , \psi_{{\bf q}}\,\psi_{-{{\bf q}}}\ ] ] with @xmath93 \label{mass}\ ] ] following wetterich we now perform a modified legendre transform " @xcite introducing the field @xmath94 conjugated to the external potential @xmath95 , which , due to the definition ( [ action ] ) coincides with the statistical average of the field @xmath86 : @xmath96 = -\log\xi_q + \frac{1}{v}\,\sum_{{\bf q}}\left ( \tilde u({{\bf q}})-kt\,\log z\right ) \,\rho_{-{{\bf q}}}- \frac{1}{2v}\,\sum_{{\bf q}}r_q({{\bf q}})\ , \rho_{{\bf q}}\,\rho_{-{{\bf q } } } \label{leg1}\ ] ] comparing this definition to eq . ( [ leg ] ) it is clear that the average action coincides with the modified free energy defined in hrt : @xmath97 , providing the link between hrt , based on the gradual inclusion of the fourier components of the regular part of the physical interaction , and the effective average action " rg method built upon the idea of adding a cut - off and momentum dependent mass term @xmath98 to the original action . from eq . ( [ mass ] ) it follows that for @xmath29 the mass term vanishes identically , while at fixed @xmath25 , @xmath98 is large only for @xmath99 providing an effective cut - off for the long wave - length fluctuations , as requested in all the implementations of the momentum space rg ideas @xcite . a difference in the initial condition however persists between the average action method and hrt : in the rg approach , @xmath98 diverges as @xmath37 , thus providing an infinite mass to all fluctuations and leading to a soluble problem . instead in hrt we get @xmath100 : for this special but finite value of the mass term the regular part of the interaction cancels exactly in @xmath87 $ ] and the model reduces to the soluble reference system . the procedure adopted in hrt is reminiscent of some recent development in functional rg methods @xcite . the solution of the infinite hierarchy of equations previously derived needs some approximation scheme . two kinds of approximations are customary in this type of problems : truncations and closures . a truncation of the hierarchy simply corresponds to neglecting @xmath74-particle correlations for @xmath74 larger than a maximum value @xmath101 ( typically @xmath102 ) . such a procedure has been followed in the early investigations of rg equations but it is rigorously justified only near a critical point in spatial dimension close to @xmath103 @xcite , while it is not expected to provide an accurate description of a system , like a liquid , where short range correlations are important . instead , most of the applications of hrt , both in the smooth and sharp cut - off formulations , adopted a non perturbative closure to the first equation of the hierarchy ( [ smooth],[sharp ] ) and only this type of approximation will be discussed here . non perturbative closures in fact allow to preserve the convexity of the free energy , providing a physically satisfactory description of phase transitions between homogeneous phases , together with a picture of the critical region fully consistent with the scaling hypothesis . the evolution equation for the free energy involves the knowledge of the two particle direct correlation function in the presence of a cut - off @xmath25 . a closure is a parametrization of the direct correlation function expressing @xmath104 in terms of the free energy density @xmath105 . as previously noticed , the compressibility relation , which follows directly from eq . ( [ second ] ) , @xmath106 plays a special role in hrt because , when implemented into the free energy evolution equation , allows to include in the very first equation some information contained into the whole hierarchy . the simplest non perturbative closure of the first hrt equation is inspired by the well known random phase approximation ( rpa ) in liquid state theory @xcite : @xmath107 the exact initial condition at @xmath37 is obtained for @xmath108 , while the compressibility relation ( [ komp ] ) can be enforced at all @xmath25 via a suitable choice of the parameter @xmath109 . when this closure is substituted into eq . ( [ smooth ] ) or ( [ sharp ] ) the resulting partial differential equation for the free energy @xmath38 as a function of density @xmath16 and cut - off @xmath25 can be solved for a given value of the temperature . interestingly , this approximate closure can be equivalently formulated starting from the evolution equations ( [ smooth ] ) or ( [ sharp ] ) for the full functional @xmath110 $ ] @xcite : by restricting its form through a suitably chosen parametrization , the functional flow equation gives rise to partial differential equations for the parameters . for instance , the rpa - like closure is obtained by use of a simple weighted density approximation @xcite for the free energy : @xmath111 = a_r[\rho({{\bf r } } ) ] + \int { \rm d}{{\bf r}}\,f_q(\bar\rho({{\bf r } } ) ) \label{wda}\ ] ] where @xmath112 $ ] is the exact free energy functional of the reference system and the function @xmath113 is the parameter whose evolution will be determined by the flow equation . the weighted density in eq . ( [ wda ] ) @xmath114 is defined by the local average : @xmath115 with a normalized weight function given by @xmath116^{1/2}$ ] @xcite . by substituting this form into the evolution equation ( [ smooth ] ) or ( [ sharp ] ) , we obtain the same expression found by use of the ansatz ( [ rpa ] ) if we identify @xmath117 . this class of closures , in the nprg framework , are known as local potential approximation ( lpa ) because of the similarity between eq . ( [ wda ] ) and the usual local density approximation of density functional theory . clearly , more refined approximations to the direct correlation function , or equivalently to the free energy functional , may be devised , leading to more accurate representation of the exact free energy flow . in fluids with a hard sphere reference system ( in the continuum or on a lattice ) , ( [ rpa ] ) can be generalized by including a further contribution , non vanishing only inside the hard sphere diameter , so as to satisfy the core condition @xmath118 . this constraint can be implemented exactly in the smooth cut - off formulation for yukawa potentials @xcite thanks to the exact analytical solution of the ornstein - zernike equation for this model @xcite . analogous closures might be implemented on the flow equation for the two particle direct correlation function ( [ csmooth ] ) or ( [ csharp ] ) . a closure to the second equation of the hierarchy amounts to expressing the three and four particle correlations in terms of the two point function . a possible closure has been proposed already in the earliest stage of hrt @xcite but it has not been implemented yet : @xmath119 \label{c4}\end{aligned}\ ] ] this class of approximations has the advantage of allowing for a non trivial anomalous dimension exponent at the critical point @xcite . a similar closure has been independently proposed in the context of functional rg @xcite and a simplified form has been numerically investigated in the same framework @xcite . up to now , all the applications of hrt to microscopic models adopted a closure of the form ( [ rpa ] ) with or without the additional optimization of the potential inside the hard sphere diameter to enforce the core condition . the self consistent ornstein zernike approximation ( scoza ) is a well known liquid state theory which proved successful in the determination of phase diagrams in fluids characterized by a pair interaction written as the sum of a singular hard sphere part plus a yukawa tail @xmath120 @xcite . originally , scoza has been formulated as a generalization of the mean spherical approximation @xcite for the direct correlation function outside the core : @xmath121 while enforcing the core condition @xmath122 to define @xmath123 inside the hard sphere diameter . the parameter @xmath124 is determined by requiring thermodynamic consistency between the internal energy and the compressibility routes to thermodynamics : @xmath125 the two exact expressions ( [ scozac ] ) , together with the core condition and the parametrization ( [ scoza ] ) outside the hard sphere diameter , give rise to a partial differential equation for the free energy as a function of @xmath16 and @xmath126 . remarkably , this equation is exactly reproduced by hrt in the smooth cut - off formulation ( [ smooth0 ] ) by use of a linear turning on of the interaction ( [ wqsmooth ] ) : @xmath127 where @xmath128 is the switching - on parameter @xcite . clearly this form does not satisfy the requirements stated in section [ sechrt ] for a selective introduction of density fluctuations : while scoza just tunes the amplitude of @xmath129 , hrt changes simultaneously both amplitude and range of the potential by imposing a cut - off @xmath25 on its long wave - length fourier components thereby qualifying as a microscopic implementation of rg ideas . in fact , scoza has been shown to provide a good description of the non universal properties in the critical region , as well as non classical critical exponents but fails in reproducing the correct scaling picture predicted by rg and hrt @xcite . an attempt to reconcile hrt and scoza has been formulated in ref . @xcite , where a further adjustable parameter is added in the rpa - like closure of the hrt equation , to enforce thermodynamic consistency between the compressibility and the internal energy route . one of the goals which motivated the formulation of hrt was to provide a description of the critical properties of realistic fluid models consistent with the rg picture . critical exponents , governing the singularities of thermodynamic quantities in the critical region , acquire universal values quoted in table [ table1 ] @xcite . at the critical density , the constant volume specific heat diverges as @xmath130 as the reduced temperature @xmath131 vanishes ; the isothermal compressibility as @xmath132 and the correlation length as @xmath133 . the shape of the coexistence curve is asymptotically given by @xmath134 , while the equation of state on the critical isotherm is given by @xmath135 . finally , precisely at the critical point , the pair correlation displays a power law tail of the form @xmath136 . in the sharp cut - off formulation , universality is indeed a direct consequence of the asymptotic form of the hrt hierarchy close to the critical point @xcite : all the microscopic details of the model disappear from the asymptotic equations and the full hierarchy acquires a universal form equivalent to a previously studied rg differential generator @xcite . scaling laws and non classical values for the critical exponents immediately follow from this identification . in the smooth cut - off formulation , the universality of critical phenomena is less apparent because the specific form of the chosen cut - off function @xmath137 ( [ wqsmooth],[mass ] ) survives also in the asymptotic critical region . universality of the critical properties can then be proved only order by order in the @xmath138 expansion and no general proof is available @xcite . here we will briefly review the critical properties of the approximate non perturbative class of closures previously introduced both in the sharp and in the smooth cut - off formulation ( [ rpa ] ) . moreover we will sketch the argument leading to the conclusion that already at the level of this simple approximation , hrt always gives a convex free energy , even at coexistence . the unphysical divergence of the compressibility at the phase boundary predicted by the sharp cut - off closure @xcite is circumvented by the smooth cut - off formulation which then provides a satisfactory description of first order transitions @xcite . other , more sophisticated closures have been proposed . as previously noticed , the second equation of the hrt hierarchy describes the evolution of the two particle direct correlation function in terms of the three and four particle correlations ( [ csmooth],[csharp ] ) . by use of the closure ( [ c3],[c4 ] ) it has been shown that the critical exponents are correctly reproduced to second order in @xmath138 , including a non vanishing anomalous dimension @xmath139 @xcite . unfortunately this quite promising non perturbative closure has not been thoroughly investigated in microscopic models , probably due to the problems posed by the numerical solution of the resulting non linear partial differential equation . recently , a similar ( although simpler ) approximation has been studied in the asymptotic regime , in the framework of nprg , leading to encouraging results for the universal quantities @xcite . in the sharp cut - off hrt , the free energy evolution equation ( [ sharp ] ) depends on the two point function @xmath71 evaluated on shell " , i.e. on the surface @xmath56 . this means that in the late stage of the integration , when @xmath25 is small , only the long wave - length physics contributes to the free energy flow . moreover , in the critical region , @xmath71 diverges at small @xmath140 and then we can disregard the additive term in eq . ( [ sharp ] ) and the sub - leading @xmath141 contribution . finally , the evolution equation acquires a universal form @xmath142 where @xmath143 is a geometric factor equal to the ratio between the volume of the unit sphere in @xmath2 dimensions and @xmath144 . when we insert a closure of the form ( [ rpa ] ) at right hand side , recalling eq . ( [ calf ] ) , we need to evaluate the direct correlation function at wave - vector @xmath39 . by exploiting the analyticity of @xmath145 and @xmath146 around @xmath147 , we can write @xmath148 where @xmath149 determines the curvature of the direct correlation function in this approximation and attains a finite , positive limit at the critical point . recalling the compressibility sum rule ( [ komp ] ) and inserting this asymptotic form into eq . ( [ sharp2 ] ) we obtain a closed partial differential equation expressing the free energy flow at long wave - length and in the critical region . the rg structure of this equation can be highlighted by a suitable rescaling , as shown in refs . @xcite , which allows the explicit calculation of the universal properties , like critical exponents and scaling functions , via the fixed point analysis . the analyticity in the momentum dependence of the direct correlation function , implied by the rpa - like ansatz ( [ rpa ] ) , immediately gives the vanishing of the anomalous dimension , or fisher exponent , @xmath150 in any spatial dimension . this result is exact for @xmath151 , when the critical exponents are classical , while for @xmath152 such a simple non perturbative approximation to the first hrt equation gives critical properties correct to first order in the @xmath138 expansion . however this closure leads to a qualitatively incorrect prediction in @xmath153 , when the anomalous dimension is crucial in ensuring the existence of a fixed point solution . few critical exponents in @xmath154 are reported in table [ table1 ] together with the accepted values obtained from extrapolations of the high temperature expansion in the ising model @xcite . 0.2 cm .hrt estimates of the critical exponents and compressibility amplitude ratio in three dimensions for the sharp and the smooth formulations of hrt compared to the exact values @xcite obtained by extrapolation of high - temperature series expansions . the smooth cut - off procedure adopted here is detailed in section [ secsmooth ] . [ cols="^,^,^,^,^,^,^,^",options="header " , ] the ising model on a cubic lattice was the first lattice system investigated by sharp cut - off hrt @xcite by use of a more sophisticated closure , belonging to the class previously discussed , which consistently preserves the spin normalization constraint @xmath155 . a thorough analysis of lattice models via the smooth cut - off version of hrt has not been attempted yet , although a very similar procedure , put forward in the framework of non perturbative renormalization group , has been recently developed @xcite . the results shown in @xcite provided a very accurate estimate of the critical temperature of the ising model and compared favourably with the coexistence curve predicted by high temperature expansion . a numerical study of the same hrt equation aimed at determining the effective parameters of a scalar field theory able to reproduce the correct pre - critical behaviour of the ising model was performed in ref . interestingly , hrt showed that a @xmath156 representation of the long wave - length spin fluctuations was inadequate for a faithful representation of the non - universal properties of the ising model , while the inclusion of a @xmath157 term allowed to considerably improve the description of the critical region of the model . in ref . @xcite the hrt formalism was generalized to the case of an uniaxial spin model where @xmath158 can assume three values : @xmath159 . this system can be easily mapped into a symmetric binary mixture on a lattice with opposite signs for like and unlike interactions . if the coupling is limited to nearest neighbours , the blume - capel model is obtained , while for long range interactions decaying as @xmath160 , the lattice restricted primitive model ( lrpm ) is recovered . the phase diagrams of both models in the symmetry plane ( zero magnetization ) were obtained by integrating the hrt flow equation @xcite . the topologies of the phase boundaries in the density - temperature plane , shown in fig . [ figlrpm ] for the lrpm , turn out to be very similar : at high temperatures a critical @xmath124-line separates a charge - ordered fluid phase at low density from a homogeneous plasma phase . the critical line ends at a tricritical point where it merges into the liquid - vapour coexistence curve . the comparison with available simulation data for the lrpm shows that the structure of the phase diagram and the universality classes of the transitions are correctly reproduced , although some quantitative discrepancy is present , suggesting that a better representation of pair correlations should be devised . plane . hrt results : the tricritical point is marked by an open star . full dots give the @xmath124-line above the tricritical temperature and the liquid - vapour binodal at lower temperatures . the scoza data for the @xmath124-line are shown by the dotted line . simulation data from refs . @xcite are represented by filled triangles and open squares respectively . , width=264,height=264 ] the hrt formalism can be easily generalized to the case of binary mixtures @xcite : the flow equations in the smooth ( [ smooth ] ) and sharp cut - off ( [ sharp ] ) formulation remain valid for mixtures provided the two body interaction @xmath146 and the pair correlations @xmath71 and @xmath61 are substituted by matrices whose indices run over the two species . the ornstein - zernike relation ( [ calf ] ) acquires the form @xmath161 = \delta^{ik } \label{calfmat}\ ] ] while the compressibility sum rule ( [ komp ] ) becomes a set of three independent equations which can be written as : @xmath162 the derivation of the hrt evolution equations straightforwardly follows the same steps sketched in section [ sec : exact ] . by adopting the smooth cut - off procedure , the resulting expression is : @xmath163^{-1 } \ , \frac{\partial \tilde\phi_q({{\bf q}})}{\partial q } \right \ } \label{smoothmat}\ ] ] where matrix products are understood at right hand side and tr " stands for the trace operation . in the sharp cut - off limit this form reduces to @xmath164 \label{sharpmat}\ ] ] the trace of the logarithm at right hand side of eq . ( [ sharpmat ] ) must be interpreted in matrix sense and coincides with the logarithm of the determinant of the @xmath165 matrix @xmath166 $ ] . by use of a rpa - like closure of the form ( [ rpa ] ) , the flow equation then becomes a partial differential equation in three independent variables : the cut - off @xmath25 and the densities of the two species @xmath167 . the analysis of criticality in binary mixtures is performed following the procedure developed in simple fluids @xcite : several universality classes are predicted by the hrt equations , but most of the possible fixed points turn out to be unstable and the critical properties at a generic point along a transition line is of the ising type . actually , two distinct fixed points correspond to critical exponents related to the case of a one component scalar order parameter . one is weakly unstable and gives rise to strictly ising - like critical properties , while the other is stable and correctly reproduces the so called phenomenological approach by fisher which , starting from the concept of field mixing , predicts the occurrence of some weak exponent renormalization @xcite : the long wavelength limit of a structure factor @xmath168 at the critical density and concentration diverges with an exponent @xmath169 ( with @xmath170 ) , a @xmath171 larger than the corresponding value for one component fluids , while the isothermal compressibility displays a weak divergence governed by the exponent @xmath172 . hrt allows to identify the thermodynamic perturbation corresponding to the crossover between these two fixed points @xcite , thereby providing a microscopic interpretation of the experimental difficulty in observing the true asymptotic critical behaviour . special points in the phase diagram might display genuinely different critical indices , like un - renormalized ising or tricritical exponents . hrt also naturally suggests an unbiased definition of the order parameter at a critical point of binary fluids @xcite , which in general can not be simply identified either as density or concentration , because critical fluctuations are expected to affect both quantities . the critical point in binary fluids is signaled by the divergence of the leading eigenvalue of the ( symmetric ) matrix of the structure factors @xmath173 for @xmath84 . the structure factors are a direct measure of density fluctuations and can be expressed as the equilibrium average of a product of density operators @xmath174 . to discriminate between demixing and liquid - vapour transitions in binary mixtures it is convenient to introduce suitable linear combinations of @xmath173 , the bathia - thornton structure factors @xcite , which properly characterize the density and concentration fluctuations : @xmath175 where @xmath176 , @xmath177 are the number concentrations of the two species and the total density , respectively . let us denote by @xmath178 leading eigenvalue of the @xmath165 matrix built with the previously defined bathia thornton structure factors in the @xmath84 limit , and @xmath179 be the corresponding normalized eigenvector . according to these definitions , the diverging eigenvalue provides a measure of the fluctuations of the particular linear combination of density and concentration @xmath180 ( for @xmath84 ) : @xmath181 , suggesting the identification of @xmath182 as the order parameter of the transition . it is convenient to pictorially represent the order parameter as an arrow in the @xmath183 plane pointing in the direction identified by @xmath40 . remarkably , the mixing angle " @xmath40 is related to well defined thermodynamic quantities @xcite : @xmath184_{coex}&=&\rho\tan\theta \quad . \label{termo2}\end{aligned}\ ] ] the first equation is written in terms of the partial molar volumes @xmath185 which attain a finite limit at the critical point . the second , equivalent , relation expresses @xmath40 in terms of the differences of concentration @xmath186 and molar volumes @xmath187 at coexistence on an isotherm isobar near the critical point . clearly , the direction of the order parameter is defined at each point of a critical line and varies continuously along that line . while in some mixture , the character of the transition remains almost unaltered by varying the thermodynamic state , in other cases it may instead experience considerable changes , passing from demixing to liquid - vapour by reducing the density . a plot of the mean field critical lines for two lennard - jones binary fluids mimicking two rare gas mixtures , shown in fig . [ figmix ] , illustrates these two possibilities . for convenience , the total packing fraction @xmath188 and the volume fraction @xmath189 are used in place of total density @xmath16 and number concentration @xmath190 . defined in the text : an arrow parallel to the density axis indicates a liquid - vapour transition , an arrow along the volume fraction axis , a demixing transition . , width=264,height=264 ] the numerical integration of the hrt equation poses some challenge due to the presence of three independent variables : cut - off @xmath25 , density @xmath16 and concentration @xmath190 . special attention must be paid to the stability requirements and to the possible occurrence of singularities induced by the presence of large coexistence regions in the phase diagram . a stable numerical code has been developed for two classes of models of binary fluids : the rare gas mixtures @xcite and the symmetric yukawa mixtures @xcite , which include the ising fluids , defined as hard spheres with an internal ising - like internal degree of freedom whose relative orientation specifies the sign of the pair interaction @xcite . in both cases , the hrt equation in the sharp cut - off formulation has been numerically integrated . here we just review few results for the model of rare gas mixtures , where extensive experimental data are available @xcite . the argon - krypton system is modeled as a mixture of particles interacting via lennard - jones ( lj ) potentials @xcite . each lj interaction has been preliminarily mapped into a reference hard sphere potential plus a lj attractive tail . the effective hard sphere diameter does not coincide with the lj parameter @xmath191 and is determined according to the standard barker - henderson prescription @xcite which gives rise to a weak temperature dependence . the bare lj parameters adopted in the numerical calculation have been fixed to reproduce the location of the critical points of the two pure components : @xmath192 , @xmath193 , @xmath194 k and @xmath195 k. the strength of the unlike attractive interaction has been set to @xmath196 k , and additivity of the effective hard sphere diameters has been assumed . plots of the coexistence regions in the packing fraction / molar concentration plane are given for two temperatures in figs . [ figarkr1-rx]-[figarkr2-rx ] , where the tie - lines , connecting thermodynamic states in equilibrium , are also shown . the argon - krypton mixture belongs to the topology of phase diagram illustrated in the left panel of fig [ figmix ] , where the critical points of the two species are connected by a line of critical points . . tie lines in the two - phase regions are shown . in the coexistence regions darker colors correspond to higher pressures . , width=226,height=226 ] . tie lines in the two - phase regions are shown . in the coexistence regions darker colors correspond to higher pressures . , width=226,height=226 ] a different view of the coexistence regions at the same temperatures can be seen in the molar concentration / pressure plane in fig . [ figarkr - xp ] : at the higher temperature pure argon is in the fluid phase , while at the lower temperature the coexistence regions of the two components merge and define a unique , connected domain . , right panel @xmath197 . experimental phase boundaries @xcite are shown as red dots . , width=264,height=264 ] a good agreement between theory and experiments @xcite can be appreciated for this system . finally , it is instructive to check that , within hrt , the full coexistence region collapses to a single line when intensive thermodynamic quantities are selected , as required by thermodynamics : in fig . [ figarkr - mup ] the same coexistence regions are plotted in a diagram where the difference between the chemical potentials of the two species and the pressure are considered as independent variables . , right panel @xmath197 . note the collapse of the full coexistence region into a single line . , width=264,height=264 ] the phase diagram of the neon - krypton mixture instead displays a different topology , belonging to the class depicted in the right panel of fig . [ figmix ] . the same lennard - jones model used for the argon - krypton mixture has been adopted with parameters appropriate for the neon - neon ( @xmath198 , @xmath199 k ) and neon - krypton interactions ( @xmath200 k ) . the coexistence regions of the two pure components remain separated in the whole phase diagram : while pure neon is in the gas phase in a large temperature range , the neon - krypton mixture displays phase coexistence between states of considerably different composition . the order parameter smoothly changes character signaling a liquid - vapour transition at low neon concentration and a demixing transition at higher pressures as already captured by the mean field calculations of fig [ figmix ] . the coexistence region in the three planes packing fraction / molar concentration , molar concentration / pressure and chemical potential / pressure are shown in figs . [ fignekr1-rx ] , [ fignekr2-rx ] , [ fignekr - xp ] and [ fignekr - mup ] respectively . . tie lines in the two - phase regions are shown . in the coexistence regions darker colors correspond to higher pressures . , width=226,height=226 ] . tie lines in the two - phase regions are shown . in the coexistence regions darker colors correspond to higher pressures . , width=226,height=226 ] , right panel @xmath201 . experimental phase boundaries @xcite are shown as red dots . , width=264,height=264 ] the comparison with the experimental data @xcite shows good agreement at not too high pressure , while some discrepancy is present at high pressures , where the repulsive part of the interaction plays an important role . as discussed in ref . @xcite , it is likely that a better representation of the soft core part of the interatomic potential must be employed for a fully quantitative description of demixing transitions at high pressures . , right panel @xmath202 . , width=264,height=264 ] hrt was originally developed in the framework of classical statistical physics but its formalism can be generalized to quantum hamiltonians . lattice spin models have been considered in ref . @xcite , where the reference system was identified as free spins in an external magnetic field which couples with the order parameter . quantum antiferromagnets on an hypercubic lattice in @xmath2 dimensions are described by the isotropic , nearest neighbour heisenberg hamiltonian @xmath203 and require the introduction of a staggered external field @xmath204 . here , @xmath205 are spin @xmath206 operators defined on the sites @xmath207 of the lattice and @xmath208 is the antiferromagnetic vector in @xmath2 dimensions . the reference system is defined by the hamiltonian ( [ heis ] ) for @xmath209 and gives rise to a solvable , single site problem . following the standard sharp cut - off formulation of hrt , the spin - spin interaction is gradually turned on , by including its fourier components starting from the boundary of the antiferromagnetic brillouin zone . the exact evolution equation for the free energy per site in sharp cut - off formulation was obtained in ref . @xcite and involves the two body dynamical spin - spin correlations in imaginary time : @xmath210 + \\ & & \log \left [ \left ( 1-{\cal f}_q^{xx}({{\bf k}},\omega_n ) \tilde j({{\bf k}})\right ) \ , \left ( 1-{\cal f}_q^{xx}({{\bf k}}+{{\bf g}},\omega_n ) \tilde j({{\bf k}}+{{\bf g}})\right ) + { \cal f}_q^{xy}({{\bf k}},\omega_n)\,{\cal f}_q^{xy}({{\bf k}}+{{\bf g}},\omega_n)\ , \tilde j({{\bf k}})\tilde j({{\bf k}}+{{\bf g } } ) \right ] \big \ } \nonumber \label{gian}\end{aligned}\ ] ] where @xmath211 is the fourier transform of the spin - spin nearest negighbour interaction and @xmath25 ranges from @xmath212 , where fluctuations are absent and the mean field approximation is recovered , to @xmath29 , where all fourier components are introduced . the structure of this equation is rather similar to its classical counterpart ( [ lsharp ] ) with two main differences : the presence of off diagonal two body correlations @xmath213 , due to the non commutativity of the spin components , and the summation over the matsubara frequencies @xmath214 . the imaginary time variable in fact naturally emerges when quantum perturbation theory is employed to evaluate the effects of an additional infinitesimal momentum shell in the spin - spin interaction . notice that in this formulation , the cut - off on fluctuations is set only on the spatial fourier components , while the matsubara frequency variable runs on the whole real axis for all @xmath25 s . the quantum hrt ( qhrt ) equation appropriate for the antiferromagnetic heisenberg model ( [ heis ] ) was studied by employing a simple closure similar in spirit to the classical rpa - like approximation ( [ lrpa ] ) : the structure of the dynamical correlations was evaluated at mean field level and only mass renormalization " was allowed by imposing thermodynamic consistency between the zero frequency limit of the dynamical correlations at wave - vector @xmath215 and the derivatives of the free energy @xmath38 with respect to the staggered magnetization . the explicit form of the adopted closure reads : @xmath216 where @xmath217 is the staggered magnetization , and the two cut - off dependent parameters @xmath218 are related to the transverse and longitudinal staggered susceptibilities : @xmath219\\ \frac{\partial^2}{\partial m_s^2 } \left ( \frac{{\cal a}_q}{n}\right ) & = & \mu_\parallel + \tilde j({{\bf g } } ) \end{aligned}\ ] ] in fig . [ figheis ] the constant field specific heat and the spontaneous magnetization of a heisenberg antiferromagnet on the cubic lattice below the critical temperature are shown for two choices of the spin @xmath206 : the extreme quantum value @xmath220 and the classical limit @xmath221 . monte carlo simulations for @xmath222 @xcite are also reported . for spin @xmath223 and @xmath220 . left panel : qhrt spontaneous magnetization for @xmath220 and @xmath223 ( classical limit ) . right panel : constant field specific heat at @xmath224 as a function of temperature compared with classical ( full dots ) @xcite and quantum ( full squares ) @xcite monte carlo simulations , respectively . in the @xmath223 limit the magnetization is normalized to @xmath206 and the specific heat to @xmath225 . , width=264,height=264 ] a more detailed comparison between theory and simulations can be found in @xcite , showing that the performances of qhrt have generally a lower quality with respect to the previously shown applications to classical models , suggesting that a better representation of static and dynamic spin correlations has to be developed : the different role of longitudinal and transverse correlations is probably not fully grasped by the adopted single mode approximation , which did not prove accurate enough to provide a consistent picture of the crossover between the renormalized classical and the quantum critical regime of the non linear sigma model @xcite . the effects of easy - axis anisotropy have been also investigated by use of qhrt @xcite , as well as the role of a weak next nearest neighbour interaction @xcite . however , a qhrt analysis of quantum ( i.e. zero temperature ) phase transitions , or of the occurrence of off - diagonal long range order has not been attempted yet . as shown in the related non perturbative rg approach for the bose - hubbard model @xcite , the investigation of bose condensation by rg methods requires a different cut - off procedure and it may be more convenient to set a cut - off on the kinetic energy term rather than on the two body interaction . according to this scheme , the reference system becomes a purely classical model , while quantum fluctuations are gradually introduced , leading to the appropriate free energy flow equation . the smooth cut - off formulation of hrt has not been applied to quantum systems yet , although the derivation of the flow equations does not present any additional difficulty . a closer connection between qhrt and non perturbative renormalization group methods may provide some clue on the structure of more refined closures to the flow equation . a proper description of first and second order phase transitions within liquid state theory requires a careful account of long wavelength fluctuations . the hierarchical reference theory of fluids was developed precisely to achieve this goal . while the general framework of the theory was set since the very first formulation of hrt , the implementations to specific physical systems required the development of approximation schemes which may vary according to the individual features of the model . in this review we tried to summarize the efforts in this direction carried out in the last ten years . among the features shared by all the non - perturbative closures of the hrt equation we may mention the correct scaling form of the thermodynamics and correlations in the critical region , the prediction of non - classical critical exponents and the exact implementation of the maxwell construction at first order transitions , showing that the inclusion of long wave - length fluctuations gives rise to coexistence regions at constant pressure and chemical potential . the latter feature is particularly valuable in the case of mixtures , where the _ ad hoc _ implementation of the equal tangent construction , required by most liquid state theories , may prove rather cumbersome . the smooth cut - off formulation of hrt , proposed several years ago , has been quantitatively investigated only recently . among its peculiar properties , it is worth mentioning the absence of the spurious divergence of the isothermal compressibility at the phase boundary : a problem which affects the sharp cut - off formulation . this allows the location of the spinodal loci within a renormalization group approach , as explictly shown for the yukawa fluid . extensions of the smooth cut - off hrt equations to the case of mixtures or quantum systems are straightforward and represent a fertile research field for future studies . the approximations to the hrt hierarchy we have been considering so far were limited to closures to the first equation , which provides the flow equation for the free energy of the model . more sophisticated closures , at the level of the second equation , which contains information on the two particle correlations , have been formulated but not quantitatively investigated yet . soft matter physics provides a stimulating environment for any microscopic theory able to capture the mechanisms leading to phase transitions . and hrt was applied to several systems of physical interest in this context . however , we feel that improved analytical representations of pair correlations in different frameworks have still to be developed to achieve a degree of accuracy comparable to simulation methods . for instance , a common feature of colloidal suspensions is the presence of short range attractive interactions , often represented by the sticky limit " . a closure of the form ( [ rpa ] ) , adopted in most of the implementations of hrt , implies a linear relation between the direct correlation function and the attractive part of the potential , which is known to provide poor results for short range interactions @xcite . extensions of hrt to lattice systems , mixtures and quantum models have been briefly covered in this review , but other interesting directions were explored in the literature , as the application of hrt to polymer physics or to disordered media @xcite . a special mention must be devoted to the relationship between hrt and non perturbative ( or functional ) renormalization group methods . the two approaches are in fact intimately related , as already noticed by caillol @xcite , and have been independently developed in different physical frameworks . we hope that the present review , which specifically devoted section [ sec : nprg ] to this problem , will contribute to clarify this issue providing a dictionary between the two , substantially equivalent , formulations . it is a pleasure to express our gratitude to all the former students and collaborators who helped us in these investigations , most notably d. pini , m. tau and f. lo verso . we acknowledge financial support by the cimaina foundation ( lr ) and by miur through a prin/08 grant ( ap ) . the flow equation for the free energy ( [ sharp ] ) , together with the definition ( [ calf ] ) and the parametrization of the direct correlation function ( [ rpa ] ) , supplemented by the compressibility sum rule ( [ komp ] ) form a closed partial differential equation ( pde ) for the free energy as a function of the density @xmath16 and the cut - off @xmath25 . as a practical example here we report the explicit equations for a model defined by a spherically symmetric interaction written as the sum of a reference potential @xmath226 ( and corresponding free energy @xmath227 ) and an attractive tail @xmath129 . as usual we set @xmath228 ; fourier transforms are denoted by tildes . the initial condition of the pde is set at some large value of @xmath229 , such that the pair interaction @xmath230 can be neglected for @xmath231 . the free energy at @xmath229 is then given by the mean field approximation : @xmath232 here and in the following , the ideal gas contribution is included in the definition of @xmath233 . the integration of the pde is then performed backwards , down to @xmath29 , where all fluctuations are included and @xmath38 coincides with the free energy of the model . a mesh is defined on the density axis @xmath234 . here @xmath235 is some relatively high value ( typically @xmath236 in units of the hard - sphere diameter ) and the free energy @xmath237 is propagated from @xmath25 to @xmath238 by discretizing the pde on the mesh . it is convenient to write the pde ( [ sharp ] ) in quasi - linear form by introducing the auxiliary quantity @xmath239 \label{defu}\ ] ] the evolution equation for this new variable can be straightforwardly obtained from eq . ( [ sharp ] ) : @xmath240 -\rho\,q^{d-1}\ , \frac{\tilde\phi(q)}{\tilde\phi(0)}\,\frac{d\,k_d}{2 } \,\frac{\partial^2 u_q}{\partial\rho^2 } \label{diff}\ ] ] where as usual @xmath241 in three dimensions ( @xmath154 ) and @xmath242 is the direct correlation function of the reference system . usually the hard sphere gas is adopted as a reference system and @xmath242 is evaluated in percus - yevick @xcite or verlet - weis approximation @xcite . this pde has the structure of a non - linear diffusion equation , where @xmath243 plays the role of time - like variable and @xmath16 of space - like variable . the stability requirement for this class of pde s is given by the positiveness of the effective diffusion coefficient . this condition is verified by eq . ( [ diff ] ) provided @xmath244 is positive . boundary conditions at @xmath245 and @xmath246 are required : it follows from the definition that at low density @xmath247 , while @xmath248 is usually approximated by its mean field expression , i.e. by eq . ( [ defu ] ) with @xmath249 $ ] . different boundary conditions at high density give very similar solutions to the pde , at least when the range of the potential is not too short . details on the discretization procedure can be found in refs . we finally remark that at low values of @xmath25 ( e.g. @xmath250 ) it is convenient to adopt a logarithmic scale in the cut - off , by setting @xmath251 with @xmath252 to slow down the inclusion of long wave - length fluctuations . note that the diagrammatic resummation quoted in eq . ( 4.17 ) of ref . @xcite is reported incorrectly . the expression bond " should be substituted with @xmath255 bond " throughout and the sentence exactly one path " in item ( ii ) should read at most one path " . p.gianinetti , a.parola and l.reatto , in _ advances in quantum many - body theory vol . 5 : 150 years of quantum many - body theory _ bishop , k.a . gernoth , n.r . walet eds . world scientific , singapore , ( 2001 ) .

ngc 1232 is a nearly face on , gas rich , sc galaxy . face on galaxies provide excellent targets for studies of radial properties , such as the chemical enrichment of the interstellar medium ( e.g. , mccall , rybski , & shields 1985 ; vila costas & edmunds 1992 ; zaritsky , kennicutt , & huchra 1994 ; ferguson , gallagher , & wyse 1998 ) or the velocity dispersion of the neutral gas ( e.g. , van der kruit & shostak 1982 , 1984 ; shostak & van der kruit 1984 ; dickey , hanson , & helou 1990 ; boulanger & viallefond 1992 ; kamphuis & briggs 1992 ; rownd , dickey , & helou 1994 ) . ngc 1232 first came to our attention as part of a study of radial trends in elemental abundances ( van zee , salzer , & haynes 1998a ; van zee 1998b ) . the outermost region observed by van zee ( 1998a , b ) has an abnormally low oxygen abundance and a higher n / o ratio than expected at that radius . such features may arise from asymmetries in the disk , or from perturbations induced by tidal interactions . we thus decided to investigate the neutral gas distribution and kinematics of ngc 1232 to determine if there were kinematic peculiarities associated with this particular region , or more generally in the outer gas disk . with a relatively low inclination angle , approximately 30 , ngc 1232 is also a prime candidate for studies of the neutral gas velocity dispersion as a function of radius . since the hi distribution typically extends a factor of 1.3 to 2 times the size of the optical disk in sc galaxies ( e.g. , broeils & rhee 1997 ) , one might naively expect a decrease in the observed velocity dispersion beyond the radius of active star formation [ star formation activity may not be truncated in spiral galaxies , however ( ferguson 1998b ) ] . previous studies of the neutral gas velocity dispersion in face on spiral galaxies have been inconclusive . while some observations indicate that the velocity dispersion may decrease with radius ( e.g. , ngc 6946 , boulanger & viallefond 1992 ) , others indicate that the velocity dispersion is constant in the outer gas disk ( e.g. , ngc 1058 , dickey 1990 ) . such studies are hampered by two main problems : ( 1 ) rotational broadening and ( 2 ) insufficient spectral resolution . the observations of ngc 1232 presented in this paper have adequate spectral resolution , but suffer from significant rotational broadening , in part due to a severe warp in the gas disk . nonetheless , these observations do provide additional insight on radial trends of the neutral gas velocity dispersion in spiral galaxies . the physical parameters of ngc 1232 are summarized in table [ tab : global ] . throughout this paper , we adopt a distance of 21.5 mpc ( van zee 1998b ) , based on an assumed h@xmath1 of 75 mpc@xmath2 and a virgocentric infall model . optically , ngc 1232 is a rather nondescript sc galaxy with well defined spiral arms and optical colors ( de vaucouleurs 1991 ) that are typical for late type spiral galaxies . its total hi mass and m@xmath3/l@xmath4 ratio are also quite typical of the class . on the other hand , it s far infrared luminosity ( soifer 1989 ) is significantly lower than typical for sc galaxies ( see , e.g. , roberts & haynes 1994 for trends as a function of hubble type ) . this paper is organized as follows . the hi synthesis observations and data reduction are presented in section [ sec : obs ] . the neutral gas distribution and kinematics of ngc 1232 are discussed in section [ sec : ana ] . a brief discussion and summary are presented in sections [ sec : disc ] and [ sec : conc ] . finally , the results of optical spectroscopy of the spatially nearby system ngc 1232a are summarized in an appendix . hi synthesis imaging observations of ngc 1232 were conducted by the nrao summer students with the dnc configuration of the very large array ( vla ) on 1992 july 4 . the observations spanned 5 hours , with a total on source integration time of 227 minutes . the correlator was used in 2ad mode with the right and left circular polarizations tuned to 1684 . the total band width was 1.56 mhz . the on line hanning smoothing option was selected , producing final spectral data cubes of 127 channels , each 2.6 wide . standard tasks in aips ( napier 1983 ) were employed for calibration and preliminary data reduction . calibration of this data set was complicated by the fact that the west arm of the array was shadowed during the observations of the primary flux density calibrator , 3c 286 . however , the observations of 3c 286 were sufficient to set the flux density scale ; phase and bandpass calibration was derived from observations of a nearby continuum source , b0237233 . with an observed flux density of 6.21 jy at 1412.5 mhz , the observations of b0237233 had adequate signal to noise for calibration of both the phase and the bandpass . after calibration , the line data was transformed to the _ x y _ plane . a robust weighting technique was employed by the aips task imagr to optimize the beam shape and noise levels of the final data cubes ( briggs 1995 ) . the `` robustness parameter '' in imagr controls the weighting of the _ u v _ data , permitting a fine tuning between sensitivity and resolution . as currently implemented , a robustness of 5 corresponds to natural weighting of the _ u v _ data ( maximizing sensitivity ) while a robustness of 5 corresponds to uniform weighting ( lower sensitivity , but better spatial resolution ) . the relevant imagr parameters for three data cubes are listed in table [ tab : maps ] . throughout this paper we will refer to the robustness of 1.0 cube as the `` low resolution cube , '' to the robustness of 0.5 cube as the `` intermediate weight cube , '' and to the robustness of 0.5 cube as the `` high resolution cube . '' all subsequent analysis of the data cubes was performed within the gipsy package ( van der hulst 1992 ) . line emission from ngc 1232 filled most of the bandpass ( channels 7 to 121 ) . since the few line free channels were at the extreme edge of the bandpass , they had poor noise characteristics and thus were insufficient to create an accurate model of the continuum emission in either the _ u v _ or _ x y _ planes . continuum subtraction was a necessary step in the data reduction process , however , since ngc 1232 has 58.5 mjy of continuum emission at l band ( condon 1987 ) . continuum subtraction was conducted in the _ x y _ plane , with a continuum image created separately for each data cube . to create the continuum image , the mean of the inner 100 channels ( 1553.8 @xmath5 1814.3 ) was computed . this image was severely contaminated by the line emission of the galaxy ; the region of galaxy emission was blotted from the image and replaced with the value 0 . next , the mean was computed for channels with velocities between 1538.2 and 1561.6 ( line emission from only the eastern half of the galaxy ) and for channels with velocities between 1811.7 and 1835.2 ( line emission from only the western half of the galaxy ) . the continuum sources coincident with the galaxy emission were recovered from these two maps . the final continuum image for each resolution was inspected to verify that all of the continuum sources were recovered ; an example of one of these continuum images ( from the intermediate weight data cube ) is shown in figure [ fig : cont ] . the continuum image for each resolution was subtracted from every channel of the data cube to produce the final continuum free data cubes . selected channels from the continuum subtracted intermediate weight data cube are presented in figure [ fig : chans ] . the continuum subtraction appears to have been successful ( see , for instance , the baseline of the integrated flux density profile , figure [ fig : flux ] ) . note , however , that even if the continuum subtraction had not been as successful , the subsequent analysis of the gas kinematics and dynamics would still be valid . in particular , the velocity field and velocity dispersion measurements are fairly robust , even in the presence of continuum emission ; only the total gas distribution would have been significantly effected by poor continuum subtraction . to determine if the total hi flux density was recovered in the hi synthesis observations , the flux density was measured in each channel of the continuum subtracted data cubes . the flux density profile , corrected for primary beam attenuation , is shown in figure [ fig : flux ] . qualitatively , the flux density profile is similar to those obtained from single dish observations ( fisher & tully 1981 ; reif 1982 ; staveley smith & davies 1988 ) . the total flux density recovered from the low resolution data cube was 122.4 @xmath0 12.2 jy , in good agreement with previous single dish measurements ( e.g. , 115.5 @xmath0 7.2 jy , green bank 43 m [ fisher & tully 1981 ] ; 127.9 @xmath0 13.8 jy , jodrell bank [ staveley smith & davies 1988 ] ) . thus , despite the calibration problems and missing short spacings , the flux density scale appears to be correct , and the total hi flux density has been recovered . the slight asymmetry between the two horns of the integrated hi profile has been ascribed to emission from the spatially nearby galaxy ngc 1232a ( eso 547g016 ) ( e.g. , reif 1982 ; becker 1988 ) . however , there is no apparent gas density enhancement or perturbation in the velocity field near this object ( see figure [ fig : mom ] ) . furthermore , a heliocentric optical velocity of @xmath6 was measured from several emission lines with the palomar 5 m telescope ; further details of the emission lines observed in this galaxy are presented in the appendix . this newly derived redshift is in basic agreement with the velocity listed in ned ; thus , this object does not appear to be physically associated with ngc 1232 . both moment analysis and gaussian fitting were used to characterize the gas distribution , velocity field , and velocity dispersion of ngc 1232 . moment maps of each data cube were computed in the following manner . first , the low resolution cube was smoothed to a resolution of twice the beam ; second , the smoothed cube was clipped at the 2@xmath7 level ; the resultant clipped cube was then interactively blanked to remove spurious noise spikes . signal was identified based on spatial continuity between channels . a conditional transfer was used to blank the corresponding locations in the low , intermediate , and high resolution data cubes ( corrected for primary beam attenuation ) based on the blanked , smoothed cube . moment maps of the blanked data cubes were created with the gipsy task moments . in addition , gaussian fits to the observed line profiles at each position in the unclipped data cubes were obtained with the gipsy task gaufit . in regions of high signal to noise , gaussian fitting is preferred over moment analysis since it is less susceptible to beam smearing . the resultant maps were qualitatively similar to the moment maps in both gas distribution and velocity field . however , significant differences were found in the velocity dispersion maps , probably due to beam smearing . thus , the moment maps were used for investigation of the gas distribution and kinematics , while interactive gaussian fitting was employed for study of the velocity dispersion as a function of position . the zeroth and first moments are presented in figure [ fig : mom ] ; the optical images used in this figure are described more fully in van zee ( 1998b ) . there are several notable features in the gas distribution of ngc 1232 . first , the hi distribution extends well beyond the optical radius . at a column density limit of 10@xmath8 atoms @xmath9 , the hi to optical diameter ratio is 1.6 , quite typical of spiral galaxies ( e.g. , cayatte 1994 ; broeils & rhee 1997 ) . second , as has been seen in many other massive galaxies ( e.g. , brinks & bajaja 1986 ; broeils & van woerden 1994 ) , the neutral gas column density is lower in the center of the galaxy . the azimuthally averaged radial profile of the gas density is shown in figure [ fig : rot]d and tabulated in table [ tab : rot ] . as seen in figure [ fig : rot]d , the eastern and western sides of the galaxy are not perfectly symmetric . the eastern half of the galaxy has a higher column density , corresponding to the prominent spiral arms on that side . on the western side , the gas distribution extends to larger radii , perhaps due to a warp of the gas disk in the outer regions . the observed asymmetry between the horns of the global hi profile is the result of the convolution of this asymmetric gas distribution with the rotation of the disk . such asymmetries are not expected to be long lived , but also are not uncommon in spiral galaxies ( e.g. , haynes 1998 ) . closed contours in the velocity field ( figure [ fig : mom]d ) suggest that the outer gas disk is warped , causing an apparent decline in the rotation curve . a model for the rotation curve of ngc 1232 was derived using tilted ring analysis of the intermediate resolution velocity field . the process of fitting a rotation curve is iterative ; we began by fitting both sides of the velocity field in order to determine the kinematic center and systemic velocity ( table [ tab : global ] ) . next , the center and systemic velocity were held fixed and each side was fit separately to determine the position and inclination angles . the position angle was well constrained by the models . however , the fit to the inclination angle was relatively unconstrained . based on the optical axial ratios , the inclination angle was initially assumed to be approximately 30 . since the rotcur fits resulted in wild variations of the inclination angle , the inclination was set to 30 in the inner regions of the galaxy and then permitted to slowly decrease in the outer regions . finally , the rotation curve for each side was determined by fixing the inclination and position angles to the average of both sides with the additional constraint that these values must change smoothly from ring to ring . the result of rotcur s fit to the approaching and receding sides are shown in figure [ fig : rot ] ; table [ tab : rot ] lists the average values for both sides . as seen in the upper panel of figure [ fig : rot ] , a warped disk model is able to recover a flat rotation curve in the outer gas disk . however , since the inclination angle was poorly constrained by the tilted ring models , we can not rule out the possibility that the inclination angle is constant at all radii ( i.e. , the gas disk may not be warped ) . thus , a rotation curve was also derived using the same parameters as the warped disk model , with the exception that the inclination angle was held fixed at 30 throughout the disk ; not unexpectedly , this model results in a falling rotation curve in the outer disk . figure [ fig : pv ] shows the position velocity diagram for a slice along the major axis of ngc 1232 . the rotation curve derived from the warped disk model is superposed . as frequently found in spiral galaxies , the warp appears to begin at the edge of the optical disk ( see e.g. , ngc 4013 , sparke & casertano 1988 ; ngc 4138 , jore , broeils , & haynes 1996 ) . both the warped disk and falling rotation curve models adequately reproduce the major features of the velocity field . however , several kinks in the iso velocity contours are found in the outer regions of the gas disk which can not be well fit by either of these models . such kinks are usually associated with spiral streaming motions , but these features occur in regions well beyond the prominent spiral arms . note , however , that such features would be difficult to detect in the inner regions due to the presence of a steeply rising rotation curve . thus , their apparent absence in the inner region of the galaxy does not eliminate the possibility that they arise from the continuation of spiral density waves to the edge of the disk . due the the steep rotational gradient , higher spatial resolution observations will be needed to investigate fully the pattern speeds in the outer gas disk . the derived rotation curves were used to determine the dark matter content of ngc 1232 . rotation curve decomposition requires knowledge of the gaseous and stellar mass distributions ( assumed to be disk like ) and an assumed dark matter distribution . the contribution of the gaseous and stellar components to the gravitational potential were calculated using the gipsy routine rotmod . the input surface densities were estimated from a series of ellipses with position angles of 270 and inclinations of 30 . the stellar distribution was taken as the observed azimuthal average to the extent of the optical emission ( figure [ fig : rot]e ) , followed by an extrapolated exponential disk . in fitting the rotation curve , the hi distribution was multiplied by 4/3 to account for the he content . the gas content was held fixed during the fitting process . note that for the observed neutral gas distribution of ngc 1232 , the net acceleration is outwards for the inner part of the gas disk ; following the usual convention , this outward acceleration is denoted by a negative velocity . the stellar disk was fit with one free parameter , the mass to light ratio ( m / l@xmath10 ) . note that since the available r band image was obtained under non photometric conditions , the observed surface brightness distribution was scaled to match the total blue luminosity ( de vaucouleurs 1991 ) . while b band and r band scale lengths are typically slightly different for spiral galaxies ( see , e.g. , de jong 1996 ) , this minor difference should be insignificant for the present mass models . a simple spherical halo model was used for the dark matter contribution ; this model has only two free parameters , the core radius , r@xmath11 , and the central density , @xmath12 : @xmath13 where @xmath14 is the dark matter density at radius @xmath15 . in practice , the maximum velocity of the halo is fit rather than the central density : @xmath16,\ ] ] where v@xmath17 is the contribution of the halo to the gravitational potential at a radius @xmath15 and v@xmath18 is the asymptotic velocity , equal to @xmath19 . to determine the parameters of the dark matter halo , the derived rotation curves were fit using a maximum disk method . in this method , the disk mass to light ratio is fit using the inner part of the rotation curve without overshooting the derived circular velocities . initially , the maximum mass to light ratio ( in the b band ) is found by assuming no dark matter is present . for the warped disk model , the rotation curve derived from only the observable mass rapidly underestimated the observed rotation curve at large radii . thus , the presence of a dark matter halo was assumed and the dark matter halo core radius , maximum velocity , and disk mass to light ratio were fit in an iterative manner . in contrast , the falling rotation curve model was well fit by the luminous matter alone and did not require a dark matter halo ( i.e. , the falling rotation curve is nearly keplarian in nature ) . the derived rotation curve decompositions are shown in figure [ fig : rotcur ] . in both cases , the derived disk mass to light ratios ( 3.0 for a warped disk , 3.7 for a falling rotation curve ) are reasonable for the observed colors ( see , e.g. , broeils 1992 ) . one significant difference between the two models is the derived dynamical mass . for a spherical mass distribution , the dynamical mass within radius @xmath15 is : @xmath20 where @xmath21 is in km s@xmath2 and @xmath15 is in kpc . however , if no dark matter halo is required , the system is disk like and the dynamical mass is significantly less than that calculated assuming a spherical halo . for the falling rotation curve model , the dynamical mass is simply the sum of the gaseous and stellar mass , @xmath22 . for the warped disk model , the dark matter halo dominates at large radii , so the dynamical mass is given by the above equation , @xmath23 at the last measured point of the rotation curve . this latter value gives a dark to luminous ratio of 1.9 , quite typical of sc galaxies ( broeils 1992 ) . as mentioned in section [ sec : obs ] , interactive gaussian fitting was used to determine the velocity dispersion throughout the galaxy . both the second moment map and the velocity dispersion map from gaufit showed an apparent trend in the velocity dispersion as a function of radius , with the inner regions of the galaxy having extremely high velocity dispersions . despite the relatively low inclination of this galaxy , however , the velocity dispersion measurements are significantly contaminated by rotational broadening throughout the galaxy . for instance , for beams between 40 and 120 arcsec from the center , the average rotational gradient is 0.3 arcsec@xmath2 ; for beams between 210 and 300 arcsec , the average rotational gradient is 0.2 arcsec@xmath2 . with a 40 beam , this corresponds to a minimum of 8 of rotational broadening per beam . only in the outer regions of the galaxy is the velocity gradient sufficiently shallow to permit detailed analysis of the line shape and width . a line profile from gas in the outer disk is shown in figure [ fig : prof ] . this profile was created by averaging line profiles from regions on the eastern and western sides of the galaxy with radii between 200 and 260 . the profile is well fit by a gaussian with a velocity dispersion of 10.0 . despite the large rotational gradient , it is possible to investigate radial trends in the velocity dispersion by deconvolving the rotational and turbulent motions . the contribution from rotational broadening can be calculated either from a rotation curve model ( e.g. , table [ tab : rot ] ) , or from the observed rotational gradient , as measured in the position velocity diagram . the latter method is more direct , but is limited to regions along the major axis . however , the major axis is also the region where the rotational broadening is minimized ; therefore , in the following analysis we limited our study to two independent slices @xmath024 from the major axis of the high resolution data cube . the velocity dispersion was measured interactively for each beam averaged region and the rotational gradient for each region was obtained directly from the position velocity diagram . figure [ fig : disp ] shows the results of the deconvolution of thermal and rotational broadening ; each slice has been folded so that all radii are positive . the observed line widths were corrected for rotational broadening by assuming that the rotational and thermal components add in quadrature . as illustrated in figure [ fig : disp ] , the velocity dispersion is approximately constant as a function of radius , with an average value of 9.9 @xmath0 1.8 . this is in remarkably good agreement with the velocity dispersion measured from the average of profiles in the outer regions of the galaxy . similar results were also obtained from more detailed rotational models . an estimate of the rotational broadening throughout the disk was obtained by creating a model system with the gipsy task galmod ; the system was assumed to have a negligible intrinsic velocity dispersion and the observed gas distribution and the derived ( no warp ) rotation curve were given as the input parameters . the derived velocity dispersion of this model system is thus indicative of the rotational broadening at any given position in the galaxy . the intrinsic velocity dispersion was recovered by subtracting , in quadrature , the model dispersion from the observed velocity dispersion of the high resolution data cube ( pixel by pixel ) . similar to the above method , no radial trend was seen in the final velocity dispersion map . the average velocity dispersion was measured to be 10.3 @xmath0 1.7 throughout the gas disk using this technique . the observed velocity dispersion is much higher than one might expect for thermal motions . for thermal broadening at 21 cm , the velocity dispersion is related to the thermal temperature by : @xmath24 thus , a velocity dispersion of 9.9 corresponds to a doppler temperature of @xmath2511900 k , which is significantly higher than the expected temperature of the warm neutral medium ( c.f . kulkarni & heiles 1988 ) . however , similarly high velocity dispersions have been observed in several other galaxies ( e.g. , ngc 3938 , van der kruit & shostak 1982 ; ngc 1058 , dickey 1990 ; ngc 6946 , boulanger & viallefond 1992 ; m101 , kamphuis 1993 ) . since neither the previous nor the present observations have adequate spatial resolution to resolve individual hi clouds , the observed velocity dispersion is the ensemble average of several clouds per beam . in fact , the mean motions of the gas clouds , rather than the temperature of the gas within each cloud , could be the dominant motion resulting in the broad velocity dispersion . for a galaxy with a high star formation rate , turbulent motion is obviously a significant contributor to the velocity dispersion in the inner regions of the galaxy , where supernovae and outflows provide large amounts of kinetic energy . one difficulty , however , is to explain how such high levels of turbulence are sustained throughout the gas disk , particularly in regions well beyond the optical edge or in systems with modest star formation rates ( such as ngc 1232 ) . recently , sellwood & balbus ( 1999 ) proposed that mhd instabilities in the differentially rotating gas disk could support a moderate amount of turbulence at all radii via dynamical heating of the disk . their model requires a modest magnetic field ( @xmath25 3 @xmath26 g ) to reproduce a velocity dispersion of 6 in the outer gas disks of spiral galaxies . while it is still unclear whether this mechanism is the dominant source of turbulence at large radii , it is an intriguing possibility . several studies report a radial decrease in the velocity dispersion , with the outer regions of a galaxy having narrower and more perfectly gaussian line shapes ( e.g. , ngc 6946 , boulanger & viallefond 1992 ; m101 , kamphuis 1993 ) . as discussed previously , ngc 1232 also has an observable radial decrease in the velocity dispersion , all of which can be accounted for by rotational broadening in the presence of a steeply rising rotation curve . while most studies acknowledge the complications of rotational broadening , it can be quite difficult to account fully for its effects on the data . for instance , when convolved with a decreasing gas density in the center of the galaxy , beam smearing can result in highly asymmetric , non gaussian spectra . a velocity dispersion measured either by moment analysis or gaussian fitting will not accurately reflect the width of the line in such spectra . thus , it is extremely difficult to study the profile shape and radial trends in the velocity dispersion unless the observations are obtained with high spectral _ and _ spatial resolution . we have investigated the gas distribution and kinematics of the nearly face on spiral galaxy ngc 1232 . our results and conclusions are summarized below . \(1 ) as is typical of spiral galaxies , the neutral gas distribution extends a factor of 1.6 beyond the optical radius . the gas distribution is slightly asymmetric , with a higher gas column density associated with the prominent spiral arms on the eastern side . \(2 ) the observed rotation curve has a nearly keplarian decline beyond the edge of the optical disk . small variations in the inclination angle ( which is poorly constrained due to the low inclination of this galaxy ) can recover a flat rotation curve for the system . assuming that the system is indeed warped , the dark to luminous ratio is 1.9 . \(3 ) the presence of a steeply rising rotation curve complicates measurements of the velocity dispersion as a function of radius . nonetheless , after correcting for rotational broadening , the velocity dispersion appears to be constant throughout the galaxy , with an average value of 9.9 @xmath0 1.8 . as mentioned in the introduction , one of the reasons for looking at the gas distribution of ngc 1232 was to see if there were unusual gas kinematics associated with the outermost hii regions . there are no obviously disturbed kinematic features in the present data set , but such features may have been washed out by the poor spatial resolution of the observations . nonetheless , to first order , the spectra are all well fit by a single gaussian ( with the exception of the central region where the rotational broadening is severe ) and are consistent with a slightly warped , differentially rotating , gas disk extending well beyond the optical radius . thus , it is unlikely that large scale kinematic peculiarities are the cause of the abnormal abundances measured in the outer hii region . analysis of this data set emphasizes the need for good quality high resolution ( both spectral _ and _ spatial ) observations of face on galaxies . the few canonical systems ( e.g. , ngc 1058 and ngc 3938 ) have been studied extensively with both wsrt and the vla , but more systems are needed to address possible dependences of the velocity dispersion on galaxy morphology and star formation rates . while the requirement for low inclination angles significantly restricts the potential target list , corrections for rotational broadening are less important with high spatial resolution observations . in fact , these studies will become more feasible within the next few years ; with a new correlator , the upgraded vla will permit high velocity resolution over a wide bandpass , thereby improving the quality of similar observations of moderately inclined spiral galaxies . we thank the nrao summer students of 1992 who elected to observe ngc 1232 as part of their summer student project : chris de pree , robbie dohm palmer , eric olson , ben oppenheimer , carlos rabaca , james sweeney , and eric wagoner . we thank michael rupen , katrina koski , andreea petric , and dave westpfahl for many informative conversations about gas in face on galaxies . martha haynes kindly provided access to the palomar 5 m telescope for the redshift determination of n1232a . we thank the anonymous referee for thoughtful comments on rotational broadening . robbie dohm palmer and katrina koski are gratefully acknowledged for providing helpful comments on an early draft of the paper . jb acknowledges funding through the research experience for undergraduates program run by the nrao and funded by the national science foundation . as noted in section [ sec : obs ] , optical spectra of ngc 1232a were obtained with the palomar 5 m telescope during the night of 1999 january 23 . the spectra were of sufficient quality that further analysis ( beyond the redshift determination ) was warranted . here , we describe the observations in more detail and present the results of oxygen and nitrogen abundance calculations for the two hii regions observed . during the observations , a long slit ( 2 ) with a 2 aperture was centered on ngc 1232a for one 1200 sec exposure . the slit was positioned at an angle of 0 ( the parallactic angle at the time of the observation ) and passed through both the center of the galaxy and an hii region 19 to the south ( figure [ fig : n1232a ] ) . a 5500 dichroic was used to split the light to the two sides ( blue and red ) , providing complete spectral coverage from 36007600 . the blue spectra were acquired with the 600 lines / mm diffraction grating ( blazed at 4000 ) . the red spectra were acquired with the 316 lines / mm diffraction grating ( blazed at 7500 ) . thinned 1024@xmath271024 tek ccds , with read noises of 8.6 e@xmath28 ( blue ) and 7.5 e@xmath28 ( red ) , were used on the two sides of the spectrograph . both ccds had a gain of 2 . the effective spectral resolution of the blue camera was 5.0 ( 1.72 / pix ) ; the effective spectral resolution of the red camera was 7.9 ( 2.47 / pix ) . the spatial scale of the long slit was 0.62 /pix on the blue and 0.48 /pix on the red side . while the night was non photometric , relative flux calibration was obtained by observations of standard stars from the list of oke ( 1990 ) . wavelength calibration was obtained by observations of arc lamps taken before and after the galaxy observations . a hollow cathode ( fe and ar ) lamp was used to calibrate the blue spectra ; a combination of he , ne , and ar lamps were used to calibrate the red spectra . the spectra were reduced and analyzed with the iraf package . the spectral reduction included bias subtraction , scattered light corrections , and flat fielding with both twilight and dome flats . the 2dimensional images were rectified based on the arc lamp observations and the trace of stars at different positions along the slit . the sky background was removed from the 2-dimensional images by fitting a low order polynomial along each row of the spectra . one dimensional spectra of the two hii regions were extracted from the rectified images using a 3.6 extraction region for the southern hii region and a 3.0 region for the central hii region ( figure [ fig : n1232a ] ) . reddening corrected line strengths for both hii regions are listed in table [ tab : lines ] . derivation of the oxygen abundance for both hii regions followed the methods described in van zee ( 1998b ) . the temperature sensitive line [ oiii ] @xmath294363 was detected in the southern hii region and thus the electron temperature could be determined directly ; for this hii region , the oxygen abundance was calculated using the emissivity coefficients from a version of the fivel program ( de robertis 1987 ) assuming a low density ( 100 @xmath30 ) and a t@xmath31 of 11050 @xmath32 . for the central hii region , the oxygen abundance was estimated based on the strong line ratios ( r@xmath33 ( [ oii ] + [ oiii])/h@xmath34 ) using the model grid of mcgaugh ( 1991 ) . an electron temperature of 6500 @xmath0 500 k reproduced this estimated oxygen abundance , and was thus adopted to derive the n / o ratio for the hii region . the derived oxygen and nitrogen abundances for both hii regions are listed in table [ tab : lines ] . not unexpectedly , the southern hii region is relatively low abundance while the central hii region is significantly higher abundance . with measurements of only two hii regions in hand , the derivation of an abundance gradient in this system is premature ; nonetheless , to first order , the abundance gradient is similar to those found in other spiral galaxies ( e.g. , vila costas & edmunds 1992 ; zaritsky 1994 ; van zee 1998b ) . becker , r. , mebold , u. , reif , k. , & van woerden , h. 1988 , , 203 , 21 boulanger , f. , & viallefond , f. 1992 , , 266 , 37 briggs , d. 1995 , phd thesis , new mexico tech brinks , e. , & bajaja , e. 1986 , , 169 , 14 broeils , a. h. 1992 , phd thesis , university of groningen broeils , a. h. , & rhee , m.h . 1997 , , 324 , 877 broeils , a. h. , & van woerden , h. 1994 , , 107 , 129 cayatte , v. , kotanyi , c. , balkowski , c. , & van gorkom , j. h. 1994 , , 107 , 1003 condon , j. j. 1987 , , 65 , 485 de jong , r. s. 1996 , , 313 , 377 de robertis , m. m. , dufour , r. j. , & hunt , r. w. 1987 , , 81 , 195 de vaucouleurs , g. , de vaucouleurs , a. , corwin , h. g. , buta , r. , paturel , g. , & fouqu , p. 1991 , third reference catalogue of bright galaxies ( springer , new york ) ( rc3 ) dickey , j. m. , hanson , m. m. , & helou , g. 1990 , , 352 , 522 fisher , j. r. & tully , r. b. 1981 , , 47 , 139 ferguson , a. m. n. , gallagher , j. s. , & wyse , r. f. g. 1998a , , 116 , 673 ferguson , a. m. n. , wyse , r. f. g. , gallagher , j. s. , & hunter , d. a. 1998b , , 506 , l19 haynes , m. p. , hogg , d. e. , maddalena , r. j. , roberts , m. s. , & van zee , l. 1998 , , 115 , 62 jore , k. p. , broeils , a. h. , & haynes , m. p. 1996 , , 112 , 438 kamphuis , j. 1993 , phd thesis , university of groningen kamphuis , j. , & briggs , f. 1992 , , 253 , 335 kulkarni , s. r. , & heiles , c. 1988 , in galactic and extragalactic radio astronomy , g.l . verschuur and k.i . kellermann , eds . , ( springer - verlag : new york ) mccall , m. l. , rybski , p. m. , & shields , g. a. 1985 , , 57 , 1 mcgaugh , s. s. 1991 , , 380 , 140 napier , p. j. , thompson , r. t. , & ekers , r. d. 1983 , proc . ieee , 71 , 1295 oke , j. b. 1990 , , 99 , 1621 reif , k. , mebold , u. , goss , w. m. , van woerden , h. , & siegman , b. 1982 , , 50 , 451 roberts , m. s. , & haynes , m. p. 1994 , , 32 , 115 rownd , b. k. , dickey , j. m. , & helou , g. 1994 , , 108 , 1638 sellwood , j. a. , & balbus , s. a. 1999 , , 511 , 660 shostak , g. s. , & van der kruit , p. c. 1984 , , 132 , 20 soifer , b. t. , boehmer , l. , neugebauer , g. & sanders , d. b. 1989 , , 98 , 766 sparke , l. s. , & casertano , s. 1988 , , 234 , 873 staveley smith , l. , & davies , r. d. 1988 , , 231 , 833 van der hulst , j. m. , terlouw , j. p. , begemen , k. , zwitser , w. , & roelfsema , p. r. 1992 , in astronomical data analysis software and systems , ed . d. worall , c. biemesderfer , and j. barnes ( asp conference series ) , 25 , 131 van der kruit , p. c. , & shostak , g. s. 1982 , , 105 , 351 van der kruit , p. c. , & shostak , g. s. 1984 , , 134 , 258 van zee , l. , salzer , j. j. , & haynes , m. p. 1998a , , 497 , l1 van zee , l. , salzer , j. j. , haynes , m. p. , odonoghue , a. a. , & balonek , t. j. 1998b , , 116 , 2805 vila costas , m. b. , & edmunds , m. g. 1992 , , 259 , 121 zaritsky , d. , kennicutt , r. c. , & huchra , j. p. 1994 , , 420 , 87 ll distance ( mpc ) & 21.5kinematic center ( b1950 ) & 03 07 30.2 , 20 46 08 hi properties : integrated h i flux density ( jy km s@xmath2 ) & 122.4 @xmath0 12.2 heliocentric velocity ( km s@xmath2 ) & 1678.9 @xmath0 1.0 profile width ( 50% ) ( km s@xmath2 ) & 234 @xmath0 2 . profile width ( 20% ) ( km s@xmath2 ) & 256 @xmath0 3 . velocity dispersion ( km s@xmath2 ) & 9.9 @xmath0 1.8 size at 10@xmath8 atoms @xmath9 ( arcmin @xmath27 arcmin ) & 11.6 @xmath27 9.7 peak h i surface density ( atoms @xmath9 ) & 1.4 @xmath27 10@xmath35 optical properties : apparent blue magnitude & 10.52 @xmath0 0.14 ( u b)@xmath1 & 0.02 @xmath0 0.10 ( b v)@xmath1 & 0.60 @xmath0 0.04 r@xmath36 ( kpc ) & 23.2 r@xmath37 ( kpc ) & 6.3 derived properties : hydrogen mass , @xmath38 ( 10@xmath39 ) & 1.33 optical luminosity , @xmath40 ( 10@xmath41 ) & 4.46 fir luminosity , @xmath42 ( 10@xmath41 ) & 0.875 dynamical mass , @xmath43 ( 10@xmath39 ) & 43.6 dynamical mass no warp , @xmath43 ( 10@xmath39 ) & 18.3 @xmath44 ( @xmath45 ) & 0.3 @xmath46 & 0.03 @xmath47 & 1.9 rcccc 10 . & 1.27 & 30 & 270 & 70.7 @xmath0 20.7 40 . & 2.20 & 30 & 270 & 184.9 @xmath0 1.4 70 . & 4.23 & 30 & 270 & 209.9 @xmath0 1.6 100 . & 5.85 & 30 & 270 & 216.6 @xmath0 0.8 130 . & 6.34 & 30 & 270 & 221.8 @xmath0 0.7 160 . & 5.91 & 30 & 270 & 220.1 @xmath0 0.6 190 . & 5.14 & 29 & 270 & 221.0 @xmath0 0.6 220 . & 3.70 & 28 & 269 & 220.9 @xmath0 0.7 250 . & 2.18 & 26 & 268 & 219.4 @xmath0 1.3 280 . & 1.19 & 23 & 268 & 218.8 @xmath0 1.7 310 . & 0.72 & 21 & 268 & 215.8 @xmath0 1.6 340 . & 0.52 & 20 & 268 & 219.1 @xmath0 1.9 370 . & 0.41 & 20 & 269 & 217.9 @xmath0 2.3 400 . & 0.32 & 20 & 269 & 217.8 @xmath0 6.4 lccc & 3728 & 2.32 @xmath0 0.17 & 2.30 @xmath0 0.14 & 3869 & & 0.19 @xmath0 0.02 hi & 4340 & 0.47 @xmath0 0.05 & 0.47 @xmath0 0.03 & 4363 & & 0.037 @xmath0 0.016 hi & 4861 & 1.00 @xmath0 0.06 & 1.00 @xmath0 0.05 & 4959 & 0.61 @xmath0 0.05 & 1.34 @xmath0 0.06 & 5007 & 1.81 @xmath0 0.11 & 4.33 @xmath0 0.19 hei & 5876 & 0.09 @xmath0 0.01 & 0.10 @xmath0 0.01 & 6300 & 0.09 @xmath0 0.12 & 0.064 @xmath0 0.005 & 6364 & 0.03 @xmath0 0.01 & 0.020 @xmath0 0.003 & 6584 & 0.11 @xmath0 0.01 & 0.045 @xmath0 0.004 hi & 6563 & 2.86 @xmath0 0.22 & 2.84 @xmath0 0.18 & 6584 & 0.38 @xmath0 0.03 & 0.14 @xmath0 0.01 hei & 6678 & & 0.038 @xmath0 0.004 & 6716 & 0.53 @xmath0 0.04 & 0.26 @xmath0 0.02 & 6731 & 0.27 @xmath0 0.02 & 0.15 @xmath0 0.01 & 7136 & & 0.079 @xmath0 0.006 c@xmath48 & & 0.02 @xmath0 0.08 & 0.24 @xmath0 0.06 f(h@xmath34)@xmath49 & & 2.05 & 4.98 ew(h@xmath34 ) [ ] & & 10.8 & 95.3 log(r@xmath50 ) & & 0.676 @xmath0 0.019 & 0.902 @xmath0 0.013 t(o@xmath51 ) & & 6500 @xmath0 500 & 11050 @xmath32 12 + log(o / h ) & & 8.79 @xmath0 0.10 & 8.21 @xmath0 0.10 log(n / o ) & & 1.24 @xmath0 0.14 & 1.42 @xmath0 0.15

wireless powered communication network has arisen as a new system with stable and self - sustainable power supplies in shaping future - generation wireless communications @xcite . the enabling technology , known as simultaneous wireless information and power transfer ( swipt ) , has particularly drawn an upsurge of interests owing to the far - field electromagnetic power carried by radio - frequency ( rf ) signals that affluently exist in wireless communications . with the transmit power , waveforms , and dimensions of resources , etc . , being all fully controllable , swipt promises to prolong the lifetime of wireless devices while delivering the essential communication functionality , as will be important for low - power applications such as rf identification ( rfid ) and wireless sensor networks ( wsns ) ( see @xcite and the references therein ) . on the other hand , privacy and authentication have increasingly become major concerns for wireless communications and physical ( phy)-layer security has emerged as a new layer of defence to realize perfect secrecy transmission in addition to the costly upper - layer techniques such as cryptography . in this regard , relay - assisted secure transmission was proposed @xcite and phy - layer security enhancements by means of cooperative communications have since attracted much attention @xcite . in particular , cooperative schemes can be mainly classified into three categories : _ decode - and - forward ( df ) _ , _ amplify - and - forward ( af ) _ , and _ cooperative jamming ( cj ) _ @xcite with cj being most relevant to phy - layer security . specifically , coordinated cj refers to the scheme of generating a common jamming signal across all single - antenna relay helpers against eavesdropping @xcite , while uncoordinated cj considers that each relay helper emits independent artificial noise ( an ) to confound the eavesdroppers @xcite . it is expected that in the scenarios where the direct link is broken between the transmitter ( tx ) and the legitimate receiver ( rx ) , some of the relays have to take on their conventional role of forwarding the information while others will perform cj @xcite . a recent paradigm that generalizes all the above - mentioned cooperation strategies is _ cooperative beamforming ( cb ) mixed with cj _ @xcite , where the available power at each relay is split into two parts : one for forwarding the confidential message and the other for cj . however , mixed cb - cj approaches may be prohibitive in applications with low power devices because idle relays with limited battery supplies would likely prefer saving power for their own traffic to assisting others communication . in light of this , swipt provides the incentive for potential helpers to perform dedicated cb mixed with cj at no expense of its own power , but opportunistically earn harvested energy . motivated by this , our work considers secrecy transmission from a tx to a legitimate rx with the aid of a set of single - antenna wireless energy harvesting ( weh)-enabled af - operated relays in the presence of multiple single - antenna eavesdroppers . as a matter of fact , cooperative schemes that involve weh - enabled relays was recently investigated in @xcite . we consider the use of the _ dynamic power splitting ( dps ) _ receiver architecture , initially proposed for swipt in @xcite , which divides the received power with an adjustable ratio for energy harvesting ( eh ) and information receiving ( ir ) . weh - enabled relays using dps receivers have also been considered in @xcite and @xcite , without ( w / o ) and with secrecy consideration , respectively . note that there is also interest in addressing the threat that weh receivers may attempt to intercept the confidential messages in swipt - enabled networks @xcite . nevertheless , we will focus on exploiting the benefits of weh - enabled relays when they are trustful . in particular , motivated by the strong interest in swipt and the vast degree - of - freedom ( dof ) achievable by cooperative relays , this paper aims to maximize the secrecy rate with the aid of weh - enabled af - operated relays , subject to the eh power constraints of individual relays by jointly optimizing the cb of the relays and the cj covariance matrix . in this paper , we assume that there is no direct link between the source and destination nodes , and perfect global channel state information ( csi ) is available for the case of centralized optimization . it is worth pointing out that although our setting may look similar to @xcite , their optimal cb - cj design is not applicable to ours due to the multiplicative nature in beamforming weights incurred by the power splitting ( ps ) ratios that intrinsically poses more intractability to our optimization problem . further , our work also differs from @xcite where an efficient algorithm was proposed to maximize the secrecy rate for the optimization of the ps ratios and af relay beamforming . the difference is twofold . first , an was not considered in the second transmission phase in @xcite and in addition , their algorithm only converged to a local optimum , as opposed to our work that gives the global optimal solutions for cb . the rest of the paper is organized as follows . section [ sec : system model ] describes two types of weh - enabled rx architecture for the af relays and defines the secrecy rate region of the relay wiretap channel . section [ sec : problem formulation ] then formulates the secrecy rate maximization problems that jointly optimize the an ( or cj ) and the af - relay cb for the weh - enabled relays operating with the two types of rx . the problems are respectively solved by centralized schemes in section [ sec : centralized ] and distributed approaches in section [ sec : distributed ] . section [ sec : numerical results ] provides simulation results to evaluate the performance of the proposed schemes . finally , section [ sec : conclusion ] concludes the paper . _ notations_we use the uppercase boldface letters for matrices and lowercase boldface letters for vectors . the superscripts @xmath0 , @xmath1 , @xmath2 and @xmath3 represent , respectively , the transpose , conjugate , conjugate transpose operations on vectors or matrices , and the optimum . in addition , @xmath4 stands for the trace of a square matrix . moreover , @xmath5_{i , j}$ ] denotes the @xmath6th entry of a matrix , while @xmath7 and @xmath8 represent the euclidean norm and the entry - wise absolute value square of a vector , respectively . also , @xmath9 denotes a diagonal matrix with its diagonal specified by the given vector and @xmath5_{i=1}^n$ ] represents an @xmath10 vector with each element indexed by @xmath11 . furthermore , @xmath12 and @xmath13 stand for product and hadamard product , respectively . @xmath14 denotes the field of complex ( real ) matrices with dimension @xmath15 and @xmath16}$ ] indicates the expectation operation . finally , @xmath17 is short for @xmath18 . in this paper , we consider secrecy transmission in a swipt - enabled wsn , where a tx ( alice ) wants to establish confidential communication with the legitimate rx ( bob ) with no direct link but with the aid of @xmath19 weh - enabled sensors operating as af relays , denoted by @xmath20 , in the presence of multiple eavesdroppers ( eves ) , denoted by @xmath21 , all equipped with single antenna . we assume that there is no direct link from the tx to any of the eves , due to , for instance , severe path loss or shadowing , as illustrated in fig . [ fig : system model ] . we consider a two - hop relaying protocol based on two equal time slots and the duration of one transmit - slot is normalized to be one unit so that the terms `` energy '' and `` power '' are interchangeable with respect to ( w.r.t . ) one transmit - slot . at the receiver of each af relay , we introduce two types of weh - enabled receiver architecture , namely , _ static power splitting ( sps ) _ ( fig . [ fig : subfig : sps rx ] ) and _ dps _ ( fig . [ fig : subfig : dps rx ] ) , both of which allow the relay to harvest energy and receive information from the same received signal . specifically , the receiver first splits a portion of @xmath22 , of the received power for eh and the rest @xmath23 for ir , @xmath24 . the @xmath22 portion of harvested power is further divided into two streams with @xmath25 used for generating the an to confound eves and @xmath26 used for amplifying the received signal , where @xmath27 is the @xmath11th element of the received signal @xmath28 , and @xmath29 denotes the eh efficiency . note that dps with adjustable @xmath22 s is presently the most general receiver operation because practical circuits can not directly decode the information from the stream used for eh @xcite and sps is just a special case of dps with @xmath30 , @xmath24 , fixed for the whole transmission duration . however , sps , advocated for its ease of implementation , is introduced separately in the sequel for its simplified relay beamforming design . in the first transmit - slot , the received signal at each individual relay can be expressed as @xmath31 where the transmit signal @xmath32 is a circularly symmetric complex gaussian ( cscg ) random variable with zero mean and unit variance , denoted by @xmath33 , @xmath34 denotes the complex channel from the tx to the @xmath11th relay , @xmath35 is the transmit power at the tx , and @xmath36 is the additive white gaussian noise ( awgn ) introduced by the receiving antenna of the @xmath11th relay , denoted by @xmath37 . as such , the linearly amplified baseband equivalent signal at the output of the @xmath11th relay is given by @xmath38 where @xmath39 denotes the complex af coefficient , and @xmath40 denotes the noise due to signal conversion from the rf band to baseband , denoted by @xmath41 . since @xmath42 is constrained by the portion of the harvested power for forwarding , i.e. , @xmath26 , @xmath39 is accordingly given by @xmath43 where @xmath44 denotes the phase of the af coefficient for the @xmath11th relay . next , we introduce the cj scheme . denote the cj signal generated from @xmath19 relays by @xmath45^t$ ] and define its covariance matrix as @xmath46 $ ] . then the coordinated cj transmission can be uniquely determined by the truncated eigenvalue decomposition ( evd ) of @xmath47 given by @xmath48 , where @xmath49)$ ] is a diagonal matrix with @xmath50 s denoting all the positive eigenvalues of @xmath47 and @xmath51 is the precoding matrix satisfying @xmath52 . note that @xmath53 denotes the rank of @xmath54 which will be designed later . as a result , the cj signal can be expressed as @xmath55 where @xmath56 s are drawn from the columns of @xmath57 , and @xmath58 s are independent and identically distributed ( i.i.d . ) complex gaussian variables denoted by @xmath59 , which is known as the optimal distribution for an @xcite . on the other hand , @xmath60 , @xmath24 , denotes the power constraint for jamming at the @xmath11th relay , which implies that @xmath61 where @xmath62 is a diagonal matrix with its diagonal @xmath63 ( a unit vector with the @xmath11th entry equal to @xmath64 and the rest equal to @xmath65 ) . note that the cj scheme proposed above is of the most general form . for the special case when @xmath66 , i.e. , @xmath67 , each relay transmits a common jamming signal @xmath68 with their respective weight drawn from @xmath69 @xcite . this case is desirable in practice since it has the lowest complexity for implementation . in summary , the transmitted signal at the @xmath11th relay is given by @xmath70 according to together with , , and , the transmit signal from all relays can be expressed in vector form as @xmath71 where @xmath72 and @xmath73 are , respectively , diagonal matrices with their diagonals composed of @xmath74 @xmath75 and @xmath76 . in addition , @xmath77_{i=1}^n$ ] , @xmath78_{i=1}^n$ ] , and @xmath79_{i=1}^n$ ] . @xmath80 in the second transmit - slot , the received signal at the desired receiver , i.e. , bob , is given by @xmath81 where @xmath82_{i=1}^n$ ] comprises complex channels from the @xmath11th relay to the rx and @xmath83 is the corresponding receiving awgn . by substituting into , @xmath84 can be expressed as @xmath85 the received signal at the @xmath86th eve , @xmath87 , is given by @xmath88 where @xmath89_{i=1}^n$ ] denotes the complex channels from the relays to the @xmath86th eve and @xmath90 is the awgn at the @xmath86th eavesdropper . the mutual information for the rx ( bob ) is given by @xmath91 , and that for the @xmath86th eve is @xmath92 , @xmath93 , where @xmath94 and @xmath95 , which denote their respective signal - to - interference - plus - noise ratios ( sinrs ) , are given at the top of next page . next , we define the secrecy rate region that consists of all the achievable secrecy rate for the relay wiretap channel given transmit power @xmath35 , denoted by @xmath96 , which is given by @xcite @xmath97 in this section , we consider the secrecy rate maximization problem by jointly optimizing the an beams , relay beam and their power allocations for weh - enabled af relays operating with sps , i.e. , @xmath30 , @xmath24 , is fixed . by replacing @xmath39 with , @xmath98 in at the next page becomes @xmath99 and @xmath100_i\triangleq h_{sr_i}h_{r_id}\sqrt{\tfrac{\eta\bar\alpha_i(1-\bar\alpha_i)|h_{sr_i}|^2p_s}{(1-\bar\alpha_i)(|h_{sr_i}|^2p_s + \sigma_{n_a}^2)+\sigma_{n_c}^2}}.\ ] ] in addition , @xmath101 and @xmath102 in can also be combined as @xmath103_{i , i},\ ] ] where @xmath104_{i , i}=\frac{\eta\bar\alpha_ip_s|h_{sr_i}|^2|h_{r_id}|^2((1-\bar\alpha_i)\sigma_{n_a}^2+\sigma_{n_c}^2 ) } { ( 1-\bar\alpha_i)(|h_{sr_i}|^2p_s+\sigma_{n_a}^2)+\sigma_{n_c}^2}.\end{aligned}\ ] ] as a result , @xmath105 can be rewritten as @xmath106 where @xmath107_{i=1}^n$ ] . similarly by letting @xmath108_i\triangleq h_{sr_i}h_{r_ie , k}\sqrt{\frac{\eta\bar\alpha_i(1-\bar\alpha_i)|h_{sr_i}|^2p_s}{(1-\bar\alpha_i)(|h_{sr_i}|^2p_s + \sigma_{n_a}^2)+\sigma_{n_c}^2}}\ ] ] and @xmath109_{i , i}\triangleq\frac{\eta\bar\alpha_ip_s|h_{sr_i}|^2|h_{r_ie , k}|^2((1-\bar\alpha_i)\sigma_{n_a}^2+\sigma_{n_c}^2 ) } { ( 1-\bar\alpha_i)(|h_{sr_i}|^2p_s+\sigma_{n_a}^2)+\sigma_{n_c}^2},\ ] ] we have @xmath110 by some simple manipulation , is reformulated as a per - relay jamming power constraint given by @xmath111 now , the secrecy rate maximization problem w.r.t . @xmath112 s , @xmath44 s and @xmath47 for sps - based relays can be formulated as @xmath113 @xmath114 here , we consider the secrecy rate maximization problem for weh - enabled af relays with adjustable ps ratios @xmath115 by jointly optimizing the an beams , relay beam , weh ps ratios @xmath115 , and an ps ratios @xmath116 . first , consider the following variable transformation : @xmath117 using this , @xmath118 can then be expressed as @xmath119 , where @xmath120_{i=1}^n$ ] and @xmath121_{i=1}^n$ ] . moreover , @xmath122 and @xmath123 can be simplified as @xmath124 and @xmath125 , respectively , where @xmath126_{i=1}^n$ ] with @xmath127 , @xmath24 , and @xmath128_{i=1}^n$ ] . similarly , we have @xmath129 , @xmath130 , and @xmath131 , where @xmath132_{i=1}^n$ ] , @xmath133 . then we apply the above transformation to @xmath134 ( c.f . ) and @xmath135 ( c.f . ) , @xmath93 , to get and ( see next page ) . now , we can recast the constraints w.r.t . @xmath47 , @xmath22 s , and @xmath112 s to those w.r.t . the transformed variables @xmath136 s and @xmath137 s . in accordance with , the optimization variables , @xmath22 s and @xmath112 s , can be alternatively given by @xmath138 where @xmath139 . replacing @xmath22 s and @xmath112 s with , is reformulated as @xmath140 on the other hand , since @xmath141 and @xmath142 , @xmath24 , after some simple manipulation , it follows from that @xmath143 as such , the secrecy rate maximization problem for dps - based relays becomes @xmath144 in this section , we resort to centralized approaches to solve problem @xmath145 and @xmath146 , respectively , assuming that there is a central optimizer that is able to collect all csis including @xmath147 , @xmath148 and @xmath149 , perform the optimization , and broadcast to relays their individual optimized parameters . to start with , we recast @xmath145 into a two - stage problem by introducing a slack variable @xmath150 . first of all , we solve the epigraph reformulation of @xmath145 with a fixed @xmath151 $ ] as @xmath152 defining @xmath153 as the optimum value of @xmath154 and denoting @xmath155 , the objective function of @xmath145 is given by @xmath156 where @xmath157 in the objective function has been omitted and we claim a zero secrecy rate if admits a negative value . as a result , @xmath145 can be equivalently given by @xmath158 note that this single - variable optimization problem allows for simple one - dimension search over @xmath159 $ ] , assuming that @xmath160 is attainable given any @xmath150 in this region . as the physical meaning of @xmath161 in @xmath154 can be interpreted as the maximum permitted sinr for the best eavesdropper s channel , feasibility for a non - zero secrecy rate implies that @xmath162 where cauchy - schwarz inequality has been applied in @xmath163 and @xmath164 follows from @xmath165 , @xmath166 . the above epigraph reformulation of non - convex problems like @xmath145 has been widely employed in the literature @xcite , and @xmath167 admits the same optimal value as @xmath145 while @xmath154 with the optimal @xmath150 provides the corresponding optimal solution to @xmath145 . we summarize the steps for solving @xmath145 here : given any @xmath159 $ ] , solve @xmath154 to obtain @xmath160 ; solve @xmath167 via a one - dimensional search over @xmath150 . before developing solutions to @xmath154 , we have the lemma below . @xmath160 is a concave function of @xmath150 . [ lemma : concavity of h_1(tau ) ] see appendix [ appendix : proof of concavity of h_1(tau ) ] . using lemma [ lemma : concavity of h_1(tau ) ] , it is easy to verify that @xmath168 is also a concave function of @xmath150 according to the composition rule @xcite , which allows for a more effective search for the optimum @xmath150 , e.g. , bi - section method , than the exhaustive search used in @xcite . moreover , although @xmath160 is not differentiable w.r.t . @xmath150 , the bi - section method can still be implemented , the algorithm involving which is similarly applied in solving @xmath146 and thus will be given later in section [ sec : proposed solutions for dps ] . in the sequel , we focus on solving @xmath154 . by introducing @xmath169 and ignoring the rank - one constraint on @xmath170 , @xmath171 can be alternatively solved by @xmath172 note that the objective function has been multiplied by @xmath150 compared with that of @xmath154 for ready computation of @xmath160 . although @xmath173 is made easier to solve than @xmath154 by rank relaxation , it is still a quasi - convex problem considering the linear fractional form of the objective function and constraints , for which charnes - cooper transformation @xcite will be applied for equivalent convex reformulation . specifically , by substituting @xmath174 and @xmath175 into @xmath173 , it follows that @xmath176 problem @xmath177 can now be optimally and efficiently solved using interior - point based methods by some off - the - shelf convex optimization toolboxes , e.g. , cvx @xcite . we have the following results : 1 . the optimal solution to @xmath177 satisfies @xmath178 ; 2 . @xmath179 , where @xmath180 is given by @xmath181 in which @xmath182 is given in appendix [ appendix : proof of rank - one solution with cj for sps ] ; 3 . @xmath183 . [ prop : rank - one solution with cj for sps ] see appendix [ appendix : proof of rank - one solution with cj for sps ] . proposition [ prop : rank - one solution with cj for sps ] implies that the rank - one relaxation of @xmath173 from @xmath154 is tight for an arbitrary given @xmath150 . the @xmath184 s and @xmath185 s can thus be retrieved from the magnitude and angle of @xmath186 , respectively , by applying evd to @xmath187 . similar to section [ sec : optimal solutions for sps ] , in this section , we aim at solving the two - stage reformulation of @xmath146 by introducing a slack variable @xmath188 $ ] . first , for a given @xmath150 , we solve @xmath189 next , denoting @xmath190 by @xmath191 , where @xmath192 is the optimum value for problem @xmath193 , we solve the following problem that attains the same optimum value as @xmath146 : @xmath194 where @xmath195 is similarly derived as @xmath196 so that we directly arrive at @xmath197 , denoted by @xmath195 . we claim that @xmath198 can be solved by bi - section for @xmath150 over the interval @xmath199 $ ] assuming that @xmath191 is valid for any given @xmath150 ( otherwise a zero secrecy rate , i.e. , @xmath200 , is returned . ) , since @xmath191 has the following property . @xmath191 is a concave function of @xmath150 . [ lemma : concavity of h_2(tau ) ] the proof is similar to that for lemma [ lemma : concavity of h_1(tau ) ] , and thus is omitted . it is also seen that how to attain @xmath191 forms the main thrust for solving @xmath146 . however , the constraints in , and are not convex w.r.t . @xmath136 and/or @xmath137 , @xmath24 , due to their high orders and multiplicative structure . @xmath193 thus turns out to be very hard to solve in general . to cope with these non - convex constraints , we introduce the following lemma . the restricted hyperbolic constraints which have the form @xmath201 , where @xmath202 , @xmath203 , are equivalent to rotated second - order cone ( soc ) constraints as follows . @xmath204[lemma : hyperbolic to soc ] for convenience , denoting @xmath205 , @xmath206 , @xmath207 by @xmath208 , @xmath209 , and @xmath210 , respectively , @xmath24 , can be rewritten as @xmath211 according to and , it is easily verified that @xmath212 . hence , is eligible for lemma [ lemma : hyperbolic to soc ] , which is reformulated into the soc constraint : @xmath213 similarly , can be simplified as @xmath214 , and after some manipulation , it is recast into a constraint of the restricted hyperbolic form as @xmath215 is thus , in line with lemma [ lemma : hyperbolic to soc ] , equivalent to an soc constraint given by @xmath216 at last , is a linear constraint w.r.t . @xmath208 and @xmath209 given by @xmath217 @xmath218 note that have so far been equivalently transformed into the soc constraints , the linear constraints , as well as , the latter two of which are jointly convex w.r.t . @xmath208 and @xmath209 , @xmath24 . however , is still not convex w.r.t . @xmath210 , @xmath24 , yet . to circumvent this , in the sequel we propose to solve problem @xmath146 by alternating optimization . the upshot of the algorithm is that first we fix @xmath47 by @xmath219 and thus @xmath210 by @xmath220 , @xmath24 , and solve problem @xmath221 ( @xmath193,@xmath198 ) with fixed @xmath47 as @xmath221 ( @xmath222,@xmath223 ) in the sequel . ] to find the optimal @xmath224 , @xmath225 and @xmath226 via @xmath222 and @xmath223 ; then with @xmath227 , @xmath24 , we devise the optimal solution derived in section [ sec : optimal solutions for sps ] to obtain the optimal cj covariance , viz @xmath228 , and thus @xmath229 , @xmath24 ; finally , by updating @xmath230 and @xmath231 , @xmath24 , problems @xmath221 and @xmath145 are iteratively solved until they converge . the remaining challenges lie in solving problem @xmath222 now that , and are all made convex w.r.t . their variables @xmath208 , @xmath209 , @xmath24 . similar to that for @xmath154 , we introduce @xmath232 and @xmath233 and exempt problem @xmath222 from @xmath234 and @xmath235 as follows : @xmath236 @xmath191 ( c.f . ) and are given at the top of next page . recalling the procedure to deal with @xmath173 , we now apply charnes - cooper transformation to convert @xmath237 into a convex problem , denoted by @xmath238 , by replacing @xmath239 and @xmath240 with @xmath241 and @xmath242 , respectively . the solution for @xmath238 is tight and characterized by the following proposition . we have the following results : 1 . the optimal solution to @xmath238 satisfies @xmath243 such that @xmath244 ; 2 . @xmath245 is given by @xmath246 where @xmath247 , @xmath248 , and @xmath249 is given in appendix [ appendix : proof of rank - one solution with cj for dps ] ; 3 . @xmath250 , of which the diagonal entries compose a vector denoted by @xmath251 , can be reconstructed by @xmath252 with rank one . [ prop : rank - one solution with cj for dps ] see appendix [ appendix : proof of rank - one solution with cj for dps ] . the @xmath253 s and @xmath112 s are thus attained according to via evd of @xmath254 and @xmath255 . the proposed algorithm for solving @xmath146 is presented in table [ table : algorithm i ] . ' '' '' @xmath228 ; @xmath256 that denotes the optimum value for @xmath145 given @xmath257 , @xmath24 @xmath258 , @xmath259 @xmath260 @xmath261 , @xmath262 , @xmath24 , and solve @xmath221 : @xmath263 , @xmath264 , @xmath265 , @xmath266 , @xmath267 @xmath268 , @xmath269 solve @xmath222 and @xmath191 @xmath270 @xmath271 , where @xmath272 and @xmath273 denotes an arbitrary small value . @xmath274 @xmath275 @xmath276 , @xmath277 , and obtain @xmath278 according to @xmath279 , @xmath24 , and solve @xmath145 via @xmath154 and @xmath167 @xmath280 , @xmath281 , and obtain @xmath282 and @xmath283 according to @xmath284 , @xmath24 update @xmath285 according to @xmath278 , @xmath282 , @xmath283 , and @xmath228 ' '' '' in this section , we investigate heuristic algorithms to solve problems @xmath145 and @xmath146 in a completely distributed fashion . note that different from the paradigm of _ distributed optimization _ that allows for certain amount of information exchange based on which iterative algorithms are developed to gradually improve the system performance , we herein assume that each individual relay can only make decision based on its local csis , namely , @xmath34 , @xmath286 , @xmath287 , @xmath24 , and there is no extra means of information acquisition for ease of implementation . the purpose for such an algorithm is twofold : on one hand , we aim to answer the question that in the least favourable situation , namely , no coordination over the relays , how to improve the achievable secrecy rate of the system ? on the other hand , it provides a lower - bound for the centralized schemes proposed in section [ sec : centralized ] , which sheds light upon the trade - off achievable between secrecy performance and complexity . besides , we emphasize the jamming scheme that is different from the cj in the centralized schemes . unlike the cj signal coordinately transmitted by all relays , in the distributed implementation , each relay is only able to generate its an locally , i.e. , @xmath288^t$ ] , in which @xmath289 s are i.i.d . an beams , denoted by @xmath290 . this type of cj is known to be uncoordinated with the covariance matrix given by @xmath291)$ ] . in this section , we assume that each relay consumes all of its remaining power from af for an , i.e. , @xmath292 , @xmath166 ( c.f . ) . hence , the an design solely depends on @xmath22 s and/or @xmath112 s . first , we propose a heuristic scheme for the @xmath11th af relay to decide on @xmath112 , @xmath24 , which is given by @xmath293 where @xmath294 is a constant controlling the relay s level of jamming . for example , a larger @xmath295 indicates that each relay prefers to splitting a larger portion of power for jamming and vice versa . the intuition behind is that if the @xmath11th relay observes that @xmath296 , which means that a nonnegative secrecy rate is achievable even if there is only itself in the system , it will shut down the an ; otherwise , it will split up to @xmath295 portion of the harvested power for jamming . in an extreme case of @xmath297 , probably when an eve is located within the very proximity of this relay , it allocates the maximum permissible portion of power , i.e. , @xmath295 , for an . next , since an individual relay can not evaluate the secrecy performance of the whole system , @xmath44 s are simply chosen to be the optimum for the multi - af relaying without security considerations , i.e. , @xmath298 , @xmath24 . following the same designs for @xmath112 s and @xmath44 s in section [ sec : distributed alg . for sps ] , the remaining task for weh - enabled relays operating with dps is to set proper values for @xmath22 s . we choose @xmath22 s that maximize the `` hypothetical sinr '' . this `` hypothetical '' sinr may not be the actual sinr for the destination , but just a criterion calculated based on the `` hypothetical '' received signal seen by the @xmath11th relay , given by @xmath299 the corresponding sinr is thus expressed as @xmath300 where @xmath301 . consequently , the maximization of w.r.t . @xmath22 , @xmath24 , is formulated as @xmath302 the optimal @xmath22 , @xmath24 , to @xmath303 is @xmath304[prop : optimal alpha to distr ] it is easy to verify that problem @xmath303 is convex and the minimum solution of its objective function derived from the first - order derivative happens to fall within the feasible region of @xmath22 , which is seen in . with @xmath112 s , @xmath44 s and @xmath22 s set , each af relay is then able to decide its relay weight and an transmission . in this section we compare our proposed schemes operating with sps or dps with some benchmarks . in the centralized case , the optimal solution for sps in section [ sec : optimal solutions for sps ] is denoted by _ cj - sps _ , while algorithm [ table : algorithm i ] in section [ sec : proposed solutions for dps ] is denoted by _ cj - dps_. the distributed schemes in section [ sec : distributed alg . for sps ] and section [ sec : distributed alg . for dps ] are referred to as _ distributed - sps _ and _ distributed - dps _ , respectively . to demonstrate the effectiveness of our an - aided secure multi - af relay beamforming algorithms , we also provide three benchmark schemes : _ nocj - sps _ , _ nocj - dps _ and _ random ps_. for _ nocj - sps _ , we solve problem @xmath145 by replacing @xmath47 with @xmath305 . similarly , for _ nocj - dps _ , we initialize @xmath306 and quit the loop in algorithm [ table : algorithm i ] after the very first time of solving problem @xmath221 . _ random ps _ , on the other hand , picks up i.i.d . @xmath22 and @xmath112 uniformly generated over @xmath307 $ ] , respectively , and co - phases @xmath308 , @xmath24 . consider that @xmath19 weh - enabled af relays and @xmath309 eavesdroppers are located within a circular area of radius @xmath310 . specifically , we assume that their respective radius and radian are drawn from uniform distributions over the interval @xmath311 $ ] and @xmath312 , respectively . we also assume that the channel models consist of both large - scale path loss and small - scale multi - path fading . the unified path loss model is given by @xmath313 where @xmath314 , @xmath315 denotes the relevant distance , @xmath316 m is a reference distance , and @xmath317 is the path loss exponent set to be @xmath318 . @xmath34 , @xmath286 , and @xmath319 , @xmath166 , @xmath133 , are generated from independent rayleigh fading with zero mean and variance specified by . the simulation parameters are set as follows unless otherwise specified : the radius defining the range is @xmath320 m ; the transmit power at the source is @xmath321dbm ; the noise variances are set as @xmath322dbm , @xmath323dbm , @xmath324 , and @xmath325 , @xmath93 ; the eh efficiency is set to @xmath326 . also , numerical results were averages over @xmath327 independent channel realizations . here , we evaluate the performance of the proposed centralized designs in section [ sec : centralized ] . the efficiency of the alternating optimization that iteratively attains numerical solution to @xmath146 is studied in fig . [ fig : alternating sec vs iter ] , which shows the increment of the achievable secrecy rate after each round of the iteration . the most rapid increase is observed after the first iteration , which illustrates that the optimization of the ps ratios , @xmath22 s , accounts for the main factor for the secrecy rate performance gains over a sps scheme that sets @xmath328 . it is seen that the alternating algorithm converges within the relative tolerance set to be @xmath329 , after an average of @xmath330@xmath331 iterations for several channel realizations , which appears reasonable in terms of complexity . dbm , @xmath332 , and @xmath333.,width=288 ] fig . [ fig : centralized compare vs n ] shows the achievable secrecy rate for the legitimate rx with the increase in the number of af relays by different schemes . as we can see , the secure multi - af relaying schemes assisted by the transmission of an outperforms that without an for both sps and dps . in addition , with the increase in @xmath19 , the role of cj gradually reduces for both schemes of sps and dps this is because as @xmath19 gets larger , the optimal designs tend to suppress the interception at the most capable eavesdropper more effectively with @xmath19 dof , enforcing the numerator of @xmath334 to a relatively low level , which can also be observed from @xmath335 in fig . [ fig : subfig : eve s rate vs n ] , and therefore the optimal amount of power allocated to an beams inclines to be little ; otherwise the jamming yielded will be detrimental to the reception at the legitimate rx . besides , given the same number of af relays , the secrecy performance gains brought by the proposed schemes with cj are more prominent in the presence of more eavesdroppers , since it is hard to reduce all the eavesdroppers channel capacity without resorting to cj properly . db for @xmath332 and @xmath336 , respectively.,width=288 ] fig . [ fig : centralized compare sec vs eves ] shows the achievable secrecy rate for the legitimate rx versus the number of eavesdroppers by different schemes . first , similar to the results shown in fig . [ fig : centralized compare vs n ] , the proposed an - aided multi - af relaying designs operating with dps - enabled relays , viz _ cj - dps _ , perform best among all the schemes . secondly , as @xmath309 goes up , the an - aided schemes , _ cj - dps _ and _ cj - sps _ , allow the secrecy rate to drop slowly , in other words , more robust against multiple eavesdroppers , while the secrecy rate of their nocj counterparts almost goes down linearly with @xmath309 . furthermore , with @xmath309 increasing , for example , more than @xmath337 , the increase in the number of relays , from @xmath332 to @xmath338 , can not replace the role of cj as shown in fig . [ fig : centralized compare vs n ] and [ fig : centralized compare sec vs eves ] , since in the presence of many eavesdroppers , more relays may also result in improved eavesdroppers decoding ability w / o the assistance of cj . it is also noteworthy that with @xmath339 , there is little use of cj by the centralized schemes , which was also observed in @xcite . for @xmath332 and @xmath336 , respectively.,width=288 ] [ fig : centralized compare sec vs ps ] provides simulation results of different schemes by varying the source transmit power . it is seen that with more power available at the source , the advantage of cj is more pronounced , since given other variables fixed , larger @xmath35 indicates larger feasible regions for @xmath145 and @xmath146 . furthermore , as similarly seen in fig . [ fig : centralized compare vs n ] , with a mild number of eavesdroppers ( @xmath333 ) , subject to the same @xmath35 , a large number of cooperative relays enables more dof in designing the optimal @xmath22 s and @xmath44 s , which alleviates the dependence on an beams to combat eves . here , we study the performance of the distributed schemes , namely , _ distributed - sps _ and _ distributed - dps _ in section [ sec : distributed ] . as mentioned earlier , these heuristics are provided as benchmarks to demonstrate what can be done under the extreme `` no - coordination '' circumstance , in comparison with _ random ps_. note that any other distributed schemes with certain level of cooperation among relays are supposed to increase the secrecy performance up to the proposed centralized algorithms , namely , _ cj - dps _ and _ cj - sps _ , at the expense of extra computational complexity and system overhead . db and @xmath333.,width=288 ] fig . [ fig : distributed compare sec vs n ] provides the results for the achievable secrecy rate of various schemes versus the number of relays . _ distributed - sps _ and _ distributed - dps _ , are observed to be outperformed by their centralized counterparts though , they are considerably superior to _ random ps_. it is also seen that the performance gap between the centralized and distributed approaches is enlarged as @xmath19 increases , which is expected , since larger @xmath19 yields more dof for cooperation that is exclusively beneficial for the centralized schemes . furthermore , compared with the centralized schemes , the distributed ones are more vulnerable to the increase in the eavesdroppers number . db and @xmath19=8.,width=288 ] in fig . [ fig : distributed compare sec vs eves ] , we investigate the relationship between the secrecy rate performance and the number of eavesdroppers by different methods . as can be observed , compared with the centralized schemes , the secrecy rates achieved by _ distributed - sps _ and _ distributed - dps _ both reduce more drastically with the increase in @xmath309 due to the lack of effective cooperation . also , the advantage of dps over sps for the distributed schemes is compromised since @xmath22 s are not jointly designed with other parameters . at last , a similar observation has been made as that for fig . [ fig : distributed compare sec vs n ] , that is , larger @xmath19 yields more visible performance gap between the centralized and distributed approaches . and @xmath333.,width=288 ] in fig . [ fig : distributed compare sec vs ps ] , we examine the effect of increasing the transmit power at the source on the secrecy performance of different schemes under the same settings as those in fig . [ fig : centralized compare sec vs ps ] . among all the presented designs , _ cj - dps _ still achieves the best secrecy rate as observed in other examples . also , the fact that larger @xmath19 benefits more from cooperative designs is corroborated again due to the same reason as that for fig . [ fig : distributed compare sec vs n ] and [ fig : distributed compare sec vs eves ] . furthermore , the secrecy rate of _ distributed - sps _ or _ distributed - dps _ is quickly saturated when @xmath340db while that for their centralized counterparts still rises at fast speed . this paper studied secure communications using multiple single - antenna weh - enabled af relays assisted by an via cj for a swipt network with multiple single - antenna eavesdroppers . using ps at the relays , the achievable secrecy rates for the relay wiretap channel were maximized by jointly optimizing the cb and the cj covariance matrix along with the ps ratios for relays operating with , respectively , sps and dps . for dps , reformulating the constraints into restricted hyperbolic forms enabled us to develop convex optimization - based solutions . further , we proposed an information - exchange - free distributed algorithm that outperforms the random decision . as the optimal solution to @xmath177 is proved to be optimal to problem @xmath154 , the optimum value for @xmath177 is @xmath341 . hence , we identify the property of @xmath160 by investigating @xmath177 , with its lagrangian given by @xmath342 where @xmath343 denotes a tuple consisting of all the primal and dual variables . specifically , @xmath344 , @xmath345 and @xmath346 are lagrangian multipliers associated with @xmath347 , @xmath348 and the first constraint of @xmath177 , respectively ; @xmath349 are the dual variables associated with the sinr constraint for the @xmath86th eve , respectively ; @xmath350_{i=1}^n)$ ] with each diagonal entry @xmath351 denoting the dual variable associated with the per - relay power constraint ; @xmath352 is the lagrangian multiplier associated with @xmath353 . in addition , @xmath354_{i=1}^n)$ ] . the karush - kuhn - tucker ( kkt ) conditions for are partially listed as follows : the associated dual problem is accordingly given by @xmath356 where @xmath344 , @xmath345 , and @xmath352 are given by , respectively . since it is easily verified that @xmath177 satisfies the slater condition , the strong duality holds @xcite . this implies that the dual optimum value in @xmath357 is @xmath160 , which turns out to be a point - wise minimum of a family of affine functions and thus concave over @xmath159 $ ] @xcite . pre- and post - multiplying @xmath359 with the left hand side ( lhs ) and right hand side ( rhs ) of , respectively , and substituting @xmath360 into , @xmath361 can be rewritten as @xmath362 introducing the notation of @xmath5_{\rm offd}$ ] to represent a square matrix with its diagonal entries removed , it follows from that @xmath363_{\rm offd}=\boldsymbol 0.\ ] ] by subtracting from , @xmath344 can be rewritten as @xmath364_{\rm d}\notag\\ & + \left[\boldsymbol w_0^{\frac{1}{2}}\boldsymbol y_2\boldsymbol w_0^{\frac{1}{2}}\right]_{\rm d}+\sum_{k=1}^k\theta_k p_s\tilde{\boldsymbol h}_{se , k}^\dag\tilde{\boldsymbol h}_{se , k}^t,\label{eq : y1 intermediate2}\end{aligned}\ ] ] where @xmath5_{\rm d}$ ] denotes a square matrix with only the diagonal remained . observing that @xmath365_{\rm d}^{-1}\left[\boldsymbol w_0^{\frac{1}{2}}\boldsymbol h_{re , k}^\dag\boldsymbol h_{re , k}^t\boldsymbol w_0^{\frac{1}{2}}\right]_{\rm d}\nonumber\\ & = { \rm diag}\left([|h_{r_ie , k}|^2\left.\middle/\right.|h_{r_id}|^2]_{i=1}^n\right)\notag\\ & \equiv \boldsymbol r_{ed , k},\label{eq : r_{ed , k}}\end{aligned}\ ] ] @xmath361 can be finally recast as @xmath366 where @xmath367_{\rm d}-\left(\left[\boldsymbol w_0^{\frac{1}{2}}\boldsymbol h_{rd}^\dag\boldsymbol h_{rd}^t\boldsymbol w_0^{\frac{1}{2}}\right]_{\rm d}-\boldsymbol d_{\hat{sd}}\right ) \left(\lambda\boldsymbol i-\sum_{k=1}^k\theta_k\left(\frac{1}{\tau}-1\right)\boldsymbol r_{ed , k}\right).\end{aligned}\ ] ] in the following , we show that @xmath368 is a positive definite matrix . note that since @xmath369 is a diagonal matrix , its definiteness is only determined by the signs of its diagonal entries , for which we commence with the discussion in three difference cases below . 1 . * case i * : @xmath370 such that @xmath371_{i , i}<0 $ ] . since @xmath372_{\rm d}-\boldsymbol d_{\hat{sd}}\right]_{i , i}=$ ] + @xmath373 , it follows from that @xmath374_{i , i}>0 $ ] in this case . * case ii * : @xmath370 such that @xmath371_{i , i}>0 $ ] . we have @xmath375_{\rm d}\right]_{i , i}-[\boldsymbol w_0]_{i , i}\vert h_{r_id}\vert^2(\lambda^\ast-\sum_{k=1}^k\theta_k\left(\frac{1}{\tau}-1\right)[\boldsymbol r_{ed , k}]_{i , i})\ge 0 $ ] in accordance with , which implies that @xmath374_{i , i}=\left[\left[\boldsymbol w_0^{\frac{1}{2}}\boldsymbol y_2\boldsymbol w_0^{\frac{1}{2}}\right]_{\rm d}\right]_{i , i}-[\boldsymbol w_0]_{i , i}\vert h_{r_id}\vert^2(\lambda-\sum_{k=1}^k\theta_k\left(\frac{1}{\tau}-1\right)[\boldsymbol r_{ed , k}]_{i , i})+[\boldsymbol d_{\hat{sd}}]_{i , i}(\lambda-\sum_{k=1}^k\theta_k\left(\frac{1}{\tau}-1\right)[\boldsymbol r_{ed , k}]_{i , i})>0 $ ] ( c.f . ) . * case iii * : @xmath370 such that @xmath371_{i , i}=0 $ ] . in this case , it follows that @xmath374_{i , i}= \left[\boldsymbol w_0^{\frac{1}{2}}\boldsymbol y_2\boldsymbol w_0^{\frac{1}{2}}\right]_{i , i}$ ] + @xmath376 . it is noteworthy that the number of @xmath11 s such that @xmath371_{i , i}=0 $ ] can not exceed one . this can be proved by contradiction as follows . if @xmath377 , @xmath378 , such that @xmath371_{i_1,i_1}=0 $ ] and @xmath371_{i_2,i_2}=0 $ ] , it implies that @xmath379_{i_1,i_1}=\sum_{k=1}^k\theta_k[\boldsymbol r_{ed , k}]_{i_2,i_2}$ ] , which contradicts to the fact that for any two independent continuously distributed random variables , the chance that they are equal is zero . in summary , @xmath374_{i , i}\ge 0 $ ] , @xmath24 . if @xmath374_{i , i } > 0 $ ] , then it is obvious that @xmath380 . next , we show that it still holds true in the case that @xmath381 , such that @xmath374_{i^\prime , i^\prime}=0 $ ] , @xmath382 , by definition . define the null - space of @xmath369 by @xmath383 and multiply @xmath384 and @xmath385 , @xmath386 , on the lhs and rhs of @xmath368 , respectively . if @xmath387 , it is straightforward to obtain @xmath388 ; otherwise , it follows that @xmath389_{i^\prime}\vert^2>0 $ ] , as @xmath390_{i^\prime}\neq 0 $ ] in probability . this completes the proof . as a result , @xmath344 in is shown to always take on a special structure , that is , a full - rank matrix minus a rank - one matrix . note that this observation plays a key role in proving the rank - one property of @xmath391 , which is also identified in ( * ? ? ? * appendix c ) . finally , multiplying both sides of by @xmath391 , as per , we obtain the following equation : @xmath393 which further implies that @xmath394 . in addition , since the optimality of @xmath177 suggests that @xmath395 , @xmath396 is thus proved . as @xmath391 can be decomposed as @xmath397 by evd , results in @xmath398 , which further implies that @xmath399 therefore , admits a unique solution @xmath182 up to a scaling factor , which is given by @xmath400 consequently , we have @xmath401 , where @xmath402 . on the other hand , by plugging @xmath401 into the equality constraint of @xmath177 , we have @xmath403 , which yields @xmath404 at last , we show @xmath405 of proposition [ prop : rank - one solution with cj for sps ] . for the case of @xmath406 , it is obvious that @xmath407 . for the case of @xmath408 , only a sketch of the proof is provided here due to the length constraint . according to , first it is provable that @xmath409 is a full - rank matrix when @xmath177 obtains its optimum value ; next , observing that @xmath410 , it follows that @xmath411 as a result of @xmath412 ; then according to , @xmath413 is thus obtained . for convenience of the proof , the optimum value for @xmath237 and @xmath415 are denoted by @xmath416 and @xmath417 , respectively . assuming that @xmath418 is the optimal solution to @xmath237 , it is easily verified to be feasible for @xmath415 as well , which implies that @xmath419 . on the other hand , if @xmath415 returns an optimal solution of @xmath420 , by defining @xmath421 and @xmath422 , @xmath24 , we show that @xmath423 is also feasible for @xmath237 as follows . as for , @xmath424 where @xmath163 is due to , and @xmath164 comes from @xmath425 and @xmath426 . similarly , @xmath427 + @xmath428 can be proved to satisfy as well . in addition , @xmath429 , @xmath24 , i.e. , holds true . these feasibility implies that @xmath430 . by combining the above two facts , we have @xmath431 and complete the proof . then , we apply the charnes - cooper transformation again to @xmath415 , the result of which is denoted by @xmath432 , to study the property of @xmath433 and @xmath250 . it is noteworthy that the charnes - cooper transformed constraint of admits the form given by @xmath434 , @xmath24 , where @xmath435 is the column vector inside @xmath7 of the lhs of and @xmath436 indicates the rhs expression . since it is easily checked that @xmath437 as a result of feasibility , we have @xmath438 , which implies that @xmath439 according to schur complement . it thus follows that holds true , @xmath440 : @xmath434 . as such , we can show that can be recast into a constraint without @xmath441 , i.e. , not related to @xmath442 , @xmath443 , following the same procedure as ( * ? ? ? * appendix iii ) by exploiting ( * ? ? ? * _ lemma 2 _ ) . similarly , the charnes - cooper transformed constraint of can also be rewritten without @xmath442 , @xmath443 . hence , the partial lagrangian for @xmath432 in terms of @xmath442 and @xmath443 can be expressed as ( [ eq : lagrangian of ( p2.1-sdp - eqv ) ] ) ( see top of next page ) , where @xmath445 denotes a tuple comprising all the associated primal and dual variables : @xmath344 , @xmath345 , and @xmath349 are lagrangian multipliers associated with @xmath442 , @xmath443 , and , @xmath93 , respectively ; @xmath346 is the dual variable associated with the only equality constraint ; @xmath446_{i=1}^n)$ ] and @xmath447_{i=1}^n)$ ] denote those associated with @xmath448 and @xmath449 , @xmath24 , respectively ; finally , the diagonal entry of @xmath450_{i=1}^n)$ ] denotes the dual variable associated with @xmath451 , @xmath24 . the kkt conditions related to are accordingly given by 1 . * case i * : @xmath456 , @xmath133 . in this case , since @xmath457 ( c.f . ) due to the strong duality , it is easily verified that @xmath458 and therefore @xmath459 . * case ii * : @xmath460 such that @xmath461 . according to , @xmath462 , which is a positive semidefinite diagonal matrix . according to the similar argument made in * case iii * of appendix [ appendix : proof of rank - one solution with cj for sps ] , it is thus shown that @xmath249 has maximum one zero diagonal entry and the positive definiteness of @xmath463 can be proved by definition without difficulty . as @xmath344 ( c.f . ) again complies with the difference between a positive definite matrix and a rank one matrix , i.e. , @xmath464 , it turns out that @xmath465 according to . then , following the same procedure as that in appendix [ appendix : proof of rank - one solution with cj for sps ] , @xmath466 of proposition [ prop : rank - one solution with cj for dps ] can be proved ( details omitted for brevity ) . finally , it is verified that @xmath238 is related to @xmath250 merely with its diagonal entries , viz @xmath467_{i , i}$ ] , @xmath166 , ( c.f . , ) . furthermore , denoting @xmath468_{i , i}^{1/2}]_{i=1}^n$ ] by @xmath251 , it is easily checked that the diagonal entries remain the same after we replace @xmath250 with @xmath252 . combining the above two facts , we arrive at the conclusion that such modification does not affect its optimality while returning a rank - one @xmath250 for @xmath238 , which completes the proof for @xmath405 . e. tekin , s. member , and a. yener , `` the general gaussian multiple - access and two - way wiretap channels : achievable rates and cooperative jamming , '' _ ieee trans . inf . theory _ 54 , no . 6 , pp . 27352751 , jun . 2008 . j. zhang , m. c. gursoy , `` relay beamforming strategies for physical - layer security , '' in _ proc . ieee annual conference on information sciences and systems ( ciss10 ) _ , princeton , usa , mar . 2010 , pp.16 . y. yang , q. li , w .- k . ma , j. ge , and p. c. ching , `` cooperative secure beamforming for af relay networks with multiple eavesdroppers , '' _ ieee signal process . 1 , pp . 3538 , jan . 2013 . h. xing , k .- k wong , z. chu , and a. nallanathan , `` to harvest and jam : a paradigm of self - sustaining friendly jammers for secure af relaying , '' _ to appear in ieee trans . signal process . _ , available on - line at arxiv:1502.07066v3 . z. ding , k. k. leung , d. l. goeckel , and d. towsley , `` opportunistic relaying for secrecy communications : cooperative jamming vs. relay chatting , '' _ ieee trans . 10 , no . 6 , pp . 17251729 , jun . 2011 . y. yang , q. li , w .- k . ma , j. ge , and m. lin , `` optimal joint cooperative beamforming and artificial noise design for secrecy rate maximization in af relay networks , '' in _ proc . ieee workshop on signal processing advances in wireless communications ( spawc ) _ , darmstadt , de , jun . 2013 , pp.360364 . q. li , y. yang , w .- k . ma , m. lin , j. ge , and j. lin , robust cooperative beamforming and artificial noise design for physical - layer secrecy in af multi - antenna multi - relay networks , " _ ieee trans . signal process . 1 , pp . 206220 , jan . k. ishibashi , `` dynamic harvest - and - forward : new cooperative diversity with rf energy harvesting , '' in _ proc . ieee international conference on wireless communications and signal processing ( wcsp ) _ , hefei , cn , oct . 2014 , pp . 15 . a. nasir , x. zhou , s. durrani , and r. kennedy , `` relaying protocols for wireless energy harvesting and information processing , '' _ ieee trans.wireless commun . 36223636 , jul . z. ding , s. perlaza , i. esnaola , and h. poor , `` power allocation strategies in energy harvesting wireless cooperative networks , '' _ ieee trans . wireless commun . _ , vol . 13 , no . 2 , pp . 846860 , feb . 2014 . s. timotheou , i. krikidis , g. zheng , and b. ottersten , `` beamforming for miso interference channels with qos and rf energy transfer , '' _ ieee trans . wireless commun . _ , vol.13 , no.5 , pp . 26462658 , may 2014 . m. zhao , x. wang , and s. feng , `` joint power splitting and secure beamforming design in the multiple non - regenerative wireless - powered relay networks , '' _ ieee commun . 9 , pp . 15401543 , sept . 2015 . d. ng , e. lo , and r. schober , `` robust beamforming for secure communication in systems with wireless information and power transfer , '' _ ieee trans . wireless commun . _ , vol . 13 , no . 8 , pp 45994615 , aug . 2014 . q. li , q. zhang , and j. qin , `` secure relay beamforming for simultaneous wireless information and power transfer in non - regenerative relay networks , '' _ ieee trans . 5 , pp . 24622467 , jun . 2014 . q. li and w .- k . ma , `` spatially selective artificial - noise aided transmit optimization for miso multi - eves secrecy rate maximization , '' _ ieee trans . signal process . 27042717 , may 2013 .

in the field of coding theory , calculations of good upper and lower bounds on the size of a code serves as a benchmark for code design . techniques of constructing subspace codes for error correction in random network coding ( also called ` projective space codes ' ) are similar to the techniques used to construct binary codes . a projective space code is defined as a subset of subspaces of a vector space . the viewpoint of lattice theory for projective space codes has already been suggested @xcite . a lattice is a partially ordered set in which any two elements have a least upper bound and a greatest lower bound . the set of all subspaces of a vector space is called a projective space and the set of all subsets of a set is called a power set . both the power set and the projective space form lattices under inclusion order . a ` lattice scheme ' is defined as a subset of a lattice . it has already been observed in @xcite that binary codes and projective space codes are lattice schemes . binary codes are power set schemes and projective space codes are projective space schemes . the lattice framework common to projective space codes and binary codes have also been observed in @xcite and @xcite . the authors of @xcite establish that the projective space codes are analogs of classical binary codes . they also investigate the notions of linearity and complements of a code . in @xcite , the relationship between binary codes and projective space codes is investigated thoroughly using the framework of lattices and a generalized notion of orthogonal complement is introduced for lattices . one of the ways to generate a bound on the code size is to capture a bare minimum structure for the bound through an appropriate mathematical abstraction and generalize the approach . for example , the introduction of ` association schemes ' captured the notion of sphere packing and sphere covering arguments in traditional coding theory @xcite and strengthened the bounds by a linear programming approach . in our paper , the notion of singleton bound is generalized using the framework of lattices . since error correcting codes for random network coding was proposed in @xcite , there has been a lot of activity in the theory of projective space codes . the authors of @xcite consider projective space codes in which all the subspaces are of the same dimension . such codes are called constant dimension codes . the koetter - kschischang singleton bound ( kks bound ) for constant dimension codes was derived and achieved asymptotically in @xcite . in the case of binary codes , hamming distance between two binary vectors is a metric on the binary coding space . similarly the subspace distance , introduced in @xcite , between two subspaces of a vector space is a metric on the projective space . in a projective space , the number of elements in a sphere of a particular radius depends on the center of that sphere @xcite . however , in the case of binary codes , the number of elements in a sphere depends only on the radius of the sphere and not on the center of the sphere . so the subspace distance on the projective space behaves differently from the hamming distance on the usual binary vectors . as noted in @xcite , the anticode bound introduced in @xcite gives a setting in which one can get tighter sphere packing bounds for constant dimension codes . a close look at the bounds for subspace distance and non - existence of nontrivial perfect codes has been given in @xcite . an appropriate generalization for singleton bounds ( analogous to association scheme for sphere packing / covering bounds ) is not known yet . in this paper , we propose that the lattice framework is appropriate for generalizing singleton bounds . we derive an upper bound on the size of a lattice scheme , which will be called lattice singleton bound ( lsb ) for the remainder of the paper . we get the kks bound and the classical singleton bound as special cases of the lsb . while most results in projective space codes assume that the dimensions of the subspaces in the code are constant , a few papers in literature have considered codes with non - constant dimension . in particular , the authors of @xcite derive a gilbert - varshamov type bound , which we will term as ` etzion vardy- gilbert varshamov bound ' ( ev - gvb ) , for non - constant dimension codes . we derive a _ new _ upper bound on the code size for non - constant dimension codes by applying the lsb to the projective lattice . to the best of our knowledge , our singleton bound is the first upper bound for non - constant dimension codes . the construction of constant dimension codes have been the main focus in the field of code constructions for projective spaces . the first constant dimension subspace codes introduced in @xcite achieve the kks bound asymptotically . and there is no known code constructions that exactly achieve the singleton bound proposed in @xcite . the question of achieving that singleton bound is an open question . a large class of constant and non - constant dimension codes using ferrer s diagrams and lifted rank metric codes are constructed in @xcite . the authors of @xcite report marginal improvements over the size of the code compared to the codes of @xcite . a plot of code sizes of the codes reported in @xcite compared to a plot of our upper bound shows that both the reported codes have code sizes that are close to optimal . this also establishes that our upper bound is tight for atleast certain coding parameters . the contributions of this paper are as follows : 1 . we derive a singleton bound in the framework of lattices . we recover the singleton bound for binary codes and the singleton bound for constant dimension subspace codes ( kks bound ) as special cases of our lsb . we obtain a new singleton bound for non - constant dimension subspace codes . we compare our bound on the code size with the code sizes of non - constant dimension codes found in the literature . we also compare our upper bound with a gilbert - varshamov type lower bound . the plots reveal that our bound is tight for certain parameters of the code . the paper is organized as follows : we introduce the preliminaries of lattices in section [ sec_lat_prelims ] . section [ sec_lat_sch ] introduces the idea of lattice schemes where it is shown that classical codes and subspace codes are both lattice schemes . section [ sec_main_res ] introduces the singleton bound in the framework of lattice schemes and contains the main theorem of this paper . section [ sec_main_res ] contains the main distinguishing feature of projective space codes which we believe make singleton type bounds weak in projective spaces . in section [ sec_upp_bou ] , we obtain a new upper bound for non - constant dimension codes and plot our upper bound with ev - gvb along with different codes constructed in the literature . finally , in section [ sec_conc ] , we conclude by summarizing our contributions and discussing the scope for future work . _ notations : _ a set is denoted by a capital letter and its elements will be denoted by small letters ( for example , @xmath0 ) . the set @xmath1 is defined as @xmath2 . all the sets considered in this paper will be finite . given a set @xmath3 , @xmath4 denotes the number of elements in the set and for a subset @xmath5 of @xmath3 , @xmath6 denotes the complement of the set @xmath5 in @xmath3 . for two sets @xmath5 and @xmath7 , @xmath8 denotes the cartesian product of the two sets , i.e. @xmath9 . @xmath10 represents the symmetric difference of sets , i.e. @xmath11 . a lattice will be denoted by @xmath12 and sometimes we will drop the join and the meet notation , simply calling it @xmath13 . @xmath14 denotes the finite field with @xmath15 elements where @xmath15 is a power of a prime number . @xmath16 denotes all the non zero elements of @xmath14 . the symbol @xmath17 denotes a vector space ( generally over @xmath14 ) . for a subset @xmath18 of @xmath17 , @xmath19 denotes the linear span of all the elements in @xmath18 . given two subspaces @xmath5 and @xmath7 , @xmath20 denotes the smallest subspace containing both @xmath5 and @xmath7 . let @xmath17 be a @xmath21 dimensional space over @xmath14 . then the symbol @xmath22 denotes the grassmanian , i.e. the set of all @xmath23 dimensional subspaces of @xmath17 . the number of elements in @xmath22 is denoted as @xmath24 . @xmath25 denotes the @xmath21 dimensional vector space of @xmath21-tuples over @xmath14 . given a vector @xmath26 , @xmath27 denotes the @xmath28-th co - ordinate of @xmath29 . the support of @xmath26 ( denoted by @xmath30 ) is defined as the set of indices where the vector is non - zero . @xmath31 denotes the vector space @xmath25 with the hamming metric , i.e. @xmath32 . this section serves as a quick introduction to lattice theory . all the lattice theory definitions and theorems required for the rest of the paper are given in this section . we follow notations and definitions from @xcite . [ def_poset_order ] a _ poset _ is a pair @xmath33 , where @xmath34 is a set and @xmath35 is a binary relation ( called the _ order relation _ ) on the set @xmath34 satisfying : 1 . ( reflexivity ) for all @xmath36 2 . ( antisymmetry ) if @xmath37 , then @xmath38 3 . ( transitivity ) if @xmath39 , then @xmath40 for the remainder of the paper , @xmath34 denotes a poset with @xmath35 as the order relation . if @xmath41 and @xmath42 , then we use the shorthand @xmath43 . @xmath41 is read as `` x is less than y '' or `` x is contained in y '' . an _ upper bound _ ( _ lower bound _ ) of a subset @xmath3 of @xmath34 is an element @xmath44 containing ( contained in ) every @xmath45 . the _ least upper bound _ ( _ greatest lower bound _ ) of @xmath3 is the element of @xmath34 contained in ( containing ) every upper bound ( lower bound ) of @xmath3 . if a least upper bound , or a greatest lower bound of a set exists , it is unique due to the antisymmetry property of the order relation ( definition [ def_poset_order ] ) . the least upper bound of a set @xmath3 is denoted by sup @xmath3 and the greatest lower bound is denoted by inf @xmath3 . a _ lattice _ @xmath13 is a poset which has the property that @xmath46 , the sup@xmath47 exists and inf@xmath47 exists . the sup@xmath47 is denoted by @xmath48 ( read as `` a _ join _ b '' ) and the inf@xmath47 is denoted by @xmath49 ( read as `` a _ meet _ b '' ) . the lattice itself is denoted by @xmath12 . we will assume , for the purposes of the paper , that all the lattices are finite and have a unique greatest element denoted by @xmath50 , and a unique least element denoted by @xmath51 . a _ sublattice _ of a lattice @xmath13 is a subset @xmath52 of @xmath13 that satisfies the following condition : @xmath53 a map @xmath54 from @xmath13 to @xmath52 is said to be a _ lattice homomorphism _ if it satisfies the following conditions : 1 . @xmath55 2 . @xmath56 further , if @xmath54 is bijective , then we say that @xmath13 and @xmath52 are _ let @xmath57 and @xmath58 denote the power set of @xmath3 . we can view @xmath59 as a poset where set inclusion is the order relation . the power set under this order is a lattice . for any subsets @xmath5 and @xmath7 , @xmath60 and @xmath61 . further , the lattice generated by subsets of @xmath62 is a sublattice . notice that @xmath63 and @xmath64 . @xmath65 one can completely specify an order of a poset ( finite ones ) by a _ hasse diagram _ , like the one shown in fig . [ fig_power_set ] . if @xmath66 and there is no @xmath67 such that @xmath68 , we say that `` @xmath69 _ covers _ @xmath70 '' . in the hasse diagram of a lattice , @xmath70 and @xmath69 are joined iff @xmath69 covers @xmath70 or @xmath70 covers @xmath69 . the diagram is drawn in such a way that if @xmath69 covers @xmath70 , then @xmath69 is written above @xmath70 . clearly , @xmath66 iff there exists a path from @xmath70 moving up to @xmath69 . the order relation is completely specified by such a diagram . naturally , @xmath50 will be the topmost element and @xmath51 will be the lowest element . ): a lattice of subsets ] for a set @xmath3 , @xmath71__pow(@xmath3)__@xmath72 will denote lattice of the power set of @xmath3 with union and intersection as join and meet of the lattice respectively . a lattice @xmath12 is _ distributive _ if the following conditions hold : 1 . for all @xmath73 2 . for all @xmath74 it is immediate that for any set @xmath3 , @xmath71__pow__(@xmath3)@xmath72 is a distributive lattice . however all lattices are not distributive , as seen in the examples below . let @xmath17 be a vector space of dimension @xmath21 over a field @xmath75 . the class of all the subspaces of @xmath17 , denoted by @xmath76 , can be ordered under inclusion in a manner similar to a set . the join of two subspaces @xmath5 and @xmath7 will then be the smallest subspace containing both @xmath5 and @xmath7 . this means that @xmath77 and the largest subspace contained in both @xmath5 and @xmath7 is @xmath78 . so the meet of @xmath5 and @xmath7 is @xmath78 . such a lattice will be denoted as @xmath79 . clearly @xmath80 and @xmath81 . this lattice will be called the _ projective lattice_. this lattice is not distributive . to see this , we consider the vector space @xmath82 over @xmath83 . the hasse diagram of @xmath84 is shown in fig . [ fig_subspace ] . here , @xmath85 , @xmath86 and @xmath87 . clearly , @xmath88 and thus @xmath84 is not distributive . @xmath84 is identified by the name _ m@xmath89_. @xmath65 : a lattice of subspaces , also called @xmath90 the above lattice is not distributive , i.e. @xmath91 fails for three distinct subspaces @xmath92 . if @xmath93 , one can prove that @xmath94 and thus it behaves partly distributively . this property is captured in the following definition : a lattice @xmath12 is _ modular _ if the following condition holds : + ( modularity ) if @xmath95 , then @xmath96 if the lattice is distributive , the modularity condition holds . this means that _ any distributive lattice is modular_. however @xmath79 is a modular lattice that is non distributive . the observation that for @xmath71__pow(@xmath3)__@xmath72 , any two elements @xmath97 satisfy @xmath98 , and in an analogous fashion , for @xmath79 , any two elements @xmath97 satisfy @xmath99 , seems to suggest that modular lattices must satisfy an equation of the form @xmath100 for some real - valued function @xmath101 on the lattice . this is indeed true and in turn , such a function helps characterise modular lattices . we need the following few definitions and results to make the characterisation precise . an _ isotone valuation _ on a lattice @xmath13 is a real valued function @xmath101 on @xmath13 that satisfies : 1 . ( valuation ) for all @xmath102 , @xmath103 2 . ( isotone ) @xmath104 additionally , the isotone valuation is called _ positive _ , if @xmath105 . @xmath106 is said to be the distance induced by @xmath101 . * pg.230 , th.1 ) [ thm_pseudo_metric ] given a lattice @xmath13 and an isotone valuation @xmath101 , the function @xmath107 is a metric iff @xmath101 is positive . in general , @xmath108 is a pseudo metric . in a lattice @xmath12 , a _ chain _ @xmath109 is a subset of @xmath13 with the property that for all @xmath110 , @xmath66 or @xmath111 . we say that in a chain , any two elements are comparable . given two elements @xmath112 , a chain @xmath113 of @xmath13 with the property @xmath114 is called a _ chain between @xmath70 and @xmath69_. the _ length _ of the chain is defined as @xmath23 . the _ height _ of an element @xmath29 in a lattice @xmath13 is the maximum length of a chain between @xmath51 and @xmath29 . it is denoted by @xmath115 . the number @xmath116 is called _ the height of the lattice _ @xmath13 or the _ dimension of _ @xmath13 . note that the chain from @xmath51 to @xmath51 contains only one element , and thus @xmath117 . we need an additional property to characterise modular lattices . a lattice is said to have the _ jordan - dedekind property _ if all maximal chains between two elements have the same finite length . all lattices need not have the jordan - dedekind property . for example , in the @xmath118 lattice , shown in fig . [ fig_hasse_diag_n_5 ] , there are two maximal chains from @xmath119 to @xmath120 . the maximal chain @xmath121 has three units of length and the other maximal chain @xmath122 has a length of two units . thus @xmath118 does not satisfy the jordan dedekind property . it turns out that modularity is closely related to this property . lattice : a lattice without jordan - dedekind property ] [ jd ] ( * ? ? ? * pg.41 , th.16 ) let @xmath12 be a lattice of finite length with the height function @xmath123 , then the following conditions are equivalent : 1 . @xmath13 is a modular lattice . 2 . @xmath13 has the jordan - dedekind property and @xmath123 is a valuation . the @xmath118 lattice shown in fig . [ fig_hasse_diag_n_5 ] does not satisfy the jordan - dedekind property and therefore , due to theorem [ jd ] , @xmath118 is non - modular . the height function @xmath123 is a positive isotone evaluation for modular lattices . therefore from theorem [ thm_pseudo_metric ] , the distance function induced by @xmath123 is a metric on the modular lattice . all the lattices , considered in this paper , will be modular . we will use this observation in the next section to define a coding metric for the definition of lattice schemes . we , therefore , have the following proposition which follows from theorem [ thm_pseudo_metric ] . [ thm_height_metric ] let @xmath13 be a modular lattice of finite length with the height function @xmath123 , then @xmath124 is a metric on @xmath13 . clearly @xmath125=1 $ ] iff @xmath29 covers @xmath51 . such elements are called _ atoms _ of the lattice . the atoms of the lattice are analogues of the singletons in lattice of sets ( one dimensional spaces in the lattice of subspaces ) . a _ geometric modular lattice _ is a modular lattice of finite height in which every element is a join of atoms . further , if the geometric modular lattice is distributive , then it will be called _ geometric distributive_. the lattice of subsets is a an example of a geometric distributive lattice and the lattice of subspaces is an example of a geometric modular lattice . is shown . in part ( b ) , a non - geometric , non - distributive modular lattice @xmath126 is depicted . ] however , all modular ( distributive ) lattices are not geometric . for example , consider the sublattice ( in fact , a chain ) @xmath127 of @xmath71__pow(@xmath128)__@xmath72 ( shown in part ( a ) of fig . [ fig_nongeometric ] ) . @xmath129 is not geometric because @xmath128 can not be obtained as a join of atoms of @xmath130 ( since there is only one atom in @xmath129 ) . @xmath129 is distributive since it is a sublattice of distributive lattice viz . therefore @xmath130 is an example of non - geometric distributive lattice . in order to construct a non - geometric non - distributive modular lattice , we will consider a sublattice of @xmath131 . for the sake of presentation , we will represent a vector @xmath132 by the natural number @xmath133 . for example , @xmath134 will be represented as @xmath135 . consider the sublattice @xmath136 of @xmath131 . @xmath126 is modular since it is a sublattice of a modular lattice viz . @xmath126 is not distributive because it contains a copy of @xmath137 as a sublattice ( @xmath137 is non - distributive ) . it can be verified that @xmath138 can not be obtained as a join of atoms of @xmath126 and thus @xmath126 is non - geometric . therefore @xmath126 is an example of non - geometric non - distributive modular lattice . henceforth all the lattices are assumed to be geometric modular unless otherwise mentioned . in order to develop a lattice based framework for singleton bounds , we need a definition of a code in this framework . in this section , we define ` lattice schemes ' which will serve as analogues of codes . we will show that hamming space , rank metric space and the projective spaces are examples of lattice schemes . although some of these observations have been mentioned in @xcite , we repeat it for the sake of continuity . a lattice scheme , which is analogous to a code , is defined as follows : let @xmath13 be a lattice and @xmath139 be the metric induced by the height function @xmath123 of the lattice @xmath13 . a _ lattice scheme _ @xmath109 in @xmath140 is a subset of @xmath13 and the _ minimum distance of @xmath109 _ , denoted by @xmath119 is defined as @xmath141 the dimension of a lattice scheme is defined as @xmath142 . a coding space @xmath143 is a metric space where @xmath3 is a set and @xmath144 is a metric on @xmath3 . a code @xmath109 in a coding space @xmath143 is a subset of @xmath3 . the connection between lattice schemes and codes is made precise in the following definition . [ def_transform ] let @xmath109 be a lattice scheme in @xmath140 and @xmath145 be a code in a coding space @xmath146 . we say that the _ code _ @xmath145 is _ equivalent to a lattice scheme _ @xmath109 , if there exists a function @xmath147 , that satisfies the following conditions : 1 . @xmath148 2 . @xmath149 for all @xmath150 @xmath151 is called a transform for the code @xmath145 . a transform @xmath151 of a code is one - one . to prove this fact , assume @xmath152 , then @xmath153 . by 2 ) of definition [ def_transform ] , this would imply @xmath154 . and since @xmath144 is a metric , we have @xmath155 . @xmath65 when a lattice scheme is equivalent to a code , we also say the code is equivalent to the scheme . a transform for a code preserves the distance between any pair of codewords . therefore whenever a lattice scheme is equivalent to a code , the lattice scheme will have the same minimum distance as the code . the following proposition follows from definition [ def_transform ] , [ prop_dist_preserve ] let @xmath109 be a lattice scheme with minimum distance @xmath119 that is equivalent to code @xmath145 with minimum distance @xmath156 , then 1 . @xmath157 2 . if @xmath158 , then there exists @xmath159 such that the lattice scheme @xmath5 is equivalent to the code @xmath160 since ` @xmath109 is equivalent to @xmath145 ' , the first part follows from the definition . for the second part , since @xmath109 is equivalent to @xmath145 , there exists a transform @xmath151 . we define @xmath161 . clearly @xmath159 and the function @xmath151 still serves as a lattice transform for the code @xmath162 . thus @xmath5 is equivalent to @xmath162 . due to the above proposition , given a lattice scheme equivalent to a code , we can talk of the minimum distance without specifying whether it is the minimum distance of the code or the minimum distance of the scheme . from the second part of the proposition [ prop_dist_preserve ] , we can infer that the subsets of codes are equivalent to certain subsets of schemes . therefore , if we prove that a coding space itself is equivalent to some lattice , then any code in the coding space is equivalent to a scheme in the lattice . we use this observation to establish that every binary code is equivalent to a scheme in the power set lattice . [ exmpl_binary_codes ] let @xmath163 , @xmath164_pow(@xmath3)_@xmath72 and @xmath165 . @xmath13 is a geometric distributive lattice and @xmath166 as seen in the previous section . consider codes in the coding space @xmath167 where @xmath168 is the hamming distance between two vectors . we claim that the entire coding space @xmath169 is equivalent to the power set lattice @xmath13 . to see this let , @xmath170 it can be verified that @xmath171 ( where @xmath172 represents the symmetric difference operator ) and that @xmath173 is onto . further , @xmath174 . the number of elements in @xmath175 will be the hamming weight of @xmath176 . thus @xmath177 . therefore , @xmath173 is the power set lattice transform for the binary code . since the map is onto , @xmath173 is a bijective map . by application of the second part of proposition [ prop_dist_preserve ] , we see that every binary code is equivalent to a power set lattice scheme . @xmath65 the lattice of subspaces , discussed in the previous section , also provide examples of lattice schemes . in a projective lattice , the metric induced by the height function is the subspace distance . therefore , any subspace code is equivalent to a lattice scheme in the projective lattice . let @xmath17 be a vector space over @xmath178 is a projective lattice with height function @xmath179 . the coding space is @xmath180 where @xmath181 represents the subspace distance defined in @xcite . for any two subspaces @xmath5 and @xmath7 , @xmath182 . since @xmath183 , the metric induced by the height function is the subspace distance . the identity map can be a transform in this case . and thus , subspace schemes are equivalent to subspace codes.@xmath65 lattice schemes in the above lattice correspond to subspace codes in projective spaces @xcite . certain constant dimension subspace code constructions , described in @xcite , make use of the rank distance gabidulin codes @xcite . the rank metric codes are actually equivalent to gabidulin codes since they are ` lifted ' from gabidulin codes . we show that rank distance codes are equivalent to certain schemes in the projective lattice . we also show that the lifting function is a transform . let @xmath184 denote the space of @xmath185 matrices over @xmath14 . for @xmath186 , define @xmath187 . this coding space is the rank distance space . we claim that the rank distance space is equivalent to a scheme in the projective lattice . consider the projective lattice @xmath188 with the height function @xmath179 . define a transform @xmath151 as follows : @xmath189 } \end{aligned}\ ] ] notice that @xmath190 for any @xmath5 in @xmath191 . therefore subspaces whose dimension is not @xmath192 are not in the range of @xmath151 . therefore @xmath151 is an example of a transform that is not onto . the function @xmath151 defined above is called the lifting of @xmath5 in @xcite . the authors of @xcite establish that @xmath193 . thus rank metric codes are equivalent to schemes in the projective lattice and @xmath151 is the projective lattice transform for the rank metric code . @xmath65 all our main results on the bounds and their behavior are derived in this section . we introduce the notion of puncturing a scheme and investigate the effects of puncturing on the minimum distance of a lattice scheme . it will be proved that , after puncturing a scheme , the maximum drop in minimum distance will be two . however , if the lattice is known to be distributive , it is shown that the maximum drop in minimum distance is one . this observation will be applied repeatedly until the minimum distance drops to zero . we will need the following definition of _ whitney number of the second kind _ , from @xcite , to state the lattice singleton bound : the _ whitney numbers _ @xmath194 of a lattice @xmath13 in a lattice with height @xmath123 and @xmath195 is defined as @xmath196 the whitney numbers of a lattice count the total number of elements in the lattice of a given height . a scheme @xmath109 is said to be _ punctured to @xmath197 _ if @xmath198 for some @xmath199 . if @xmath200 has a height of @xmath201 , the scheme @xmath109 is said to be _ punctured by a dimension_. note that for a @xmath199 , @xmath202 is a sublattice of @xmath13 @xcite . for two lattice elements @xmath200 and @xmath203 of equal height , we have @xmath204 . this means that the whitney numbers of two different punctured lattices remain the same , as long as the elements that puncture the scheme are of the same height . we need the following lemma ( called the ` distance drop lemma ' ) to establish the proof of the main theorem later . [ lem_drop_dist ] let @xmath109 be a scheme in @xmath140 with minimum distance @xmath119 , and let @xmath109 be punctured by a dimension to @xmath197 , then 1 . @xmath13 is distributive @xmath205 @xmath206 2 . in general , @xmath207 proof of 1 ) : we assume @xmath13 is distributive . let @xmath208 , @xmath209 and @xmath210 . we have to show that @xmath211 for all @xmath212 . by the definition of @xmath139 , @xmath213 - h[\tilde{a } \wedge \tilde{b } ] . \end{aligned}\ ] ] since the lattice is distributive , we can write it as , @xmath214 - h[a \wedge b \wedge w ] . \end{aligned}\ ] ] the height function is a valuation and thus satisfies @xmath215 + h[x \wedge y ] = h[x ] + h[y]$ ] . we use this in the above equation to obtain , @xmath216 + h[w ] - h[a \vee b \vee w ] - h[a \wedge b \wedge w ] . \end{aligned}\ ] ] since @xmath217 , it must be that @xmath218 \leq h[i ] = n$ ] . using this inequality and @xmath219 = n-1 $ ] , we get @xmath220 + ( n-1 ) - n - h[a \wedge b \wedge w ] . \end{aligned}\ ] ] clearly @xmath221 and therefore @xmath222 \leq h[a \wedge b]$ ] . using this , the definition of @xmath223 and the fact that @xmath119 is the minimum distance of the scheme @xmath109 , we finally get @xmath224 proof of 2 ) : we have to show that @xmath225 for all @xmath212 . again by the definition of @xmath139 , @xmath226 - h[\tilde{a } \wedge \tilde{b}].\end{aligned}\ ] ] repeatedly using the fact that the height function is a valuation and thus satisfies @xmath215 + h[x \wedge y ] = h[x ] + h[y]$ ] , we get , @xmath227 + h[b ] + 2h[w ] - h[a \vee w ] \\ - h[b \vee w ] - 2h[a \wedge b \wedge w ] . \end{aligned}\ ] ] clearly @xmath221 and therefore @xmath222 \leq h[a \wedge b]$ ] . additionally @xmath228 and @xmath229 , which means @xmath230 , h[\tilde{b } ] \leq h[i ] = n$ ] . using this inequality and @xmath219 = n-1 $ ] , we get the following : @xmath231 + h[b ] + 2(n-1 ) - 2n - 2h[a \wedge b ] \end{aligned}\ ] ] using the definition of @xmath223 and the fact that @xmath119 is the minimum distance of the scheme @xmath109 , we finally get , @xmath232 the distance drop lemma states that in a non - distributive lattice , the drop in the minimum distance after puncturing a dimension , can be at most two units . so it would be interesting to know if it is possible that the drop of two units is exhibited by some scheme in a non - distributive lattice . the following example constructs such a scheme . [ example_proj_space ] let @xmath17 be a three dimensional space , over @xmath233 , spanned by @xmath234 . let @xmath235 @xmath236 , @xmath237 @xmath238 and @xmath239 @xmath240 . we have @xmath241 but @xmath242 . note that @xmath243 ( that is , the sublattice generated by @xmath244 , @xmath245 and @xmath246 is not distributive ) as expected . if our scheme contained @xmath247 and @xmath246 , then puncturing the scheme with @xmath246 would have left @xmath246 as it is and punctured only the remaining two subspaces . the above example clearly illustrates that the lack of distributivity in a lattice can drop the distance of a punctured scheme by two units . we will now derive a singleton bound for lattice schemes , that establishes an upper bound on the cardinality of the scheme , for a given minimum distance for geometric modular lattices . [ theorem_lsb ] if @xmath248 is a geometric modular lattice with height @xmath123 and @xmath249 , @xmath139 is the metric induced by the height function , and @xmath109 is a scheme of @xmath13 with minimum distance @xmath119 , then @xmath250 where 1 . @xmath251 , when @xmath13 is _ distributive _ , 2 . @xmath252 , when @xmath13 is _ modular_. given the lattice @xmath13 , pick an element @xmath200 of height @xmath253 ( which always exists in a geometric modular lattice ) . now puncture the lattice by a dimension to get a new geometric modular lattice @xmath129 with height @xmath253 and suppose that drop in minimum distance of the scheme , after puncturing a dimension , is at most @xmath254 . if @xmath255 is the minimum distance of the punctured scheme @xmath197 , then @xmath256 . if @xmath257 , then all elements in the punctured scheme @xmath197 are still distinct . we repeat the puncturing operating on the new lattice . the puncturing is repeated maximum number of times so that the minimum distance of the punctured scheme does not drop to zero . in other words , we keep puncturing the lattice , until the minimum distance of the punctured scheme is just short of zero . let us say that the original lattice was punctured @xmath258 times . since the minimum distance of the punctured scheme is still non zero , the punctured scheme contains exactly the same number of elements as @xmath109 . at this stage , the minimum distance of the scheme is non - zero and all the elements of the punctured scheme are in a lattice of height @xmath259 . clearly the number of elements in the scheme @xmath109 is upper bounded by the total number of elements in the punctured lattice . thus @xmath260 if the minimum distance of the scheme , after being punctured @xmath258 times , is @xmath261 , then @xmath262 and @xmath261 is the smallest number such that @xmath263 . this implies that @xmath264 . from the lemma [ lem_drop_dist ] , we know that @xmath265 for a distributive lattice and @xmath266 for a modular lattice . thus @xmath267 for a distributive lattice and for a modular lattice , @xmath252 . this completes the proof . we remark that the bound does not equal @xmath268 , for the case @xmath269 , which should yield only two elements . the bound is weaker for large minimum distance because we have no information about the range of heights of the elements of the scheme . we can tighten the bound by systematically puncturing and projecting a dimension as done in @xcite . also , as noted in example [ example_proj_space ] , puncturing a dimension of a scheme does not reduce the heights of the elements of the scheme in a uniform manner . so we need the following definition , given two elements @xmath270 and @xmath200 in a lattice @xmath13 , define @xmath271 to be equal to an element of height @xmath272 that is contained in @xmath273 . in particular , if @xmath200 is fixed and @xmath274 , the set @xmath275 is said to be _ punctured and projected by a dimension_. thus when @xmath276 , @xmath271 is arbitrarily chosen to be an element immediately below @xmath270 in the hasse diagram of the lattice . for all other cases , @xmath277 . the following example illustrates that puncturing a dimension reduces the heights of different elements of a scheme by different amounts . it also demonstrates that puncturing and projecting by a dimension equally reduces the heights of all elements of a scheme . in example [ exmpl_punc_proj ] is shown here . the dark arrow shows that after puncturing and projecting by @xmath246 both @xmath244 and @xmath245 can drop to the same subspace @xmath278 . however , puncturing by @xmath246 alone drops @xmath245 to @xmath278 but not @xmath244 . ] [ exmpl_punc_proj ] let @xmath17 be a three dimensional space , over @xmath233 , spanned by @xmath234 . let @xmath235 @xmath279 , @xmath237 @xmath280 and @xmath239 @xmath279 . [ fig_puncture ] shows how puncturing and projecting drops the heights of all elements by one unit . thus the dimension of @xmath244 remains the same after puncturing by @xmath246 but the dimension of @xmath245 drops by one unit . however , after puncturing and projecting the lattice by @xmath246 , both @xmath244 and @xmath245 drop by a height of one unit . to be precise , define @xmath281 and observe that @xmath282 . but @xmath283 . to define @xmath284 we need to choose a one dimensional subspace of @xmath244 . for instance , we can choose @xmath285 . @xmath65 we can now establish that puncturing and projecting a dimension of a scheme can drop the minimum distance of the scheme by at most two units . this lemma is useful since it tells us that every time we puncture and project by a dimension , all the elements in the scheme drop once in height . [ lemma_punctured_drop ] let @xmath109 be a scheme in a lattice @xmath140 with minimum distance @xmath119 , and let @xmath249 and let @xmath199 such that @xmath286 . if the maximum height of @xmath109 was @xmath287 and the minimum height of @xmath109 was @xmath192 , then the maximum height of @xmath288 is @xmath289 and the minimum height is @xmath290 . additionally @xmath291 by definition of @xmath288 , all the elements of @xmath288 have lost a unit of height . so the only thing to establish is that @xmath292 . the proof is very similar to the previous lemma . by the definition of @xmath139 , @xmath293 + h[\overline{b \wedge w } ] - 2h[\overline{a \wedge w } \wedge \overline { b \wedge w } ] . \end{aligned}\ ] ] from the definition of @xmath294 , it follows that @xmath295 and thus @xmath296 \leq h[x \wedge w]$ ] for all @xmath297 and thus , @xmath298 - 1 ) + ( h[b ] - 1 ) - 2h[a \wedge b ] \end{aligned}\ ] ] and therefore from the formula for height we get , @xmath299 we will need lemma [ lemma_punctured_drop ] to get the following improvement on the code size , [ thm_tighter_bound ] let @xmath109 be a scheme in a geometric modular lattice @xmath13 with minimum distance @xmath119 . let @xmath192 denote the smallest height of all the elements in @xmath109 and @xmath287 denote the largest height of all the elements in @xmath109 . then , @xmath300 let @xmath200 be an element of height @xmath253 and @xmath301 . by lemma [ lemma_punctured_drop ] , the punctured and projected scheme in @xmath129 has a maximum height of @xmath302 , a minimum height of @xmath290 and a minimum distance drop of at most two units . now the proof proceeds in a manner similar to the proof of the theorem [ theorem_lsb ] . after puncturing @xmath303 times , the elements of the scheme will have a maximum height of @xmath304 and a minimum height of @xmath305 and the theorem is proved . we will now apply theorem [ theorem_lsb ] and theorem [ thm_tighter_bound ] to two important lattice schemes ( namely the power set lattice and the projective lattice ) and derive the corresponding lsb . the lsb that we obtain coincides with the singleton bound found in the literature . first , we derive the classical singleton bound as a corollary to the lsb theorem . let @xmath109 be a code in @xmath306 , with minimum distance @xmath119 , then @xmath307 in example [ exmpl_binary_codes ] , @xmath308 is the number of subsets of size @xmath309 in the scheme @xmath109 , given that @xmath310 . using the fact that @xmath13 is geometric and distributive , @xmath251 . first we observe that , @xmath311 and @xmath312 applying the lsb theorem , we get , @xmath313 which is the singleton bound for @xmath306 . now we derive the kks bound reported in @xcite . it follows as a direct corollary to theorem [ thm_tighter_bound ] when the lattice is chosen to be the lattice of subspaces . if @xmath314 and @xmath315 , then @xmath316 let @xmath317 where @xmath109 is chosen to be a subset of the grassmanian @xmath22 . then @xmath314 and the lattice coefficients @xmath318 . now apply theorem [ thm_tighter_bound ] to @xmath109 . the distance drop lemma , used in the main theorem , along with example [ example_proj_space ] clearly shows that at every step of puncturing the distance can drop by at most two units . on the other hand , one can construct examples where distance drop is one unit . now suppose some scheme achieves the lsb ( and thus the kks bound ) . then that scheme has elements which consistently drop by a distance of two units _ irrespective _ of which element is puncturing it . and this should happen at every puncturing step until the minimum distance drops to zero . since this seems highly unlikely , we conjecture the following : _ conjecture : _ there are no schemes that can achieve the lsb for projective spaces . in particular , there are no constant dimension codes that can achieve the kks bound . it is clear that many examples fit into the framework of lattice schemes and the proposed lsb specializes to the known singleton bounds in literature . in this section , we will show that the lsb gives a new upper bound when applied to projective spaces . a family of non - constant dimension code constructions in projective spaces have been reported by @xcite and @xcite . our lsb plots show that the code sizes , of both reported codes , are near optimal for certain fixed values of the coding parameters . singleton bound in projective spaces will be specified using gaussian numbers ( they are the @xmath15-analogues of binomials @xcite ) . lower bounds on code sizes based on sphere covering methods can be found in @xcite . upper bounds on code sizes for projective space codes , found in the literature , are for constant dimension codes . the following bound is an upper bound on the code size of a non - constant dimension code in a projective space . let @xmath109 be a code in @xmath319 ( where @xmath17 is vector space over @xmath14 ) , with minimum distance @xmath119 , then @xmath320 where @xmath321 denotes the gaussian number . the @xmath322 lattice with @xmath323 is a modular geometric lattice . we note that @xmath324 the gaussian number @xmath325 is the total number of @xmath52-dimensional subspaces of an @xmath326-dimensional space over @xmath14 and thus by theorem [ theorem_lsb ] , @xmath327 where @xmath328 . the bound is weak when the minimum distance @xmath119 is comparable to the code dimension @xmath21 . however when the minimum distance is smaller than code dimension , the tightness of the bound is not apparent . therefore we investigate the behaviour of our bounds , at such ranges , by plotting it with other bounds in the literature . we will compare our bound ( lsb ) to the gilbert varshamov bound ( ev - gvb ) proposed in @xcite for various values of the minimum distance . further , we will plot points achieved by various codes in the literature . the plot is shown in fig . [ fig_drawing ] . the minimum distance of the projective code and the finite field size has been fixed at @xmath329 and @xmath268 respectively , throughout this section . the plot clearly shows our upper bound above the ev - gv lower bound . the points marked etzsilb codes and khks code are the code parameters reported in @xcite and @xcite respectively , for @xmath330 and @xmath331 . [ fig_drawing ] shows that the etzsilb codes and khks codes are close to optimal for @xmath330 and @xmath331 . the khks code sizes are roughly @xmath332 bits away from the upper bound . it has already been observed that the distance drop in projective spaces is variable and thus we believe that our bound is weak for larger values of @xmath119 . in this paper , we introduced the notion of lattice schemes which serve as analogues of codes . we derived a general notion of singleton bound for which the classical singleton bound for binary codes and the singleton - like bound for constant dimension subspace codes were special cases . we proved that puncturing a dimension gives a tight bound for distributive lattices but not for projective lattices . however , it was seen that puncturing and projecting a dimension gives a tighter bound , for projective lattices . this leads us to conjecture that it is not possible to achieve the kks bound . the upper bound derived in the paper seems to be tighter when the minimum distance is much smaller than the code dimension . a next step would be to tighten the bounds for large minimum distances . it would be interesting to investigate whether there exists lattice schemes that achieve the lsb with equality . in particular , it would be very interesting to know if there exists a reed solomon - like scheme which specializes to the different counterparts in binary and subspace codes . also , whether quantum codes are lattice schemes can be investigated .

since the observation and explanation of the gunn - effect in the early 60 s it has been known that traveling field inhomogeneities are generated by negative differential velocity ( ndv).@xcite the properties of these traveling gunn - domains have been studied extensively theoretically and experimentally @xcite leading to several proposals for applications . the most important application is probably the frequency generator . although oscillator devices based on the gunn - effect operating in the range of 100 ghz have been reported,@xcite gunn - oscillators have not led to the expected breakthrough in technology . this is largely due to the difficulties in tuning the oscillation frequency . furthermore , it is impossible to vary the drift velocity vs field characteristics ( @xmath0 curve ) except by using different bulk materials ( e.g. , gaas , ingaas or inp ) . these problems may be overcome by using an entirely different class of ndv exhibiting materials formed by semiconductor superlattices . in an applied electric field perpendicular to the two - dimensional layers , several different transport mechanisms such as miniband transport , resonant tunneling , and real - space transfer from well to barriers ( @xmath1 transport ) give rise to complex @xmath0 curves with several regions of ndv as shown in fig . [ slwfield ] . in contrast to the above mentioned materials , the tunability of the barrier and well thickness as well as the control over the barrier height in the superlattice can be used to tailor the drift velocity vs field @xmath0 characteristics . by proper engineering , samples with several ndv regions can be manufactured with control over previously inaccessible features such as the actual shape of the ndv ( cf . [ slwfield ] ) . the frequency of the oscillations also depends on the number of moving charges in the superlattice , which may be controlled by varying the doping in the quantum wells . currently , three different types of superlattice oscillations due to ndv have to be distinguished : * the most prominent type of oscillations is found , when carriers within the miniband are accelerated beyond the brillouin zone boundary , where their drift velocity becomes negative . in the absence of scattering , the electron wave packet then oscillates with the bloch frequency @xmath2 , where @xmath3 and @xmath4 denote the superlattice period and planck s constant , respectively.@xcite these so - called bloch oscillations were predicted for superlattices by esaki and tsu in 1970.@xcite inspired by the prospect of thz frequency generators based on superlattice bloch oscillators , a long series of investigations targeting the miniband regime followed leading to the observation of damped bloch oscillations in experiments with pulsed optical excitation a few years ago.@xcite * a different type of oscillations occurs in the miniband regime , when the scattering times are shorter than the tunneling times . in this case transient charge accumulations traveling through the superlattice may lead to oscillations in the current . experiments using pulsed optical excitation of carriers in superlattices with wide minibands showed damped oscillations with frequencies up to 20 ghz.@xcite more recently , reflection gain up to 60 ghz in doped superlattices with miniband widths of up to 80 mev was demonstrated.@xcite * very recently a third type of oscillations has been shown to exist in weakly coupled superlattices in field regions where tunneling into higher subbands takes place.@xcite as a consequence of domain formation , this type of oscillations may also be observed in field regions where the electrons tunnel between the lowest subbands . while the first two oscillation types have never been observed under constant bias conditions , this new type is shown to operate over a broad frequency range practically independent of the external circuit and without any external triggering . in this paper we investigate experimentally and theoretically the third type of oscillations in lightly and moderately doped superlattices ( sl ) . this type of oscillations may be generally observable in the ndv regime of the @xmath0 characteristics of such superlattices . in a companion paper , @xcite we have shown that self - sustained oscillations may exist if the dimensionless doping parameter , @xmath5 ( @xmath6 is the 3d sl doping , and @xmath7 is the field at the resonance we consider ) lies between two values @xmath8 which can be calculated numerically from the local @xmath0 characteristics ( @xmath9 is the number of sl periods ) . if several tunneling resonances exist ( cf . fig . [ slwfield](a ) and ( b ) ) the frequency of the oscillations increases with increasing index of the subband involved in the tunneling resonance . this observation is attributed to the larger drift velocities associated with the resonances involving higher subbands . furthermore , several new oscillatory modes without the typically well - defined domains are shown to exist . these oscillatory modes can be tuned in frequency simply by changing the applied voltage . because the oscillations are observable even at room temperature , we suggest to use superlattice oscillators based on resonant tunneling into higher subbands as a tunable source for high frequencies . four gaas / alas superlattice structures , grown by molecular beam epitaxy , are discussed in this paper . the samples are hereafter referred to as 9.0/4.0 ( doped and undoped ) , 9.0/1.5 , and 13.3/2.7 , where the first number refers to the gaas well width and the second to the alas barrier thickness in nm ( cf . table [ samples ] ) . the superlattices are embedded between two highly doped al@xmath10ga@xmath10as contact layers with a doping density n@xmath11=2@xmath12 @xmath13 ( si for @xmath14- and be for @xmath15-type doping ) forming @xmath14-@xmath16-@xmath14 and @xmath15-@xmath17-@xmath14diodes . after evaporating auge / ni contacts onto the top and substrate side and alloying them for ohmic connections , the samples are wet - etched into mesas of 120 @xmath18 m diameter . the circular top contacts of 70 @xmath18 m diameter leave a large part of the mesas uncovered to allow for optical access . all experiments are performed at 5 k in a he - flow cryostat using high - frequency coaxial cables with a bandwidth of 20 ghz . the time - averaged current - voltage data are recorded with a keithley smu236 . the time - resolved current is detected with a tektronix sampling oscilloscope csa 803 using the ghz sampling head sd-32 . [ iv6750 ] shows the current - voltage ( i - v ) characteristics of the doped 9.0/4.0 sample in forward bias . clearly visible are two plateau regions with almost constant current between 0.5 and 4 v and 6.5 and 8.5 v. to prove that these plateaus are related to the subband resonances , time - of - flight ( tof ) measurements were performed on an undoped 9.0/4.0 reference sample ( @xmath15-@xmath17-@xmath14 diode).@xcite the peak photocurrent is thereby taken as a measure for the inverse transport time , which at low excitation densities is directly proportional to the drift velocity for homogeneous fields . note that for the presentation in fig . [ iv6750 ] the built - in voltage ( @xmath19 v ) of the @xmath15-@xmath17-@xmath14 diode has been subtracted from the reverse bias voltage applied to the @xmath15-@xmath17-@xmath14 diode and the sign has been inverted . the resonances for tunneling from the first into the second ( @xmath20 ) as well as into the third ( @xmath21 ) conduction subband are clearly observed at 6 and 15 v , respectively . the low - field transmission maximum ( @xmath22 ) , which in the case of a strongly coupled superlattice corresponds to miniband conduction , is not resolved in this sample , because the transport time is too long in this field range to be resolved with this method . however , it has been shown previously that even weakly coupled superlattices exhibit a negative differential drift velocity in this regime.@xcite in addition to the @xmath23 resonances due to @xmath24-subbands in the gaas wells and the exponential nonresonant background , a pronounced step - like increase of the peak photocurrent is observed at about 10 v , which is assigned to enhanced transport through the lowest @xmath25-level in the alas barriers . the lowest @xmath25-states of the alas barriers are energetically located in such a way that transport through these @xmath25 states becomes possible at voltages between the @xmath20 and @xmath21 resonances . in the drift velocity vs field characteristics , this @xmath1 resonance leads to a step - like increase of the drift velocity as indicated by the dotted line in fig . [ slwfield](b ) . the voltages of the observed resonances can be compared to the calculated energy levels listed in table [ samples ] . for example , the bias voltages @xmath26 for @xmath27 resonances are calculated using @xmath28 where @xmath29 denotes the energy of the injecting conduction subband and @xmath30 the energy of the conduction subband , in which the carriers are tunneling into . @xmath9 refers to the number of periods and @xmath31 to the electron charge . in the case of @xmath1 resonances , however , the distance between the levels is reduced to half the superlattice period . therefore , an additional factor of 2 has to be included , i.e. , @xmath32 where @xmath33 denotes the energy of the receiving state in the barrier . the voltages for the @xmath24 resonances @xmath20 and @xmath21 are calculated using eq . ( [ resonances1 ] ) resulting in 5.4 and 14.6 v for the 9.0/4.0 sample , respectively . using eq . ( [ resonances2 ] ) , we obtain a resonance voltage of 8.2 v for the @xmath34 resonance in this sample . all values agree well with the observed values . note that all observed resonances appear at slightly higher voltages due to field inhomogeneities and screening effects . to investigate the electric field inside the superlattice , photoluminescence ( pl ) measurements have been performed.@xcite due to the quantum - confined stark effect ( qcse ) , the pl from regions with strong electric fields is red - shifted with respect to the pl from low - field regions . the experiments reveal that in both plateau regions of this sample the pl consists of two peaks with their respective intensity depending on the applied bias voltage . this stark - splitting of the pl signal is taken as direct evidence for the existence of two distinct electric - field domains , which are formed because of ndv and current conservation . for the first plateau the consequence of this conservation law is depicted by the dashed line in fig . [ slwfield](b ) . clearly , the electric field of the low - field domain ( @xmath35 ) corresponds to the @xmath22 resonance peak , while the field strength of the high field domain ( @xmath36 ) lies on the next rising branch of the @xmath0 characteristics slightly below the @xmath20 resonance.@xcite applied to the second plateau , the current conservation law ( dashed lines in fig . [ iv6750 ] ) shows that the high field domain in this voltage region forms as a consequence of resonant transport through the @xmath37 level in the barriers ( @xmath34 transport ) . in contrast to previous work,@xcite where this new formation mechanism was considered for the first time , this measurements provide the first direct evidence relating @xmath25-levels in the barriers to domain formation . however , it should be mentioned that it is not very surprising to observe domain formation due to transport through the @xmath37 level . for domain formation to occur , only a _ minimum of the @xmath0 curve _ is required in conjunction with a sufficiently high carrier density.@xcite in that sense the subject of domain formation and conditions for oscillations can be discussed simply by looking at the @xmath0 curve as discussed in a companion paper.@xcite in the 13.3/2.7 sample the calculated @xmath34 resonance ( @xmath38v ) is located between the @xmath21 and @xmath39 resonances , which are calculated to be at 9.5 and 17.9 v , respectively . consequently , the i - v characteristic exhibits three plateaus as can be seen in fig . [ iv6173 ] . the thin dashed line has been added to indicate the positions of the transport maxima for homogeneous fields . unlike in the 9.0/4.0 sample , the pl in the plateau regions shows no stark splitting indicating that the field distribution inside the sl is less sharply defined . a possible explanation of this behavior is the existence of rather broad resonances in this weakly doped superlattice . however , the i - v characteristics shows some clear signs of domain formation ( e.g. , current branches@xcite ) , but the differences between the electric field values of these domains seem to be rather small . since the qcse is quadratic in the electric field , the peaks of the two domains can not be resolved in the pl experiments . because stable electric field domains have been investigated intensively over the last years , we will not discuss the stationary regime any further . we will rather focus on the non - stationary behavior instead . the absence of a regular pattern of current branches in the plateaus is a first hint that the field distribution inside the sl may not always be stable . indeed , the current , e.g. , in the lower plateau of the 9.0/4.0 sample ( fig . [ trace6750 ] ) , shows a strongly time - dependent component , which has been attributed to an oscillatory instability of the field distribution inside the superlattice.@xcite during one cycle of the oscillation a monopole forms near the cathode , drifts through the superlattice towards the anode and dissolves , while a new monopole is formed.@xcite this type of oscillation is denoted the first oscillation mode . it is interesting to note that the displayed trace also contains a number of small current spikes . supported by simulations , voltage turn - on measurements have shown that these small spikes are due to well - to - well hopping of the domain boundary moving through the superlattice.@xcite thus , the number of current spikes indicates how many superlattice periods are involved in the oscillation . for the displayed oscillation this leads to a number close to thirty , which allows to speculate that for this particular voltage the domain boundary crosses about three quarters of the superlattice during one cycle of the oscillation . to analyze the oscillations in more detail , we have measured the voltage dependence of the oscillation frequency within the plateaus . the fourier transform@xcite exhibits a series of frequency maxima , of which we take the fundamental frequency as the oscillation frequency of the field distribution . the voltage dependence of the oscillation frequency in the two plateaus of the 9.0/4.0 sample is plotted in fig . [ freq6750 ] showing that the observed frequencies in the first plateau are weakly voltage dependent , while in the second plateau there is no dependence on the applied voltage at all . the noticeable increase of the frequency with voltage at the end of the first plateau is due to the charge monopole having less room to travel from the beginning of the superlattice to the position of the high field domain . in both plateaus there are regions , where sinusoidal oscillations could not be observed . in these regions the current contains a time dependent component , which is not periodic and therefore difficult to measure . in particular the discontinuities in the frequency at the beginning of the first plateau suggest that the oscillations may become chaotic , when the charge monopole oscillates near the contacts . while chaotic behavior has been shown to exist theoretically in the case of an external driving frequency , @xcite the boundary regions of the plateaus are still under investigation in this respect . very recently driven and undriven chaos have been observed in this system.@xcite note that chaotic behavior has also been observed at the onset of trap - dominated gunn oscillations in ultra - pure p - ge.@xcite this behavior is due to intermittent switching between small and large amplitude current oscillations,@xcite although its theoretical interpretation remains uncertain.@xcite in contrast to the 9.0/4.0 sample , the frequencies of the oscillations in the 13.3/2.7 sample exhibit a pronounced voltage dependence as shown in fig . [ hf6173 ] for three different temperatures in the third plateau . the oscillations begin to appear close to the resonance maximum and decrease monotonically with increasing voltage . furthermore , in this sample oscillations are only detected near the beginning of the plateaus in the region of ndv . this observation not only applies to the third plateau , but also to the first and second plateaus . thus , there is a clear difference between the oscillations observed in these two samples . unlike in the 9.0/4.0 sample , the time - resolved pl does not show an oscillating two - domain solution,@xcite but merely an intensity modulated broad line . these two differences are taken as a signature for the existence of a different oscillatory mode in this sample , which is not related to well - defined electric field domains . in fig . [ ghz465 ] the experimental results on the 9.0/1.5 sample are shown . in this case we used a 25 @xmath18 m mesa in order to keep the current at a reasonable level . due to the stronger coupling , ghz oscillations are already observed in the lowest plateau . in addition to a strong voltage dependence of the frequency , there is also an abrupt jump to lower values . higher plateaus could not be measured due to the large increase in current at about 8 v. time - resolved pl spectra on a 120 @xmath18 m mesa do not show two well - separated lines . however , intensity oscillations with the same frequencies are clearly observed . current oscillations on these large mesas are also measured using pulsed electric fields to avoid the heating of the sample . [ iv6173 ] shows the current voltage characteristic of the 13.3/2.7 sample at room temperature . from the fact that it still shows the plateau - like features related to resonant tunneling one can speculate that the oscillations may also be found at higher temperatures . in fig . [ hf6173 ] the oscillations in the third plateau are observed up to 200 k , while in fig . [ rt6173 ] tunable oscillations indeed exist even at 295 k in the case of the second plateau . these results are very encouraging , because they prove that a device such as a frequency generator based on the recently found tunable oscillatory mode can in principle be manufactured . however , in order to obtain well - defined resonances even at room temperature , one probably has to switch to another material system with stronger confinement . for example , superlattices based on ingaas / inalas material system exhibit pronounced plateaus even at room temperature.@xcite therefore , it is expected that with properly chosen parameters oscillators based on ingaas / inalas superlattices can be operated at room temperature . in order to explain why the frequency depends in one case on the applied voltage , while it is voltage independent in the other case , the condition for the two oscillatory modes has to be investigated in more detail . numerical calculations on the following discrete drift model have been performed for the electric field @xmath40 and 3d electron density @xmath41 of the @xmath42 sl period as well as for the total current density @xmath43 using the following equations,@xcite @xmath44 here @xmath45 is the period index , @xmath6 the 3d doping density , @xmath3 the sl period , @xmath46 the average permittivity , and @xmath47 the applied voltage between the two sl ends . eqn . ( [ poisson])-([bias ] ) have to be supplemented using the appropriate initial conditions for the field profile @xmath40 as well as with a boundary condition for @xmath48 , which we take as @xmath49 where @xmath50 denotes a dimensionless charge accumulation within the first sl period due to the excess doping outside the sl ( @xmath51-@xmath16-@xmath51 structure ) . we expect that the narrower the barriers are , the larger @xmath50 is , since the overlap between wavefunctions of adjacent quantum wells is larger for narrower barriers . to find a precise relation between the structure of the @xmath51 region before the sl and @xmath52 , microscopic modeling of the full @xmath51-@xmath16-@xmath51 structure ( characterized in ref . ) is needed . this is outside the scope of the present work . however , we can estimate the values of @xmath52 by comparing experimental and calculated frequency versus voltage bias curves . thus comparing figs . [ freq6750 ] , [ hf6173 ] and [ ghz465 ] with fig . [ fig5.4](b ) , we obtain that @xmath52 should be very small ( @xmath53 ) for samples 9.0/4.0 and 13.3/2.7 , whereas @xmath52 should be from 10 to 100 times larger for sample 9.0/1.5 ( see the discussion below ) . the @xmath54 function is a phenomenological electron velocity , which is proportional to the tunneling probability . the main transport mechanism in weakly coupled superlattices is sequential resonant tunneling ( srt ) . thus , @xmath54 has peaks at the applied voltages or electric field strengths corresponding to the alignment of the energy levels of adjacent quantum wells as given in eq . ( [ resonances1 ] ) , which are always separated by a region of ndv . the experimentally observed voltage dependence of the frequency in samples 9.0/4.0 and 13.3/2.7 ( cf . figs . 5 and 6 ) can be explained in terms of the actual shape of the @xmath54 curve , the doping density in the wells and the charge accumulation within the first period . the crucial features in the @xmath54 curve are the separation between the peaks as well as their width . the position of the peaks can be approximated by eq . ( [ resonances1 ] ) . the width of the peaks and the values of the minima of @xmath54 depend on the particular scattering mechanisms , which are present in the samples . however , it is reasonable to assume that the peaks of the @xmath54 curve for the 13.3/2.7 sample are wider than those of the 9.0/4.0 sample , since the narrower the barriers , the wider the energy intervals for resonant tunneling.@xcite therefore , we will compare simulations of the same model using two different @xmath54 curves and different doping densities . we shall use curves @xmath54 with only one peak , corresponding to a @xmath55 resonance , with @xmath56 . thus we are only dealing with the self - sustained oscillations occurring in the plateau after the chosen resonance . adding more peaks does not appreciably change the self - sustained oscillations on each different plateau , although new features may appear for highly doped sls . in fact , there are stationary nonuniform solution branches with electric field profiles presenting simultaneous coexistence of three or more domains if a @xmath54 curve with several peaks is considered.@xcite to compare experimental results belonging to three different samples to our data , it is convenient to render the eqs . ( [ poisson])-([bc ] ) dimensionless . suppose we want to analyze time - periodic current oscillations at a certain plateau of the current - voltage characteristic . we adopt as the units of electric field and velocity the corresponding values at the peak of the velocity curve @xmath54 prior to the plateau , @xmath57 and @xmath58 , respectively . thus , we have to scale the magnitudes in the figures by a different factor for each sample and plateau . the time scale factors , @xmath59 , for the 9.0/4.0 ( second plateau , after the @xmath20 resonance ) , 13.3/2.7 ( second plateau , after the @xmath20 resonance ) , and 9.0/1.5 ( first plateau , after the @xmath22 peak ) , samples are 3.57 , 0.98 , and 0.30 ns , respectively . the electric field scale factors , @xmath60 , are on the same order of magnitude , @xmath61 , @xmath62 , and @xmath63 v / cm . we set@xcite @xmath64 where the dimensionless parameter @xmath65 is defined as @xmath66 a typical self - sustained current oscillation ( far from the voltage bias corresponding to the onset or the end of the instability ) is caused by the generation , motion and annihilation of domain walls , which are charge monopoles . @xcite we want to estimate the dependence of the oscillation frequency or period @xmath67 on the bias @xmath68 . there exist three different oscillation modes depending on the value of the bias : * _ first mode : _ oscillation of the electric field profile around an almost uniform stationary state ( fig . [ profiles](a ) ) . it occurs when the bias is just above the onset of the instability . it can be shown that the oscillation frequency decreases as the bias increases by numerically solving the problem of linear stability of the stationary electric field profile as discussed in a companion paper.@xcite * _ second mode : _ monopole recycling ( fig . [ profiles](b ) ) , which typically occurs when the bias is larger . in this mode of oscillation , the period is the sum of the formation time of a monopole ( the time it takes each new charge accumulation injected at the first sl period to form a monopole ) plus the time elapsed until another charge accumulation is injected . new monopoles are formed when the current is close to its maximum value ( @xmath69 , or @xmath70 in dimensional units ) and the monopole formation time is small compared to the period of the oscillation . the bias interval , over which the second oscillatory mode exists , can be shown to increase with sample doping or with the number of sl periods . * _ third mode : _ oscillation of the electric field profile about a nonuniform stationary state containing two electric field domains ( fig . [ profiles](c ) ) . this mode is sometimes present in weakly doped samples , which oscillate for larger biases . it then occurs near the end of the oscillatory branch . in this third mode of oscillation , a domain wall remains pinned at a given location and the field values at the low and high field domains oscillate in antiphase . the shape of the current oscillation is almost sinusoidal and the maximum current is clearly below 1 ( where the accumulation layers were injected from the injecting contact for the mode discussed before ) . a similar oscillation mode has been found in simulations of imperfect superlattices.@xcite next we report the results of numerical simulations of the self - sustained oscillations of the current in the discrete model for different values of the dimensionless parameters . the simulations yield self - sustained current - oscillations when the doping @xmath65 is such that the corresponding stationary field profile is clearly inhomogeneous , e.g. , as in fig . [ fig5.3 ] . the current on the middle branch of stationary solutions is significantly larger than @xmath71 . in these cases the bias has to be within a certain interval @xmath72 ( @xmath73 is on the ndv branch of @xmath74 , and @xmath75 may or may not be larger than the minimum @xmath76 ) to generate self - sustained current oscillations . both @xmath77 and @xmath75 are increasing functions of @xmath52 . in fig . [ fig5.3](a ) , the oscillatory behavior begins at @xmath78 via a supercritical hopf bifurcation . the amplitude of the oscillation increases with bias following a square - root law and the frequency is almost constant . at @xmath79 the branch of oscillatory solutions disappears via a second hopf bifurcation . for longer sls with @xmath80 and the same doping , fig . [ fig5.3](e ) , or for a 50-well sl with larger doping , [ fig5.3](c ) , the end of the oscillatory branch is different : a limit cycle collides with the unstable fixed point from the middle branch of the z - shaped current - voltage characteristics and disappears . this bifurcation scenario results in a decrease of the frequency down to zero , when the collision takes place , while the amplitude is unchanged to lowest order . the bias interval where the oscillatory branch exists , @xmath72 , shrinks as @xmath65 decreases , and for @xmath81 ( @xmath82 for @xmath83 , @xmath84 ) , there is no oscillatory solution . close to the hopf bifurcations the first and third modes of oscillation are found . we no longer see recycling and motion of domain walls . instead , in all these examples , there is an interval of bistability between the self - sustained oscillations and the stable stationary lower branch of the z - shaped characteristic when @xmath85 . the bistability may be hard to observe experimentally because the basin of attraction of the self - sustained current oscillations is very small and most initial field profiles evolve to stationary situations . a similar type of bistability can be found when ( i ) wiggles in the static current - voltage characteristic @xmath86 appear , and ( ii ) the upper branch of a wiggle is unstable against oscillatory behavior while the lower branch is stable . see fig . 1 of ref . for @xmath87 . [ fig5.3 ] shows the variation in magnitude and frequency of the current oscillations as a function of the bias for different doping values and number of sl periods . to clarify the nature of the oscillations let us describe the _ second mode _ of oscillations , which occurs for most biases . [ fig5.1](a ) shows the field profiles at different times of a given period of the current oscillations for a 50-well sl . we can identify a field profile consisting of two well - formed field domains at time ( 1 ) . we define the field of the domain to be the electric field @xmath88 at the position where @xmath89 has a local minimum , i.e. , the variation in the field is minimal . the domain wall between two domains is a charge monopole containing an excess of electrons ( @xmath90 ) . its evolution towards the receiving contact ( until it disappears ) is captured by snapshots at times ( 2 ) to ( 4 ) in fig . [ fig5.1](a ) . let @xmath91 be the instantaneous position of the domain wall defined as the sl period for which @xmath89 is a local maximum . except for the time intervals , in which two domain walls are present simultaneously , the constant voltage condition implies that @xmath92 where @xmath93 and @xmath94 are the values of the field in the low and high field domains , respectively . within the low field domain there is a tiny inhomogeneity ( contact layer ) close to the injecting contact , which is due to the boundary condition.@xcite as long as the low field domain is in the region of positive differential velocity ( @xmath95 ) , the field profile is stable and the contact layer follows adiabatically the current . the position @xmath91 of the domain wall moves to the right with a certain velocity @xmath96 , where @xmath97 is always significantly less than 1 ( the peak velocity in dimensionless units ) . ( [ eqel ] ) implies that @xmath93 , @xmath94 , and @xmath98 must increase with time in order to fulfill the fixed voltage condition . when @xmath93 surpasses 1 , a small charge accumulation is injected at @xmath99 , which moves to the right and grows until it becomes a well - formed charge monopole . see the snapshots at times ( 3 ) and ( 4 ) in fig . [ fig5.1](a ) . the process is repeated in time after the old monopole has disappeared and the new one is well established . note that two different monopoles are present simultaneously during a certain part of the oscillation period . for a long sl the displacement current is quite small during most of each oscillation period , and we have @xmath100 and @xmath101 [ @xmath102 are the three solutions of the equation @xmath103 for @xmath104 . note that the solutions @xmath105 and @xmath106 disappear for @xmath107 . see fig . [ fig5.2 ] for a 200-period sl . notice that now the domains are well - established at times for which the values of the current are different from those depicted in fig . [ fig5.1](a ) . even for a smaller sl as the one in fig . [ fig5.1 ] , the displacement current is small , when the current is near its maximum value for each period [ see the time intervals between ( 1 ) and ( 2 ) in fig . [ fig5.1](b ) and between ( 2 ) and ( 3 ) in fig . [ fig5.2](b ) ] . notice that the field value on the low field domain is near the peak of the @xmath74 curve during these time intervals . this will be useful in what follows . it is important to understand the bias dependence of the frequency which could be used to tune the frequency of an eventual device in agreement with the experimental observations of section ii . the frequency might increase or decrease with bias depending on the sample parameters @xmath9 and @xmath65 , as shown in figs . [ fig5.3](b , d , f ) . let us start at the time @xmath108 when @xmath109 [ then @xmath110 , @xmath111 , the contact layer looses its stability , and a charge accumulation is injected at @xmath99 ( @xmath112 in fig . [ fig5.2 ] ) . one period of duration @xmath67 is completed at @xmath113 when the next charge accumulation is injected . at this instant , the position of the charge monopole is given by eq . ( [ eqel ] ) with @xmath114 and @xmath115 which yields @xmath116 thus , the charge accumulation has to travel a distance @xmath117 in the time @xmath67 . during the monopole formation time @xmath118 , @xmath119 , and the mean velocity of the charge accumulation is @xmath120 . on the other hand , the same charge accumulation moves at a smaller mean velocity equal to @xmath96 ( with @xmath121 ) for the rest of the period , @xmath122 , once it has become a charge monopole . we therefore have @xmath123 this gives @xmath124 we now estimate the monopole formation time @xmath118 . the new charge accumulation @xmath125 travels with a velocity of the order of @xmath126 and is amplified in time via @xmath127 with an initial value @xmath128 . at time ( 3 ) of fig . [ fig5.2 ] , @xmath129 , @xmath125 is so large that it separates two different well - formed field domains . then we may consider that a new charge monopole has been born . this charge monopole travels at the velocity @xmath96 mentioned above and it sharpens as it moves ; see the point marked by ( 4 ) in the fig . [ fig5.2](a ) . this stage lasts until the fields before and after the monopole reach the values @xmath130 and @xmath131 , respectively , and we are back to situation ( 1 ) having completed one period . mathematically , this behavior can be well described by an asymptotic analysis in the continuum limit ( @xmath132 ) . @xcite in order to estimate @xmath118 , we note that it is determined by the condition @xmath133 where @xmath134 . then eq . ( [ eqamplification ] ) yields @xmath135 now we obtain @xmath136 from eq . ( [ eqel ] ) with @xmath137 . up to the first order in @xmath138 , this yields @xmath139}{\nu m^2 } \ , ( t - t_0 ) .\ ] ] linearizing @xmath140 , we obtain @xmath141}}\ , .\ ] ] now @xmath117 decreases as @xmath68 increases , which can be seen from eq . ( [ eqm ] ) . as @xmath121 , eq . ( [ eqperiod ] ) indicates that there are two competing mechanisms influencing the dependence of @xmath67 with @xmath68 . in general the monopole formation time @xmath118 is negligible compared to @xmath142 in the following cases : ( i ) for long sls ( @xmath117 in eq . ( [ eqperiod ] ) is then large ) , ( ii ) when the values of @xmath52 are large ( i.e. , @xmath143 small ) , ( iii ) for large doping @xmath65 . in these cases , @xmath142 should be a decreasing function . we find that the frequency increases with @xmath68 for large values of the quantities @xmath144 and @xmath52 , while it decreases otherwise . these behaviors are illustrated in figs . [ fig5.3 ] and [ fig5.4 ] . we can now interpret the experimental observations according to our theoretical results . * sample 9.0/4.0 has narrow peaks , high doping with respect to the others , and is short ( n=40 ) . therefore , the second oscillation mode dominates for most biases with a short stage of monopole formation ( @xmath145 is quite large ) . during most of each oscillation period , the electric field on the low field domain takes on values close to that of the resonant peak . this makes it possible to distinguish two peaks in the pl data . @xcite the frequency is practically independent on the bias.@xcite this sample is modeled by the velocity curve shown in the inset of fig . [ depen](a ) . the results of the numerical simulation for current oscillations and field profiles are also shown in fig . [ depen](a ) . * sample 13.3/2.7 has wide peaks , low doping , and is also short ( @xmath83 ) . we would expect the first mode of oscillation to dominate , which agrees with the experimental observation that the frequency is a decreasing function of the bias . in this mode the high field domain is never well established , and the low field domain is off - resonance during most parts of each oscillation period ( cf . [ depen](b ) ) . this explains that no stark - splitting of the pl signal is observed in the experiments for this sample . * sample 9.0/1.5 is similar to sample 13.3/2.7 , except that its doping is 2.5 times larger and the sl is shorter ( @xmath146 ) . the velocity curve used in the simulations for this sample is shown in the inset of fig . for this sample , fig . [ ghz465 ] seems to indicate that the first two oscillation modes are important : for low bias the first mode dominates , whereas for larger bias the second mode does . [ ghz465 ] also shows that the frequency increases with bias abruptly after reaching its minimum value . numerical calculations suggest that this could correspond to having a larger excess charge in the first sl period than in the case of sample 9.0/4.0 . this could be expected for a sl with narrower barriers ( cf . [ fig5.4 ] ) . until now we have studied eqs . ( [ poisson ] ) - ( [ bias ] ) with the boundary condition of eq . ( [ bc ] ) and @xmath52 being positive . now we are going to describe what happens if @xmath147 , i.e. , there are less electrons in the first well than the doping density ( @xmath148 ; cf . fig . [ fig6.2 ] ) . as in the case of positive @xmath52 , the oscillations are due to the generation , motion and annihilation of domain walls connecting domains ( which are regions of almost uniform electric field approximately given by the zeros of @xmath149 ) . the difference is that now the electric field profile is a monotone decreasing function of the qw index : the high - field domain is close to the beginning of the sl and the low - field domain extends to the end of the sl . the domain walls are now charge - depletion layers , having less electrons than the doping density . let us describe one period of the current oscillations for a long sl as the one in fig . [ fig6.2 ] with @xmath80 . we will assume that initially ( point marked with 1 in fig . [ fig6.2](a ) ) the field profile has two domains connected by a domain wall ( @xmath150 to the left of the domain wall and @xmath151 to the right of the domain wall with an initial value of the current @xmath152 ) . the domain wall is approximately centered at @xmath153 so that the voltage bias condition of eq . ( [ bias ] ) holds . the domain wall then moves towards the end of the superlattice with a speed close to the instantaneous value of the current . the current decreases until a certain minimum value slightly smaller than @xmath154 is reached . then a new high - field domain is created ( close to the beginning of the sl ) and the current rises sharply as the two domain walls move towards the end of the sl . when the current is near its maximum value , the old domain wall disappears as the values of the field in the intermediate and rightmost domains coalesce . then one period of the oscillation is completed . all these features can be understood by means of an asymptotic analysis to be reported elsewhere . @xcite some of the basic ideas were sketched in ref . . clearly , a superlattice oscillator based on the above described types of oscillations may be exploited in a device , e.g. , by operating the superlattice as a frequency generator . in order to establish the use of the newly found oscillations , it is important to know what maximum frequency can be achieved . while experimental investigations in this direction are still missing , it is clear that the ultimate frequency limit will be connected with the maximum drift velocity , which can be achieved without destroying the shape of the @xmath0 curve . to first order we assume that the oscillation frequency is determined _ only _ by the transition time of the carriers through the sample . this approach neglects a possible oscillation of the domain boundary over a few periods only , but allows to give an estimate for the oscillation frequencies from the current density @xmath155 , the 3d doping density @xmath156 , and the length @xmath157 of the sample using @xmath158 in fig . [ frequencies ] the measured oscillation frequencies are plotted versus the frequencies calculated using eq . ( [ frequency ] ) . for each plateau in the 9.0/4.0 sample only one measured frequency is shown , while for the 13.3/2.7 and 9.0/1.5 samples the measured minimum and maximum frequencies are plotted . the figure shows a strong correlation between drift velocity and oscillation frequency over several orders of magnitude . the highest observed _ miniband _ drift velocities are on the order of 10@xmath159 cm / s which , as sibille et al.@xcite pointed out , for a superlattice of 500@xmath160 nm length translates into a fundamental frequency of 200 ghz . however , the electric fields in the reported case can not be increased arbitrarily without breaking up the miniband , while for the subband field - oscillator this limitation does not apply . accordingly , the observed frequencies become larger for higher subband resonances , which for sufficiently thin barriers will push the drift velocities and the frequencies to new limits . looking at the oscillation frequency as a function of barrier width , we note that a change of a factor of 2.67 in barrier width leads to a change in the oscillation frequency in the first plateau of a factor of about 4000 . projecting this behavior for other material systems to feasible barrier widths , we conclude that the frequencies could reach the 100 ghz range . hence , in contrast to gunn - effect devices , the oscillation frequency of the superlattice oscillator is not only defined by the length of the sample , but also by the parameters of the heterostructure . in this respect the superlattice oscillator leaves much more room for optimization . additionally , a frequency generator based on a superlattice oscillator with several plateaus may cover several orders of magnitude with one device . in particular the tunability _ within the plateaus _ at high frequencies allows to envision a tunable frequency generator based on resonant carrier transport covering the whole ghz regime , which to our knowledge no other device is capable of . we have demonstrated that doped , weakly coupled semiconductor superlattices exhibit current self - oscillations with ghz frequencies , which can be tuned by simply changing the applied voltage . the oscillations , which are observed in gaas - alas superlattices , are not limited to low temperatures , but in one case have been observed at room temperature . they occur due to an unstable electric - field domain formation . a theoretical modeling , both analytical and numerical , demonstrates that three different oscillatory modes can occur , which depend on the detailed shape of the drift velocity curve , the doping density , the boundary condition , and the length of the superlattice . in contrast to the bloch - oscillator proposed by esaki and tsu,@xcite the self - oscillations of the field distribution are in most cases due to a formation , motion , and recycling of a charge accumulation layer inside the superlattice . in some cases the domain boundary is localized over a few quantum wells , while the field strengths of the low and high field domain oscillate . we would like to stress that the observed oscillations are also different from the damped miniband oscillations observed by le person et al.,@xcite because they occur only at field strengths , where the miniband is already destroyed by wannier - stark localization . considered from a macroscopic point of view in the sense that a traveling charge accumulation layer generates the oscillation in the current , the oscillations in doped superlattices are similar to the moving field domains in gunn diodes . in both cases the nonlinearity , which is responsible for the instability , is generated by negative differential velocity . the microscopic origin , however , is entirely different with important consequences for the frequency limit and the tunability . in conclusion , the subband superlattice oscillator has great potential for applications as a tunable ghz oscillator . in order to achieve such a device , it is necessary to fully understand the origin of the oscillations and determine all possible oscillation modes . furthermore , other material systems such as ingaas / inalas should be investigated in order to generate even higher ghz oscillations at room temperature . the authors would like to thank d. bertram for the time - of - flight experiments , r. klann and s. venakides for valuable discussions , a. fischer for sample growth , and e. wiebecke and h. kostial for sample preparation . the work was supported in part by the dfg within the framework of sfb 296 , by the dgicyt grant pb94 - 0375 , and by the eu human capital and mobility programme contract erbchrxct930413 . one of us ( aw ) acknowledges financial support from the deutsche forschungsgemeinschaft ( dfg ) . .parameters of the four investigated samples . @xmath9 denotes the number of periods and @xmath161 the two - dimensional carrier density introduced by doping . @xmath162 and @xmath163 indicate the energy positions of the lowest @xmath24-subbands in the wells and the lowest @xmath25-subbands in the barriers , respectively . the width of the lowest conduction miniband is given by @xmath164 . the energies of the subbands measured from the bottom of the gaas conduction band have been obtained by conventional envelope function calculations.[samples ] [ cols="^,^,^,^,^,^,^,^,^ " , ]

given the diversity of mass and length scales , redshifts , initial conditions and physical processes that take part in the hierarchical formation of galaxies , the fact that most galaxies exhibit some common key properties , such as an exponential disk profile e.g. @xcite , and follow common scaling relations such as star formation rate - stellar mass relation ( e.g. @xcite ) is remarkable . such simple relations may imply that despite the diversity of initial and end states , there is a subset " of universal processes that effectively dominate these relations . even though many constituents are necessary to model correctly galaxy formation to its last detail , a toy model that involves as few parameters and processes as possible , and can nevertheless yield those key properties , is powerful in its ability to offer intuition towards their understanding . this approach has motivated the _ ` equilibrium - model ' _ ( or ` bathtub - model ' ) @xcite that has been used to investigate the content of galaxies , both gaseous and stellar , the associated star formation rate ( sfr ) , metallicities , and the relations between them , especially for the ( blue ) main sequence " of galaxies . these models typically assume an equilibrium relation ( see for example @xcite ) , @xmath9 , where @xmath10 is the sfr , @xmath11 is the gas mass of the disk and @xmath12 is the gas - to - star depletion time ( or turn - over " time , @xcite ) . in addition , the outflowing baryonic mass out of the galaxy , back to the circum- or intergalactic medium ( cgm / igm ) in these models is proportional to the sfr , @xmath13 , where @xmath14 is the mass - loading factor , set by energy and momentum injection back into the interstellar medium ( ism ) . to date , all of the equilibrium models have been applied to galaxies as a whole , ignoring the internal disk structures . in contrast , in this paper we formulate a radially resolved model , in which the equilibrium relations are local " and functions of ( planar ) distance from the disk centres , and assuming that the gas dynamics is set by the mass - continuity equation and global angular momentum conservation . our model is self - consistent , as star formation , disk rotation , outflows and the disk size are all set dynamically with the disk evolution . we study in detail how systems that follow these simple assumptions evolve , reach a ( quasi-)steady state , accumulate stellar mass , and analyze dynamically their surface density and toomre-@xmath15 profiles . in our presentation , we show that many of our numerical results can be derived analytically , by solving simplified cases of the full ( partial differential ) equations , for our single phase ism . the paper is organised as follows . in [ sec : setup ] we write the evolution equations and the assumptions of our model , and discuss a dimensionless representation for it . in [ sec : cossnap ] we consider a cosmological snapshot in which we hold the cosmological conditions constant , and analyze the consequent steady state and the convergence to it . next , in [ sec : cosevol ] we generalise the cosmological snapshot , by accounting for the evolution of the halos , and varying baryon accretion rates . in [ sec : results ] we analyze the results of our model for the cosmologically evolving system , and present a parameter study . we summarise in [ sec : discussion ] . we consider an evolving , rotationally supported , star forming disk , within a cosmologically growing dm halo . the halo accretes baryonic matter from its cosmological neighbourhood , at some given rate @xmath16 , forms stars continuously , and expels matter back to the cgm / igm at a rate proportional to the star formation rate . our radially resolved equilibrium model is based on the dynamics set by mass - continuity and angular momentum conservation , together with local equilibrium relations . the mass continuity equation in cylindrical coordinates ( assuming cylindrical symmetry ) includes cosmological accretion , and star formation and outflows as source and sink terms , and follows the gas surface density in the disk , @xmath17 . it can therefore be written as @xmath18 where @xmath19 and @xmath20 are the local " sfr and outflow rates respectively , @xmath21 is the radial gas velocity within the disk ( inflow velocity is @xmath22 ) , and where for the second equality we have assumed ` local ' equilibrium relations , @xmath23 to solve the mass - continuity equation , we will have to specify the boundary condition at some outer radius @xmath24 , and the gas radial inflow velocity @xmath25 at each radius . we do that as follows : * we assume that the entire ( thin ) baryonic disk , gas+stars , is confined to a radius @xmath26 , so that @xmath27 . we determine @xmath26 by setting the specific angular momentum of the accreted gas onto the disk , at any time @xmath28 , equal to that of the parent halo with a given ( cosmological ) spin parameter @xmath29 where @xmath30 and @xmath31 are the angular momentum , energy and the virial mass of the halo , respectively , and @xmath32 is newton s gravitational constant . observations and simulations suggest that the specific angular momentum of the baryonic disks indeed remain globally equal to the parent halos as the disks evolve despite the complex inflow and outflow patterns ( e.g. @xcite ) . we therefore do not impose cosmological initial conditions for the cumulative mass distribution of specific angular momentum found in dark matter only simulations ( e.g. * ? ? ? * in realistic disks , the radial gas inflows depend in a complex way on local conditions and global tidal torques and bar instabilities . in our model we assume an azimuthally averaged gas inflow velocity , @xmath25 , at each radius @xmath33 , and we consider three cases . the first two are ( _ a _ ) constant radial inflow velocity , @xmath34 , and ( _ b _ ) radial inflow velocity that satisfies @xmath35 , where @xmath36 is the inflow velocity at the outer disk radius @xmath26 . the third case is motivated by a viscous disk in which the dissipated energy in one rotational timescale , which is of order @xmath37 , where @xmath38 and @xmath39 is the rotational velocity at each radius , is compensated by the gain in potential energy , @xmath40 @xcite , leading to ( _ c _ ) @xmath41 . our case ( _ c _ ) in combination with our mass conservation eqn . ( [ eq : dimen ] ) is similar to the viscous evolution models presented by @xcite and @xcite but with the inclusion of inflow and outflows , and self consistent star formation . with the gas surface density @xmath42 we can keep track of the gas mass in the disk out to distance @xmath43 as a function of time @xmath44 the total gas mass in the disk is @xmath45 . to obtain the accumulated stellar mass @xmath46 one has to integrate the sfr over time , and assume some stellar migration profile . we will assume in this paper that once formed , stars do not migrate radially , but rotate around the disk center together with the inflowing gas , so that @xmath47 where @xmath48 denotes a time before star formation has begun . solving for @xmath42 will also enable us to examine the local toomre-@xmath15 parameter @xcite , that quantifies the stability of the gas in the disk with respect to star formation , as a function of the galactocentric distance . by definition , @xmath49 where @xmath50 is the gas sound speed ( thermal plus turbulent ) and @xmath51 is the epicyclic frequency @xmath52 we approximate @xmath39 by the keplerian rotational velocity for a spherical mass distribution @xmath53 , that is composed of the gaseous and stellar disk components , plus the extended dark matter distribution @xmath54 we assume that the rotation curves for our two - dimensional disks differ only slightly than for spherical mass distributions ( see @xcite ) . we ignore pressure support along the disk plane ( c.f . @xcite ) . it is useful to rescale to dimensionless form . we define dimensionless time and distance coordinates , @xmath55 where @xmath56 is the crossing time , @xmath57 the normalised inflow velocity is @xmath58 we then define the dimensionless accumulation " parameter @xmath59 and the normalised surface density @xmath60 thus in dimensionless form the mass - continuity equation is @xmath61 this dimensionless representation is valid only if @xmath16 and @xmath36 are constant , which is a reasonable assumption only if the disk reaches a steady - state faster than the time it takes @xmath16 to change significantly . we assume this in [ sec : cossnap ] , but relax this assumption in [ sec : cosevol ] , which will force us back to eqn . ( [ eq : dimen ] ) . in terms of this dimensionless notation , the fixed constant inflow velocity is written as @xmath62 , the @xmath63 case is written as @xmath64 and the @xmath65 case can be written as @xmath66 , where @xmath67 is a dimensionless mass " , and the coefficient @xmath68 is set by the input parameters . we start by solving the disk evolution equation for a cosmological snapshot " , for which the halo mass and cosmological inflow rates are constant , and the dimensionless version of our model is valid . we focus first on steady state solutions with ( [ sec : ysf ] ) and without ( [ sec : nsf ] ) star formation . we demonstrate that simple mass conservation naturally leads to disks with exponential profiles over many length scales . finally , in [ sec : cca ] we examine the timescale for the convergence to steady - state . quantitatively , the case of no star formation ( i.e. infinite depletion time ) , is considered simply by setting @xmath69 in eqn . ( [ eq : cont ] ) . in steady state , the inflow velocity is independent of time , and the continuity equation can be easily solved , yielding @xmath70 this solution is valid for @xmath71 or @xmath72 ( cases ( _ a _ ) and ( _ b _ ) ) discussed above . for @xmath66 ( case ( _ c _ ) ) eqn . ( [ eq : yqss ] ) is also satisfied but depends on @xmath73 , so this is actually a differential equation on its own , rather than a solution for the surface density . ignoring the dm halo , i.e. , assuming that the entire rotation curve , and hence the inflow velocity are determined solely by the baryonic matter in the disk , this differential equation can also be solved analytically . using eqn . ( [ eq : m ] ) the differential equation for the surface density can be written as @xmath74 solving this differential equation , setting the constant of integration so that @xmath75 , and writing the solution in terms of @xmath76 , we get @xmath77 comparing this solution to eqn . ( [ eq : yqss ] ) shows that in the no - star formation no - dm case , the constant inflow velocity is a better fit to the physically motivated case ( _ c _ ) , in which the inflow velocity changes by @xmath78 per cent across the entire disk . the dimensionless gas profiles , for all three inflow models , are generic results for @xmath69 , and do not depend on the accretion rate , the size of the disk or the sound speed . it is instructive to display these solutions dimensionally for a specific ( characteristic ) cosmological accretion rate @xmath16 and disk inflow velocity @xmath36 . in fig . [ fig : yqss ] we plot the steady state solution for these three @xmath79-cases , in a dimensionless form ( bottom - left axes ) , and dimensional setting ( top - right axes ) , for @xmath80 , @xmath81 and @xmath82 , a cosmological accretion rate that corresponds to a uniformly distributed cosmological accretion by a @xmath83 median halo at @xmath3 @xcite . it can be seen , that the surface density in all three cases , fits an exponential profile ( dashed lines , same colours ) quite well for the bulk of the disk . at the inner part of the disk , however , the exponential profile ceases to fit the steady state solution , and in fact , the surface density diverges towards the center of the disk . however , the enclosed mass @xmath84 for these cases does not diverge . the gas profile shapes we find and display in figs . [ fig : yqss ] and [ fig : y ] resemble those found assuming the `` strong form '' of angular momentum conservation , such that the cumulative distribution of mass versus specific angular momentum , @xmath85 , is conserved and equal to that of a uniform density sphere in solid body rotation ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? in contrast , our profile shapes are essentially set by radial inflow and mass conservation only ( eqn . ( [ eq : dimen ] ) ) . we interpret the divergence of the surface density towards the center as material that forms a bulge ( including a central black hole ) . in order to study the bulge to disk mass ratio ( b / d ) , defined as @xmath86 of our solutions , we first define the bulge radius , as the distance from the center of the disk at which the gas surface density is larger than the surface density predicted by the best fitting exponential profile by a factor of two . the calculated disk scale radii , bulge radii and the corresponding b / d for the three @xmath79 test - cases are summarised in table [ tab : thetable ] , .a summary the disk scale radius , bulge radius and the bulge to disk mass ratios for the three @xmath79-test cases discussed in the test , in steady state , with no star formation . [ cols="^,^,^,^",options="header " , ] where it can be seen that the ( _ a _ ) and ( _ c _ ) disks , are indeed very similar , as predicted by comparing eqns . ( [ eq : yqss ] ) and ( [ eq : wow ] ) . enhanced stellar feedback efficiencies could remove the accumulated `` bulge material '' near the disk centres ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) but we do not include such feedback effects in our computations . so far we have considered the gas steady state solution without any star formation or outflows . thus the only parameter that determines the gas surface density profile is the inflow velocity . we now turn on the star formation circuit , which is also responsible for the outflowing gas from the disk back to the cgm / igm . in this section we will focus on cases ( _ a _ ) and ( _ b _ ) , and will return to case ( _ c _ ) only when we take the dm component into account . in the current treatment , fixing @xmath79 to the two limiting cases , there is no difference between star formation and outflows . from a physical point of view , though , these are different in essence , since while outflowing gas escapes the gravitational potential of the disk , forming stars stay in the disk , and contribute to the gravitational potential , the rotational velocity and the toomre-@xmath15 parameter . we will distinguish between those two processes in [ sec : cosevol ] . solving the steady - state equation ( @xmath88 ) with @xmath87 for cases ( _ a _ ) and ( _ b _ ) allows us to quantify the effect of star formation on the steady state solutions for the surface density . although tedious , both cases are analytically solvable , and converge continuously to eqn . ( [ eq : yqss ] ) as @xmath89 . we present the analytic solution for case ( _ a _ ) in appendix [ sec : sssan0 ] . by plugging in physically motivated values for @xmath90 and @xmath91 , we can get a feel of reasonable @xmath92-values . considering a @xmath93 disk , inflow velocity @xmath94 , depletion time of @xmath95 at @xmath3 @xcite , and @xmath96 ( see for example @xcite ) we get @xmath97 . in fig . [ fig : y ] ( _ left _ ) we plot the steady state solutions @xmath98 for @xmath99 and @xmath100 of case ( _ a _ ) . as @xmath92 increases the gas surface density of the bulk of the disk has a shallower slope . this is because larger @xmath92 means more efficient star formation and less efficient inflow , or @xmath101 . in this limit the evolution at every radius @xmath102 is almost independent , thus resulting in a more uniform disk . for small @xmath92 , as the depletion time becomes larger , gas flows towards the center rapidly , not staying long enough to form stars , until there is practically no star formation at all , and eqn . ( [ eq : yqss ] ) is relevant . [ fig : y ] shows that ( quasi-)exponential disks form naturally when the radial gas crossing time is short compared to the star - formation gas depletion times . the effect of @xmath103 on case ( _ b _ ) is similar . moreover , even though the steady state profiles of the two @xmath79-cases with @xmath69 are significantly different across the disk , as @xmath92 increases both surface densities diminish , until they are barely distinguishable for @xmath104 . having analysed the steady state solutions of eqn . ( [ eq : cont ] ) , we turn to solving its ( dimensionless ) time evolution , but still for a cosmological snapshot , i.e. assuming constant cosmological inflow rate and non - varying halo parameters . for simplicity , we focus on the @xmath69 case . by carrying out this analysis , we will be able to estimate the dimensionless time @xmath105 it takes the disk to reach a steady state . as we already noted , the cosmological snapshot solution should be a reasonable approximation , if the steady state convergence timescale is short compared to the time during which the cosmological conditions change significantly . in fig . [ fig : y ] ( _ right _ ) we plot the dimensionless surface density @xmath106 , for case ( _ a _ ) , at a few @xmath107 intervals . the gas surface density @xmath108 reaches a steady state by @xmath109 ( the @xmath110 case reaches a steady state even sooner , within @xmath111 ) , that is within one crossing time , converging to the analytic solution of eqn . ( [ eq : yqss ] ) , starting off with an empty disk . considering , for example , a @xmath112 and @xmath113 disk , means that the system reaches a steady state within half a gyr . at @xmath114 , the dm halo ( and therefore the cosmological accretion onto the disk as well ) change significantly within half a gyr . at later times , however , the cosmological accretion rate changes only slightly , meaning that a quasi steady state is maintained . in addition , we also find that the convergence to the steady state surface density is an outside - in " process . as a consequence , the fact that the radial surface density profile is well approximated by an exponential profile is not true throughout the evolution of the disk . this approximation becomes better as the evolution approaches its quasi steady state . so far we have considered halos of constant mass , and thus constant baryonic accretion rate , to which we referred to as a cosmological snapshot . as we have shown , within this picture the disk evolution equation can be written in a purely dimensionless form and solved analytically . we now generalise our model by accounting for the evolution of those cosmological conditions as well as other galactic parameters and local star formation , which is a necessary step for the model to be self consistent : * * dm halo * the dm halo is a crucial ingredient in disk formation and evolution for a couple of reasons . first , the halo growth rate , that depends on @xmath115 sets the baryonic accretion rate onto the disk . in our model we follow @xcite , taking the baryonic accretion rate as @xmath116 , where the halo growth rate is @xmath117 and can be integrated to give the exact halo mass at redshift @xmath118 , assuming that its mass at some @xmath119 is known , as we show in appendix [ sec : mhofz ] . + second , the dm halo also affects the rotation curve of the baryonic disk . as a result not only the mere total halo mass is important , but also its spatial distribution @xmath120 . furthermore , due to the dependence of the rotation curve on the halo mass distribution , the inflow velocity and the disk size will also be affected . therefore , to account for the mass distribution correctly , we must specify the halo concentration @xmath121 as a function of @xmath115 and @xmath118 . we assume continuously evolving median concentration halos following @xcite . we neglect adiabatic contraction of the dark matter halos * * sound speed * since in our model the gas inflow velocity is @xmath122 and @xmath123 , the gas sound speed is an essential parameter . higher sound speed means gas that flows inwards more efficiently , resulting in low @xmath42-values at large @xmath33 , with the addition of a larger numerator in @xmath124 . therefore , we expect that the larger the gas sound speed is , the harder it gets for the disk to form stars , and the smaller the stellar disk becomes . + we follow the redshift evolution of the sound speed , as in @xcite , i.e. @xmath125 as noted in @xcite , this relation is valid over a wide range of redshifts . however , due to measurement uncertainties , @xmath50 shows a substantial scatter around this value ( for example @xmath126 at @xmath127 , and @xmath128 at @xmath3 ) . therefore , we will use eqn . ( [ eq : cs ] ) as our fiducial sound speed in [ sec : fid ] , and study how our model predictions change when considering different @xmath50 in [ sec : bfm ] , keeping the @xmath129 dependence on redshift . * * disk size * to constrain the disk size , we assume that the specific angular momentum of the accreted matter , @xmath130 , equals that of the parent dm halo , @xmath131 , where the accreted gas is assumed to be in centrifugal equilibrium with the disk , i.e. it starts rotating around the center of the disk at a rotational velocity @xmath132 immediately as it enters the disk . not making this assumption , necessarily means that other assumptions , about the time it takes the newly accreted gas to reach the disks rotational velocity and the amount of energy that is dissipated in this process , have to be made . it would also cause a situation in which at each radius @xmath33 , the gas rotation is inhomogeneous . since we assume that the cosmological accretion at each time is uniform across the disk , then the constraint on the conservation of the angular momentum reads @xmath133 therefore for a given @xmath134 , since we know the halo mass at each redshift , we can solve eqn . ( [ eq : amconstraint ] ) for the disk size @xmath135 . with eqn . ( [ eq : amconstraint ] ) we are essentially adopting the results of both observations ( e.g. @xcite ) and cosmological hydrodynamic simulations ( e.g. * ? ? ? * ; * ? ? ? * ) that show that specific angular momentum is conserved from halo to disk . this despite the different angular momentum histories of the gas and dm . thus , we only specify the spin parameter @xmath134 , but not the cumulative mass distribution of the specific angular momentum , @xmath136 , found in dark matter only simulations as cosmological initial conditions ( e.g. * ? ? ? * ; * ? ? ? * * depletion time * at the base of our radially resolved equilibrium model lies the equilibrium relation ( [ eq : eq ] ) , that relates the sfr to the gas content of the disk . the proportionality , @xmath91 , quantifies the efficiency of star formation the shorter @xmath91 is , the more efficient star formation becomes , which can proceed even if the gas in the disk flows rapidly inwards due to a high inflow velocity . for the analysis below , we take @xmath137 following @xcite . * * star formation * it has already been noted by @xcite that @xmath15 in the disk will self - regulate to maintain @xmath138 . the intuitive explanation for this self - regulation in @xcite is generally valid in the spatially resolved case as well . nevertheless , resolving the disk radially , means that @xmath139 and @xmath15 are all calculated as a function of the distance from the center of the disk . wherever the ratio between the epicyclic frequency and @xmath42 is small enough , @xmath15 will reach its threshold value , and self regulation will proceed . at the outer regions of the disk , however , the low gas surface density will dominate @xmath124 , which will consequently be @xmath140 , and therefore star formation will not be initiated as expected for the outer lower surface density portions of galaxy disks ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? this does not imply that regions in the disk where no star formation is taking place at some redshift will remain quiescent . however , due to the cosmological evolution of the parent halo , and as a result of the baryonic accretion rate and the size of the disk , star formation can begin at a later time . + in order to account for star formation self - consistently in the analysis below , we calculate @xmath124 , and turn the star formation - circuit on only at those regions of the disk in which @xmath1 . in this section we consider cosmological evolution of the halo and the accretion rate , employ a self consistent star formation criterion , and solve numerically the evolution equations for the gas and the stars in the disk . first , in [ sec : fid ] , we define and analyze thoroughly our fiducial model . next , in [ sec : bfm ] we examine how our results depend on model parameters , @xmath141 and @xmath14 for a fixed ( at @xmath3 ) halo mass . finally , in [ sec : bfm2 ] we study how the model results change when the halo mass is modified . as our fiducial model we assume a @xmath142cdm cosmology , with @xmath143 , @xmath144 , and baryon fraction @xmath145 @xcite . we also assume that the dm halo profile follows a navarro - frenk - white profile @xcite with median concentration parameter that varies with redshift , following @xcite . the fiducial dm halo we consider is of mass @xmath146 at @xmath3 and spin parameter @xmath147 . in addition , we assume @xmath148 , no instantaneous recycling , and @xmath149 at @xmath3 within the disk . our massive fiducial halo represents the tip of the galaxy main - sequence at @xmath150 . solving for the cosmologically time - dependent fiducial halo , in which the star formation term is turned on only when and where @xmath1 , we plot in fig . [ fig : fid ] ( _ top left _ ) the resulting surface density of the two disk components at @xmath3 , and the corresponding @xmath124 in fig . [ fig : fid ] ( _ top right _ ) . as can be seen in the figure , for the fiducial model both gas and the stellar profiles are well approximated by exponential profiles ( dashed lines ) , for a wide range of radii . for the stars and gas distributions , exponential profiles with scale lengths equal to @xmath151 and @xmath152 respectively , represent the computed distributions within a factor of two , from 0.5 to 9 @xmath153 for the stars , and from 1 to 11 @xmath153 for the gas . remarkably , though somewhat fortuitously , the stellar length scale precisely equals the length scale given by @xmath154 for assumed exponential disk in an isothermal halo @xcite . according to our bulge - criterion , the fiducial model bulge extends to @xmath155 , and the bulge to disk mass ratio is @xmath156 . in addition , while the gaseous disk extends roughly to @xmath157 , the stellar disk is somewhat more compact , stretching out to @xmath158 . the gas half mass surface density , defined as @xmath159 is @xmath160 , where @xmath161 is the the distance form the center of the disk such that @xmath162 the stellar half mass surface density , defined in a similar way is @xmath163 . at @xmath3 we find @xmath164 , consistent with observed sizes for this stellar mass and redshift @xcite . the star formation rate ( at @xmath3 ) for the fiducial model is @xmath165 , and the gas fraction is @xmath166 . comparing our predicted sfr to the sfr - main sequence , we find that the predicted sfr for our fiducial halo , @xmath167 , is in very good agreement with the sfr according to @xcite , @xmath168 , for the same redshift @xmath3 , and stellar mass @xmath169 . using our model we can also follow the actual evolution of the disk , as a function of redshift . in fig . [ fig : fid ] ( _ bottom row _ ) we plot the surface density of the gaseous ( _ left _ ) and stellar disks ( _ right _ ) for @xmath170 and @xmath171 . it can be seen that even though that the characteristic gas surface density reaches its final " ( @xmath3 ) value already at @xmath172 , the disk growth continues even at smaller redshifts . studying the stellar disk surface density evolution , we find that star formation begins in the inner regions of the disk at @xmath172 , and propagates outwards with time , until at @xmath3 the stellar disk is almost double in size with respect to @xmath173 , or @xmath174 later . in fig . [ fig : msofz ] ( _ left _ ) we plot the redshift evolution of the total mass for each of the disk components ( gas and stars ) , together with the dm halo and hot gas for our fiducial model , where for the hot gas we sum all the gas that flows through the disk and out from the center , plus the accumulated outflow that is proportional to the total stellar mass . most of the baryons remain as hot halo gas throughout the evolution . at @xmath172 , the onset of star formation in the fiducial model case , the disk gas mass growth slows down , as is to be expected from the equilibrium relations . the effect on the hot gas mass , however , is insignificant since most of the hot gas is ejected from the disk center . in fig . [ fig : msofz ] ( _ right _ ) we plot the rotation curve @xmath175 ( black curve ) together with the baryonic and dark matter contributions to the total mass ( solid and dashed red curves ) . the disk is baryon dominated throughout . it is also of interest to plot the resulting cumulative mass versus specific angular momentum , @xmath85 , for our fiducial model . in fig . [ fig : mlessthanj ] versus @xmath176 for our fiducial model ( _ solid blue _ ) and @xmath177 ( _ dashed black _ ) for a sphere in solid body rotation . ] we plot @xmath178 versus @xmath176 for our fiducial model ( solid curve ) together with the relation @xmath177 for a sphere in solid body rotation . the two curves differ only slightly . this is consistent with our resulting ( quasi)-exponential disks , and the known similarity between the angular momentum distributions of centrifugally supported exponential disks and rotating spheres ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . that allows star formation at @xmath3 for a @xmath83 ( _ blue _ ) and a @xmath179 ( _ dark - blue _ ) halo as a function of the gas sound speed . the fiducial model ( @xmath180 ) is designated by a small black circle only the regions below the curves are star forming . no star formation occurs above the blue curves . _ top - right axes : _ @xmath134 dependence of the total sfr for the @xmath181 case . it can be seen that the sfr dependence on @xmath134 is weak , changing by less than @xmath182 within the @xmath183 range . ] we wish to examine how the disk properties are modified when the halo or disk parameters are altered . we note that a key advantage of our approach a simple toy - model of disk evolution is that it enables scanning through a large parameter - space at a low computational cost . we focus on @xmath83 halos at @xmath3 , as for our fiducial model . we examine how the redshift at which star formation begins , @xmath184 , depends on @xmath50 and @xmath134 . since the initial conditions for our disk evolution are of a median halo and an empty disk that is fed by cosmological accretion , star formation can not take place until enough gas has been accreted and @xmath42 is large enough . the growing disk size , on the other hand , makes @xmath42 smaller , thus delaying the onset of star formation . the impact of having a large @xmath50 is threefold . first , the large inflow rate empties the gas at each radius more efficiently , resulting in reduced @xmath42 . second , it delays star formation due to the @xmath15-barrier it induces . third , as we show below , although counter - intuitive , larger @xmath50 results in larger disks , and hence smaller @xmath42 , thus delaying star formation . even though larger @xmath134 can also delay star formation , through the disk size , for @xmath134 from @xmath185 to @xmath186 , which we consider , its effect is weaker , as seen in fig . [ fig : cs_l_first ] ( _ top left _ ) . interestingly , after star formation has begun , @xmath134 hardly has an effect on sfr , despite its effect on the disk size . qualitatively , increasing @xmath134 means increasing the angular momentum of the halo , and hence also of the accreted baryons . therefore , as @xmath134 increases , the disk size @xmath24 increases too . however , since the baryonic accretion onto the disk @xmath187 is independent of @xmath134 , then to first order the total baryonic mass in the disk will also be independent of @xmath134 . consequently , even though large @xmath134 means smaller characteristic @xmath42 , as long as it is high enough to maintain @xmath138 , star formation , which is @xmath188 should proceed at more or less the same rate , as seen in fig . [ fig : cs_l_first ] ( _ top right _ ) . nevertheless , there should exist a @xmath189 , that once reached , renders @xmath42 too low and @xmath15 too high , which will not allow any more star formation . in fig [ fig : lambdamax ] ( _ bottom - left axes _ ) we plot @xmath189 that allows star formation at @xmath3 , as a function of the gas sound speed , for our fiducial halo ( @xmath83 at @xmath3 ) , and for a slightly less massive halo ( @xmath179 ) . it can be seen that @xmath190 , as preferred by observations and simulations is well within the star forming portion of the parameter space , and becomes borderline only as the gas sound speed increases significantly . in fig [ fig : lambdamax ] ( _ top - right axes _ ) we plot the total sfr for the fiducial sound speed , as a function of @xmath134 . it can be seen that for @xmath134 altered by more than an order of magnitude , the total sfr changes by less than @xmath100 per cent , following our qualitative argument . studying the disk size , one might expect that since @xmath191 , then for higher sound speeds , the gaseous disk will be more compact , as the radial migration of the gas becomes more efficient . however , the disk size is primarily set by the baryonic accretion onto the disk , whose extent , in our model , is set by conservation of angular momentum . let us consider a halo of mass @xmath115 , with some specific angular momentum , @xmath131 that is , of course , independent of the sound speed . a larger @xmath50 at the early stages of the disk evolution , will result in a mass profile that is more concentrated near the center of the disk , and thus lower baryonic specific angular momentum . therefore , to satisfy the angular momentum constraint , eqn . ( [ eq : amconstraint ] ) , the disk must grow for higher sound speeds . this is seen in fig . [ fig : cs_l_first ] ( _ bottom left _ ) , where we plot @xmath26 ( the truncation radius of the baryonic disk ) as a function of the gas sound speed and @xmath134 , for our fiducial halo at @xmath3 . for the stellar disk size dependence on @xmath50 , which is plotted in fig . [ fig : cs_l_first ] ( _ bottom right _ ) , we note that there are two competing effects . as @xmath50 increases , eqn . ( [ eq : q ] ) implies that in the outer regions of the disk , where @xmath42 is lower , star formation is suppressed . but , since the gaseous disk grows as @xmath50 grows , then the absolute size of the stellar disk is potentially larger , at least as long as overall @xmath15 is small enough . in fig . [ fig : cs_l_more ] we plot the corresponding @xmath192 ( _ top row _ ) and @xmath193 ( _ bottom row _ ) , the half mass radius and surface density of the gas ( _ left column _ ) and stars ( _ right column _ ) . it can be seen that regions in parameter space that result in small @xmath26 and @xmath194 have large @xmath193 . this is because small disks get a larger mass contribution from the central divergence of the surface density . the opposite is not exactly satisfied the largest disks do not have the lowest half mass surface densities . for the gas component , larger disks contain more baryons , and hence their characteristic surface density is higher , while for the stellar component not all of the parameter space results in disks that form stars , resulting in lower characteristic surface densities . owing to these effects , the total disk mass , as well as the relative gaseous and stellar contributions to it depend on @xmath50 and @xmath134 in a non - trivial way . as seen in fig . [ fig : cs_l_msandmore ] ( _ top left _ ) , the total disk mass increases as the sound speed decreases , as one would expect , primarily due to the fact that higher @xmath50 results in a rapidly emptying disk , where the excess gas that inevitably reaches the center of the disk and ekected back to the hot halo . the hot halo is also fed by the outflowing baryons , whose mass @xmath195 sfr . the total hot gas mass due to these contributions , is plotted as a function of @xmath196 for our fiducial halo at @xmath3 in fig . [ fig : cs_l_msandmore ] ( _ top right _ ) . it can be seen that the hot gas mass depends very weakly on @xmath134 , similarly to the sfr . this implies that outflows constitute a significant part of the total hot gas mass . the relative contributions of gas and stars to the total disk mass can be read off fig . [ fig : cs_l_msandmore ] ( _ bottom left _ and _ bottom right _ ) , where we plot the gas fraction @xmath197 , and @xmath198 , respectively . it can be seen that except for the fact that the gas fraction is lower where sfr is higher , it exhibits a stronger dependence on @xmath134 than the sfr , since as argued above higher @xmath134 generates larger disks , that results in a smaller portion of it that forms stars . apart from @xmath134 and @xmath50 , another free parameter in our model is the mass loading factor @xmath14 . considering variation in @xmath14 from @xmath199 to @xmath200 results in insignificant modifications to the results reported above , and we thus do not pursue this @xmath14 analysis further . we note , however , that analysis of the extreme case @xmath201 , i.e. no outflows at all , shows a stronger dependence of sfr and @xmath202 on @xmath134 , verifying our argument about the importance of outflows to the hot gas mass . as we have already mentioned , the tight scaling relation between the sfr in disk galaxies and their stellar mass is remarkable . this scaling relation has been repeatedly observed , in many independent observations and different surveys . therefore , it is interesting to compare our model predictions to actual observations of sfr , and how well those predictions fall within the galaxy main sequence . in particular , we compare our model predictions to the best - fitting sfr-@xmath203 relation , as found in @xcite , for star forming galaxies between @xmath127 and @xmath171 @xmath204 where @xmath205 since the initial conditions of our simulated disks are of empty disk , and since , as we have shown , in our model star formation is initiated at @xmath172 , it is clear that the correspondence between our model predictions and eqn . ( [ eq : whit ] ) can not hold for any high redshift , and thus we are interested in comparing these sfr s at a lower redshift , so that the disk has already reached an equilibrium . in fig . [ fig : sfrscatter ] relation ( or `` main sequence '' , as predicted by our model for various model parameters ( @xmath206 ) at @xmath3 , and the corresponding observed relation ( dashed ) according to @xcite . ] we plot the whitaker relation , eqn . ( [ eq : whit ] ) , as a function of @xmath203 for @xmath3 ( _ black dashed _ ) compared to our model predictions , all at @xmath3 and for a halo of mass @xmath83 , and various model parameters , @xmath207 and @xmath14 . it can be seen from this plot that our model predicts very well the sfr-@xmath203 relation , even as @xmath196 and @xmath14 are varied . so far we have only considered the evolution of disks within a dm halo of mass @xmath83 at @xmath3 , and examined how perturbations to model parameters affect the evolution of those disks . we now wish to consider less massive halos . these imply smaller baryonic accretion rates , and hence less massive baryonic disks . this affects the disk size and sfr , and consequently also the gas fractions in the disks . reduced halo masses lead to smaller disks as less gas is accreted , but as shown in fig . [ fig : changem ] ( _ top left _ ) , however , the dependence of the disk size on the halo mass weakens as @xmath50 decreases . since a less massive halo means reduced baryonic accretion , and hence lighter gaseous disks , it is also to be expected that a smaller @xmath115 implies smaller sfr , for a given @xmath50 . as we have already shown , as @xmath50 increases , the characteristic gas surface density has to become larger to sustain star formation . therefore , for each @xmath50 there is the corresponding @xmath115 below which star formation is no longer possible , since @xmath208 at all times for all radii . this forms a star - formation threshold , which is visible in fig . [ fig : changem ] ( _ bottom left _ ) of the sfr as a function of @xmath50 and @xmath115 at @xmath3 . this naturally points out a validity range for our model . examining the stellar disk size as a function of @xmath115 and @xmath50 , as in fig . [ fig : changem ] ( _ top right _ ) , we find that the behaviour we found in the analysis of fig . [ fig : cs_l_first ] ( _ middle right _ ) , of our fiducial halo , is modified for less massive halos . for halos of mass @xmath209 the stellar disk size is monotonically decreasing as @xmath50 increases . this implies that for the less massive halos the fact that the total size of the gaseous disk increases as @xmath50 increases is subdominant to the @xmath15-barrier to star formation due to the increase of the gas sound speed . finally , in fig . [ fig : changem ] ( _ bottom right _ ) we plot the total stellar mass in the disk as a function of @xmath115 and @xmath50 at @xmath3 . we find that even though the total stellar mass increases monotonically with @xmath115 , as might be expected , it does not follow the general behavior of the stellar disk size discussed above , which means that maximising the star forming region does not guarantee a higher total sfr through that region . in this paper we have presented a radially - resolved - equilibrium - model " for the growth of baryonic disks in cosmologically accreting massive halos . we have solved the time - dependent equations that govern the radially dependent gaseous and stellar contents of disks , as well as their star - formation rates , inflows and outflows from and to the inter- and circum - galactic medium , and inward radial gas flows within the disks , while imposing mass and angular momentum conservation . we have focused on the evolution up to redshifts @xmath0 . the equilibrium model for galaxy evolution has been widely employed to account for the interrelated observed parameters of the ( blue ) main - sequence @xcite . in this paper we have generalised the analytic model to account for the global disk structures their gaseous and stellar surface densities , radially resolved sfr and study the impact of the variation of various parameters on this structure . we have shown that : * the dynamical time scales required for both components of baryonic disks to reach their equilibrium configuration are much shorter than the those of cosmological variations in the dm halo mass and profile . this feature guarantees that the disk structures and star formation rates remain close to equilibrium during the cosmological evolution . * radial ( in plane ) gas inflows together with mass conservation naturally give rise to ( quasi-)exponential disks over many length scales . in our models the disk sizes , surface densities , and star formation rates are self - consistently computed assuming global conservation of the halo specific angular momentum . * the stellar disk size is determined by the sfr , which is self consistently calculated according to the condition toomre @xmath1 . for a roughly flat rotation curve , this condition is not satisfied when the surface density becomes too small , as in the outer parts of the gaseous disk , which naturally leads to stellar disks that are more compact than the corresponding gaseous disks . * focusing on cosmological evolution up to @xmath3 , and on massive , @xmath2 halos ( at @xmath3 ) , our model reproduces the star formation main sequence , @xmath210 . in particular , our fiducial model predicts sfrs of @xmath4 , stellar masses @xmath5 , gas contents @xmath6 , half mass sizes of 4.5 and 5.8 kpc for the stars and gas , and characteristic surface densities @xmath211 . our analytic models may be used to study the interrelationships between spin parameter , halo mass , and redshift , on the disk gas and stellar surface densities and radially dependent star - formation rates . we thank andreas burkert , reinhard genzel and linda tacconi for many helpful discussions . we thank the referee for very helpful comments that improved our presentation . this work was supported by the dfg via german - israel project cooperation grant ste1869/1 - 1/ge625/15 - 1 . we also acknowledge support by the raymond and beverly sackler tel aviv university - harvard / itc astronomy program , and israel science foundation i - core grant 1829/12 . for the velocity case ( _ a _ ) , the steady state solution of the @xmath212 ( that is , the star forming ) scenario may be obtained analytically . to do that , we first rewrite the differential equation , as a classical non - homogeneous first order ordinary differential equation of the form @xmath213 : @xmath214 identifying the coefficients @xmath215 we can immediately find the solution to this equation , with the constant of integration chosen so that @xmath216 , to yield @xmath217 it can be easily verified that for @xmath218 @xmath219 which converges continuously to the no - star formation case . another notable fact is that having @xmath87 does not change the functional form of the divergence of @xmath108 towards the center of the disk , where the surface density satisfies @xmath220 where @xmath221 can be read off eqn . ( [ eq : yofalpha ] ) and satisfies @xmath222 . finally , we note that the same is true for case ( _ b _ ) : the dimensionless surface density may be calculated analytically , and shown to converge continuously to the @xmath69 scenario . let us write eqn . ( [ eq : dekel ] ) in a more general way @xmath224 we note that @xmath225 where @xmath226 is the cosmological scale factor and @xmath227 are the cosmological redshift and time as usual . therefore eqn . ( [ eq : dekelgeneral ] ) can be written as @xmath228 where @xmath229 is the hubble constant at @xmath118 . this equation can be solved numerically , as we do for the quantitative analysis in this paper . it is illustrative though to assume a matter dominated universe ( @xmath230 ) , so that @xmath231 , where @xmath232 is the present day hubble constant , and solve this differential equation analytically . in a matter dominated universe then @xmath233 where @xmath234 .

the monte carlo method is a fundamental computational tool in science . its goal is to sample configurations @xmath0 in a given state space from a probability distribution @xmath1 . this can usually not be achieved directly for multi - dimensional distributions . markov - chain monte carlo methods @xcite overcome this problem by generating configurations @xmath2 starting from an initial configuration @xmath3 which belongs to a simpler distribution @xmath4 ( often a fixed initial condition , or some ad - hoc random choice ) . configurations @xmath5 are then generated from @xmath6 according to a stochastic algorithm which guarantees , as time moves on , that @xmath7 departs from the initial condition and converges for @xmath8 towards the equilibrium distribution @xmath9 . the markov - chain approach can be implemented for arbitrary distributions @xmath10 , using for example the metropolis and the heat - bath algorithms . for many applications , enormous effort has gone into designing fast algorithms for which one reaches @xmath11 for reasonable running times @xmath12 . in this paper , we are concerned with a related problem : rather than to find the fastest algorithm for a given problem , we are interested in quantifying the speed of a given markov - chain algorithm . this is , we want to prove after which time @xmath12 the sample @xmath5 is equilibrated . it then reflects the equilibrium distribution and no longer the initial configurations . in many practical applications , it is extremely difficult to decide from within the simulation whether it has indeed equilibrated @xcite . instead , one must validate the simulation results with other approaches , from exact solutions to experimental data . the correct characterization of the convergence towards equilibrium from within the simulation has remained a serious conceptual and practical problem of the monte carlo method . from a fundamental viewpoint , the problem of rigorously proving convergence of a simulation was solved , at least in principle , through a paradigm called `` exact sampling '' , which allows to generate , with markov chains , samples @xmath0 directly from the equilibrium distribution @xmath10 without any influence of the initial configuration@xcite . in practice , however , it has not been possible to implement exact sampling for many complicated problems , as for example disordered systems , for which standard methods of evaluating equilibration times fail . the reason for this difficulty is as follows : exact sampling proves for a given markov - chain simulation that the correlation of the initial configuration with the configuration at time @xmath12 strictly vanishes . this is done by showing explicitly that all possible initial configurations @xmath13 yield the same output under coupled monte carlo dynamics . in many simple models , one can prove this coupling property indirectly . in general , however , one must indeed survey the entire configuration space . this is usually too complicated to be achieved . we recently developed a local - patch algorithm@xcite which indeed monitors the entire configuration space of complicated systems , even for very large sizes . the approach uses local information , concentrated on so - called `` patches '' . the scale of these patches increases during the simulation . information on patches can then be combined for the entire system to provide a crucial upper bound for the ( global ) coupling time , and to generate an exact sample . the algorithm was demonstrated to work for spin glasses at lower temperatures than previous methods @xcite , even though the physically interesting regime has still not been reached yet . the local - patch algorithm is quite general : in addition to spin glasses , we implement it in this paper for hard disks and improve on previous results @xcite . the successful application of exact sampling to hard - sphere systems is remarquable because the configuration space is continuous so that , naively , its complete survey appears out of reach . a markov chain is fully characterized by the so - called `` transfer matrix '' of transition probabilities between any two configurations @xmath14 and @xmath15 . as will be illustrated shortly ( section [ s : coupling_one_d ] ) in a specific example , the largest eigenvalue of the transfer matrix is @xmath16 and the corresponding eigenvector @xmath17 describes the equilibrium state . the convergence towards equilibrium is governed by the spectrum of the transfer matrix and by the overlap of the eigenvectors @xmath18 with the initial configuration : @xmath19 in the limit of infinite simulation time , the second - largest eigenvalue determines the exponential convergence of the probability distribution towards equilibrium . this eigenvalue sets a time scale @xmath20 and the convergence is as @xmath21 the rigorous determination of convergence properties of markov chains has been undertaken in many cases , from urn models to card - shuffling ( see @xcite ) , diffusion processes , and many more ( see @xcite ) . efficient algorithms , as for example the bunching method @xcite are commonly used to perform an empirical error analysis of monte carlo data in more complicated cases , where rigorous calculations are out of the question . however , these methods are not failsafe . in practice , it is often difficult to extract @xmath22 from the large number of physically relevant time scales . in disordered systems , for example , there is often no reliable way to ascertain that the simulation has run long enough , and @xmath22 may be much larger than assumed ( see e.g. @xcite , sect . 1.5 ) . in the limit of infinite times @xmath23 , the markov chain converges towards the equilibrium distribution , and the positions @xmath5 become independent of the initial condition . the loss of correlation with the initial condition is evident for markov chains that couple , that is , which for each possible initial condition @xmath3 produce the same output @xmath5 . in many cases of interest this happens after a finite global `` coupling time '' @xmath24 , which depends on the realization of the markov chain . propp and wilson @xcite realized that this coupling property allows one to draw `` exact '' samples from the distribution @xmath10 . this approach , called `` coupling from the past '' , eliminates the problem of analyzing the convergence properties . however , to establish that a markov chain has coupled , the entire state space of the system must be supervised . this was believed infeasible except for special problems where the dynamics conserves a certain ( partial ) ordering relation on configurations . a partial order is conserved in heat - bath dynamics of the ferromagnetic ising model , whereas the frustration in the spin - glass model foils this simplification . [ s : coupling_one_d ] we first discuss convergence and coupling for a markov chain describing the hopping of a single particle on a simple @xmath25-site lattice with periodic boundary conditions ( see fig . [ f : one_d_single ] ) . in one time step , the particle hops with probability @xmath26 from one site to its two neighbors : @xmath27 in addition , we have @xmath28 . the equal hopping probabilities imply via the detailed balance condition @xmath29 that the stationary probability distribution @xmath30 of this problem is independent of @xmath14 . this system s monte carlo algorithm is encoded in the @xmath31 transfer matrix @xmath32 : @xmath33 the eigenvalues of @xmath32 are @xmath34 + 1 $ ] ( with multiplicities ) that is , for @xmath35 , @xmath36 . the largest eigenvalue , @xmath37 corresponds to the conservation of probabilities . by construction , it is associated with the equilibrium solution @xmath38 . the second - largest eigenvalue is @xmath39 . for @xmath35 we have @xmath40 . this eigenvalue controls the long - time corrections to the stationary solution , which vanish as @xmath41^{t } = { \exp { \left[}- t/{\tau_{\text{corr}}}{\right]}}$ ] , with @xmath42 we note that the time scale @xmath22 only describes the asymptotic behavior of the correlation . the calculation of the time @xmath12 at which the probability distribution @xmath7 itself is within a suitably chosen @xmath43 of a the equilibrium distribution @xmath10 is more involved ( see , for example , @xcite ) . [ s : coupling ] as illustrated in fig . [ f : one_d_diffusion ] , the monte carlo algorithm can be formulated in terms of random maps . in our example , this means that instead of prescribing one move per time step , as in fig . [ f : one_d_single ] , we now sample moves for all times @xmath12 and all sites @xmath14 , in such a way that the dynamics of a single particle again satisfies the detailed balance condition of eq . ( [ e : detailed_balance ] ) . the most natural implementation of this approach is illustrated in fig . [ f : one_d_diffusion ] : arrows are chosen independently for all times @xmath12 and all sites @xmath14 . at time @xmath44 , for example , the particle should move down from sites @xmath45 , @xmath46 , @xmath47 and @xmath48 and straight from site @xmath49 . we can now check the outcome of the monte carlo calculation . in the example of fig . [ f : one_d_diffusion ] , from time @xmath50 on , all initial configurations of the single particle yield the same output . this is remarkable because , evidently , at this time the initial conditions are completely forgotten . the coupling time @xmath51 is a random variable ( @xmath52 in fig . [ f : one_d_diffusion ] ) which depends on the realization of the full monte carlo simulation from time @xmath44 onwards ( until coupling has been reached ) . the independence of random maps on different time steps implies that the probability for not coupling vanishes at least exponentially fast in the limit @xmath8 . under the random - map dynamics , an initial state with @xmath25 particles eventually evolves into a state with one particle ( in later sections , spin - glass configurations will take the place of the single - particle positions ) . more generally , a state with @xmath14 configurations can evolve at each time step into a state with @xmath53 configurations . figure [ f : one_d_diffusion ] displays a sequence of random maps and illustrates the associated time - forward search of the coupling time . this extended monte carlo dynamics on @xmath14-configuration states can again be described by a transfer matrix : @xmath54 where the block @xmath55 ( of sizes @xmath56 ) concerns all the processes which lead from a state at time @xmath12 with @xmath14 configurations to a state with @xmath57 configurations at time @xmath58 . the upper left block of this matrix , @xmath59 , is the original matrix from eq . ( [ e : one_particle_mc ] ) . as an example , we find from eq . ( [ e : algo_probabilities ] ) the following elements of this transfer matrix : @xmath60 etc . the matrix @xmath61 describes a physical system with variable particle number ( from @xmath45 to @xmath25 ) and a space comprising @xmath62 states , the number of non - empty states in this new simulation ( for a problem of @xmath25 spins , the number of configurations is @xmath63 and the total number of @xmath14-configuration states ( states with @xmath14 configurations ) is @xmath64 ) . the `` forward '' transfer matrix @xmath65 allows us to compute the coupling probabilities as a function of time in fig . [ f : coup_proba ] . the matrix @xmath61 is block - triangular in the number of particles @xmath66 , with the @xmath67 block given by @xmath59 . therefore , all the eigenvalues of @xmath59 are also eigenvalues of @xmath61 . in particular , the largest eigenvalue of @xmath61 is again @xmath68 , with corresponding right eigenvector @xmath69 . the second - largest eigenvalue of @xmath61 belongs to the @xmath70 block and leads to the time scale of the coupling , @xmath71 . it is given by @xmath72 , larger than @xmath73 . this second - largest eigenvalue @xmath74 governs the coupling probability @xmath75 for large times . it follows from the block - triangular form of the forward transfer matrix @xmath61 that the time scales satisfy @xmath76 . a general argument allows us to better understand this result : for any running time @xmath12 we may separate all the markov chains into those that have already coupled and those that have not . only the non - coupled chains ( whose number vanishes as @xmath77 ) contribute to connected correlation functions : @xmath78 here , @xmath79 is an observable whose mean value is zero and @xmath80 the configuration of the system at time @xmath12 . for the chains which have coupled by the time @xmath12 , @xmath80 does not depend on @xmath81 , and the contribution to the correlation function vanishes . for the other chains , we find @xmath82 where we suppose that even the non - coupled chains converge towards equilibrium on a time scale @xmath83 , and use that the probability for a chain not to have coupled behaves as @xmath77 in the long - time limit . equation ( [ e : correlation_coupling_convergence ] ) shows that the difference between @xmath71 and @xmath22 is caused by the convergence taking place within non - coupling chains : @xmath84 this relation is illustrated in fig . [ f : coup_proba ] for the one - dimensional diffusion model with @xmath35 . a later figure , fig . [ f : spin_spin_corr ] , will illustrate for the case of spin glasses the split between the general spin spin correlation function and that same object computed for non - coupling chains only . [ s : forward_and_backward ] the probability distribution of coupling times in the forward direction can be obtained from the transfer matrix @xmath61 as we discussed in section [ s : coupling ] . here we analyze the distribution of coupling times in the backward direction for the application of the `` coupling from the past '' protocol which , as we will see , is the same as the one in the forward direction . the backward coupling process leads to a generalized transfer matrix , @xmath85 , which again describes an extended monte carlo simulation . we consider a hypothetical simulation which has run since time @xmath86 up to time @xmath87 ( see fig . [ f : one_d_cftp ] ) . it follows from the discussion in section [ s : coupling ] that the simulation has coupled . furthermore , because of the infinite separation between the infinitely remote initial condition and the final one , the resulting configuration ( at @xmath87 ) is in equilibrium . but it remains to be seen which one of the five configurations at @xmath87 is generated . in the example of fig . [ f : one_d_cftp ] , a one - step backtrack to time @xmath88 allows us to see that the configuration at @xmath87 can be neither @xmath89 nor @xmath90 nor @xmath91 . likewise , for @xmath92 output positions on the @xmath25-site ring this back - propagation leads to a dead end , and only a single position yields a full set of possibilities at some time @xmath93 . thus to find the output configuration of the simulation , one is interested in the first time in the past for which the simulation couples , that is one searches the `` backward '' coupling time . the implementation of this backward simulation , as defined for many particles , can again be described by a transfer matrix . for any distribution of arrows , any occupied site @xmath14 at time @xmath12 propagates its occupation back to all the sites at time @xmath94 which have arrows pointing towards @xmath14 . the matrix element of @xmath95 between two states is given by the statistical weights of all the arrows connecting the two states . for example , we find for @xmath85 @xmath96 this is a non - trivial variant of the forward simulation as , for example , the matrix @xmath85 is not block - triangular as @xmath61 , but it is particle hole symmetric ( as we see in the above example ) . formally , a random map @xmath97 ( here a set of arrows ) associates configurations which are connected under the monte carlo dynamics . a @xmath14-configuration state @xmath98 is by definition a set of configurations . at time @xmath12 the state @xmath98 can be associated with the state @xmath99 at time @xmath58 via the forward matrix if and only if @xmath99 is the ( set ) image of @xmath98 by an allowed mapping @xmath97 ( i.e. @xmath100 ) . the same holds for the backward matrix but @xmath98 must be the reciprocal ( set ) image of @xmath99 ( i.e. @xmath101 ) . the backward transfer matrix @xmath85 manifestly differs from the forward matrix @xmath61 . however , we construct explicitly in appendix [ s : similarity ] the similarity transform that maps @xmath61 onto @xmath85 . this means that @xmath102 ( see appendix [ s : similarity ] ) . the similarity transformation @xmath103 associates a @xmath14-configuration state @xmath98 with the sum of states that share at least one configuration with @xmath98 , included itself . the spectrum of the backward transfer matrix thus agrees with the one of the forward matrix and the distribution of coupling times @xmath75 is identical for backward and forward dynamics . this result is natural because the probabilities for not coupling for @xmath12 time steps are identical in both forward and backward direction : @xmath104 ( see @xcite ) . the probability distribution @xmath105 measures the weight of the configuration @xmath106 under repeated application of the backward transfer matrix from the configuration at @xmath87 : @xmath107 . transition probabilities of the forward transfer matrix must satisfy the markov chain transition probabilities for single particles , but the choice of random maps is otherwise unrestricted . the one - particle sector is trivially correct for independent moves as in our diffusion model of section [ s : coupling ] . we now discuss several alternative random maps for the one - dimensional diffusion , which may lower ( or increase ) the coupling time ( with , however @xmath76 ) or achieve a rapid reduction of the number of configurations for smaller time scales . a naive example for the one - dimensional diffusion example consists of arrows , such as in fig . [ f : one_d_diffusion ] , but which for one time @xmath12 all point into the same direction , straight , up , and down , each with probability @xmath108 , so that single - particle moves satisfy the detailed balance condition . evidently , this random map does not couple , and the non - coupling markov chains , in eq . ( [ e : coupling_convergence ] ) , converge in a time @xmath109 . we now modify this rigid algorithm by allowing arrows to change direction with probability @xmath43 . this makes the markov chain couple on a time scale @xmath110 , much larger than the correlation time @xmath22 , for small @xmath43 . the choice of independent random moves ( @xmath111 ) is optimal in this class of maps , but it is not the choice minimizing @xmath71 among all random maps . for example we may choose correlated moves for selected neighboring pairs of sites say , for sites @xmath112 and @xmath113 and let the move from site @xmath48 be independent ( see fig . [ f : one_d_corr_pairs ] ) . elements of @xmath114 are , for example , @xmath115 the single - particle sector of this algorithm is as before , but the second - largest eigenvalue of the transfer matrix @xmath61 with such correlated pair moves becomes smaller , indicating faster coupling : @xmath116 coupling times of both `` independent - arrow '' random mapping and `` correlated - pair '' random mapping scale alike for large @xmath25 . we note that in applications , as in our patch algorithm of section [ s : local_patch ] , it might be not so much of interest to speed up the coupling than to rapidly decrease the number of possible configurations at times @xmath117 . therefore , one goal could be to decrease the eigenvalues of the matrix @xmath118 , whose time scales correspond to the rapid reduction of the number of configurations towards more manageable numbers . as discussed in section [ s : forward_and_backward ] , the coupling of markov chains allows one to produce exact samples of the equilibrium distribution : in the diffusion example , we were able to run the monte carlo simulation backwards in time using @xmath85 , but this matrix can usually not be constructed . to find the sample at time @xmath87 , one may tentatively set a time @xmath119 and produce all the random maps between time @xmath44 and @xmath12 . one can then check explicitly whether all the possible initial conditions at time @xmath44 have coupled , that is , for the diffusion problem , whether the initial @xmath25-particle configuration @xmath106 has yielded one of the one - particle configurations . if this goal has not been reached , one must complement the random maps already computed with random maps for earlier times ( see fig . [ f : one_d_cftp ] ) . the one - dimensional diffusion problem without periodic boundary conditions illustrates an algorithm which determines the coupling time with much less effort . we consider odd times at which only sites @xmath120 and @xmath48 and even time steps at which only sites @xmath49 and @xmath47 may flip ( see fig . [ f : one_d_alternating ] ) . this preserves the correct stationary probability distribution , but the trajectories no longer cross each other ( as at time @xmath121 in fig . [ f : one_d_diffusion ] ) . as a consequence , it suffices to follow the two extremal configurations , which start at sites @xmath89 and @xmath122 , from time @xmath44 on in order to determine the coupling time for a given full monte carlo simulation . the multiple - particle monte carlo simulation starts with the state @xmath123 until it yields a single - particle state . the above strategy of following extremal configurations can be applied to the ferromagnetic ising model ( but not to spin glasses , see @xcite ) . in this case , the two configurations with all spins up and all spins down , respectively , are extremal . this idea also holds for the heat - bath algorithm of two - dimensional directed polymers in a random medium . we study the edwards anderson @xmath124 ising spin glass on a @xmath125-dimensional square lattice , where each site is randomly coupled to all its @xmath126 neighbors . the energy of a configuration @xmath127 with @xmath128 is : @xmath129 where @xmath130 indicates the sum over nearest neighbors . this model has a phase transition at finite temperature in @xmath131 and at zero temperature in @xmath126 . sampling spin - glass configurations with markov - chain algorithms is extremely difficult in @xmath132 dimensions below the critical temperature , but it is also non - trivial in the two - dimensional case @xcite . we concentrate here on the study of a local heat - bath monte carlo algorithm , for which we apply the coupling - from - the - past protocol and obtain exact samples . we note that in two dimensions , the exact partition function of the ising model on a finite lattice can be determined exactly for any choice of couplings @xcite . this makes possible a direct - sampling algorithm , which is completely unrelated to the material presented here , but which we sketch , for the sake of completeness , in appendix [ s : onsager_method ] . the heat - bath monte carlo algorithm for spin models updates at each time step a randomly chosen site @xmath133 of a spin configuration by comparing a function of the local field on site @xmath133 with a uniform random number @xmath134}$ ] : @xmath135}^{-1 } \\ -1 & \text{else } \end{cases } , { \label{e : heat_bath}}\ ] ] where @xmath136 is the local field on site @xmath133 . a realization of the markov chain corresponds to sampling the real - valued random numbers @xmath137 and the random integers @xmath138 . the unit of `` physical '' time ( one `` sweep '' ) corresponds to @xmath25 individual updates . the situation is now much more complicated than for the @xmath139 diffusion , as the role of the five initial configurations in fig . [ f : one_d_diffusion ] is taken up by the @xmath63 possible spin configurations . to prove coupling one must show to which configuration they all converge at the coupling time . the state space is huge and one must find strategies to avoid enumerating and surveying @xmath63 configurations . [ s : partial_survey ] in @xcite , we presented an exact - sampling algorithm which works down to quite low temperatures in the two - dimensional ising spin glass , and which is also operational in three dimensions . we found that practically the same results could be obtained by starting the simulation at time @xmath44 not from all the @xmath63 initial configurations , but from a more manageable number @xmath140 of randomly chosen configurations . we show in fig . [ f : ultra_naive_16x16_0.5 ] that such `` partial survey '' calculations yield useful lower bounds for the coupling time scale @xmath71 . each curve in the figure represents the mean number of distinct configurations remaining after coupled monte carlo simulations ( that is , with the same random numbers @xmath141 for all configurations ) for different values of @xmath140 . increasing @xmath140 within this partial - survey approximation naturally improves the lower bound on the coupling time but , in practice , the value obtained saturates quite quickly . as discussed previously , @xmath142 is always larger than @xmath22 because only non - coupling chains contribute to correlation functions ( see eq . ( [ e : coupling_convergence ] ) ) . to again illustrate the relation between coupling and convergence times , we separate in fig . [ f : spin_spin_corr ] non - coupling chains from the calculation of the spin spin correlation function of a @xmath143 spin glass at inverse temperature @xmath144 . indeed , even if the chain has not coupled , the configurations @xmath80 may lose the dependence on the initial configuration @xmath81 . in this section we discuss the application of the `` coupling from the past '' protocol to hard - sphere systems . the study of monte carlo algorithms for hard - sphere systems goes back a long time , as the metropolis algorithm was first implemented for hard disks , that is , two - dimensional spheres@xcite . even today , the physics of the hard - disk system is not well understood , and monte carlo algorithms have not been developed as successfully as , say , for the ising model . in this very constrained system , the estimation of correlation times is quite controversial , especially at high densities@xcite , and rigorous results from exact - sampling approaches would be extremely welcome . we first discuss the birth death formulation of the markov - chain monte carlo algorithm for this system and then compute lower bounds on coupling times using the partial - survey algorithm . its empirical coupling time saturates ( for increasing @xmath140 ) to much smaller values than the coupling times obtained by the summary - state method @xcite . this suggests that these previous algorithms are not optimal , an impression which is confirmed by our local - patch algorithm of section [ s : patch_disk ] . the partition function of hard spheres in the grand - canonical ensemble , with fugacity @xmath145 , is given by a weighted sum over legal configurations of spheres : @xmath146 here , configurations of @xmath25 spheres are written as : @xmath147 where @xmath148 denotes the centers of the spheres . in eq . ( [ e : partition_function ] ) , @xmath149 equals one if spheres of the configuration @xmath150 do no overlap and zero otherwise . we again use periodic boundary conditions . + the spatial poisson birth death process allows us to apply coupling from the past to hard - sphere systems ( see @xcite ) : disks of radius @xmath151 arrive ( `` are born '' ) randomly on a two - dimensional unit square with constant rate @xmath145 . once born , they disappear ( `` die '' ) with unit rate . the probability for a disk to arrive within an infinitesimal time @xmath152 in a small box of area @xmath153 centered at @xmath154 is @xmath155 . this disk is added to the configuration only if it overlaps with no other disk present . each disk disappears with probability @xmath152 within the time interval @xmath152 . sphere are added at point @xmath154 or removed from the configuration @xmath150 according to the detailed balance condition . with the notation @xmath156 we have : @xmath157 in fig . [ f : birth_death_w ] , we illustrate the time - evolution of accepted and rejected birth - death events on a one - dimensional hard - sphere problem , starting from an empty initial condition at time @xmath44 . in the hard - sphere algorithm , the probability distribution of time intervals between successive births is an exponential with parameter @xmath145 : @xmath158 . in fig . [ f : birth_death_w ] , the life time of a sphere is represented by a horizontal extension of the box , irrespective of whether it has been accepted or not ( the vertical dimension denotes the diameter ) . life times are exponentially distributed as well . for the exponential distribution , the time before the next death of a system of @xmath25 spheres follows an exponential distribution with parameter @xmath25 . likewise the time before any event , ( birth or death ) , follows an exponential distribution with parameter @xmath159 . the probability for the next event to be a birth is then @xmath160 . [ s : hard_disk_coup ] coupling from the past applies to hard - sphere systems even though the space of configurations is continuous ( unlike in lattice simulations ) . to apply the protocol , one considers a time evolution , as in fig . [ f : birth_death_w ] , but stretching back to time @xmath161 . two special aspects must now be handled : first , we must determine which boxes ( corresponding to spheres ) are indeed placed ( `` true '' ) , and which ones are rejected ( `` false '' ) . this is difficult to decide at high density . however , in the low - density case presented in fig . [ f : time_space ] , several spheres are `` true '' , simply because they do not overlap with already present `` true '' or `` false '' spheres . this allows the status of other spheres to be fixed and , finally , the configuration to be constructed . in the limit @xmath162 , the approach works up to a constant density @xcite . this density is much higher than the density @xmath163 direct - sampling algorithm can achieve@xcite . this approach @xcite is equivalent to deciding whether a given spin is up or down in the `` summary state '' algorithm for ising systems @xcite , which , in the thermodynamic limit works down to a fixed constant temperature . second , one must fix the initial condition at time @xmath164 , because spheres born at times smaller than @xmath164 may still be alive at time @xmath164 . this is solved through the sampling of a second time , @xmath165 , after which we know that all spheres present at time @xmath164 have disappeared . the time interval @xmath166 is sampled as the maximum of @xmath167 life times , where @xmath167 is an upper bound on the number of spheres in a legal configuration . figure [ f : time_space ] sketches the time evolution of a monte carlo simulation for the one - dimensional hard - sphere problem which has started at time @xmath161 . boxes are drawn starting from time @xmath164 , but the simulation is picked up at time @xmath168 . it is straightforward to complement the simulation shown ( between times @xmath169 and @xmath164 , for example ) , in case it does not couple in the interval shown . however , we must show that it couples between @xmath169 and @xmath164 or at least results in less than @xmath167 spheres . in the monte carlo simulation in fig . [ f : time_space ] , the status of the boxes at later times can be easily decided , because at later times all spheres belong to clusters which are disjoint from the initial condition . however , this possibility disappears at higher densities . a simple example of this is shown in fig . [ f : example_bd ] . as one can not decide on the status of the initial sphere ( which crosses the line at @xmath168 ) , we should initialize the simulation with the two configurations , one corresponding to a `` true '' state and one to a `` false '' state . after several steps of the time evolution , we arrive in both cases at the same physical configuration ( the two dark spheres , which are both `` true '' ) . for all times @xmath170 , we consider the set @xmath171 of all `` true '' or `` false '' spheres crossing the time line at @xmath12 ( see fig . [ f : time_space ] ) . from the set @xmath172 , one can in principle construct all the possible initial configurations , but their number remains huge . as in the spin - glass case , we may also select @xmath173 among these configurations , and propagate these . this is again the partial survey approximation . in fig . [ f : patches_vs_wilson ] we compare average lower bounds on the coupling time from this approximation with results from the summary state algorithm@xcite . in the time - evolution of fig . [ f : time_space ] , one can determine the number of remaining configurations at any time @xmath174 and detect when exactly the coupling occurs . [ s : local_patch ] in the present section , we discuss our local - patch algorithm , which performs the heat - bath dynamics for a general @xmath125-dimensional ising spin glass on an @xmath25-site hyper - cubic lattice . this algorithm allows us to control all the @xmath63 initial configurations even for very large lattices and to eventually prove that the system has coupled . the python script implementing this algorithm has less than 300 lines . it is available electronically and a listing of the code is contained in appendix [ s : listing_python ] . the ( non - rigorous ) partial - survey algorithm of section [ s : partial_survey ] determines the coupling time for a subset of all the configurations at time @xmath12 . the ( rigorous ) patch algorithm , in contrast , works with a superset of all configurations at time @xmath12 : by restricting the configurations to the smaller region of a patch , one severely limits their number , at the price of introducing compatibility problems between neighboring patches . for the two - dimensional spin glass , we use @xmath25 rectangular patches of same shape and orientation , with @xmath175 sites , and initially at @xmath176 , we have @xmath177 spin configurations on each patch . likewise , the set of global spin configurations is broken up into a list @xmath178 $ ] of sets @xmath179 of spin configurations restricted to patches @xmath14 . we can recover a superset @xmath180 of all relevant spin configurations from the direct product @xmath181 here , each configuration of @xmath180 is pieced together from configurations on all patches , with compatible spins on all lattice sites . ( two compatible spin configurations , on patches @xmath14 and @xmath15 , are shown in fig . [ f : patches_schema ] ) . on large lattices , the direct product in eq . ( [ e : omega ] ) can be performed only if the number of spin configurations per patch is small . if there is only one configuration per patch , we can construct a unique global configuration on the whole lattice . for each time step @xmath12 of the heat - bath algorithm , we choose a random lattice site @xmath133 and a random number @xmath182}$ ] and then update the spin @xmath183 for all configurations on all patches containing @xmath133 . the site @xmath133 may be in the center of a patch @xmath14 ( all the neighbors of site @xmath133 also belong to @xmath14 , as in fig . [ f : patches_schema ] ) . in this case , each configuration of @xmath179 yields one configuration of @xmath184 . several configurations in @xmath179 may yield the same configuration in @xmath184 , so that the size of @xmath185 does not increase in this case . if the site @xmath133 is on the boundary of a patch @xmath15 , we only know upper and lower bounds for the field on the site @xmath133 and , depending on the value of the random number @xmath186 , may be unable to update @xmath183 . in this case , we add two configurations to the set @xmath184 , corresponding to @xmath187 as well as @xmath188 . the set @xmath184 may then contain more configurations than @xmath179 . [ s : pruning ] besides updating configurations on patches , we also perform a `` pruning '' operation : figure [ f : patches_schema ] presents two `` compatible '' configurations on patches @xmath14 and @xmath15 . these could possibly be pieced together into a global configuration , together with configurations on other patches . on the other hand , if the set @xmath189 contains no configuration compatible with a configuration @xmath190 on patch @xmath14 , we can eliminate ( prune ) @xmath190 from @xmath185 . pruning may be implemented through a dictionary ( hash table ) , using as `` key '' the part of the patch configuration in the overlap region between @xmath14 and @xmath15 , and as `` value '' the list of patch configurations sharing this key ( see fig . [ f : dictionary ] ) . this is programmed very easily in the python programming language ( see appendix [ s : listing_python ] ) . pruning can be iterated until all the sets @xmath179 are pairwise compatible . to achieve this goal , it suffices to prune nearest - neighbor patches only . we have found it useful to perform one pruning operation for each pair of nearest - neighbor patches after a certain number of updates ( see fig . [ f : patches_straight_16x16_1.0 ] ) . the average number of configurations per patch saturates to a value which depends on the temperature and also the size of the patch . this is due to the balance between the decrease of the number of configurations induced by the coupling and its increase caused by the noise at the patch boundaries . this noise is reduced through the crucial pruning step of the algorithm . as illustrated in fig . [ f : patches_straight_16x16_1.0 ] , the number of configurations per patch does not necessarily drop to one at large times , even if the underlying heat - bath dynamics couples . the entropy per spin is smaller for larger patches , because the influence of the boundaries is reduced . however , one can not start the computation with large patches because of the large number of possible configurations . a merging procedure allows to increase the patch size in a rigorous way . merging is implemented analogously to pruning : for overlapping patches @xmath14 and @xmath15 , dictionaries are again computed with the same keys and values ( see again fig . [ f : patches_schema ] ) . for a given key configuration on the overlap region , we assemble the corresponding values in the dictionary of patch @xmath14 with all corresponding values in the dictionary of patch @xmath15 . all these couples of configuration must be taken into account for the larger patch @xmath191 . the merging of the configurations on neighboring patches can be implemented very efficiently in the python programming language ( see appendix [ s : listing_python ] ) . in our computations , we start with small square patches , say of size @xmath192 , and then pass to the size @xmath193 , after a few sweeps , then to size @xmath194 , etc . an analogous procedure is followed in higher dimensions . results obtained with this `` jump - start '' approach are shown in fig . [ f : fig_jump_32x32_0.5_5gen ] for the two - dimensional ising spin glass at temperature @xmath195 , with a disorder average performed over about @xmath196 samples . in the patch algorithm , the pruning procedure detects inconsistent configurations in a particular patch . two configurations on different patches are considered compatible if their spins match in the overlap region ( at time @xmath12 ) . more generally , we can keep track of the past evolution of patch configurations and may then declare them compatible only if they have matched for all times up to @xmath12 . otherwise , they can not belong to a unique global spin configuration . in @xcite , the bipartite nature of the square lattice was used to update one entire sub - lattice at a time . in this approach , only one sub - lattice is stored at a time . this allows one to start with larger patches , but the compatibilities between configurations are less well conserved . by contrast , in the present algorithm , we keep the information on both sub - lattices , and one of the sub - lattices is the past configuration . two configurations are compatible in this new version if they are compatible on both sub - lattices . likewise one can use past values of a configuration @xmath197 to restrict its compatibilities with configurations on other patches . other generalizations are more straightforward , one can for example optimize the shape of patches in order to minimize the number of spins on the boundary , and work with more than @xmath25 patches in order to increase the chance for detecting incompatibilities of configurations . [ s : patch_disk ] in this section , we adapt the local - patch algorithm for the classical model of hard disks in a two - dimensional box with periodic boundary conditions . as mentioned , it works for large system sizes at higher density than previous method of exact sampling . in section [ s : hard_disk_coup ] , we introduced the set @xmath171 of all disks the dynamics has tried to add in the box and which have not disappeared at time @xmath12 . the coupled monte carlo simulations start with all possible configurations that are allowed by the set @xmath172 at time @xmath168 . because of possible overlap between disks some of the @xmath198 possible configurations are invalid but there may be far too many of them in practice . to reduce the number of configurations which must be handled , we introduce a regular square lattice with @xmath25 sites covering the simulation box . likewise , a superset of all feasible configurations on a patch at time @xmath168 can be deduced from the set @xmath172 , restricted to disks ( true or false ) with centers inside the patch . from then on , whenever disks appear in the simulation box , we can decide whether they are accepted on a particular patch configuration by checking overlaps into the patch only . disks that disappear are simply removed from all the concerned patch configurations . at birth time , if the disk to be placed on a patch configuration may overlap with a disk outside the patch , the configuration is split into two : one configuration with the new disk and one without ( as for spin systems ) . to detect and prune irrelevant configurations we check that the updated sets of configurations are compatible with other patches by a pruning procedure analogous to the one of section [ s : pruning ] . after several updates , the pruning is performed for most of overlapping patches several times . for any pair of overlapping patches @xmath14 and @xmath15 , the pruning eliminates patch configurations with are inconsistent with all other configurations on a neighboring patch . as in the spin glass case , this process can be implemented with dictionaries ( hash tables ) ( see fig . [ f : dictionary_disks ] ) and can be used to merge configurations on neighboring patches into larger local configurations . figure [ f : patches_vs_wilson ] displays the mean coupling times of a hard - disk birth death simulation for several choices of the fugacity @xmath145 . the radius of the disks is @xmath199 in a unit square box with periodic boundary conditions . results are compared to the partial - survey approximation algorithm , with @xmath200 initial configurations and to the results of the summary - state algorithm . we concentrated on determining the coupling times of the birth death dynamics for hard disks . however , the regime of operation of this algorithm is far in the liquid phase ( see fig . [ f : patches_vs_wilson ] ) , and the physically interesting regime , around the liquid solid transition density , @xmath201 @xcite is still out of reach for exact sampling methods . for hard disks , it remains a challenge to set up a working partial - survey algorithm with correlation times comparable to those of the usual metropolis algorithm @xcite . in this paper , we have discussed exact - sampling algorithms which allow one to totally eliminate the influence of the initial condition from a markov - chain monte carlo simulation . this overcomes one of the main limitations of the method , namely the rigorous estimation of the correlation time . we discussed central subjects , such as the relation between coupling times and convergence times , in a simple example of one - dimensional diffusion , before applying them to ising spin glasses and to hard - sphere simulations . algorithmically , the exact - sampling framework obliges one to follow the entire state space of a system . in the absence of simplifications , such as the half - order discussed in section [ s : coupling_one_d ] , this can be done approximately through a partial survey of @xmath140 initial conditions . one can also restrict the configuration in size onto so - called patches , thereby restricting their number . a superset of the set of global configurations can in principle be reconstituted from the patches . this is easier when the patches are large , and we showed how pruning and merging operations allow one to increase the size of patches during the simulation and to finally prove coupling . our exact - sampling algorithm works both for spin glasses and hard - disk systems , and were able to go to lower temperatures , and higher densities than previous methods . the partial survey algorithm , which can be implemented easily , allowed us to prove that our local - patch algorithm is optimal for the local dynamics for both spin - glass and hard - disk systems . we have provided a number of new idea in order to allow exact - sampling methods to reach the phases transitions of the three - dimensional ising spin glass and the critical density of hard disks . [ s : similarity ] forward and backward transfer matrices completely describe the coupling dynamic of general markov chains ( on a finite state space ) , that is their elements are the coupled transition probabilities between sets of configurations . forward and backward matrices represent `` extended '' monte carlo dynamics , in the two time directions . these two formulations are equivalent , even though the forward dynamics , starting from @xmath87 , does not generate exact samples . here , we demonstrate similarity between forward and backward transfer matrices , and construct the similarity transformation between the two . let @xmath202 be the finite space of configurations of the problem of interest . @xmath202 may contain all @xmath25 positions on the @xmath25-site diffusion problem or the @xmath63 configurations of a @xmath25-site spin system . the @xmath14-configuration states build up an `` extended '' state space . they provide a natural basis for the forward and backward matrices . this basis is @xmath203 , the set of non trivial parts of @xmath202 . for any state @xmath204 we define @xmath205 as the set of states of the basis @xmath206 that has at least one configuration in common with @xmath207 ( @xmath208 ) . the similarity matrix @xmath103 is then defined as : @xmath209 a random mapping arrows for the case of 1@xmath210-diffusion is a mapping on @xmath202 : @xmath211 . it defines a time step of the markov chain for every configuration and satisfies @xmath212 and its weight is noted @xmath213 . in the case of 1@xmath125-diffusion with independent arrows , or any `` independent '' random map in general , we naturally define the weight of the random map as a product of elements of the monte carlo transfer matrix as @xmath214 the forward matrix associates any state @xmath207 to all states that it is connected to by a mapping : @xmath215 using eq . ( [ e : similarity ] ) we find @xmath216 the backward matrix @xmath85 has different rules but we will show that the similarity @xmath217 holds . using a random map @xmath97 , a state @xmath206 at time @xmath12 evolves to another state @xmath218 at time @xmath58 in the backward process if and only if @xmath219 . for example in the 1d - diffusion a hole goes to a hole and a particle goes to a particle . therefore : @xmath220 and finally : @xmath221 in fact eqs ( [ e : bp ] ) and ( [ e : pf ] ) are equivalent because @xmath222 overlaps @xmath207 if and only if @xmath218 overlaps @xmath223 ( @xmath224 ) . this proves the similarity of the backward and forward matrices . [ s : onsager_method ] in this appendix , we sketch for completeness an unrelated direct - sampling algorithm for the two - dimensional ising spin glass . to generate exact samples , this algorithm does not use markov chains . it rather relies on the fact that the partition function of the two - dimensional ising model or of one sample of the spin glass on a planar lattice with @xmath25 sites can be expressed as the square root of the determinant of one @xmath225 matrix ( for open boundary conditions ) or of four such matrices ( for periodic boundary conditions ) @xcite . this relation has been much used in the recent literature , in order to study the physics of the two - dimensional ising spin glass at low temperature @xcite . the partition function yields the thermodynamics of the system , but the knowledge of entire configurations gives for example access to complicated spatial configuration functions . the sampling algorithm for two - dimensional spin - glass configurations constructs the sample one site after another . let us suppose that the gray spins in the left panel of fig . [ f : onsager_direct ] are already fixed , as shown . we can now set a fictitious coupling @xmath226 either to @xmath227 or two @xmath228 and recalculate the partition function @xmath229 with both choices . the statistical weight of all configurations in the original partition function with spin `` @xmath230 '' is then given by @xmath231 and this two - valued distribution can be sampled with one random number . equation ( [ e : direct_sampling ] ) resembles the heat - bath algorithm of eq . ( [ e : heat_bath ] ) , but it is not part of a markov chain : after obtaining the value of the spin on site @xmath14 , we keep the fictitious coupling , and add more sites . going over all sites , we can generate direct spin - glass samples at any temperature . we note that this algorithm is polynomial , and the effort is basically temperature - independent , both for the two - dimensional ising model and the ising spin glass ( see also @xcite ) . [ s : listing_python ] the following python code has produced all the spin - glass data presented in this article . an analogous program was used for the hard - disk system . an electronic version of the code is available from the authors . .... # ! /usr / bin / python # # --------------------------------------------------------------------- # # program : pruning_nd.py # # purpose : this program performs the heat - bath and the pruning on a # # n - dimensional hypercubic lattice with periodic boundary # # conditions and with cuboid patches . # # version with jump - start capability . # # output : mean number of configurations per patch vs. time # # ( can be modified ) # # version : 04-oct-2009 # # author : w. krauth , c. chanal # # language : python 2.5 # # --------------------------------------------------------------------- from random import uniform , randint , seed , shuffle , choice from operator import itemgetter import time , math , os , sys # # --------------------------------------------------------------------- # # sample geometry # # --------------------------------------------------------------------- def torus_neighbors(n_dim , l ) : n = l**n_dim site_dic = { } coord_dic = { } for j in range(n ) : j_d = j x_d = j//(l**(n_dim-1 ) ) coord = [ x_d ] for d in range(n_dim-1 ) : # this loop does not run anything for n_dim=1 j_d = ( j_d - x_d*l**(n_dim - d-1 ) ) x_d = j_d//(l**(n_dim - d-2 ) ) coord.append(x_d ) coord.reverse ( ) # optional , to set the usual order of the directions coord = tuple(coord ) site_dic[coord ] = j coord_dic[j ] = coord nbr = [ ] for j in range(n ) : coord = list(coord_dic[j ] ) nbr_list = [ ] for d in range(n_dim ) : coord[d ] = ( coord[d]+1)%l coord_p = tuple(coord ) nbr_list.append(site_dic[coord_p ] ) coord[d ] = ( coord[d]-1+l)%l for d in range(n_dim ) : coord[d ] = ( coord[d]-1+l)%l coord_p = tuple(coord ) nbr_list.append(site_dic[coord_p ] ) coord[d ] = ( coord[d]+1)%l nbr_list = tuple(nbr_list ) nbr.append(nbr_list ) nbr = tuple(nbr ) return nbr , site_dic , coord_dic # # --------------------------------------------------------------------- # # initial patch geometry # # --------------------------------------------------------------------- def patch_set(n_dim , l , m , site_dic , coord_dic ) : patch = [ ] for j in range(n ) : coord = list(coord_dic[j ] ) dummy_list = [ ] for k in range(m**n_dim ) : coord_p = [ ] for d in range(n_dim ) : l = ( k//(m**d))%m coord_p.append((coord[d]+ l)%l ) coord_p = tuple(coord_p ) dummy_list.append(site_dic[coord_p ] ) patch.append(tuple(dummy_list ) ) patch = tuple(patch ) return patch # # --------------------------------------------------------------------- # # neighbor relations on patches # # --------------------------------------------------------------------- def nbr_patch_set(example_patch , nbr ) : nbr_patch = [ ] for j in example_patch : dummy = [ ] for k in nbr[j ] : if k in example_patch : dummy.append(example_patch.index(k ) ) dummy = tuple(dummy ) nbr_patch.append(dummy ) nbr_patch = tuple(nbr_patch ) return nbr_patch # # --------------------------------------------------------------------- # # initial patch configurations # # --------------------------------------------------------------------- def patch_init(patch ) : def bin(n , conf_length ) : # # # # convert n to binary number with conf_length digits # # q = -1 bin_conf = '' n_digits = 0 while q ! = 0 : q = n//2 r = n%2 bin_conf = ` r`+bin_conf n = q n_digits = n_digits+1 n_digits = conf_length - n_digits for i in range(n_digits ) : bin_conf = ` 0`+bin_conf return bin_conf configs = [ ] dummy_list = [ ] number = len(patch[0 ] ) for j in range(2**number ) : x = bin(j , number ) dummy_list.append(x ) for i in range(len(patch ) ) : configs.append(set(dummy_list ) ) return configs # # --------------------------------------------------------------------- # # updating configurations on all patches # # --------------------------------------------------------------------- def confs_update(n_dim , configs , nbr_patch , jij , patch , i_site , upsilon ) : for k in range(n ) : if i_site in patch[k ] : dummy = list(patch[k ] ) pos = dummy.index(i_site ) llist = [ patch[k][x ] for x in nbr_patch[pos ] ] jij_list = [ jij[(i_site , m ) ] for m in llist ] config_k = configs[k ] config_kp = set ( [ ] ) for c in config_k : field_char = itemgetter(*nbr_patch[pos])(c ) field_sum = sum((2*eval(m)-1)*j for ( m , j ) in zip(field_char , jij_list ) ) field_min = field_sum-2*n_dim+len(field_char ) field_max = field_sum+2*n_dim - len(field_char ) b_one = c[:pos]+'1'+c[pos+1 : ] # configuration with ' 1 ' at position ' pos ' b_zero = c[:pos]+'0'+c[pos+1 : ] # configuration with ' 0 ' at position ' pos ' if upsilon < phplus[field_min ] : config_kp.add(b_one ) elif upsilon > phplus[field_max ] : config_kp.add(b_zero ) else : config_kp.add(b_zero ) config_kp.add(b_one ) configs[k ] = config_kp return configs # # --------------------------------------------------------------------- # # pruning of two patches # # --------------------------------------------------------------------- def prune_pair(config_k , config_l , key_k , key_l ) : f = itemgetter(*key_k ) dic_k = { } for x in config_k : a = f(x ) if dic_k.has_key(a ) : dic_k[a].add(x ) else : dic_k[a ] = set([x ] ) set_k = set(dic_k.keys ( ) ) f = itemgetter(*key_l ) dic_l = { } for x in config_l : a = f(x ) if dic_l.has_key(a ) : dic_l[a].add(x ) else : dic_l[a ] = set([x ] ) set_l = set(dic_l.keys ( ) ) set_kl = set.intersection(set_k,set_l ) config_k = set ( ) config_l = set ( ) for x in set_kl : config_k.update(dic_k[x ] ) config_l.update(dic_l[x ] ) return config_k , config_l # # --------------------------------------------------------------------- # # merging of two patches # # --------------------------------------------------------------------- def merge_pair(config_k , config_l , key_k , key_l , key_add ) : f = itemgetter(*key_k ) dic_k = { } for x in config_k : a = f(x ) if dic_k.has_key(a ) : dic_k[a].add(x ) else : dic_k[a ] = set([x ] ) set_k = set(dic_k.keys ( ) ) f = itemgetter(*key_l ) dic_l = { } for x in config_l : a = f(x ) if dic_l.has_key(a ) : dic_l[a].add(x ) else : dic_l[a ] = set([x ] ) set_l = set(dic_l.keys ( ) ) set_kl = set.intersection(set_k,set_l ) config_kl = set ( ) # merged set for x in set_kl : for y in dic_k[x ] : for z in dic_l[x ] : zprime = ' ' .join(itemgetter(*key_add)(z ) ) config_kl.add(y+zprime ) return config_kl # # --------------------------------------------------------------------- # # main program starts here # # --------------------------------------------------------------------- # seed(13 ) beta = 0.25 # inverse of temperature # # --------------------------------------------------------------------- # # heat bath definitions ( see smac fig . 5.20 , and smac eq . 5.18 ) # # --------------------------------------------------------------------- n_dim = 3 phplus = { } for d in range(n_dim+1 ) : field = 2*d phplus[field ] = 1/(1+math.exp(-2*field*beta ) ) phplus[-field ] = 1/(1+math.exp(2*field*beta ) ) # # --------------------------------------------------------------------- # # loop over samples starts here # # --------------------------------------------------------------------- l = 6 os.system('echo ` hostname ` ` date ` ' ) print beta , l , ' beta l ' for nsamp in range(100 ) : n = l**n_dim m = 2 # m^n_dim is the initial size of patches over_min = ( m-1)*m**(n_dim-1 ) # minimum overlap for the pruning del_t = 40 # time lap on each patch size n_gen = 7 # number of generations t_max = 2000 # total number of sweeps n_frac = 6 # do n / n_frac spin updates between prunings nbr , site_dic , coord_dic = torus_neighbors(n_dim , l ) # # --------------------------------------------------------------------- # # jij : a dictionary ( i , j ) - > j_ij # # --------------------------------------------------------------------- jij = { } for k in range(n ) : for d in range(n_dim ) : jij[(k , nbr[k][d ] ) ] = choice([-1,1 ] ) jij[(nbr[k][d],k ) ] = jij[(k , nbr[k][d ] ) ] patch = patch_set(n_dim , l , m , site_dic , coord_dic ) nbr_patch = nbr_patch_set(list(patch[0]),nbr ) permut = [ k for k in range(n ) ] configs = patch_init(patch ) # # --------------------------------------------------------------------- # # loop over generations in the jump - start procedure # # --------------------------------------------------------------------- for iter1 in range(n_gen ) : i_dir = iter1 % n_dim tot_it = 0 . patch_size = len(patch[0 ] ) print patch_size , ' size of patch ' while ( tot_it < del_t and iter1 < n_gen-1 ) or ( iter1 = = n_gen-1 and \ tot_it + del_t*iter1 < t_max ) : quality = sum([len(configs[k ] ) for k in range(n)])/float(n ) sys.stdout.flush ( ) print tot_it+del_t*iter1 , quality # if ( iter1==n_gen-1 ) : n_frac=18 # do n / n_frac spin updates between prunings for iter3 in range(n / n_frac ) : tot_it + = 1./n i_site = randint(0,n-1 ) upsilon = uniform(0,1 ) configs = confs_update(n_dim , configs , nbr_patch , jij , patch , i_site , upsilon ) # # --------------------------------------------------------------------- # # pruning starts here # # --------------------------------------------------------------------- shuffle(permut ) for kk in range(n ) : k = permut[kk ] for ll in range(kk+1,n ) : l = permut[ll ] inter_set = set(patch[k ] ) & set(patch[l ] ) if len(inter_set ) > = over_min : # minimum overlap key_k = [ list(patch[k]).index(i ) for i in inter_set ] key_l = [ list(patch[l]).index(i ) for i in inter_set ] configs[k],configs[l ] = prune_pair(configs[k],configs[l],key_k , key_l ) if ( quality==1 ) : break if ( quality==1):break # # --------------------------------------------------------------------- # # merging starts here : producing new patches , and computing configurations on them # # --------------------------------------------------------------------- patch_new = [ ] configs_new = [ ] for k in range(n ) : key_add = [ ] dummy_new = list(patch[k ] ) l = nbr[k][i_dir ] # l is the neighbor of k in the ' i_dir ' direction for x in patch[l ] : if x not in patch[k ] : dummy_new.append(x ) key_add.append(list(patch[l]).index(x ) ) patch_new.append(tuple(dummy_new ) ) inter_set = set(patch[k ] ) & set(patch[l ] ) key_k = [ list(patch[k]).index(i ) for i in inter_set ] key_l = [ list(patch[l]).index(i ) for i in inter_set ] config_k_merge = merge_pair(configs[k],configs[l],key_k , key_l , key_add ) configs_new.append(config_k_merge ) # # --------------------------------------------------------------------- # # merging ( final steps ) : computing neighbor relations on the new patches # # --------------------------------------------------------------------- patch = tuple(patch_new ) nbr_patch = nbr_patch_set(list(patch[0]),nbr ) configs = configs_new if i_dir = = n_dim-1 : m + = 1 over_min = ( m-1)*m**(n_dim-1 ) ....

research on quantum statistical properties of light has been a subject of intense investigations and discussions . in view of the very wide - spread potential applications , the influence of material bodies on the quantum features of light must be thoroughly investigated . typical quantum effects that are closely related to the change of the quantum vacuum due to the presence of material bodies are the casimir effect @xcite ( for a review , see @xcite ) and the purcell effect @xcite ( see also @xcite ) . though there has been a large body of work on the problem of quantization of the electromagnetic field in dielectric media ( for a review , see @xcite ) , most work has been concentrated on nonabsorbing media . roughly speaking , there have been two routes to treat radiation in absorbing media . in the first , attention has been restricted to equilibrium field correlation functions that are calculated by employing the dissipation - fluctuation theorem @xcite . in the second , explicit field quantization has been performed ( for a review , see @xcite ) . a consistent quantization scheme for the electromagnetic field in bulk material has been given by huttner and barnett @xcite . using the mesoscopic hopfield model @xcite for a dielectric , they have diagonalized the bilinear hamiltonian of the system that consists of the electromagnetic field , a harmonic - oscillator polarization field , and an infinite set of harmonic - oscillator reservoir fields responsible for absorption . by starting directly with the phenomenological maxwell equations for the macroscopic electromagnetic field , the scheme can naturally be extended to arbitrary inhomogeneous media @xcite characterized by a space- and frequency - dependent complex permittivity that satisfies the kramers - kronig relations . the theory is based on a source - quantity representation of the electromagnetic field , where the field operators are expressed in terms of a continuous set of fundamental bose fields via the green tensor of the classical problem ( for details , see also @xcite ) . the green - tensor formalism is well suited to study the behavior of the quantized electromagnetic field in the presence of dispersing and absorbing bodies . in particular , it has successfully been applied to the study of input - output relations @xcite , the spontaneous emission @xcite , the resonant energy exchange @xcite , the resonant dipole - dipole interaction @xcite , and the casimir effect @xcite , and also various geometries ( including multilayer structures ) have been considered . it is worth noting that the theory not only takes into account material absorption in a consistent way but it also includes automatically evanescent - field contributions generated , e.g. , by radiating atoms that are very close to the bodies under consideration . for the particular case of a dispersing and absorbing dielectric structure in vacuum , without active light sources at any finite distance from the structure ( and without active light sources inside the structure ) , a ratherinvolved hybrid quantization scheme has been proposed @xcite and used to study the problem of three - dimensional input - output relations @xcite . in the scheme , the whole space is divided into two regions , namely a dielectric scattering region and a vacuum region that surrounds the scattering region . the quantization in the vacuum region is performed on the basis of a mode decomposition of the electromagnetic - field operators in a similar way as in vacuum quantum electrodynamics , whereas the contribution of the dielectric scattering region is taken into account by applying the green tensor formalism @xcite . the argument is @xcite that the green tensor formalism alone is not complete , because it does not explicitly treat the boundary of the medium with free space and thus does not describe explicitly the scattering and emission processes which are usually experimentally investigated . this is of course not the case as the above mentioned applications @xcite and the following study of the three - dimensional input - output relations at multilayer plates clearly show . moreover , when there are active light sources at _ finite distances _ from the scattering region , then the incoming field contains both propagating and evanescent components . the latter ones , which are typically observed for small distances , are not included in the hybrid quantization scheme @xcite and the input - output relations derived from it @xcite . however , they are automatically included in the green tensor formalism @xcite . in the following we consider the problem of the three - dimensional input - output relations for the electromagnetic - field operators at the boundaries of a dielectric planar multilayer structure in more detail , by applying the green tensor formalism and extending our previous one - dimensional analysis @xcite to three dimensions . to be quite general , we allow ( i ) for the presence of active light sources at arbitrary positions inside and/or outside the multilayer plate and thus ( ii ) for both propagating - field components and evanescent - field components . in particular , in cavity qed the active sources are typically inside the plate , and the outgoing fields are not only determined with the incoming fields and the noise fields ( unavoidably associated with material absorption ) but also with the fields generated by the sources inside the plate . on the contrary , in optical near - field microscopy the active ( probe ) sources are typically outside the plate , but near its surface . employing the well - known three - dimensional green tensor for a multilayer plate @xcite , we introduce amplitude operators for the input and output fields as well as for the fields inside the plate and derive input - output relations both in the two - dimensional fourier space and in the coordinate space . finally , the problem of introduction of bosonic input and output operators in the two - dimensional fourier space is considered in detail . the paper is organized as follows . in sec . [ sec : ba ] the quantization scheme is briefly summarized and the green tensor for a multilayer plate is introduced . in sec . [ sec : gr ] the generally valid input - output relations are derived , and the problem of formulating input - output relations for bosonic field operators is studied . a summary and some concluding remarks are given in sec . [ sec : co ] followed by an appendix , in which relevant commutation relations are derived . to study optical fields interacting with active sources in the presence of dispersing and absorbing ( linear ) dielectric bodies , we first note that on a length scale that is large compared with interatomic distances in the bodies , the effect of the bodies can be described within the frame of macroscopic maxwell equations in terms of a spatially varying permittivity which is a complex function of frequency . this concept , which is widely used in classical optics also applies in quantum optics . let us assume that the active light sources are neutral atoms and consider an arbitrarily inhomogeneous medium characterized by a permittivity @xmath1 where the real part @xmath2 and the imaginary part @xmath3 are necessarily related to each other through the kramers - kronig relations , due to the causality principle . the motion of the atoms and the medium - assisted electromagnetic field is then governed by the multipolar - coupling hamiltonian @xcite @xmath4 \!\!\bigg\}^2 \nonumber \\[1ex ] & & + \,\frac{1}{2\epsilon_0 } \sum_{a , a ' } \int { \mbox{\rm d}}^3{r } \ , \hat{\bf p}_{a}({\bf r})\cdot \hat{\bf p}_{a'}({\bf r } ) -\sum_a \int { \mbox{\rm d}}^3{r}\ , \hat{{\bf p}}_{a}({\bf r})\cdot \hat{\bf e}({\bf r } ) , \end{aligned}\ ] ] where @xmath5 numbers the atoms , and @xmath6 numbers the charged particles inside the @xmath5-th atom . further , @xmath7 and @xmath8 are respectively the operators of coordinates and canonical momenta of the particles , and @xmath9\ ] ] is the operator of the polarization of the @xmath5-th atom at position @xmath10 . the @xmath11 [ and @xmath12 are bosonic field operators that play the role of fundamental variables of the electromagnetic field and the medium , including a reservoir necessarily associated with material absorption . the commutation relations for the operators @xmath13 and @xmath14 are @xmath15 = \delta _ { \mu \mu'}\delta ( \omega - \omega ' ) \delta ^{(3)}({\bf r } - { \bf r } ' ) , \ ] ] @xmath16 = 0,\ ] ] where the greek letters label the cartesian coordinates @xmath17 . in eq . ( [ 2 ] ) , the operators @xmath18 and @xmath19 of the medium - assisted electromagnetic field are expressed in terms of the fundamental variables as follows : @xmath20 @xmath21 @xmath22 @xmath23 where the integration should be performed over all space , and @xmath24 is the classical green tensor , which can be found from the equation @xmath25 together with appropriate boundary conditions at infinity . equations ( [ 6 ] - [ 9 ] ) can be considered as generalization of the familiar mode decomposition that would apply if dispersion and absorption could be disregarded . instead of dealing with equations of motion for mode operators , equations of motion for the fields @xmath11 must be treated . it should be pointed out that since the real part @xmath2 and the imaginary part @xmath3 of the permittivity are related to each other through the kramers - kronig relations , @xmath3 can not vanish identically for really existing media . clearly , @xmath26 can be very small , so that @xmath27 in eq . ( [ 9 ] ) is very small in certain areas of space ( @xmath28 ) . however , this statement says nothing about the magnitude of the total integral over the coordinate @xmath28 , because of the following integral relation for the classical green tensor @xmath29 , \end{aligned}\ ] ] where we have adopted the convention of summation over repeated vector - component indices . in particular , this relation enables one to include also vacuum - like areas in the consideration . thus , all the calculations are to be performed by assuming a permittivity close to unity with a small but finite imaginary part in those areas , and at the end the permittivity may be set equal to unity . in practice , experimental realization of ( macroscopic ) vacuum areas is of course fictional . scheme of the multilayer dielectric plate . the hatched regions indicate the presence of active light sources . , width=326 ] as mentioned , the quantization scheme is valid for an arbitrary space dependence of the permittivity . here , we consider a multilayered planar structure ( fig . [ fig ] ) , whose permittivity is defined in a stepwise fashion ( the @xmath30-direction is chosen to be perpendicular to the layers ) : @xmath31 where @xmath32 0 , \ , & \mbox{otherwise } , \end{array } \right.\ ] ] and @xmath33 is the ( complex ) permittivity of the @xmath34-th layer . in the above , the index @xmath34 labels the region on the left of the plate , the region on the right of the plate , and the layers of the plate . for simplicity , we express the @xmath30-coordinate dependence in shifted coordinate systems , introduced in each layer separately , so that the range of the @xmath30-coordinate is taken to be for the region on the left of the plate , for the region on the right of the plate , and for the @xmath34-th layer of the plate with thickness @xmath35 . exploiting the translational symmetry in the lane , we may represent the green tensor as a two - dimensional fourier integral @xmath36 where , and is the wave vector parallel to the layers . the notations @xmath37 and @xmath38 indicate that @xmath30 varies in the @xmath34-th layer and @xmath39 in the @xmath40-th layer . inserting eq . ( [ 14 ] ) into eq . ( [ 9 ] ) , we may write the electric - field operator @xmath41 as a two - fold fourier transform , @xmath42 where @xmath43 & & \times \sum _ { j'=0}^n \int _ { [ j ' ] } { \mbox{\rm d}}z'\ , { \mbox{\boldmath $ g$}}^{(jj')}(z , z ' , { \bf k } , \omega)\cdot \hat { \bf j}^{(j ' ) } ( z ' , { \bf k } , \omega ) \end{aligned}\ ] ] ( @xmath44 $ ] indicates integration over the @xmath40-th region ) . here , the fourier transformed current operators @xmath45 have been introduced , where the operators @xmath46 are the fourier transforms of the original bosonic field operators @xmath47 , @xmath48 the green tensor @xmath49 taken from ref . @xcite may be written as @xmath50 & & = - \,{\bf e}_z \,\frac { \delta_{jj ' } } { k_j ^2 } \,{\bf e } _ z\ , \delta ( z - z ' ) + { \mbox{\boldmath $ g$}}^{(jj')}(z , z ' , { \bf k } , \omega ) , \end{aligned}\ ] ] where @xmath51 & & = \frac { i } { 2 } \sum _ { q = p , s}\sigma _ q \left [ \bm { \mathcal{e } } ^{j > } _ { q } ( z , { \bf k } , \omega ) \ , \xi^{jj ' } _ { q } \bm{\mathcal{e } } ^{j ' < } _ { q } ( z ' , -{\bf k } , \omega ) \theta ( j\!-\!j ' ) \right . \nonumber \\[1ex ] & & \left . \,+\ , \bm{\mathcal{e}}^{j < } _ { q } ( z , { \bf k } , \omega ) \ , \xi^{j'j } _ { q } \bm{\mathcal{e } } ^{j ' > } _ { q } ( z ' , -{\bf k } , \omega ) \theta ( j'-j ) \right ] \end{aligned}\ ] ] ( , ) . note that for one should write instead of [ and instead of ] . ( [ 20 ] ) , the functions @xmath52 and @xmath53 denote waves of unit strength traveling , respectively , rightward and leftward in the @xmath34-th layer and being reflected at the boundary , @xmath54 @xmath55 and @xmath56 where @xmath57 . here , @xmath58 ( @xmath59 @xmath60 @xmath61 ) , where @xmath62 and and @xmath63 are , respectively , the transmission and reflection coefficients between the layers @xmath40 and @xmath34 . finally , the unit vectors @xmath64 in eqs . ( [ 21 ] ) and ( [ 22 ] ) are the polarization unit vectors for te and tm waves , @xmath65 \label{28 } { \bf e}^{(j)}_{p\pm}({\bf k } ) & = & \frac { 1 } { k_j } \left(\mp \beta _ j\frac{{\bf k}}{k } + k { \bf e } _ z \right ) . \end{aligned}\ ] ] the ` propagation constant ' @xmath66 determines the propagation behavior in @xmath30-direction of the waves in the @xmath34-th region . note that in case of vacuum the waves are propagating only for ( i.e. , ) . they are evanescent for ( i.e. , ) . the equations of motion which follow from the hamiltonian ( [ 2 ] ) in sec . [ sec : ba.1 ] together with the green tensor as given in sec . [ sec : ba.2 ] can be used to study various effects of the atom - light interaction in the presence of dispersing and absorbing multilayered planar structures ( see the examples mentioned in sec . [ sec1 ] ) . here , we restrict our attention to the general relation between the fields outside and inside a multilayer plate regardless of the generation of the fields by the active light sources . the theory thus includes the general case , where sources are present in the both regions , outside and inside the plate . if the incoming fields , incident on the two boundary planes of the plate ( i.e. , @xmath30 @xmath60 @xmath67 for and @xmath30 @xmath60 @xmath68 for ; cf . fig . [ fig ] ) , are known as well as the fields generated inside the plate , one can calculate the fields outgoing from the two boundary planes by means of input - output relations . note that any two planes @xmath30 @xmath60 @xmath69 @xmath70 @xmath71 and @xmath30 @xmath60 @xmath72 @xmath73 @xmath74 for and , respectively , also can be used in principle . as we shall see , these input - output relations are valid for both passive and active devices , for arbitrary layer materials and arbitrary media surrounding the plate and for both propagating - field components and evanescent - field components . here and in the following we restrict our attention to the electric field , noting that the corresponding expressions for the magnetic field can readily be obtained by applying eq . ( [ 8 ] ) . following the line suggested in ref . @xcite and substituting the green tensor ( [ 19 ] ) for and into eq . ( [ 16 ] ) , we may decompose the field operators @xmath76 and @xmath77 in the form of @xmath78 & & \hspace{15ex } \left . + \ , { \bf e}^{(0)}_{q- } ( { \bf k } ) \hat{\mathrm{e}}_{q\,{\rm out } } ^{(0 ) } ( z,{\bf k } , \omega)\right],\end{aligned}\ ] ] @xmath79 & & \hspace{15ex } \left . + \ , { \bf e}^{(n)}_{q+ } ( { \bf k } ) \hat{\mathrm{e}}_{q\,{\rm out}}^{(n ) } ( z,{\bf k } , \omega)\right ] . \end{aligned}\ ] ] here , the operators @xmath80 & & \times\ ; \int _ { -\infty } ^{z}\ ! \ ! { \mbox{\rm d}}z'\ , e^{- i \beta _ 0 z ' } \,\hat { \bf j } ^{(0 ) } ( z ' , { \bf k } , \omega ) \cdot { \bf e}^{(0)}_{q+}({\bf k } ) , \\[1ex ] \label{32 } \lefteqn { \hat{\mathrm{e}}_{q\,{\rm in}}^{(n ) } ( z , { \bf k } , \omega ) = - \frac { \mu _ 0 \omega } { 2 \beta _ n } e^{- i \beta _ n z } } \nonumber \\[1ex ] & & \times\ ; \int _ { z } ^{\infty } { \mbox{\rm d}}z'\ , e^ { i \beta _ n z ' } \,\hat { \bf j } ^{(n)}(z ' , { \bf k } , \omega ) \cdot { \bf e}^{(n)}_{q-}({\bf k } ) \end{aligned}\ ] ] and @xmath81 & & + \,e^ { - i \beta _ 0 z } \int _ { z } ^{0 } { \mbox{\rm d}}z'\,e^ { i \beta _ 0 z ' } \,\hat { \bf j } ^{(0 ) } ( z ' , { \bf k } , \omega ) \cdot { \bf e}^{(0)}_{q-}({\bf k } ) , \quad \\[1ex ] \label{34 } \lefteqn { \hat{\mathrm{e}}_{q\,{\rm out}}^{(n ) } ( z,{\bf k } , \omega ) = e^ { i \beta _ n z } \hat{\mathrm{e}}_{q\,{\rm out}}^{(n ) } ( { \bf k } , \omega ) } \nonumber\\[1ex ] & & + \,e^ { i \beta _ n z } \int _ { 0 } ^{z } { \mbox{\rm d}}z'\ , e^{- i \beta _ n z ' } \,\hat { \bf j } ^{(n ) } ( z ' , { \bf k } , \omega)\cdot { \bf e}^{(n)}_{q+}({\bf k } ) , \qquad \end{aligned}\ ] ] respectively , play the role of input and output amplitude operators , where the input - output relations @xmath82 \hat{\mathrm{e}}_{q\,{\rm out}}^{(n)}({\bf k } , \omega ) \end{array } \right ) = \left ( \begin{array}{cc } r^q_{0/n}({\bf k},\omega ) & t_{n/0 } ^q({\bf k},\omega)\\[1ex ] t_{0/n } ^q({\bf k},\omega ) & r ^q _ { n/0}({\bf k},\omega ) \end{array } \right ) \left ( \begin{array}{c } \hat{\mathrm{e}}_{q\,{\rm in } } ^{(0)}({\bf k } , \omega ) \\[1ex ] \hat{\mathrm{e}}_{q\,{\rm in}}^{(n)}({\bf k } , \omega ) \end{array } \right ) + \sum _ { j = 1 } ^{n-1 } \left ( \begin{array}{cc } \phi _ { q\ , 0 + } ^{(j)}({\bf k},\omega ) & \phi _ { q\ , 0- } ^{(j)}({\bf k},\omega ) \\[1ex ] \phi _ { q\ , n+ } ^{(j)}({\bf k},\omega ) & \phi _ { q\ , n- } ^{(j)}({\bf k},\omega ) \end{array } \right ) \left ( \begin{array}{c } \hat{\mathrm{e}}_{q+}^{(j)}({\bf k } , \omega)\\[1ex ] \hat{\mathrm{e}}_{q-}^{(j)}({\bf k } , \omega ) \end{array } \right)\!,\quad \end{aligned}\ ] ] are valid ( for the commutation relations , see the appendix ) . they relate the output amplitude operators at the boundary planes of the multilayer plate to the input amplitude operators at the boundary planes , @xmath83 \label{37 } \hat{\mathrm{e}}_{q\,{\rm in , out}}^{(n ) } ( { \bf k } , \omega ) & = & \left . \hat{\mathrm{e}}_{q\,{\rm in , out}}^{(n ) } ( z , { \bf k } , \omega ) \right|_{z=0^+ } , \end{aligned}\ ] ] and the intraplate amplitude operators @xmath84 & & \times\ ; \int _ { 0 } ^{d_j } { \mbox{\rm d}}z ' \ , e^{\mp i \beta _ j z ' } \,\hat { \bf j}^{(j ) } ( z ' , { \bf k } , \omega)\cdot { \bf e}^{(j ) } _ { q\pm}({\bf k})\end{aligned}\ ] ] ( @xmath34 @xmath60 @xmath85 ) , which are associated with the excitation inside the layers of the plate . in eq . ( [ 35 ] ) , the @xmath86-coefficients read @xmath87 @xmath88 it should be pointed out that the first term on the right - hand side in eq . ( [ 19 ] ) , which gives rise to a local contribution to the electric field , has been omitted in eqs . ( [ 29 ] ) and ( [ 30 ] ) . though this contribution is irrelevant for the incoming and outgoing fields , it must be included in the overall field operator in general , even if there are effectively no sources at the points of observations . it is not difficult to prove that ( similar to the one - dimensional case @xcite ) the @xmath30-dependent amplitude operators ( [ 31 ] ) ( [ 34 ] ) obey quantum langevin equations , @xmath89 & & \hspace{10ex } -\,\frac { \mu _ 0 \omega } { 2 \beta _ 0 } \,\hat { \bf j } ^{(0 ) } ( z , { \bf k } , \omega)\cdot { \bf e}^{(0)}_{q+}({\bf k } ) , \end{aligned}\ ] ] @xmath90 & & \hspace{10ex } + \,\frac { \mu _ 0 \omega } { 2 \beta _ 0 } \,\hat { \bf j } ^{(0 ) } ( z , { \bf k } , \omega)\cdot { \bf e}^{(0)}_{q-}({\bf k } ) , \end{aligned}\ ] ] and similar equations are valid for @xmath91 and @xmath92 . these equations together with eq . ( [ 35 ] ) render it possible to easily calculate the input and output fields at any position outside the plate . needless to say that other than the boundary planes and of the plate can be chosen as reference planes for formulating the input - output relations . equation ( [ 35 ] ) represents the basic input - output relations for the amplitude operators in the fourier space . the coefficients therein are determined only by the ( complex ) permittivities and thicknesses of the layers of the plate and the permittivities of the surrounding media . in particular , for ( i.e. , ) eq . ( [ 35 ] ) immediately yields the input - output relations for the special case of the plate being surrounded by vacuum . the input - output relations in the form of eq . ( [ 35 ] ) are generally valid , independent of the mechanism of creation of the incoming fields and the fields inside the layers of the plate . it is worth noting that they take into account both propagating waves and evanescent waves . in particular in the case when the plate is surrounded by vacuum , then the input and output amplitude operators are associated with propagating waves or evanescent waves in @xmath30-direction . the temporal evolution ( in the heisenberg picture ) of the amplitude operators is determined by the time dependence of the basic variables @xmath11 [ or equivalently , @xmath93 , which is governed by the hamiltonian ( [ 2 ] ) . in the special case when the plate is free of active atomic sources , then the fields inside the layers of the plate represent langevin noise sources associated with material absorption . if absorption is disregarded and the plate is surrounded by vacuum that may contain active sources , then eq ( [ 35 ] ) solves a scattering problem of the type considered in ref . @xcite , namely scattering of fields containing evanescent components . in general , absorption can not be disregarded and both atomic sources and langevin noise sources contribute to a field ( e.g. , in cavity qed ) , which can contain both propagating and evanescent components ( e.g. , in near - field scanning probe microscopy ) . the input - output relations ( [ 35 ] ) enable one to calculate correlation functions of the output field amplitudes in terms of those of the input field amplitudes and the amplitudes of the fields inside the plate . the simplest case is the calculation of the expectation values of the field amplitudes . for example , in a typical scattering arrangement the active light sources are located outside the plate , so that the field inside the plate is the absorption - assisted random field whose thermal - equilibrium expectation value vanishes . application of eq . ( [ 35 ] ) thus leads to the expectation - value relations @xmath94 @xmath95 which exactly correspond to standard results in classical optics . on the other hand , when the active sources are located ( in a cavity - like system ) inside the plate , we find that @xmath96 , \end{split}\ ] ] @xmath97 . \qquad \ ; \end{split}\ ] ] needless to say that when there are no active sources inside the plate , then is valid and thus . clearly , higher - order correlation functions of the outgoing amplitude operators do not vanish in this case in general . for example , the spectral intensity ( in the @xmath0-space ) of the radiation outgoing from a plane of a plate in thermal equilibrium at temperature @xmath98 , for chosen polarization , is proportional to @xmath99 , where @xmath100 & & = w_{q\,{\rm out}}^{(0,n)}({\bf k},\omega)\,\delta(\omega-\omega ' ) \delta({\bf k}-{\bf k}').\end{aligned}\ ] ] applying eq . ( [ 35 ] ) together with eqs . ( [ 38 ] ) and ( [ 17 ] ) and making use of @xmath101 & & = n(\omega , t)\ , \delta_{\mu\mu ' } \delta(\omega-\omega')\delta({\bf k}-{\bf k}')\end{aligned}\ ] ] ( @xmath34 @xmath102 @xmath103 ) , where @xmath104^{-1}\ ] ] ( @xmath105 , boltzmann constant ) is the well - known bose - einstein distribution function , we derive @xmath106 & & \left . + \ , \bigl|r_{j/0}^q\bigr|^2 c_{q--}^{(j)}({\bf k},\omega ) + \left[r_{j/0}^q c_{q-+}^{(j)}({\bf k},\omega ) + { \rm c.c.}\right]\right\}\qquad\end{aligned}\ ] ] ( and @xmath107 accordingly ) , with the coefficients @xmath108 being given in eqs . ( [ a15 ] ) and ( [ a16 ] ) in the appendix . the input - output relations ( [ 35 ] ) in the @xmath110-space can be transformed , according to eq . ( [ 15 ] ) , into the @xmath109-space in a straightforward manner . writing the electric - field operator at the boundaries of the plate as @xmath111 & = & \hat{\bf{e}}_{q\,{\rm in } } ^{(0 ) } ( { \bm \rho } , \omega ) + \hat{\bf{e}}_{q\,{\rm out } } ^{(0 ) } ( { \bm \rho } , \omega ) , \\[1ex ] \nonumber\\ \label{48 } \hat{\bf e } ^{(n)}({\bm \rho } , \omega ) & = & \left.\hat{\bf e } ^{(n)}(z , { \bm \rho } , \omega ) \right|_{z=0^+ } \nonumber\\*[1ex ] & = & \ \hat{\bf{e}}_{q\,{\rm in}}^{(n)}({\bm \rho } , \omega ) + \hat{\bf{e}}_{q\,{\rm out}}^{(n ) } ( { \bm \rho } , \omega ) , \end{aligned}\ ] ] we derive @xmath112&&\hspace{10ex } + \sum _ { j=1 } ^{n-1 } \int { \mbox{\rm d}}^2 { \rho}'\ , \left\lbrace { \mbox{\boldmath $ \phi$ } } _ { 0 + } ^{(j ) } ( { \bm \rho } , { \bm \rho ' } , \omega)\cdot \hat{\bf e } _ { + } ^{(j)}({\bm \rho ' } , \omega ) + { \mbox{\boldmath $ \phi$ } } _ { 0- } ^{(j ) } ( { \bm \rho } , { \bm \rho ' } , \omega)\cdot \hat{\bf e } _ { - } ^{(j)}({\bm \rho ' } , \omega ) \right\rbrace , \\[1ex ] \label{50 } \lefteqn { \hat{\bf e } _ { { \rm out } } ^{(n)}({\bm \rho } , \omega ) = \int { \mbox{\rm d}}^2 { \rho}'\ , { \mbox{\boldmath $ r$}}_{n/0 } ( { \bm \rho } , { \bm \rho ' } , \omega)\cdot \hat{\bf e } _ { { \rm in } } ^{(n)}({\bm \rho ' } , \omega ) + \int { \mbox{\rm d}}^2 { \rho}'\ , { \mbox{\boldmath $ t$}}_{0/n } ( { \bm \rho } , { \bm \rho ' } , \omega)\cdot \hat{\bf e } _ { { \rm in } } ^{(0)}({\bm \rho ' } , \omega ) } \nonumber\\[1ex]&&\hspace{10ex } + \sum _ { j=1 } ^{n-1 } \int { \mbox{\rm d}}^2 { \rho}'\ , \left\lbrace { \mbox{\boldmath $ \phi$ } } _ { n+ } ^{(j ) } ( { \bm \rho } , { \bm \rho ' } , \omega)\cdot \hat{\bf e } _ { + } ^{(j)}({\bm \rho ' } , \omega ) + { \mbox{\boldmath $ \phi$ } } _ { n- } ^{(j ) } ( { \bm \rho } , { \bm \rho ' } , \omega)\cdot \hat{\bf e } _ { - } ^{(j)}({\bm \rho ' } , \omega ) \right\rbrace,\end{aligned}\ ] ] where the electric - field operators @xmath113 and @xmath114 are respectively related to the amplitude operators @xmath115 and @xmath116 as @xmath117 \label{52 } \hat{\bf e } _ { { \rm in } } ^{(n)}({\bm \rho } , \omega ) \!=\ ! \frac { 1 } { ( 2 \pi ) ^2}\!\int \!{\mbox{\rm d}}^2 { k}\sum _ { q = p , s}\ ! { \bf e}^{(n)}_{q- } ( { \bf k } ) \hat{\mathrm{e}}_{q\,{\rm in}}^{(n ) } ( { \bf k } , \omega ) e^{i{\bf k } { \bm { \rho } } } \ ! , \quad \\[1ex ] \label{53 } \hat{\bf e } _ { { \rm \pm } } ^{(j)}({\bm \rho } , \omega ) \!=\ ! \frac { 1 } { ( 2 \pi ) ^2}\!\int \!{\mbox{\rm d}}^2 { k}\sum _ { q = p , s}\ ! { \bf e}^{(j)}_{q\pm } ( { \bf k } ) \hat{\mathrm{e}}_{q\,{\rm \pm}}^{(j ) } ( { \bf k } , \omega ) e^{i{\bf k } { \bm { \rho } } } \!. \quad \;\end{aligned}\ ] ] in the above , the tensor - valued reflection and transmission integral kernels , respectively , read as @xmath118 \label{55 } \lefteqn { { \bf r}_{n/0 } ( { \bm \rho } , { \bm \rho ' } , \omega ) } \nonumber \\ & & = \frac { 1 } { ( 2 \pi)^2}\int { \mbox{\rm d}}^2 { k}\sum _ { q = p , s } { \bf{e}}^{(n)}_{q+ } ( { \bf k } ) r^q _ { n/0}{\bf{e}}^{(n)}_{q- } ( { \bf k } ) e^{i{\bf k } ( { \bm \rho}-{\bm \rho'})}\qquad \end{aligned}\ ] ] and @xmath119 \label{57 } \lefteqn { { \bf t}_{n/0 } ( { \bm \rho } , { \bm \rho ' } , \omega ) } \nonumber \\ & & = \frac { 1 } { ( 2 \pi)^2}\int { \mbox{\rm d}}^2 { k } \sum _ { q = p , s } { \bf{e}}^{(0)}_{q- } ( { \bf k } ) t^q _ { n/0}{\bf{e}}^{(n)}_{q-}({\bf k } ) e^{i{\bf k } ( { \bm \rho}-{\bm \rho'})},\qquad\end{aligned}\ ] ] and the integral kernels that relate the outgoing fields to the intraplate fields are given by @xmath120 \lefteqn { \label{59 } { \bf \phi}_{n\pm } ( { \bm \rho } , { \bm \rho ' } , \omega ) } \nonumber \\ & & = \frac { 1 } { ( 2 \pi)^2}\int { \mbox{\rm d}}^2 { k } \sum _ { q = p , s } { \bf e}^{(j)}_{q+ } ( { \bf k } ) \phi^{(j ) } _ { q\ , n \pm } { \bf e}^{(j)}_{q\pm}({\bf k } ) e^{i{\bf k } ( { \bm \rho}-{\bm \rho'})}.\qquad\end{aligned}\ ] ] note that the integral kernels depend only on the difference , because of the translational symmetry in the @xmath121- and @xmath122-directions . the input - output relations in the form of eqs . ( [ 49 ] ) , ( [ 50 ] ) can be considered as a special realization of the general relations given for a dielectric scattering region of arbitrary shape in ref . @xcite . in many applications it may be advantageously to express the incoming and the outgoing field operators in terms of appropriately chosen bosonic operators . recalling the basic commutation relations ( [ 4 ] ) and ( [ 5 ] ) and the orthogonality of polarization unit vectors as given by eqs . ( [ 27 ] ) and ( [ 28 ] ) , from eqs . ( [ 31 ] ) and ( [ 32 ] ) together with eq . ( [ 17 ] ) and ( [ 18 ] ) and eqs . ( [ 36 ] ) and ( [ 37 ] ) we find that the input amplitude operators satisfy the commutation relations @xmath123 } \nonumber \\[1ex ] & & \,=\ , c_{q\,{\rm in}}^{(0,n)}({\bf k},\omega)\ , \delta_{qq'}\delta(\omega-\omega')\delta({\bf k}-{\bf k}'),\end{aligned}\ ] ] where @xmath124 \label{62 } c_{q\,{\rm in}}^{(n ) } = \frac{\pi \hbar}{\epsilon_0 } \left(\frac{\omega}{c}\right)^2 \frac{\beta_{n}'}{|\beta_{n}|^2 } \,{\bf e}^{(n)}_{q- } ( { \bf k } ) \cdot{\bf e}^{(n)*}_{q- } ( { \bf k } ) . \end{aligned}\ ] ] similarly , the output amplitude operators and the intraplate amplitude operators can be shown to satisfy commutation relations of the same type , i.e. , @xmath125 } \nonumber \\[1ex ] & & = \ , c_{q\,{\rm out}}^{(0,n)}({\bf k},\omega)\ , \delta_{qq'}\delta(\omega-\omega')\delta({\bf k}-{\bf k } ' ) , \\[1ex ] \label{64 } \lefteqn { \left[\hat{\mathrm{e}}_{q\lambda}^{(j ) } ( { \bf k } , \omega ) , \hat{\mathrm{e}}_{q'\lambda'}^{(j)\dagger } ( { \bf k } ' , \omega')\right ] } \nonumber \\[1ex ] & & = \,c_{q\lambda\lambda'}^{(j)}({\bf k},\omega)\ , \delta_{qq'}\delta(\omega-\omega')\delta({\bf k}-{\bf k}')\end{aligned}\ ] ] [ for the coefficients @xmath126 and @xmath108 , respectively , see eqs . ( [ a12 ] ) , ( [ a13 ] ) and eqs . ( [ a15 ] ) , ( [ a16 ] ) in the appendix ] . needless to say that input amplitude operators , that refer to different sides of the plate , commute , input amplitude operators and intraplate amplitude operators commute , and intraplate amplitude operators that refer to different layers also commute . note that output amplitude operators that refer to different sides of the plate do not commute in general [ see eqs . ( [ a9 ] ) and ( [ a10 ] ) in the appendix ] , which is similar as in the one - dimensional situation ( cf . @xcite ) . from eqs . ( [ 60 ] ) and ( [ 63 ] ) it is seen that bosonic input and output operators can be introduced according to @xmath127 thus @xmath128 } \nonumber \\[1ex]&&\hspace{6ex } = \ , \delta_{qq'}\delta(\omega-\omega')\delta({\bf k}-{\bf k } ' ) . \end{aligned}\ ] ] since for in eq . ( [ 64 ] ) is valid in general , it is useful to introduce intraplate bosonic operators according to the superposition @xmath129 and choosing the coefficients @xmath130 [ eqs . ( [ a20 ] ) and ( [ a21 ] ) in the appendix ] in such a way that @xmath131 } \nonumber\\&&\hspace{6ex } = \,\delta_{\lambda\lambda'}\delta_{qq ' } \delta ( \omega -\omega ' ) \delta({\bf k } - { \bf k ' } ) . \end{aligned}\ ] ] substituting eqs . ( [ 65 ] ) and ( [ 67 ] ) into eq . ( [ 35 ] ) , we may express the input - output relations in the @xmath0-space in terms of bosonic operators , @xmath132 \hat{a}_{q\,{\rm out}}^{(n ) } ( { \bf k } , \omega ) \end{array } \right ) = \left ( \begin{array}{cc } \tilde{r}^q_{0/n}({\bf k},\omega ) & \tilde{t}_{n/0 } ^q({\bf k},\omega)\\[1ex ] \tilde{t}_{0/n } ^q({\bf k},\omega ) & \tilde{r } ^q _ { n/0}({\bf k},\omega ) \end{array } \right ) \left ( \begin{array}{c } \hat{a}_{q\,{\rm in}}^{(0 ) } ( { \bf k } , \omega ) \\[1ex ] \hat{a}_{q\,{\rm in}}^{(n ) } ( { \bf k } , \omega ) \end{array } \right ) + \sum _ { j = 1 } ^{n-1 } \left ( \begin{array}{cc } \tilde{\phi } _ { q\ , 0 + } ^{(j)}({\bf k},\omega ) & \tilde{\phi } _ { q\ , 0- } ^{(j)}({\bf k},\omega ) \\[1ex ] \tilde{\phi } _ { q\ , n+ } ^{(j)}({\bf k},\omega ) & \tilde{\phi } _ { q\ , n- } ^{(j)}({\bf k},\omega ) \end{array } \right ) \left ( \begin{array}{c } \hat { a}_{q+}^{(j ) } ( { \bf k } , \omega)\\[1ex ] \hat { a}_{q-}^{(j ) } ( { \bf k } , \omega ) \end{array } \right)\!,\quad \end{aligned}\ ] ] where the coefficients [ modified in comparison with eq . ( [ 35 ] ) ] read @xmath133 \label{71 } \tilde{t } ^q _ { 0/n } = \sqrt { \frac{c_{q\,{\rm in}}^{(0 ) } } { c_{q\,{\rm out}}^{(n ) } } } \,t ^q _ { n/0 } \ , , \qquad \tilde{r}_{n/0}^q = \sqrt { \frac{c_{q\,{\rm in}}^{(n ) } } { c_{q\,{\rm out}}^{(n ) } } } \,r_{n/0 } ^q \,,\end{aligned}\ ] ] and @xmath134 note that for ( i.e. , perpendicular incidence of light on the plate ) , the input and output relations ( [ 69 ] ) exactly agree with previous results @xcite obtained within a one - dimensional treatment . let us turn to the limiting case that the space outside the plate except for possible active atomic sources may be regarded as being vacuum , i.e. , @xmath135 @xmath136 \sqrt{\omega^2/c^2 - k^2 } & \omega / c > k. \end{array } \right.\ ] ] for the propagating - field components observed for , the coefficients @xmath137 and @xmath138 , respectively , read @xmath139 and @xmath140 as it can be seen from eqs . ( [ 61 ] ) and ( [ 62 ] ) and eqs . ( [ a23 ] ) and ( [ a24 ] ) in the appendix . hence , bosonic input and output operators can be defined according to eq . ( [ 65 ] ) . moreover , the output amplitude operators that refer to different sides of the plate commute [ eq . ( [ a22 ] ) in the appendix ] , which implies that the associated bosonic operators commute as well . making use of eqs . ( [ 75 ] ) and ( [ 76 ] ) , we see that the modified reflection and transmission coefficients ( [ 70 ] ) and ( [ 71 ] ) , respectively , become identical with the original ones . for , eq . ( [ 69 ] ) thus reduces to @xmath141 \hat{a}_{q\,{\rm out}}^{(n ) } ( { \bf k } , \omega ) \end{array } \right ) = \biggl ( \begin{array}{cc } r^q_{0/n}({\bf k } , \omega ) & t_{n/0 } ^q({\bf k } , \omega)\\[1ex ] t_{0/n } ^q({\bf k } , \omega ) & r ^q _ { n/0}({\bf k } , \omega ) \end{array } \biggr ) \left ( \begin{array}{c } \hat{a}_{q\,{\rm in}}^{(0 ) } ( { \bf k } , \omega ) \\[1ex ] \hat{a}_{q\,{\rm in}}^{(n ) } ( { \bf k } , \omega ) \end{array } \right ) + \sum _ { j = 1 } ^{n-1 } \left ( \begin{array}{cc } \tilde{\phi } _ { q\ , 0 + } ^{(j)}({\bf k } , \omega ) & \tilde{\phi } _ { q\ , 0- } ^{(j)}({\bf k } , \omega ) \\[1ex ] \tilde{\phi } _ { q\ , n+ } ^{(j)}({\bf k } , \omega ) & \tilde{\phi } _ { q\ , n- } ^{(j)}({\bf k } , \omega ) \end{array } \right ) \left ( \begin{array}{c } \hat { a}_{q+}^{(j ) } ( { \bf k } , \omega)\\[1ex ] \hat { a}_{q-}^{(j ) } ( { \bf k } , \omega ) \end{array } \right),\quad \end{aligned}\ ] ] where the transformation matrix connecting the bosonic output operators with the bosonic input operators is exactly the same as that for the corresponding amplitude operators in eq . ( [ 35 ] ) , and the matrix equation @xmath142 t_{0/n } ^q & r ^q _ { n/0 } \end{array } \right ) \left ( \begin{array}{cc } r^{q\ast}_{0/n } & t_{0/n } ^{q\ast}\\[1ex ] t_{n/0 } ^{q\ast } & r ^{q\ast } _ { n/0 } \end{array } \right ) + \sum _ { j = 1 } ^{n-1 } \left ( \begin{array}{cc } \tilde{\phi } _ { q\ , 0 + } ^{(j ) } & \tilde{\phi } _ { q\ , 0- } ^{(j ) } \\[1ex ] \tilde{\phi } _ { q\ , n+ } ^{(j ) } & \tilde{\phi } _ { q\ , n- } ^{(j ) } \end{array } \right ) \left ( \begin{array}{cc } \tilde{\phi } _ { q\ , 0 + } ^{(j)\ast } & \tilde{\phi } _ { q\ , n+ } ^{(j)\ast } \\[1ex ] \tilde{\phi } _ { q\ , 0- } ^{(j)\ast } & \tilde{\phi } _ { q\ , n- } ^{(j)\ast } \end{array } \right ) = \left ( \begin{array}{cc } 1 & 0 \\[1ex ] 0 & 1 \end{array } \right ) \end{aligned}\ ] ] is valid . in this case , the field operators @xmath143 and @xmath144 at the boundaries of the multilayer plate [ see eqs . ( [ 29 ] ) , ( [ 30 ] ) , ( [ 36 ] ) , and ( [ 37 ] ) ] can be represented as @xmath145 , \label{79}\\[1ex ] { \bf \hat{e } } ^{(n)}(0^+,{\bf k } , \omega ) = -\frac { \omega } { c } \sqrt{\frac { \pi \hbar } { \beta _ n\epsilon _ 0 } } \sum _ { q = p , s}\ , \left [ { \bf{e}}_{q-}^{(n ) } ( { \bf k } ) \hat{a}_{q\,{\rm in}}^{(n ) } ( { \bf k } , \omega ) + { \bf{e}}_{q+}^{(n ) } ( { \bf k } ) \hat{a}_{q\,{\rm out}}^{(n ) } ( { \bf k } , \omega)\right ] \label{80}\end{aligned}\ ] ] ( @xmath146 @xmath60 @xmath147 @xmath60 @xmath148 , @xmath149 @xmath150 @xmath151 ) . note that in eqs . ( [ 79 ] ) and ( [ 80 ] ) is nothing said about the location of the active light sources . clearly , if there are no active light sources outside the plate , then @xmath143 and @xmath144 evolve effectively freely there [ cf . ( [ 33 ] ) and ( [ 34 ] ) ] . for the evanescent - field components observed for , the coefficients @xmath152 [ eq . ( [ 61 ] ) ] and @xmath153 [ eq . ( [ 62 ] ) ] identically vanish , because of @xmath154 @xmath60 @xmath155 @xmath60 @xmath156 . recalling eq . ( [ 65 ] ) , we see that bosonic input operators can not be introduced for the evanescent - field components . hence , it is impossible to extend the validity of eq . ( [ 77 ] ) to the evanescent - field components . to treat them , one has therefore to go back to the generally valid input - output relations ( [ 35 ] ) for the amplitude operators . clearly , when there are no active light sources at any finite distance from the plate , then the input field can be regarded as being effectively a propagating field [ see eqs . ( [ 31 ] ) and ( [ 32 ] ) ] , that is to say , a field that does not contain evanescent input components . however , the output field may contain evanescent output components resulting from the intraplate field . applying the green - tensor formalism of the quantization of the electromagnetic field in the presence of inhomogeneous , causal , linear dielectric bodies , we have derived the most general three - dimensional quantum input - output relations for the electromagnetic - field operators at an arbitrary multilayer dielectric plate , and we have given the commutation relations needed . taking into account ( i ) dispersion and absorption of the plate and the surrounding medium , ( ii ) possible active light sources inside and/or outside the plate , and ( iii ) both propagating - field components and evanescent - field components , the input - output relations connect the output field operators at the boundaries of the plate with the input field operators at the boundaries of the plate and the fields inside the layers of the plate . the outgoing fields outside the plate can then be obtained from the ones at the boundaries of the plate via quantum langevin equations , with the langevin noise sources being determined by the active and passive light sources in the respective half - space outside the plate . we have given the input - output relations in terms of both algebraic relations for amplitude operators in the @xmath0-space and integral relations for field operators in the coordinate space . further , we have shown that in the @xmath0-space it is possible , provided that the incoming fields contain effectively propagating components , to rewrite the input - output relations in order to obtain them in terms of relations between bosonic operators instead of amplitude operators . therefore the input - output relations for the amplitude operators are more general than those for the bosonic operators . finally , we have studied the limiting case of the plate being surrounded by vacuum . in this case , the input - output relations in terms of bosonic operators only apply to the propagating - field components , for which is valid . an application to the evanescent - field components would run into contradictions . to treat them , one has therefore to go back to the amplitude operators . the derived input - output relations generalize previous results obtained from a one - dimensional calculation @xcite ) . in the special case of a plate that is embedded in vacuum , without active light sources at any finite distance from the plate ( and without active light inside the plate ) , they reduce to those given in ref . @xcite , if evanescent - field components are disregarded . in conclusion , the derived input - output relations are suited for studying the quantum statistical properties of electromagnetic fields in the presence of planar multilayered structures ( such as cavity - like systems or photonic crystals ) , with special emphasis on absorption - assisted quantum decoherence , including evanescent - field effects . we would like to thank ho trung dung and c. raabe for useful discussions . let us begin with the calculation of the commutators of the output amplitude operators @xmath157 . from eq . ( [ 35 ] ) it follows that @xmath158&&\hspace{6ex } + \,t_{n/0}^q \hat{\mathrm{e}}_{q\,{\rm in}}^{(n ) } ( { \bf k } , \omega ) + \hat{\mathrm{f}}_{q}^{(0)}({\bf k } , \omega ) \end{aligned}\ ] ] and @xmath159&&\hspace{6ex } + \ , r_{n/0}^q \hat{\mathrm{e}}_{q\,{\rm in}}^{(n ) } ( { \bf k } , \omega ) + \hat{\mathrm{f}}_{q}^{(n)}({\bf k } , \omega ) , \end{aligned}\ ] ] where @xmath160 \\ & = i \omega \mu _ 0 { \bf e}^{(0)}_{q-}({\bf k } ) \cdot\sum _ { j=1}^{n-1 } \int _ { [ j ] } { \mbox{\rm d}}z\ , { \mbox{\boldmath $ g$}}^{(0j)}(0 , z , { \bf k } , \omega)\cdot \hat { \bf j}^{(j ) } ( z , { \bf k } , \omega ) , \end{split}\ ] ] @xmath161\\ & = i \omega \mu _ 0 \,{\bf e}^{(n)}_{q+}({\bf k } ) \cdot\sum _ { j=1}^{n-1 } \int _ { [ j ] } { \mbox{\rm d}}z\ , { \mbox{\boldmath $ g$}}^{(nj)}(0 , z , { \bf k } , \omega)\cdot \hat { \bf j}^{(j ) } ( z , { \bf k } , \omega ) . \end{split}\ ] ] hence , the ( relevant ) commutators of the output amplitude operators that refer to different sides of the plate can be given by @xmath162 } \nonumber\\[1ex ] & & = \,r^q_{0/n } t_{0/n } ^{q\ , * } \left [ \hat{\mathrm{e}}_{q\,{\rm in } } ^{(0 ) } ( { \bf k } , \omega ) , \hat{\mathrm{e}}_{q'\,{\rm in } } ^{(0)\dagger } ( { \bf k } ' , \omega ' ) \right ] + t_{n/0 } ^q r^{q\,*}_{n/0 } \left [ \hat{\mathrm{e}}_{q\,{\rm in } } ^{(n ) } ( { \bf k } , \omega ) , \hat{\mathrm{e}}_{q'\,{\rm in } } ^{(n)\dagger } ( { \bf k } ' , \omega ' ) \right ] + \left [ \hat{\mathrm{f}}_{q } ^{(0 ) } ( { \bf k } , \omega ) , \hat{\mathrm{f}}_{q ' } ^{(n)\dagger } ( { \bf k } ' , \omega ' ) \right ] \nonumber\\[1ex ] & & = \,\delta_{qq'}\ , \delta(\omega-\omega')\,\delta({\bf k}-{\bf k}')\ , \left ( c_{q\,{\rm in}}^{(0 ) } r^q_{0/n } t_{0/n } ^{q\ , * } + c_{q\,{\rm in}}^{(n ) } t_{n/0 } ^q r^{q\,*}_{n/0 } \right ) + \left [ \hat{\mathrm{f}}_{q } ^{(0 ) } ( { \bf k } , \omega ) , \hat{\mathrm{f}}_{q ' } ^{(n)\dagger } ( { \bf k } ' , \omega ' ) \right ] , \end{aligned}\ ] ] where eqs . ( [ 60 ] ) ( [ 62 ] ) have been used . making use of eqs . ( [ 17 ] ) and ( [ 18 ] ) and recalling the basic commutation relations ( [ 4 ] ) and ( [ 5 ] ) , we derive @xmath163 } \nonumber\\*[1ex ] & & = \,\delta_{qq'}\,\delta(\omega-\omega')\ , \delta({\bf k}-{\bf k}')\ , \frac{4\pi \hbar}{\epsilon _ 0}\left(\frac { \omega}{c}\right ) ^2 \ , e^{(0)}_{q-\ , \mu } ( { \bf k } ) \ , e^{(n)\ , * } _ { q+\ , \mu ' } ( { \bf k } ) \sum_{j=1}^{n-1}\,\int_{[j ] } { \mbox{\rm d}}z\ , g^{(0j)}_{\mu\nu}(0 , z , { \bf k } , \omega ) \ , \frac { \omega ^2 } { c^2 } \varepsilon_{j}''\ , g^{(nj)*}_{\mu'\nu}(0 , z , { \bf k } , \omega ) \nonumber\\[1ex ] & & = \ , \delta_{qq'}\,\delta(\omega-\omega')\ , \delta({\bf k}-{\bf k}')\ , \frac{4\pi \hbar}{\epsilon _ 0}\left(\frac { \omega}{c}\right ) ^2 \ , e^{(0)}_{q-\ , \mu } ( { \bf k } ) \ , e^{(n)\ , * } _ { q+\ , \mu ' } ( { \bf k } ) \ , \biggl\lbrace \sum_{j=0}^{n}\,\int_{[j ] } { \mbox{\rm d}}z\ , g^{(0j)}_{\mu\nu}(0 , z , { \bf k } , \omega ) \ , \frac { \omega ^2 } { c^2 } \varepsilon_{j}''\ , g^{(nj)*}_{\mu'\nu}(0 , z , { \bf k } , \omega ) \biggr . \nonumber\\[1ex]&&\qquad \biggl . - \int_{-\infty}^{0 } { \mbox{\rm d}}z\ , g^{(00)}_{\mu\nu}(0 , z , { \bf k } , \omega ) \ , \frac { \omega ^2 } { c^2 } \varepsilon_{0}''\ , g^{(n0)*}_{\mu'\nu}(0 , z , { \bf k } , \omega ) - \int^{\infty}_{0 } { \mbox{\rm d}}z\ , g^{(0n)}_{\mu\nu}(0 , z , { \bf k } , \omega ) \ , \frac { \omega ^2 } { c^2 } \varepsilon_{n}''\ , g^{(nn)*}_{\mu'\nu}(0 , z , { \bf k } , \omega ) \biggr\rbrace.\end{aligned}\ ] ] it is not difficult to calculate the last two integrals in eq . ( [ a6 ] ) , by using the explicit expression ( [ 20 ] ) for @xmath164 . to calculate the sum of integrals @xmath165 } { \mbox{\rm d}}z\,\ldots$ ] , we employ the integral relation ( [ 11 ] ) for the classical green tensor @xmath166 , rewritten in terms of @xmath167 , @xmath168 } { \mbox{\rm d}}z''\ , g^{(jj'')}_{\mu\nu}(z , z '' , { \bf k } , \omega ) \ , \frac { \omega ^2 } { c^2 } \varepsilon_{j''}''\ , g^{(j'j'')*}_{\mu'\nu}(z ' , z '' , { \bf k } , \omega ) = \frac{1}{2i } g^{(jj')}_{\mu\mu'}(z , z ' , { \bf k } , \omega ) } \nonumber\\[1ex ] & & -\,\frac{1}{2i } g^{(j'j)*}_{\mu'\mu}(z ' , z , { \bf k } , \omega ) + g^{(jj')}_{\mu\nu}(z , z ' , { \bf k } , \omega ) e_{z\,\nu } \,\frac{\varepsilon_{j'}''}{\varepsilon_{j'}^*}\,e_{z\ , \mu ' } + e_{z\,\mu } \ , \frac{\varepsilon_{j}''}{\varepsilon_{j}}\ , g^{(j'j)*}_{\mu'\nu}(z ' , z , { \bf k } , \omega ) e_{z\,\nu } \ , . \end{aligned}\ ] ] after lengthy , but straightforward calculations we then derive , on using eqs . ( [ 61 ] ) and ( [ 62 ] ) , @xmath169 } \nonumber\\[1ex ] & & = \left [ \hat{\mathrm{e}}_{p\,{\rm out } } ^{(0 ) } ( { \bf k } , \omega ) , \hat{\mathrm{e}}_{s\,{\rm out } } ^{(n)\dagger } ( { \bf k } ' , \omega ' ) \right]=0,\end{aligned}\ ] ] @xmath170 = \delta(\omega-\omega')\,\delta({\bf k}-{\bf k } ' ) } \qquad \nonumber\\[1ex]&&\times\ ; \frac{\pi \hbar}{\epsilon _ 0}\left(\frac { \omega}{c}\right)^2 \left ( \frac{i \beta_0 ' ' } { { |\beta _ 0 |}^2 } t^{s\,*}_{0/n}- \frac { i \beta _ n''}{{|\beta _ n |}^2 } t_{n/0}^{s}\right ) , \qquad\end{aligned}\ ] ] @xmath171= \frac{\pi \hbar}{\epsilon _ 0}\left(\frac { \omega}{c}\right ) ^2 \left\ { \frac { t_{n/0 } ^p}{\beta _ n { |k_n| } ^2 } \left(k^2\frac{k_n ^2}{k_n^{*\,2 } } - { |\beta _ n |}^2 \right ) \right . } \nonumber\\[1ex ] & & \left . + \,\frac { t_{0/n } ^{p\ , * } } { \beta _ 0 ^ * { |k_0| } ^2 } \left(k^2\frac{k_0^{*\,2}}{k_0 ^2 } - { |\beta _ 0 |}^2 \right ) - \frac{\beta_n'}{{|\beta_n|}^2}\,t^{p}_{n/0 } \left[{\bf e}^{(n)}_{p- } ( { \bf k } ) \cdot{\bf e}^{(n)*}_{p- } ( { \bf k } ) \right ] \left[{\bf e}^{(n)*}_{p- } ( { \bf k})\cdot\,{\bf e}^{(n)*}_{p+ } ( { \bf k } ) \right ] \right . \nonumber\\[1ex ] & & -\ , \frac{\beta_0'}{{|\beta_0|}^2}\,t^{q*}_{0/n}\ , \left[{\bf e}^{(0)*}_{q+ } ( { \bf k } ) \cdot{\bf e}^{(0)}_{q+ } ( { \bf k } ) \right ] \left[{\bf e}^{(0)}_{q+ } ( { \bf k } ) \cdot{\bf e}^{(0)}_{q- } ( { \bf k } ) \right ] \biggr\}\ , \delta(\omega-\omega')\,\delta({\bf k}-{\bf k}')\ , . \end{aligned}\ ] ] next , let us consider the commutators of the output amplitude operators that refer to the same sides of the plate . performing the same steps as before , we now arrive at @xmath172 } \nonumber\\[1ex ] & & = \left [ \hat{\mathrm{e}}_{p\,{\rm out } } ^{(0 ) } ( { \bf k } , \omega ) , \hat{\mathrm{e}}_{s\,{\rm out } } ^{(0)\dagger } ( { \bf k } ' , \omega ' ) \right ] = 0,\end{aligned}\ ] ] @xmath173 = \delta(\omega-\omega')\,\delta({\bf k}-{\bf k } ' ) } \qquad \nonumber\\[1ex]&&\times\ ; \frac{\pi \hbar}{\epsilon _ 0}\left(\frac { \omega}{c}\right ) ^2 \left ( \frac{\beta_0 ' } { { |\beta _ 0 |}^2 } + \frac { 2 \beta _ 0''}{{|\beta _ 0 |}^2 } r_{0/n}^{s\,\prime \prime}\right ) , \qquad\end{aligned}\ ] ] @xmath174= \frac{\pi \hbar}{\epsilon _ 0}\left(\frac{\omega}{c}\right ) ^2 \left\ { \frac { r_{0/n } ^p}{\beta _ 0 { |k_0| } ^2 } \left(k^2\frac{k_0 ^2}{k_0^{*\,2 } } - { |\beta _ 0 |}^2 \right ) \right . } \nonumber\\[1ex ] & & \left . + \,\frac { r_{0/n } ^{p\ , * } } { \beta _ 0 ^ * { |k_0| } ^2 } \left(k^2\frac{k_0^{*\,2}}{k_0 ^2 } - { |\beta _ 0 |}^2 \right ) - \frac { k ^2 } { { |k_0| } ^2}\left(\frac{\beta _ 0}{k_0^{*\,2}}+ \frac{\beta _ 0 ^*}{k_0^{2 } } \right ) + \frac { 2 \beta _ 0 ' } { { |k_0| } ^2}\left ( \frac { k ^4 } { { |\beta_0| } ^2 { |k_0| } ^2 } + 1 \right ) \right . \nonumber\\[1ex ] & & + \,\frac{\beta_0 ' } { { |\beta _ 0 |}^2 } \left [ { \bf e}^{(0)}_{p+ } ( { \bf k } ) \cdot{\bf e}^{(0)*}_{p+ } ( { \bf k } ) \right ] \left [ \left| r_{0/n}^q\right|^2 - \left|{\bf e}^{(0)}_{p- } ( { \bf k } ) \cdot { \bf e}^{(0)}_{p+ } ( { \bf k } ) + r_{0/n } ^p\right| ^2 \right ] \biggr\}\ , \delta(\omega-\omega')\,\delta({\bf k}-{\bf k}').\end{aligned}\ ] ] the commutators @xmath175 $ ] are obtained from eqs . ( [ a11 ] ) ( [ a13 ] ) , by making the replacements @xmath146 @xmath176 @xmath147 , @xmath177 @xmath176 @xmath178 , @xmath179 @xmath176 @xmath180 , and @xmath181 @xmath176 @xmath182 . finally , it can easily be proved that the intraplate amplitude operators ( [ 38 ] ) satisfy the commutation relations (; ) @xmath183 } \nonumber\\[1ex ] & & = c_{q\lambda\lambda'}^{(j)}({\bf k},\omega)\ , \delta_{qq'}\,\delta(\omega-\omega')\,\delta({\bf k}-{\bf k}'),\end{aligned}\ ] ] where @xmath184 & & \times\ ; \left ( e^{\pm 2\beta _ j '' d_j } - 1 \right ) { \bf e}^{(j)}_{q \pm } ( { \bf k } ) \cdot{\bf e}^{(j)*}_{q\pm } ( { \bf k } ) , \\[1ex ] \label{a16 } \lefteqn { c_{q\,{\rm \pm \mp } } ^{(j ) } = \pm i\frac{\pi \hbar}{\epsilon_0 } \left(\frac{\omega}{c}\right)^2 \frac{\beta_{j}''}{|\beta_{j}|^2 } } \nonumber\\[1ex ] & & \times\ ; \left ( e^{\mp 2i \beta _ j ' d_j } - 1 \right ) { \bf e}^{(j)}_{q \pm } ( { \bf k } ) \cdot{\bf e}^{(j)*}_{q\mp } ( { \bf k}).\end{aligned}\ ] ] using eqs . ( [ a14 ] ) ( [ a16 ] ) , we find that the operators @xmath185&&\times\ ; \left [ e^{i \beta_{j } d_{j } } \hat{\mathrm{e}}_{q+}^{(j ) } ( { \bf k } , \omega ) \pm \hat{\mathrm{e}}_{q-}^{(j ) } ( { \bf k } , \omega ) \right ] , \label{a17}\end{aligned}\ ] ] where @xmath186&&\,\times\ ; \left\ { \beta _ j ' \sinh(\beta_j ' ' d_j)\ , { \bf{e}}^{(j)}_{q+}({\bf k } ) \cdot { \bf{e}}^{(j)\,*}_{q+}({\bf k } ) \right . \nonumber\\[1ex]&&\hspace{2ex } \left . \pm\ , \beta _ j '' \sin(\beta_j ' d_j)\ , { \bf{e}}^{(j)}_{q+}({\bf k } ) \cdot { \bf{e}}^{(j)\,*}_{q-}({\bf k } ) \right\ } ^{1/2}\!\qquad\end{aligned}\ ] ] satisfy bosonic commutation relations . from eq . ( [ a17 ] ) it then follows that @xmath187 where @xmath188 \label{a21 } & & \tau_{q - \pm}^{(j ) } = \pm { \textstyle\frac { 1 } { 2 } } \xi^{(j)}_{q \pm}({\bf k } , \omega).\end{aligned}\ ] ] in the limiting case ( when the plate is surrounded by vacuum ) one has to distinguish between ( propagating - field components ) and ( evanescent - field components ) . in the first case we have , so that eqs . ( [ a8 ] ) ( [ a10 ] ) reduce to @xmath189 = 0,\ ] ] and eqs . ( [ a11 ] ) ( [ a13 ] ) simplify to @xmath190 } \nonumber\\[1ex ] & & = \,\frac{\pi \hbar}{\epsilon_0}\left(\frac{\omega}{c}\right)^2 \frac{1}{\beta_0}\,\delta_{qq'}\,\delta(\omega-\omega')\ , \delta({\bf k}-{\bf k}').\qquad \end{aligned}\ ] ] @xmath191 } \nonumber\\[1ex ] & & = \,\frac{\pi \hbar}{\epsilon_0}\left(\frac{\omega}{c}\right)^2 \frac{1}{\beta_n}\,\delta_{qq'}\ , \delta(\omega-\omega')\,\delta({\bf k}-{\bf k}').\qquad \end{aligned}\ ] ] in the second case we have . equations ( [ a8 ] ) ( [ a10 ] ) then lead to @xmath192 } \nonumber\\[1ex ] & & = \,\frac{\pi \hbar}{\epsilon_0 } \left(\frac{\omega}{c}\right)^2 \frac{2 t^{q\ , \prime \prime}_{0/n}}{{|\beta_0| } } \,\delta_{qq'}\ , \delta(\omega-\omega')\,\delta({\bf k}-{\bf k } ' ) , \qquad\end{aligned}\ ] ] and from eqs . ( [ a11 ] ) ( [ a13 ] ) we find @xmath193 } \nonumber\\[1ex ] & & = \,\frac{\pi \hbar}{\epsilon_0 } \left(\frac{\omega}{c}\right)^2 \frac{2 r_{0/n}^{q\,\prime \prime}}{{|\beta_0|}}\ , \delta_{qq'}\,\delta(\omega-\omega')\,\delta({\bf k}-{\bf k } ' ) , \qquad\end{aligned}\ ] ] and similarly @xmath194 } \nonumber\\[1ex ] & & = \,\frac{\pi \hbar}{\epsilon_0 } \left(\frac{\omega}{c}\right)^2 \frac{2 r_{n/0}^{q\,\prime \prime}}{{|\beta_n|}}\ , \delta_{qq'}\,\delta(\omega-\omega')\,\delta({\bf k}-{\bf k } ' ) . \qquad\end{aligned}\ ] ] note that the commutation relations ( 42 ) in ref . @xcite for the bosonic output operators are only valid for propagating waves , not for evanescent ones . it should also be noted that the free - space light modes @xmath195 used in ref . @xcite are introduced by explicitly requiring that , where @xmath196 corresponds to @xmath0 in the present paper . hence , these modes solely represent propagating waves .

neutrinos are produced in states of given flavors electron , muon or tau that are mixtures of mass eigenstates . as different masses propagate with different phases , the mixture at some distance from a neutrino source may be different from the initial one , resulting in a disappearance of the initial flavor and an appearance of other flavors in the neutrino beam . quantitatively , the phase of the state of mass @xmath0 and momentum @xmath1 at the time @xmath2 from the production is @xmath3 at the distance @xmath4 from the source and , therefore , the disappearance and appearance probabilities depend on the squared mass differences , @xmath5 , ratio @xmath6 , @xmath7 being the neutrino energy , and mixing angles , @xmath8 , describing how mass eigenstates mix to form states of definite flavor . this phenomenon , called the neutrino oscillations , has been well established in the past two decades and is now considered one of the major discoveries in particle physics , celebrated by two recent nobel prizes @xcite . the values of the oscillation parameters two squared mass differences and three mixing angles are being measured with an increasing precision , thanks to a global effort that produced a number of experiments currently taking data or planned in the near future . within the next two decades , the precision is expected to be sufficient to begin testing the unitarity of the mixing matrix @xcite , discover the neutrino - mass hierarchy @xcite and unambiguously determine the value of the dirac phase , @xmath9 , violating the charge - parity ( cp ) symmetry of neutrino mixing @xcite . these results are going to have profound consequences for possible extensions of the standard model , building models of the neutrino masses and our understanding of the matter - antimatter asymmetry in the universe . in this topical review , we discuss systematic uncertainties in ongoing and future long - baseline ( @xmath103001300 km ) neutrino - oscillation experiments using conventional beams @xcite . such neutrino beams are tertiary products , originating from the decay of mesons predominantly pions produced in interactions of the primary proton beam with a target . as the resulting neutrinos are not monoenergetic , to extract the oscillation parameters from the collected event distribution , their energies have to be reconstructed on an event - by - event basis from the measured kinematics of particles in the final state . depending on the beamline , the relevant neutrino energies extend from a few hundred mev to a few gev , at which the dominant interaction mechanisms change from quasielastic to resonant and nonresonant meson production @xcite . because the oscillation parameters are extracted from the energy dependence of event distributions , their accurate determination requires an accurate reconstruction of neutrino energy . while the reconstruction would involve smallest uncertainties for scattering off free protons and deuteron @xcite , owing to the low cross sections involved , it is necessary to employ nuclear targets as detector materials to ensure high statistics of collected events . as a consequence , however , description of nuclear effects turns out to be one of the largest sources of systematic uncertainties in the oscillation analysis of modern long - baseline experiments . additionally , when energy reconstruction requires the kinematics of produced hadrons to be measured , detector effects may play an important role and uncertainties of the detector response contribute to the systematic uncertainties of the extracted oscillation parameters @xcite . over the past decade , the extensive neutrino - scattering program has yielded a wealth of experimental cross sections for carbon or hydrocarbon , ch @xcite , and much fewer results for other targets , such as water@xcite , argon @xcite , iron @xcite and lead @xcite . it is important to note that there seem to remain puzzling tensions between different measurements such as @xcite or @xcite which have attracted a sizable attention of the theoretical community . a number of approaches developed to describe nuclear response to other probes has been extended to neutrino interactions , and the understanding of nuclear effects relevant to neutrino - oscillation experiments has clearly improved @xcite ; see also recent reviews @xcite . however , many problems still await a quantitative explanation . while for detector targets containing hydrogen the pion - production events on free protons can be separated out @xcite , current and future neutrino - oscillation experiments are going to collect data predominantly for interactions with the targets of atomic numbers @xmath11 ranging from 12 to 40 , for which accurate and complete nuclear models permitting fully trustable data analysis are not available yet . the main difficulty of developing theoretical models useful to oscillation experiments stems from the flux average over polychromatic beams @xcite . as any data bin may receive contributions from a range of neutrino energies and different interaction mechanisms , understanding the source of discrepancies between theoretical results and experimental data is a formidable task . for example , resonant pion production followed by pion absorption in medium leads to the same final state as single - nucleon knockout . at the kinematics of the long - baseline oscillation experiments , nuclear models must be able to describe relativistic products of interaction . additionally , experiments employing tracking detectors require predictions of exclusive cross sections for different hadronic final states . this topical review is organized as follows . after general remarks regarding systematic uncertainties of measurements of cp violation in , we present our procedure of the oscillation analysis in . in sections [ sec : nucleareffects ] and [ sec : detectoreffects ] we illustrate the importance of nuclear and detector effects for an accurate reconstruction of neutrino energy and an unbiased extraction of oscillation parameters , discussing selected results from our previous publications @xcite . finally in we summarize this review . in recent and ongoing oscillation experiments , it has been a common practice to use the spread between different theoretical descriptions of nuclear effects , implementations of the same model in monte carlo generators , detector response parametrizations etc . as an estimate of the associated systematic uncertainties . while this method seems very effective , a word of caution is in order . non - negligible correlations between the ingredients used to find the uncertainties such as common assumptions of nuclear models or fine - tuning of event generators to the same data may lead to sizable underestimates , making this procedure insufficient and unreliable . for the next 20 years , the long - baseline neutrino - oscillation program is most likely going to employ conventional beams @xcite . this technique of producing neutrino beams has been used for a few decades : an intense proton beam is impinged on a target typically made of beryllium or graphite in order to produce mesons , mainly pions . the target is embedded within a horn an electromagnet producing a toroidal magnetic field that focuses secondary particles of a selected charge and defocuses those of the opposite charge . the horn polarization determines whether the resulting beam will be made of neutrinos or antineutrinos . it is important to point out that the beams and their contaminations by wrong - sign mesons in the neutrino and antineutrino modes are not related in a meaningful way . therefore , while using the same beamline , target and horn , the neutrino or antineutrino runs of an experiment could be considered two separate experiments . an accurate determination of the neutrino flux its flavor composition and energy spectrum exclusively from the proton - beam parameters , meson - production data for thick targets and horn configuration is a challenging task . state - of - the - art methods of controlling systematic uncertainties of a neutrino beam have been developed for the minos @xcite and minerva @xcite experiments , largely thanks to using two horns of adjustable positions at the numi beamline @xcite . it is not clear whether such a feature difficult to realize for off - axis beams is going to be available for future oscillation experiments . therefore , the beam normalization uncertainty of about 5% achieved in the minos and minerva experiment , can be expected also in future oscillation experiments employing conventional beams . note , however , that recently proposed new concepts of neutrino beams , such as @xmath12storm @xcite , would allow the neutrino flux to be determined with an accuracy of the order of 1% . to extract the probability of oscillation between the neutrino flavors @xmath13 and @xmath14 , experiments collect event distributions @xmath15 with respect to a set of observables @xmath16 , such as the charged lepton s energy and cosine of its production angle , @xmath17 in the above equation , the normalization factor @xmath18 depends on the beam power , data collecting time , fiducial mass etc . the flux expected in the detector , @xmath19 , and the oscillation probability @xmath20 are both functions of the true neutrino energy @xmath7 . the differential cross section @xmath21 describes likelihood for a neutrino of the flavor @xmath14 and energy @xmath7 to produce an event of kinematics @xmath16 . the detection efficiency is denoted as @xmath22 . systematic uncertainties in neutrino - oscillation experiments can be effectively reduced by using an unoscillated event distribution measured in the near detector system to predict the distribution expected in the far detector . this method of exploiting the cancelation of correlated uncertainties giving the best results when the near and far detectors are functionally identical has been used with great success in the reactor experiments daya bay @xcite , reno @xcite and double chooz @xcite to measure the mixing angle @xmath23 with remarkably high precision . barring the differences in the backgrounds and in the geometric acceptances of the beam , the ratio of the energy - unfolded event distributions in a disappearance experiment , @xmath24 , has in this case the simple interpretation @xmath25 where the far - to - near flux ratio , @xmath26 , reduces to the ratio of the squared distances from the neutrino source , @xmath27 and the oscillation probability in the near detector is @xmath28 . unlike in the reactor experiments , in long - baseline studies , the relevant cross sections are currently known with an accuracy of 1020% and the procedure of energy unfolding in much more involved . additionally , in the near and far sites , different detectors are usually employed and the differences in backgrounds and beam acceptances are non - negligible . for a disappearance measurement , minos can serve as a benchmark of how well the near - to - far comparison does reduce systematic uncertainties @xcite . in @xmath29 appearance measurements , the situation is even more involved . the final and initial flavors of neutrino are different , which leads to a dependence of the near - to - far event distributions ratio on the cross sections ratio @xmath30 . as the beams are designed to minimize the @xmath31 contamination in order to reduce the background in the far detector , direct measurements of the @xmath31 cross sections suffer from the statistics lower typically by 23 orders of magnitude than in the @xmath32 case and from larger uncertainties coming from both the flux and detector response @xcite . for example , in the recent t2k analysis @xcite , the total systematic uncertainty is estimated to be 16.6% , with the flux and detector - response contributions of 12.9 and 8.4% , respectively . there are no obvious methods to reduce those uncertainties and measurements for electron antineutrinos are going to be even more challenging . however , it is important to note that all available data are consistent with lepton universality . to emphasize the importance of reduction of systematic uncertainties in appearance measurements , let us make a simple estimate of those needed for an experiment to probe cp violation at 3@xmath33 confidence level for 75% of the possible @xmath9 values , the goal set by the p5 advisory panel @xcite . we consider the cp asymmetry defined as @xcite @xmath34 the energy average @xmath35 is taken over the full width at half the first oscillation maximum . in vacuum @xmath36 is proportional to @xmath37 and , therefore , for @xmath9 values in the vicinity of 0 and @xmath38 , the uncertainties of @xmath36 are very similar to those of @xmath9 . baseline dependence of the cp asymmetry for different @xmath9 values.,scaledwidth=80.0% ] shows the cp asymmetry @xmath36 as a function of the baseline for the @xmath9 values of @xmath39 , 0 , and @xmath40 and the remaining oscillation parameters from a recent global fit @xcite . for the 1500-km baseline , matter effects introduce the asymmetry of @xmath1047% and @xmath9 can modify it by @xmath41 . for 75% of the @xmath9 values , the cp asymmetry can become as small as 5% , which at 3@xmath33 confidence level translates into @xmath101.5% accuracy . for equal contributions of statistical and systematic uncertainties , the latter are required not to exceed @xmath101% . fortunately , if the maximal cp violation is realized in nature , as suggested by recent results of the t2k @xcite and nova @xcite experiments , less stringent conditions are going to be sufficient for the @xmath9 discovery . discussing the role of nuclear and detector effects in the oscillation analysis , in sections [ sec : nucleareffects ] and [ sec : detectoreffects ] we present the results obtained using the software package globes @xcite . simulating data according to , we keep all the oscillation parameters fixed , setting their values to @xmath42 unless specified otherwise , we perform the @xmath43 disappearance analysis for a setup similar to that of the t2k experiment @xcite with a narrow - band off - axis beam @xcite , considering carbon as the detector material . the main motivation for selecting the @xmath44 channel is that it can effectively be treated as two - flavor oscillations with the probability @xcite @xmath45 where @xmath46 which allows for a clear interpretation of the obtained results . as detailed in , the experimental setup consists of the near and far detectors simulated following @xcite . the exposure and beam intensity are chosen to ensure that the statistics of unoscillated events is high enough to constrain the analyzed systematic uncertainty . our treatment of systematic uncertainties , the @xmath47 implementation and the method of determination of the confidence regions is the same as in @xcite . we assume 20% systematic uncertainties for the shape ( bin - to - bin uncorrelated ) and for the overall normalization ( bin - to - bin correlated ) . adding a gaussian prior for each systematic error , the pull method is used and the final @xmath47 profile is obtained after the minimization over the nuisance parameters . @lrr & baseline & fiducial mass + far detector & 295.0 km & 22.5 kt + near detector & 1.0 km & 1.0 kt + as neutral - current background is expected not to play an important role in the disappearance analysis , we neglect it unless specified otherwise and consider the following mechanisms of charged - current ( cc ) neutrino interaction : quasielastic ( qe ) involving a single nucleon or more nucleons ( @xmath48 ) in the final state , resonance excitation , and deep - inelastic scattering . in all cases , meson - production events are obtained using the monte carlo generator genie 2.8.0 @xcite . qe events are simulated using either the relativistic fermi gas ( rfg ) model from genie or the realistic spectral function ( sf ) approach @xcite implemented in the @xmath49 package of additional modules @xcite . for the @xmath48 contribution , we employ the empirical procedure developed by dytman @xcite or the effective approach of increasing the value of the axial mass @xmath50 in the qe cross sections . performing the kinematic energy reconstruction , we use the well - known formula @xcite @xmath51 setting the invariant hadronic mass @xmath52 to the nucleon mass @xmath53 for mesonless events and fixing the separation energy @xmath54 to 34 mev . the same value of @xmath54 is added for every detected nucleon in the calorimetric method , @xmath55 in which the neutrino energy is reconstructed summing the charged - lepton s energy @xmath56 , the kinetic energies of the knocked - out nucleons @xmath57 and the total energy of any other particle produced @xmath58 . in an idealized scenario in which nuclear and detector effects would not play a significant role , the energy reconstruction would be perfect . in a realistic case , however , there is a non - negligible probability of energy misreconstruction @xcite . in our calculations , both nuclear and detector effects are encoded in the migration matrices , @xmath59 , the elements of which describe the probability that an event of type @xmath60 with a true energy in the bin @xmath61 is reconstructed with an energy in the bin @xmath62 . the simulated ( true and fitted ) event distributions are then obtained as @xmath63 where @xmath60 runs over the four types of interactions considered , @xmath62 and @xmath61 label the energy bins and @xmath64 stands for the true number of @xmath60 events in the bin @xmath61 . as the oscillation parameters are extracted from the energy distribution of collected events , a precise reconstruction of neutrino energy is a prerequisite for a precise oscillation analysis . it is important to note that the kinematic reconstruction method , valid for particular final states , relies on a selection of event subsample and requires an accurate simulation of both the signal events and the background contribution @xcite . analyzing various aspects of qe interactions with one or more nucleons in the final state @xcite , in this section we discuss how uncertainties in the description of nuclear effects may affect the outcome of @xmath32 disappearance experiments employing the kinematic method of energy reconstruction . while modern neutrino - scattering measurements have been predominantly performed for the carbon target , the future oscillation experiments are going to require detailed knowledge of the cross sections for oxygen and argon . to demonstrate the subtleties of extrapolating the results between nuclei , we are going to neglect differences between the migration matrices for carbon ( @xmath65 ) and oxygen ( @xmath66 ) in the @xmath32 disappearance analysis @xcite . we consider a t2k - like experimental setup with water cherenkov detectors , and model nuclear effects relying exclusively on genie . in particular , in this analysis qe scattering is described within the rfg model of bodek and ritchie @xcite , with a high - momentum tail inspired by the effect of nucleon - nucleon correlations added to the nucleon momentum distribution given by the step function . the globes analysis for 5 years of collecting data with the 750 kw beam @xcite yields @xmath10880 cc qe events , @xmath10480 pionless events resulting from other mechanisms of interaction and 275 neutral - current background events . impact of the nuclear target dependence of the cross sections in genie on the oscillation analysis . left : event distributions for oxygen ( solid line ) and carbon ( dashed line ) expected for the flux @xcite . right : 1 , 2 and 3@xmath33 confidence regions in the @xmath67 plane obtained when the simulated data for oxygen are fitted using the migration matrices for carbon ( solid lines ) and oxygen ( shaded areas ) . the triangle and circle show the best fit point and true values of the oscillation parameters , respectively.,title="fig:",scaledwidth=39.0% ] impact of the nuclear target dependence of the cross sections in genie on the oscillation analysis . left : event distributions for oxygen ( solid line ) and carbon ( dashed line ) expected for the flux @xcite . right : 1 , 2 and 3@xmath33 confidence regions in the @xmath67 plane obtained when the simulated data for oxygen are fitted using the migration matrices for carbon ( solid lines ) and oxygen ( shaded areas ) . the triangle and circle show the best fit point and true values of the oscillation parameters , respectively.,title="fig:",scaledwidth=39.0% ] the @xmath68 distributions of events obtained using the oxygen and carbon migration matrices are shown in the left panel of . the carbon result turns out to be somewhat shifted toward lower energies with respect to the oxygen one , which translates into a noticeable effect for the oscillation analysis . the right panel of shows the 1 , 2 and 3@xmath33 confidence regions in the @xmath69 plane for the true event rates simulated using the oxygen migration matrices . the shaded areas represent the analysis in which nuclear effects are accurately taken into account by using the oxygen migration matrices to calculate the fitted rates . on the other hand , the solid lines correspond to the fitted rates for the carbon migration matrices and show the effect of an inaccurate description of nuclear effects . the extracted and true values of the oscillation parameters turn out to differ at a @xmath101@xmath33 confidence level . we observe a non - negligible difference between the energy reconstruction for the oxygen and carbon targets described by means of the rfg model in genie . should the differences in their shell structures and density distributions be accounted for within a more sophisticated model , the effect can be expected to be even larger . note also that carbon and oxygen are considered rather similar nuclei compared to argon . the result of shows the importance of developing an accurate theoretical description of nuclear effects for the targets employed in oscillation experiments and of their extensive testing against scattering data . it also suggests that employing the same target in the near and far detectors can be an effective way to reduce systematic uncertainties of an oscillation experiment . as neutrino scattering data are collected using polychromatic beams , every data bin receives contributions from a range of neutrino energies . moreover , the same detected final state may be produced by different interaction mechanisms . therefore , an accurate description of experimental cross sections is a challenging task , requiring a theoretical model to cover broad kinematic region and describe multiple scattering processes @xcite . with different probe s energies and reaction mechanisms intertwined , it may be very difficult to identify the origin of discrepancies between theoretical and experimental results for neutrinos . on the other hand , when the interaction process involves only one nucleon with the remaining @xmath70 ones acting as spectators neutrino and electron scattering are subject to the same nuclear effects . the nuclear cross section in this regime is a convolution of the elementary cross section ( different for electrons and neutrinos ) with the hole and particle spectral functions ( common for electrons and neutrinos ) , describing the ground - state properties of the target and propagation of the struck nucleon , respectively @xcite . as a consequence , models of _ nuclear effects _ in neutrino interactions can be validated by systematic comparisons to electron scattering data at the kinematics of interest . we want to emphasize that it is highly improbable that a theoretical approach unable to reproduce electron scattering data would be able to describe nuclear effects in neutrino interactions . electron scattering cross sections double differential and for monoenergetic beams allow for much easier understanding of discrepancies between theoretical results and experimental data than neutrino ones . importance of an accurate description of nuclear effects on the example of the double differential c@xmath71 cross sections . left : comparison of the rfg ( dashed lines ) and sf @xcite ( solid lines ) calculations with the data for scattering angle @xmath72 and beam energies 1.35 , 1.50 and 1.65 gev @xcite . right : comparison of the sf results obtained using genie @xcite and genie@xmath73 t @xcite with the experimental data for scattering angle @xmath74 and beam energy 961 mev @xcite.,title="fig : " ] importance of an accurate description of nuclear effects on the example of the double differential c@xmath71 cross sections . left : comparison of the rfg ( dashed lines ) and sf @xcite ( solid lines ) calculations with the data for scattering angle @xmath72 and beam energies 1.35 , 1.50 and 1.65 gev @xcite . right : comparison of the sf results obtained using genie @xcite and genie@xmath73 t @xcite with the experimental data for scattering angle @xmath74 and beam energy 961 mev @xcite.,title="fig : " ] as an illustrative example , in the left panel of we compare the qe predictions of the sf calculations @xcite and the rfg model of smith and moniz @xcite with the data @xcite . whereas the rfg results can not reproduce heights , shapes and positions of the qe peaks , the sf calculations turn out to be in very good agreement with the experimental points , although they do not involve any adjustable parameters . using the realistic ground - state description @xcite , the sf calculations take into account both the shell structure and correlations between nucleons . the latter cause partial depletion of the shells lowering the qe peaks and give rise to deeply bound nucleons of high momenta , producing the tails at low values of the final electron energy @xmath75 and rendering the qe peaks asymmetric . in addition , the sf calculations @xcite include the effect of final - state interactions , essential to reproduce the positions of the qe peaks accurately . all these effects are neglected in the rfg model , which treat the nucleus as a fragment of noninteracting nuclear matter of uniform density in a constant potential . the sf approach is available in genie 2.8.0 , however , comparisons with electron scattering data have revealed some issues with the energy and momentum conservations in its implementation that affect both the position and width of the qe peak , as shown in the right panel of for the experimental cross sections @xcite . on the other hand , the sf implementation in the @xmath49 package @xcite of additional modules to genie is in good agreement with electron scattering data and , therefore , from now on we are going to rely on it to generate qe events . to explore how the difference between the rfg model and the sf approach in the qe interaction channel influences the oscillation analysis @xcite , let us consider a @xmath32 disappearance experiment similar to t2k , employing carbon as the target material . assuming 5 years of data taking at the beam power of 750 kw in the globes analysis leads to @xmath10650 ( @xmath10730 ) cc qe events in the sf approach ( rfg model ) , together with @xmath10410 cc events of qe topology coming from other interaction mechanisms and @xmath10250 neutral - current background events . the difference between the cc qe event numbers in the two considered approaches , resulting from the difference between the corresponding cross sections presented in the left panel of , contributes to a non - negligible effect for the oscillation analysis . impact of differences between the quasielastic @xmath76 cross sections in the sf approach and the rfg model on the oscillation analysis . left : mesonless cross sections employed in the analysis . right : confidence regions in the @xmath67 plane obtained when the data simulated within the sf approach are fitted using the migration matrices from the rfg ( solid lines ) and sf ( shaded areas ) calculations . ] the right panel of shows the 1 , 2 and 3@xmath33 confidence regions in the @xmath69 plane for the true event rates from the sf approach . the shaded areas are obtained for the fitted rates from the same approach , while the solid lines correspond to those from the rfg model . we observe a shift between the true values of the oscillation parameters represented by the dot and the extracted best - fit point marked by the triangle at a @xmath101@xmath33 confidence level , both in the mixing angle and the squared mass difference . as we analyze the difference between the rfg model and the sf approach only for the cc qe predictions , the presented result should be regarded as a lower limit of the influence of nuclear - model uncertainties . for resonance excitation , similar differences can be expected based on the available theoretical results @xcite . in the deep - inelastic region , the current modeling of nuclear effects has been shown inaccurate by the recent results from the minerva experiment @xcite . from now on , we rely on the sf approach to simulate cc qe events and employ a generalization of the standard formula for the kinematic energy reconstruction @xcite : in the case of mesonless events , we employ the invariant hadronic mass @xmath77 , regardless of the number of knocked - out nucleons , but when at least one meson is detected in the final state , we set @xmath52 to the @xmath78 resonance mass @xmath79 gev . to discuss the effect of uncertainties of the @xmath48 cross sections on the oscillation analysis , we consider a t2k - like disappearance experiment running in the neutrino and antineutrino mode with the same flux @xcite and compare the results obtained using two data - driven phenomenological methods to account for @xmath48 processes @xcite . in the genie@xmath73 t approach , the @xmath48 estimate from genie 2.8.0 @xcite is added to the qe contributions obtained using the @xmath49 package @xcite . in the effective approach , the modifications of the qe cross sections due to @xmath48 mechanisms are accounted for by applying in the sf calculations an effective value of the axial mass @xmath80 gev , as suggested by a number of experimental results for nuclear targets ranging from carbon to iron @xcite . the rationale for considering these two methods is that the spread of the cross sections they predict is a good representation of the spread between the available experimental data . shows that the effective calculations are in good agreement with the cc qe results from nomad @xcite and minerva @xcite , for both neutrinos and antineutrinos . they also reproduce the energy dependence of the neutrino and antineutrino data from miniboone @xcite , but not their absolute normalization . to better illustrate this feature , we also present the miniboone cross sections divided by a factor of 1.2 , consistent with the ratio of the detected to predicted events @xmath81 in the first miniboone analysis @xcite . the genie@xmath73 t approach is in very good agreement with the miniboone data for neutrinos , which were used to determine the strength of the @xmath48 contribution in genie @xcite , but this is not the case for antineutrinos . owing to their large uncertainties , the t2k data @xcite can not discriminate between the two calculations . comparison of the total cc inclusive @xmath32 cross sections in the left panel of shows that the two considered methods yield the results in good agreement with the nomad data @xcite , collected in the region dominated by deep inelastic scattering . while the flux - averaged on - axis measurement from t2k @xcite does not distinguish the two approaches , the off - axis one @xcite shows a distinct preference for the effective calculations . on the other hand , the sciboone point @xcite clearly favors the genie@xmath73 t calculations . note that the t2k @xcite and sciboone @xcite measurements do not involve energy unfolding or selection of particular final states and , therefore , they can be expected to involve the lowest uncertainties . the puzzling difference between the t2k and sciboone data has important consequences for @xmath32 disappearance studies , as illustrated in the right panel of by the confidence regions obtained for the true event rates from the genie@xmath73 t calculations and the expected number of @xmath106000 unoscillated events . the shaded areas represent the results for the genie@xmath73 t fitted rates and the solid lines correspond to the fitted rates from the effective calculations . the high value of @xmath47 per degree of freedom in the best - fit point , 66.0/26 , clearly indicates that the differences between the two considered approaches are too large to be neglected in a precise oscillation analysis . same as figure [ fig:2p2h_nu ] but for muon antineutrinos.,title="fig:",scaledwidth=53.0% ] same as figure [ fig:2p2h_nu ] but for muon antineutrinos.,title="fig:",scaledwidth=39.0% ] for completeness , in we present the results for muon antineutrinos . owing to the large differences between the @xmath82 cc qe cross sections , the two considered approaches yield very different predictions for the inclusive cc @xmath82 cross section , as shown in the left panel of . however , there are no experimental data available to discriminate between them . the observed differences translate into a large effect for the oscillation analysis . using the fitted rates from the effective calculations and the true event rates from genie@xmath73 t leads to a severe discrepancy between the extracted and true values of the oscillation parameters , see the right panel of . because treating the normalization of the qe event sample with any number of nucleons as arbitrary yields very similar results , the observed behavior can be traced back to the different shapes the two approaches predict for the migration matrices and for the energy dependence of the cc qe cross section . in final remarks of this section we note that the significance of nuclear effects , analyzed here for the carbon nucleus ( @xmath65 ) , can be expected to increase further for the argon target @xmath83 . although the considered experimental setup and our procedure of data analysis do not closely follow those of any existing experiment , the discussed results point toward the crucial role that an accurate description of nuclear effects plays in monte carlo simulations of precise oscillation experiments . for the success of precise oscillation studies it is , therefore , of great importance to continue efforts aiming to provide new cross - section measurements , improve existing monte carlo generators , develop more accurate nuclear models and determine their uncertainties in comparisons with available data . in the oscillation analysis of long - baseline experiments , neutrino energy is inferred from the kinematics of particles produced in the interaction . as a consequence , the reconstructed energy may be affected by finite detection capabilities energy resolutions , efficiencies and thresholds and uncertainties they are known with . we explore this issue within a realistic scenario , in which particles are detected according to their efficiencies and thresholds , and measured energies and angles are smeared by finite detector resolutions . to analyze how uncertainties of these detector effects may affect the oscillation analysis , we obtain simulated event distributions in the far detector within the realistic scenario and analyze them partly neglecting the detector effects implemented in the migration matrices . accounting for the effect of finite detector resolution , we smear observables according to the normal distributions centered at their true values . for muons , this procedure is applied both to the momentum and production angle , using the realistic parameters @xcite @xmath84 the energy resolutions of for @xmath85 s producing electromagnetic showers and other hadrons are set to @xmath86 respectively . their values , as well as those of the energies appearing in the above equation , are expressed in units of gev . in our calculations , the detection thresholds correspond to the measured kinetic energy of 20 mev for mesons and 40 mev for protons . for the sake of simplicity , the efficiencies are treated as energy - independent and set to 60% for @xmath85 s , 80% for other mesons , and 50% for protons . owing to difficulties of their reconstruction in neutrino events , we assume that neutrons always escape detection . in the context of the @xmath32 disappearance analysis @xcite , we analyze the t2k - like setup described in , following the same treatment of systematic uncertainties and applying the same method of @xmath47 calculations . our calculations are performed for the expected number of unoscillated events @xmath104900 . instead of varying individual parameters involved in the detector description , we vary the migration matrices making a linear combinations of the those calculated with and without detector effects . in this manner , we estimate the general sensitivity of the oscillation analysis to the detector performance , which can be expected to characterize a broad class of experiments rather than a particular ( highly idealized ) setup . impact of detector - related uncertainties on the oscillation analysis employing the kinematic energy reconstruction . left : comparison of the event distributions for the detector performance overestimated by 10% ( dashed line ) and 30% ( solid line ) with an accurate estimate of detector effects ( shaded histogram ) . right : @xmath87 confidence regions in the @xmath67 plane obtained when the data simulated with detector effects are fitted using the migration matrices neglecting them at the 10% , 20% , 30% level ( lines ) and accounting for them accurately ( shaded area ) . , scaledwidth=39.0% ] as shown in , the kinematic method of energy reconstruction based only on the muon s energy and its production angle turns out to be largely insensitive to detector effects and even the uncertainties at the level of 30% do not significantly affect the outcome of the oscillation analysis . it is a consequence of very precise reconstruction of muon kinematics achieved in modern experiments and assumed in this analysis , see . same as figure [ fig : deteff_kin ] but for the calorimetric energy reconstruction.,scaledwidth=39.0% ] on the other hand , we observe that detector effects play an important role in the calorimetric reconstruction method , in which the neutrino energy is inferred from the energies deposited in detector by all interaction products . shows that in order to avoid a significant bias in the extracted values of the oscillation parameters , the detector response has to be determined with an accuracy of at least 10% . this behavior can be traced back to much larger uncertainties in the hadron - energy determination , see , entering the calorimetric method of neutrino energy reconstruction . while the results of figures [ fig : deteff_kin ] and [ fig : deteff_cal ] are obtained for the narrow beam peaked at @xmath100.6 gev @xcite , the conclusions on influence of detector effects seem to be valid in more general case , as suggested by the findings for a wide - band beam with the peak at @xmath101.6 gev and a non - negligible contribution of energies above 3 gev @xcite . owing to the very complicated oscillation probability , an accurate neutrino - energy reconstruction is of particular importance for appearance experiments , observing electron ( anti)neutrinos from oscillations of muon ( anti)neutrinos . in this context , we analyze the role of missing energy @xcite carried away from event by undetected particles on the @xmath9 sensitivity for an experiment similar to dune @xcite . in the considered setup , a wide - band neutrino beam is produced in interactions of the initial 1.08-mw proton beam with the target material and aimed at the far detector of fiducial mass 40 kton , located 1300 km from the source . we assume 6 years of collecting data , 3 in @xmath12 mode and 3 in @xmath88 mode . for the signal , we consider 2% uncertainties of normalization ( bin - to - bin correlated ) and shape ( bin - to - bin uncorrelated ) . for the background , only a global normalization uncertainty of 5% is taken into account @xcite . the simulated event distributions are obtained accounting for all detector effects resolutions , efficiencies , and thresholds . on the other hand , extracting the oscillation parameters , we construct the migration matrices as linear combinations of those calculated with and without the shift resulting from the missing energy . in this way , for the purpose of this analysis , we single out the role of missing energy from other detector effects . impact of an underestimation of the missing energy on the oscillation analysis employing the calorimetric energy reconstruction . left : reconstructed energy distributions for deep - inelastic scattering of electron neutrino at true energy 2.95 gev , obtained accounting for 0% , 50% and 100% of the shift due to the missing energy . right : @xmath87 confidence regions in @xmath89 plane obtained when the simulated data are fitted using the migration matrices accounting for 90% , 80% and 70% ( lines ) of the missing energy and when all the missing energy is accounted for ( shaded area ) . the dot shows the assumed true values of the oscillation parameters.,scaledwidth=39.0% ] its influence on the reconstructed - energy distribution is illustrated in the left panel of for the example of deep - inelastic scattering of electron neutrino . it clearly appears that owing to undetected particles in the final state , the maximum of the distribution is shifted to an energy lower than the true neutrino energy . as the composition of particles in the final state differs for neutrinos and antineutrinos and depends on the interaction channel and the probe s energy , this is also the case for the energy carried away by undetected particles . this can be expected to have important consequences for the oscillation analysis . as presented in the right panel of , even a 20% underestimation of the missing energy introduces a sizable bias in the extracted @xmath9 value . should the missing energy be underestimated by 30% , the analysis would exclude the true value of @xmath9 at a confidence level between 2 and 3@xmath33 . this result illustrates the importance of an accurate determination of detector response in test - beam exposures and the relevance of a realistic simulation of nuclear effects in neutrino interactions , including intranuclear cascade . we would like to emphasize that while our results are meant to point out the importance of various aspects of detector effects , much more detailed studies are necessary to draw truly quantitative conclusions for specific experiments . in particular , our treatment of the missing - energy uncertainty assumed to be equal for neutrinos and antineutrinos and independent of the energy and interaction channel may be regarded as simplistic . however , as more realistic sensitivity estimates would require an accurate knowledge of the detector response and inclusion of nuclear - model uncertainties , out of necessity , we leave them for future investigations within experimental collaborations . thanks to tremendous progress in experimental neutrino physics over the last two decades , systematic uncertainties in oscillation studies have been greatly reduced . as a consequence , however , those arising from description of nuclear effects and related to determination of detector response have become very important in current appearance measurements . this is going to be even more so in future experiments aiming to unambiguously discover the lepton - sector contribution to violation of charge - particle symmetry . in this topical review , we have discussed a few examples illustrating the relevance of an accurate description of quasielastic interactions with any number of knocked - out nucleons and argued that for pion production a similar effect is expected . we pointed out that comparisons to electron scattering data provide opportunity to test nuclear models and estimate their uncertainties ; they also allow for much clearer interpretation of discrepancies than neutrino data . discussing detector effects , we have drawn attention to their larger importance for experiments employing the calorimetric method of energy reconstruction and emphasized the key role of an accurate estimate of the missing energy for precise oscillation studies . as a final remark we reiterate that future experiments are going to require coordinated efforts toward improving theoretical descriptions of nuclear effects , implementing them in monte carlo generators and determining detector response in extensive test - beam exposures . however , the discoveries coming in the next two decades have the potential to overshadow those of the past two decades . this work is supported by the national science foundation under grant phy-1352106 . 999 kajita t 2016 030501 mcdonald a b 2016 030502 kim s - b 2015 93 an f ( juno collaboration ) 2016 030401 patterson r b 2015 _ ann . nucl . part . sci . _ * 65 * 177 acciarri r ( dune collaboration ) 2015 long - baseline neutrino facility ( lbnf ) and deep underground neutrino experiment ( dune ) . conceptual design report . volume 2 : the physics program for dune at lbnf ( arxiv:1512.06148 ) abe k ( hyper - kamiokande proto - collaboration ) 2015 _ prog . phys . _ * 2015 * 053c02 abe k ( t2k collaboration ) 2011 a * 659 * 106 ayres d s ( nova collaboration ) 2007 the nova technical design report , fermilab - design-2007 - 01 acciarri r ( dune collaboration ) 2016 long - baseline neutrino facility ( lbnf ) and deep underground neutrino experiment ( dune ) . conceptual design report . volume 1 : the lbnf and dune projects ( arxiv:1601.05471 ) formaggio j a and zeller g p 2012 1307 moreno o , donnelly t w , van orden j w and ford w p 2015 d * 92 * 053006 ankowski a m , benhar o , coloma p , huber p , jen c - m , mariani c , meloni d , and vagnoni e 2015 d * 92 * 073014 ankowski a m , coloma p , huber p , mariani c and vagnoni e 2015 d * 92 * 091301(r ) wu q ( nomad collaboration ) 2008 b * 660 * 19 lyubushkin v v ( nomad collaboration ) 2009 _ eur . phys . j. _ c * 63 * 355 aguilar - arvalo a a ( miniboone collaboration ) 2010 d * 81 * 092005 aguilar - arvalo a a ( miniboone collaboration ) 2010 d * 82 * 092005 nakajima y ( sciboone collaboration ) 2011 d * 83 * 012005 aguilar - arvalo a a ( miniboone collaboration ) 2011 d * 83 * 052007 aguilar - arvalo a a ( miniboone collaboration ) 2011 d * 83 * 052009 fields l ( minerva collaboration ) 2013 022501 fiorentini g a ( minerva collaboration ) 2013 022502 abe k ( t2k collaboration ) 2013 d * 87 * 092003 aguilar - arvalo a a ( miniboone collaboration ) 2013 d * 88 * 032001 abe k ( t2k collaboration ) 2014 d * 90 * 052010 tice b g ( minerva collaboration ) 2014 231801 higuera a ( minerva collaboration ) 2014 261802 aguilar - arvalo a a ( miniboone collaboration ) 2015 d * 91 * 012004 walton t ( minerva collaboration ) 2015 d * 91 * 071301(r ) abe k ( t2k collaboration ) 2015 d * 91 * 112002 eberly b ( minerva collaboration ) 2015 d * 92 * 092008 le t ( minerva collaboration ) 2015 b * 749 * 130 abe k ( t2k collaboration ) 2015 d * 92 * 112003 rodrigues p a ( minerva collaboration ) 2016 071802 mousseau j ( minerva collaboration ) 2016 d * 93 * 071101 abe k ( t2k collaboration ) 2016 d * 93 * , 112012 marshall c m ( minerva collaboration ) 2016 d * 94 * 012002 abe k ( t2k collaboration ) 2014 d * 90 * 072012 abe k ( t2k collaboration ) 2016 first measurement of the muon neutrino charged current single pion production cross section on water with the t2k near detector ( arxiv:1605.07964 ) anderson c ( argoneut collaboration ) 2012 161802 acciarri r ( argoneut collaboration ) 2014 d * 89 * 112003 acciarri r ( argoneut collaboration ) 2014 d * 90 * 012008 acciarri r ( argoneut collaboration ) 2014 261801 tzanov m ( nutev collaboration ) 2006 d * 74 * 012008 adamson p ( minos collaboration ) 2010 d * 81 * 072002 abe k ( t2k collaboration ) 2016 d * 93 * 072002 adamson p ( minos collaboration ) 2016 measurement of single @xmath85 production by coherent neutral - current @xmath12 fe interactions in the minos near detector ( arxiv:1608.05702 ) benhar o and meloni d 2009 d * 80 * 073003 leitner t and mosel u 2010 c * 81 * 064614 martini m , ericson m , and chanfray g 2012 d * 85 * 093012 nieves j , sanchez f , ruiz simo i , and vicente vacas m j 2012 d * 85 * 113008 meloni d and martini m 2012 b * 716 * 186 lalakulich o , and mosel u , and gallmeister k 2012 c * 86 * 054606 erratum : 2014 c * 90 * 029902(e ) martini m , ericson m , and chanfray g 2012 d * 87 * 013009 coloma p and huber p 2013 221802 coloma p , huber p , jen c - m and mariani c 2014 d * 89 * 073015 mosel u , and lalakulich o , and gallmeister k 2014 151802 jen c - m , ankowski a m , benhar o , furmanski a p , kalousis l n , and mariani c 2014 d * 90 * 093004 ankowski a m , benhar o and sakuda m 2015 d * 91 * 033005 ericson m , garzelli m v , giunti c and martini m 2016 d * 93 * 073008 ankowski a m , benhar o , mariani c and vagnoni e 2016 d * 93 * 113004 alvarez - ruso l , hayato y and nieves j 2014 075015 benhar o , huber p , mariani c and meloni d 2015 neutrino - nucleus interactions and the determination of oscillation parameters ( arxiv:1501.06448 ) lu x - g , coplowe d , shah r , barr g , wark d , and weber a 2015 d * 92 * 051302 benhar o , coletti p and meloni d 2010 105 132301 michael d g ( minos collaboration ) 2008 a * 596 * 190 aliaga l ( minerva collaboration ) 2014 a * 743 * 130 abramov a g , galyaev n a , garkusha v i , hylen j , novoskoltsev f n , ryabov a d and zarucheisky v g 2002 a * 485 * 209 adey d ( @xmath12storm collaboration ) 2013 @xmath12storm neutrinos from stored muons : proposal to the fermilab pac , fermilab - proposal-1028 ( arxiv:1308.6822 ) an f p ( daya bay collaboration ) 2015 111802 choi j h ( reno collaboration ) 2016 211801 abe y ( doublechooz collaboration ) 2014 jhep*10*(2014)086 michael d g ( minos collaboration ) 2006 191801 abe k ( t2k collaboration ) 2014 d * 89 * 092003 abe k ( t2k collaboration ) 2014 241803 abe k ( t2k collaboration ) 2015 d * 91 * 112010 wolcott j ( minerva collaboration ) 2016 081802 ritz s ( hepap subcommittee ) 2014 building for discovery : strategic plan for u.s . particle physics in the global context ( http://science.energy.gov/~/media/hep/hepap/pdf/may-2014/final_p5_report_interactive_060214.pdf ) gonzalez - garcia m c , maltoni m , salvado j and schwetz t 2012 jhep*12*(2012)123 abe k ( t2k collaboration ) 2016 181801 adamson p ( nova collaboration ) 2016 d * 93 * 051104 huber p , lindner m and winter w 2005 195 huber p , kopp j , lindner m , rolinec m and winter w 2007 432 coloma p , huber p , kopp j and winter w 2013 d * 87 * 033004 huber p , lindner m , schwetz t and winter w 2009 jhep*11*(2009)044 nunokawa h , parke s j and zukanovich funchal r 2005 d * 72 * 013009 andreopoulos c 2010 a * 614 * 87 benhar o , fabrocini a , fantoni s and sick i 1994 a * 579 * 493 katori t 2015 _ aip conf . * 1663 * 030001 ahn m h ( k2k collaboration ) 2003 041801 bodek a and ritchie j l 1981 d * 23 * 1070 benhar o , day d and sick i 2008 80 189 smith r a and moniz e j 1972 b * 43 * 605 erratum : 1975 b * 101 * 547 baran d t 1988 400 sealock r m 1989 1350 benhar o and meloni d 2007 a * 789 * 379 gran r ( k2k collaboration ) 2006 d * 74 * 052002 aguilar - arvalo a a ( miniboone collaboration ) 2008 032301 mariani c ( k2k collaboration ) 2008 _ aip conf . * 981 * 247 adamson p ( minos collaboration ) 2015 d * 91 * 012005

the resonant excitation by coherent optical radiation of an electronic transition in a semiconductor , e.g. a valence- to conduction - band excitation , creates a quantum - mechanical coherent superposition of the initial and final states of the transition , called optical polarization . the non - linear optical properties of this coherent superposition , together with its time evolution , can be used to provide a sensitive measurement of many fundamental parameters of the semiconductor material @xcite including elastic and inelastic scattering , energy level splittings between nearly degenerate states , energy relaxation , as well as associated information such as land g - factors and band - mixing . in general , non - linear laser spectroscopy is a very established and well understood means to probe these phenomena @xcite . however , the description of the non - linear optical response of a semiconductor crystal can be considerably more complex than for simple isolated and non - interacting atoms @xcite . this is particularly true for the case of the ultrafast optical spectroscopy used for the study of the sub - picosecond dynamics of photoexcited carriers in bulk systems as well as in semiconductor heterostructures @xcite . the life - time of the coherent quantum - mechanical superposition generated by an ultrafast laser excitation , called `` dephasing time '' , determines the typical time - scale on which coherent phenomena can be observed . such dephasing time reflects the role played by the various `` incoherent '' , i.e. phase - breaking , mechanisms in destroying the phase coherence induced by the laser photoexcitation . since semiconductors are characterized by very short electron - hole dephasing times , of the order of few hundreds of femtoseconds @xcite , coherent phenomena manifest themselves only through ultrafast optical experiments with sub - picosecond time resolutions @xcite . on this time - scale , the ultrafast evolution of photoexcited electron - hole pairs will reflect the strong coupling between coherent and incoherent dynamics , thus providing invaluable information on the non - equilibrium relaxation processes occurring in the semiconductor , e.g. carrier - carrier and carrier - phonon scattering . the aim of this paper is to review the basic aspects related to coherent phenomena in semiconductors . in particular , we will focus on the ultrafast coherent vs. incoherent carrier dynamics as well as on coulomb - correlation effects in photoexcited semiconductors . the paper is organized as follows . in the remainder of this section , after a brief historical account of coherent experiments in solids , we will try to gain more insight into the concept of coherence by introducing a simplified description of the light - matter interaction in terms of a two - level model . in section [ s.tb ] we will discuss the theoretical approach commonly used for a realistic description of both coherent and incoherent phenomena in various semiconductor structures , e.g. bulk systems , semiconductor superlattices , quantum wells and wires . section [ s.bulk ] deals with ultrafast carrier photoexcitation and relaxation in bulk semiconductors ; in particular , we will discuss the dominant role played by coherence on the carrier photogeneration process . in section [ s.sl ] we will review and discuss typical field - induced phenomena in superlattices , namely bloch oscillations and wannier - stark localization , as well as their dephasing dynamics . section [ s.qwr ] is devoted to the analysis of the coherent optical response of quasi - one - dimensional systems ; more specifically , we will discuss the strong modifications induced by coulomb correlation on the linear and non - linear optical spectra of realistic quantum - wire structures . finally , in section [ s.suco ] we will summarize and draw some conclusions . coherent phenomena in atomic and molecular systems have been investigated for a long time @xcite . the first spin echo experiment @xcite was performed in 1950 on protons in a water solution of fe@xmath0 ions . pulses in the radio frequency range were generated by means of a gated oscillator with pulse widths between 20 @xmath1s and a few milliseconds . with these pulses dephasing times of the order of 10 ms have been measured . in the 1960s echo experiments were brought into the visible range @xcite . a q - switched ruby laser produced pulses of approximately 10 ns duration which were used to observe photon echoes from ruby . in this case the dephasing times were of the order of 100 ns . as already pointed out , for the observation of any coherent dynamics the pulse width has to be shorter than the typical dephasing time . since in semiconductors electron - hole dephasing times are much shorter they are in the range of a few picoseconds down to some femtoseconds , coherent experiments in semiconductors had to wait until the development of suitable lasers able to generate sub - picosecond pulses . the physical phenomena governing the ultrafast carrier dynamics in photoexcited semiconductors can be divided into two classes : _ coherent phenomena _ , i.e. phenomena related to the quantum - mechanical phase coherence induced by the laser photoexcitation , and _ incoherent phenomena _ , i.e. phenomena induced by the various phase - breaking scattering mechanisms . the above classification in terms of coherent and incoherent phenomena is not purely academic ; it corresponds to rather different experimental techniques for the investigation of these two different regimes . from an historical point of view , the optical spectroscopy in semiconductors started with the analysis of relatively slow incoherent phenomena ( compared to the electron - hole dephasing time - scale ) . the investigation of nonequilibrium carriers started with the analysis of the incoherent energy - relaxation processes in the late 1960s using cw - lasers @xcite . in the 1970s pulse sources for the study of photoexcited carriers became available @xcite and this initiated time - resolved studies of the energy - relaxation process . many experiments based on different techniques have been performed @xcite : band - to - band luminescence @xcite which monitors the product of electron and hole distribution functions , band - to - acceptor luminescence @xcite which provides information on the electron distribution functions only , and pump - and - probe measurements @xcite where the measured differential transmission is proportional to the sum of electron and hole distribution functions . the theoretical analysis of these relaxation phenomena is commonly based on the semiclassical boltzmann theory . the boltzmann equations for both electron and hole distribution functions are commonly solved by means of semiclassical monte carlo simulations @xcite . in addition to the analysis of incoherent energy - relaxation processes , the ultrafast optical spectroscopy has allowed the investigation of coherent phenomena . as already pointed out , a coherent laser field creates , in addition to a non - equilibrium carrier distribution , a coherent polarization . the investigation of the coherent dynamics in semiconductors started in the 1980s . different aspects have been investigated @xcite : the optical stark effect @xcite , the dephasing of free carriers @xcite and excitons @xcite , quantum beats related to various types of level splittings @xcite , charge oscillations in double - quantum - well systems @xcite and superlattices @xcite , many - particle effects @xcite , and the emission of coherent thz radiation @xcite . these experiments can not be analyzed within the framework of the boltzmann equation . the reason is that this coherent polarization reflects a well - defined phase relation between electrons and holes , which is neglected within the semiclassical boltzmann theory . any proper description requires a quantum - mechanical treatment where , in addition to the distribution functions of electrons and holes , also the interband polarization is taken into account as an independent variable . several approaches have been used in the literature @xcite : bogoliubov transformations @xcite , nonequilibrium green s functions @xcite , band - edge equations based on the real - space density matrix @xcite , and the density matrix formalism in momentum space @xcite . during the last decade the time resolution has been further improved down to few tens of femtoseconds @xcite . on such extremely short time - scale coherent effects can no longer be neglected and are found to play a dominant role also for the case of typical incoherent measurements such as time - resolved and time - integrated luminescence . in such conditions , the carrier dynamics is the result of a strong interplay between coherent and incoherent phenomena . therefore , the traditional separation between coherent and incoherent approaches for the theoretical investigation of photoexcited semiconductors is no longer valid . what is needed is a comprehensive theoretical framework able to describe on the same kinetic level both classes of phenomena as well as their mutual coupling . to this purpose , a generalized monte carlo method for the analysis of both coherent and incoherent phenomena has been recently proposed @xcite . the spirit of the method is to combine the advantages of the conventional monte carlo approach @xcite in treating the incoherent , i.e. phase - breaking , dynamics with the strength of a quantum - kinetic approach in describing coherent phenomena @xcite . in particular , the coherent contributions are evaluated by means of a direct numerical integration while the incoherent ones are `` sampled '' by means of a conventional monte carlo simulation in the three - dimensional @xmath2-space . this theoretical approach has ben applied successfully to the analysis of various ultrafast optical experiments , e.g. four - wave - mixing studies of many - body effects in bulk gaas @xcite and luminescence studies of the hot - carrier photogeneration process @xcite . in order to clarify the concept of phase coherence , let us consider the simplest physical model for the description of light - matter interaction , i.e. an optically driven two - level system @xcite . within a two - level picture , a ground state @xmath3 with energy @xmath4 and an excited state @xmath5 with energy @xmath6 are mutually coupled by a driving force , e.g. an external field , and/or by their mutual interaction , e.g. coulomb correlation . the two - level - system hamiltonian @xmath7 is the sum of a free - level contribution @xmath8 and of a coupling term @xmath9 here , the usual second - quantization picture in terms of creation ( @xmath10 ) and destruction ( @xmath11 ) operators has been introduced @xcite . the two terms forming the coupling hamiltonian @xmath12 will induce transitions from state @xmath3 to @xmath5 and vice versa according to the coupling constant @xmath13 . as a starting point , let us consider a single - electron system . in the absence of interlevel coupling ( @xmath14 ) , we have two stationary states , @xmath15 ( @xmath16 being the vacuum state ) , corresponding to a single electron in level @xmath3 or @xmath5 , respectively . on the contrary , in the presence of interlevel coupling , the state of the system is , in general , a linear superposition of the non - interacting states in equation ( [ eq1.4 ] ) : @xmath17 whose coefficients obey the following equations of motion : @xmath18 again , we see that in the absence of interlevel coupling ( @xmath19 ) there is no time variation of the coefficients , i.e. if the system is prepared in state @xmath20 or @xmath21 it will remain in such eigenstate . on the contrary , the interlevel coupling induces a time variation of the coefficients . the above two - level model provides the simplest description of light - matter interaction in atomic and molecular systems @xcite as well as in solids @xcite . the coupling term in equation ( [ eq1.3 ] ) has the same structure of a field - dipole interaction hamiltonian and the driving force @xmath22 can be regarded as a coherent light field . we will now consider two limiting cases : the ultrashort- and continuous - excitation regimes . an ultrashort optical excitation can be described in terms of a delta - like light pulse : @xmath23 . in this case , the equations of motion ( [ eq1.6 ] ) can be solved analytically : due to this excitation , the two - level system will undergo an instantaneous transition from its ground state @xmath24 to the excited state @xmath25 , with @xmath26 . therefore , after the pulse the system will remain in the excited state @xmath27 which is a coherent quantum - mechanical superposition of the two non - interacting states . in addition to a finite occupation probability @xmath28 of the excited state , there exists a well defined phase coherence between the ground- and the excited - state contributions , i.e. a part from their amplitudes , the coefficients @xmath29 and @xmath30 differ in phase by @xmath31 . this is what is generally meant by optically induced phase coherence . let us now consider a continuous optical excitation resonant with our two - level system : @xmath32 with @xmath33 . also for this case , equations ( [ eq1.6 ] ) can be solved analytically . in particular , for the initial condition at @xmath34 given by the ground state @xmath24 , the solution is given by @xmath35 where @xmath36 is the so - called rabi frequency . compared to the previous case , the continuous excitation gives rise to a periodic population and depopulation of the excited state according to @xmath37 . this purely coherent phenomenon is known as `` rabi - oscillation regime '' . as for the previous case , the excited state in equation ( [ eq1.8 ] ) reflects a well - defined optically - induced phase coherence . as mentioned above , the interlevel - coupling hamiltonian in equation ( [ eq1.3 ] ) may also describe , in addition to an optical excitation , some interlevel coupling due , e.g. to coulomb interaction . also in this case , in the presence of this interlevel coupling the quantum state of the system will result in a coherent superposition of the two stationary states in ( [ eq1.4 ] ) . again , this coupling will induce a well - defined phase coherence between the two levels . from the above considerations we see that coherent phenomena can be divided into two basic classes : _ optically - induced _ and _ coulomb - induced _ phenomena . in order to describe the optical response of an electron gas within a semiconductor crystal , the above analysis , based on a single two - level system , has to be replaced by a statistical - ensemble approach , i.e. a description based on a collection of independent two - level systems . any statistical ensemble of quantum systems is properly described by its single - particle density matrix @xcite @xmath38 where @xmath39 denotes a generic set of quantum numbers . its diagonal elements @xmath40 provide the average occupation numbers while the non - diagonal terms describe the degree of phase coherence between states @xmath39 and @xmath41 . for the case of our two - level system , @xmath42 reduces to a two - by - two matrix : the diagonal elements @xmath43 and @xmath44 describe , respectively , the average occupation of levels @xmath3 and @xmath5 , while the non - diagonal terms @xmath45 and @xmath46 reflect the average degree of phase coherence between the ground and the excited state . starting from the heisenberg equations of motion for the creation and destruction operators , i.e. @xmath47 \ , \qquad i\hbar { d\over dt } a^ { } _ n = [ a^ { } _ n,{\bf h}]\ , \ ] ] we obtain the following equations of motion for the above density - matrix elements : @xmath48 they are known as optical bloch equations @xcite in analogy with the equations first derived by bloch @xcite for the spin systems . this statistical - ensemble description reduces to the previous one for the case of a so called `` pure state '' , i.e. the case in which all the two - level systems forming the ensemble are in the same quantum - mechanical state @xmath49 . in this case , the above set of kinetic equations and the equations of motion ( [ eq1.6 ] ) are totally equivalent . however , as we will discuss in section [ s.tb ] , the density - matrix approach introduced so far allows , in addition to the study of coherent phenomena , the analysis of incoherent phenomena , which is not possible within a simple schrdinger - equation formalism . the optical bloch equations ( [ eq1.10 ] ) provide the simplest description of light - matter interaction . let us consider again the case of a continuous optical excitation resonant with our two - level system : @xmath32 . if we choose as initial condition at time @xmath34 the state @xmath50 , the solution of the above optical bloch equations is given by : @xmath51\ , \end{aligned}\ ] ] where @xmath52 . as for the case of a single two - level system , the above solution describes a rabi - oscillation regime . in particular , the interlevel density - matrix element @xmath53 originates from the superposition of the two frequency components @xmath54 and @xmath55 . they differ from @xmath56 by the rabi frequency @xmath57 . such modification of the two - level frequency @xmath56 due to its coupling with the external field is known as `` rabi splitting '' @xcite . if we now rewrite the interlevel density - matrix element @xmath53 in ( [ eq1.12 ] ) as @xmath58 we see that , a part from the quantum - mechanical phase factor corresponding to the energy separation @xmath33 , its amplitude exhibits rabi oscillations according to @xmath59 . more specifically , we obtain : @xmath60 i.e. the quantity @xmath61 is proportional to the product of the two occupation numbers , thus reflecting the total ( or macroscopic ) dipole moment of the two - level system at time @xmath62 . this elucidates the link between optically - induced phase coherence and polarization : a coherent optical excitation gives rise to a coherent quantum - mechanical superposition of the two states which results in a macroscopic polarization of the system . such polarization field is fully described by the non - diagonal matrix element @xmath53 in ( [ eq1.12 ] ) . the above simplified description of light - matter interaction neglects any kind of incoherent , i.e. phase - breaking , phenomena . as we will see in section [ s.tb ] , such dephasing processes will lead to a decay of the interband polarization @xmath53 , thus destroying the optically induced phase coherence in the carrier system . as a final remark , let us discuss the concept of phase coherence in connection with the choice of representation . on the basis of the density - matrix formalism considered so far , the phase coherence is described by the non - diagonal density - matrix elements @xmath42 in ( [ eq1.9 ] ) . however , this separation in _ diagonal _ and _ non - diagonal _ terms is clearly representation dependent : what is diagonal in a given basis is in general not diagonal in a different basis and vice versa . if one considers , as basis set , the eigenstates of the total hamiltonian @xmath63 ( which includes the external driving force ) , the density matrix @xmath64 in ( [ eq1.9 ] ) is always diagonal , i.e. no phase coherence . thus , in order to speak of phase coherence , we need to regard the total hamiltonian @xmath63 as the sum of the system hamiltonian @xmath65 plus a driving - force term @xmath12 ( see equation ( [ eq1.1 ] ) ) . provided such separation between _ system of interest _ and _ driving force _ , the non - diagonal density - matrix elements within the representation given by the eigenstates of @xmath65 will describe the degree of coherence induced in the system by the driving force @xmath12 . two typical cases , discussed respectively in sections [ s.sl ] and [ s.qwr ] , may help in clarifying the meaning of coherence . the first case is that of a semiconductor superlattice in the presence of a constant and homogeneous electric field : as we will see , within the wannier - stark representation the bloch - oscillation dynamics is the result of phase coherence between the various wannier - stark states ; on the contrary , within the bloch representation the same phenomenon is purely described in terms of carrier populations , i.e. no off - diagonal density - matrix elements . the second case is that of coulomb - induced phase coherence : within a free - particle representation , even at the simplest hartree - fock level , coulomb interaction induces phase coherence between free - carrier states . this is clearly not the case within an exciton basis , where the density matrix is diagonal . in this section we will review in a systematic way the basic ideas used in the theoretical analysis of ultrafast carrier dynamics in semiconductors . the approach we are going to present is based on the density - matrix formalism introduced in section [ ss.moc ] . in order to study the optical and transport properties of semiconductor bulk and heterostructures , let us consider a gas of carriers in a crystal under the action of an applied electromagnetic field . the carriers will experience their mutual interaction as well as the interaction with the phonon modes of the crystal . such physical system can be described by the following hamiltonian : @xmath66 the first term describes the noninteracting - carrier system in the presence of the external electromagnetic field while the second one refers to the free - phonon system . the last three terms describe many - body contributions : they refer , respectively , to carrier - carrier , carrier - phonon , and phonon - phonon interactions . in order to discuss their explicit form , let us introduce the usual second - quantization field operators @xmath67 and @xmath68 . they describe , respectively , the creation and the destruction of a carrier in @xmath69 . in terms of the above field operators the carrier hamiltonian @xmath70 can be written as @xmath71 { \bf \psi}({\bf r})\ .\ ] ] here , @xmath72 denotes the periodic potential due to the perfect crystal while @xmath73 and @xmath74 denote , respectively , the vector and scalar potentials corresponding to the external electromagnetic field . since we are interested in the electrooptical properties as well as in the ultrafast dynamics of photoexcited carriers , the electromagnetic field acting on the crystal and the corresponding electromagnetic potentials will be the sum of two different contributions : the high - frequency laser field responsible for the ultrafast optical excitation and the additional electromagnetic field acting on the photoexcited carriers on a longer time - scale . more specifically , by denoting with the labels @xmath75 and @xmath76 these two contributions , we can write @xmath77 and recalling that @xmath78 we have @xmath79 equation ( [ eq4 ] ) , which gives the electromagnetic fields in terms of the corresponding vector and scalar potentials , reflects the well known gauge freedom : there is an infinite number of possible combinations of @xmath80 and @xmath81 which give rise to the same electromagnetic field @xmath82 . we will use such freedom of choice for the laser field ( term @xmath75 ) : we assume a homogeneous ( space - independent ) laser field @xmath83 fully described by the scalar potential @xmath84 this assumption , which corresponds to the well known dipole approximation , is well justified as long as the space - scale of interest is small compared to the light wavelength . the explicit form of the laser field considered here is @xmath85 where @xmath86 is the amplitude of the light field and @xmath56 denotes its central frequency . with this particular choice of the electromagnetic potentials describing the laser field , the hamiltonian in ( [ eq2 ] ) can be rewritten as @xmath87 where @xmath88 { \bf \psi}({\bf r})\ ] ] describes the carrier system in the crystal under the action of the electromagnetic field @xmath76 only , while @xmath89 describes the carrier - light ( cl ) interaction due to the laser photoexcitation . in analogy with the carrier system , by denoting with @xmath90 and @xmath91 the creation and destruction operators for a phonon of mode @xmath92 and wavevector @xmath93 , the free - phonon hamiltonian takes the form @xmath94 where @xmath95 is the dispersion relation for the phonon mode @xmath92 . let us now discuss the explicit form of the many - body contributions . the carrier - carrier interaction is described by the two - body hamiltonian @xmath96 where @xmath97 denotes the coulomb potential . let us now introduce the carrier - phonon interaction hamiltonian @xmath98 where @xmath99\ ] ] is the electrostatic phonon potential induced by the lattice vibrations . here , the explicit form of the coupling function @xmath100 depends on the particular phonon mode @xmath92 ( acoustic , optical , etc . ) as well as on the coupling mechanism considered ( deformation potential , polar coupling , etc . ) . let us finally discuss the phonon - phonon contribution @xmath101 . the free - phonon hamiltonian @xmath102 in ( [ eq11 ] ) , which describes a system of noninteracting phonons , by definition accounts only for the harmonic part of the lattice potential . however , non - harmonic contributions of the interatomic potential can play an important role in determining the lattice dynamics in highly excited systems @xcite , since they are responsible for the decay of optical phonons into phonons of lower frequency . in our second - quantization picture , these non - harmonic contributions can be described in terms of a phonon - phonon interaction which induces , in general , transitions between free - phonon states . here , we will not discuss the explicit form of the phonon - phonon hamiltonian @xmath101 responsible for such a decay . we will simply assume that such phonon - phonon interaction is so efficient to maintain the phonon system in thermal equilibrium . this corresponds to neglecting hot - phonon effects @xcite . it is well known that the coordinate representation used so far is not the most convenient one in describing the electron dynamics within a periodic crystal . in general , it is more convenient to employ the representation given by the eigenstates of the noninteracting - carrier hamiltonian or a part of it since it automatically accounts for some of the symmetries of the system . for the moment we will simply consider an orthonormal basis set @xmath103 without specifying which part of the hamiltonian is diagonal in such representation this will allow us to write down equations valid in any quantum - mechanical representation . since the noninteracting - carrier hamiltonian is , in general , a function of time , also the basis functions @xmath104 may be time - dependent . here , the label @xmath39 denotes , in general , a set of discrete and/or continuous quantum numbers . in the absence of electromagnetic field , the above wavefunctions will correspond to the well known bloch states of the crystal and the index @xmath39 will reduce to the wavevector @xmath2 plus the band ( or subband ) index @xmath105 . in the presence of a homogeneous magnetic field the eigenfunctions @xmath104 may instead correspond to landau states . finally , for the case of a constant and homogeneous electric field , there exist two equivalent representations : the accelerated bloch states and the wannier - stark picture . such equivalence results to be of crucial importance in understanding the relationship between bloch oscillations and wannier - stark localization and , for this reason , it will be discussed in more detail in section [ ss.qmp ] . let us now reconsider the system hamiltonian introduced so far in terms of such @xmath104 representation . as a starting point , we may expand the second - quantization field operators in terms of the new wavefunctions : @xmath106 the above expansion defines the new set of second - quantization operators @xmath107 and @xmath108 ; they describe , respectively , the creation and destruction of a carrier in state @xmath39 . for the case of a semiconductor structure ( the only one considered here ) , the energy spectrum of the noninteracting - carrier hamiltonian ( [ eq9 ] ) or a part of it is always characterized by two well - separated energy regions : the valence and the conduction band . also in the presence of an applied electromagnetic field , the periodic lattice potential @xmath109 gives rise to a large energy gap . therefore , we deal with two energetically well - separated regions , which suggests the introduction of the usual electron - hole picture . this corresponds to a separation of the set of states @xmath110 into conduction states @xmath111 and valence states @xmath112 . thus , also the creation ( destruction ) operators @xmath107 ( @xmath108 ) introduced in equation ( [ eq15 ] ) will be divided into creation ( destruction ) electron and hole operators : @xmath113 ( @xmath114 ) and @xmath115 ( @xmath116 ) . in terms of the new electron - hole picture , the expansion ( [ eq15 ] ) is given by : @xmath117 { \bf \psi}^\dagger({\bf r } ) = & & \sum_i \phi^{e * } _ i({\bf r } ) c^\dagger_i + \sum_j \phi^h_j({\bf r } ) d^ { } _ j \ .\end{aligned}\ ] ] if we now insert the above expansion into equation ( [ eq9 ] ) , the noninteracting - carrier hamiltonian takes the form @xmath118 where @xmath119 \phi^{e / h}_{l'}({\bf r})\ ] ] are just the matrix elements of the hamiltonian in the @xmath120-representation . the @xmath121 sign refers , respectively , to electrons and holes while @xmath122 denotes the conduction - band edge . here , we neglect any valence - to - conduction band coupling due to the external electromagnetic field and vice versa . this is well fulfilled for the systems and field - regimes we are going to discuss in this paper . as already pointed out , the above hamiltonian may be time - dependent . we will discuss this aspect in the following section , where we will derive our set of kinetic equations . let us now write in terms of our electron - hole representation the carrier - light interaction hamiltonian ( [ eq10 ] ) : @xmath123\ .\ ] ] the above expression has been obtained within the well known rotating - wave approximation @xcite by neglecting intraband transitions , absent for the case of optical excitations . here , @xmath124 denotes the optical dipole matrix element between states @xmath125 and @xmath126 . similarly , the carrier - carrier hamiltonian ( [ eq12 ] ) can be rewritten as : @xmath127 where @xmath128 are the coulomb matrix elements within our @xmath120-representation . the first two terms describe the repulsive electron - electron and hole - hole interaction while the last one describes the attractive electron - hole interaction . here , we neglect terms that do not conserve the number of electron - hole pairs , i.e. impact - ionization and auger - recombination processes @xcite , as well as the interband exchange interaction . this monopole - monopole approximation is justified as long as the exciton binding energy is small compared to the energy gap . finally , let us rewrite the carrier - phonon interaction hamiltonian introduced in equation ( [ eq13 ] ) : @xmath129 \nonumber \\ & & - \sum\limits_{jj',{\bf q}\lambda } \left [ g^h_{jj',{\bf q}\lambda } d^\dagger_j b^ { } _ { { \bf q}\lambda } d^ { } _ { j ' } + g^{h * } _ { jj',{\bf q}\lambda } d^\dagger_{j ' } b^\dagger_{{\bf q}\lambda } d^ { } _ j \right]\end{aligned}\ ] ] with @xmath130 in equation ( [ eq22 ] ) we can clearly recognize four different contributions corresponding to electron and hole phonon absorption and emission . our kinetic description of the ultrafast carrier dynamics in semiconductors is based on the density - matrix formalism introduced in section [ ss.moc ] . since this approach has been reviewed and discussed in several papers @xcite and text - books @xcite , here we will simply recall in our notation the kinetic equations relevant for the analysis of carrier dynamics in semiconductor heterostructures , generalizing the approach presented in @xcite to the case of a time - dependent quantum - mechanical representation . the set of kinetic variables is the same considered in @xcite . given our electron - hole representation @xmath131 , we will consider the intraband electron and hole single - particle density matrices @xmath132 as well as the corresponding interband density matrix @xmath133 here , the diagonal elements @xmath134 and @xmath135 correspond to the electron and hole distribution functions of the boltzmann theory while the non - diagonal terms describe intraband polarizations . on the contrary , the interband density - matrix elements @xmath136 describe interband ( or optical ) polarizations . in order to derive the set of kinetic equations , i.e. the equations of motion for the above kinetic variables , the standard procedure starts by deriving the equations of motion for the electron and hole operators introduced in ( [ eq16 ] ) : @xmath137 by applying the heisenberg equation of motion for the field operator @xmath138 , i.e. @xmath139\ , \ ] ] it is easy to obtain the following equations of motion : @xmath140 + \frac{1}{i\hbar}\sum_{i ' } z^e_{ii ' } c^ { } _ { i ' } \nonumber \\[1ex ] \frac{d}{dt } d^ { } _ j = & & \frac{1}{i\hbar } \left[d^ { } _ j,{\bf h}\right ] + \frac{1}{i\hbar}\sum_{j ' } z^h_{jj ' } d^ { } _ { j'}\\end{aligned}\ ] ] with @xmath141 as for the case of equation ( [ eq17 ] ) , here we neglect again valence - to - conduction band coupling and vice versa . compared to the more conventional heisenberg equations of motion , they contain an extra - term , the last one . it accounts for the possible time dependence of our @xmath120-representation which will induce transitions between different states according to the matrix elements @xmath142 . by combining the above equations of motion with the definitions of the kinetic variables in ( [ eq24]-[eq25 ] ) , the resulting set of kinetic equations can be schematically written as : @xmath143 \frac{d}{dt } f^h_{j_1j_2 } = & & \frac{d}{dt } f^h_{j_1j_2}\biggl|_{\bf h } + \frac{d}{dt } f^h_{j_1j_2}\biggl|_\phi \nonumber \\[1ex ] \frac{d}{dt } p^ { } _ { j_1i_1 } = & & \frac{d}{dt } p^ { } _ { j_1i_1}\biggl|_{\bf h } + \frac{d}{dt } p^ { } _ { j_1i_1}\biggl|_\phi \ .\end{aligned}\ ] ] they exhibit the same structure of the equations of motion ( [ eq28 ] ) for the electron and hole creation and destruction operators : a first term induced by the system hamiltonian @xmath63 ( which does not account for the time variation of the basis states ) and a second one induced by the time dependence of the basis functions @xmath120 . let us start discussing this second term , whose explicit form is : @xmath144 f^e_{i_3i_4 } \nonumber \\[1ex ] \frac{d}{dt } f^h_{j_1j_2}\biggl|_\phi = & & \frac{1}{i\hbar } \sum\limits_{j_3j_4}\left[z^h_{j_2j_4}\delta_{j_1j_3 } - z^h_{j_3j_1}\delta_{j_2j_4 } \right ] f^h_{j_3j_4 } \nonumber \\[1ex ] \frac{d}{dt } p^ { } _ { j_1i_1}\biggl|_\phi = & & \frac{1}{i\hbar } \sum\limits_{i_2j_2}\left[z^h_{j_1j_2}\delta_{i_1i_2 } + z^e_{i_1i_2}\delta_{j_1j_2 } \right ] p^ { } _ { j_2i_2 } \ .\end{aligned}\ ] ] as we will see in section [ ss.qmp ] , these contributions play a central role for the description of zener tunneling within the vector - potential representation . let us now come to the first term . this , in turn , is the sum of different contributions , corresponding to the various parts of the hamiltonian . the total hamiltonian can be regarded as the sum of two terms , a single - particle contribution plus a many - body one : @xmath145 the explicit form of the time evolution due to the single - particle hamiltonian @xmath146 ( non - interacting carriers plus carrier - light interaction plus free phonons ) is given by : @xmath147 f^e_{i_3i_4 } \nonumber \\ & & + \sum\limits_{j_1}\left[u_{i_2j_1 } p^*_{j_1i_1 } -u^*_{i_1j_1 } p^ { } _ { j_1i_2}\right ] \bigg\ } \nonumber \\ \frac{d}{dt } f^h_{j_1j_2}\biggl|_{sp } = & & \frac{1}{i\hbar } \bigg\ { \sum\limits_{j_3j_4}\left[\epsilon^h_{j_2j_4}\delta_{j_1j_3 } - \epsilon^h_{j_3j_1}\delta_{j_2j_4 } \right ] f^h_{j_3j_4 } \nonumber \\ & & + \sum\limits_{i_1}\left[u_{i_1j_2 } p^*_{j_1i_1 } -u^*_{i_1j_1 } p^ { } _ { j_2i_1 } \right ] \bigg\ } \nonumber \\ \frac{d}{dt } p^ { } _ { j_1i_1}\biggl|_{sp } = & & \frac{1}{i\hbar } \bigg\ { \sum\limits_{i_2j_2}\left[\epsilon^h_{j_1j_2}\delta_{i_1i_2 } + \epsilon^e_{i_1i_2}\delta_{j_1j_2 } \right ] p^ { } _ { j_2i_2 } \nonumber \\ & & + \sum\limits_{i_2j_2 } u^ { } _ { i_2j_2 } \left[\delta_{i_1i_2}\delta_{j_1j_2 } - f^e_{i_2i_1}\delta_{j_1j_2 } - f^h_{j_2j_1}\delta_{i_1i_2 } \right ] \bigg\}\end{aligned}\ ] ] with @xmath148 . this is a closed set of equations , which is a consequence of the single - particle nature of @xmath146 . in addition , we stress that the structure of the two contributions entering equation ( [ eq30 ] ) is very similar : one can include the contribution ( [ eq31 ] ) into ( [ eq33 ] ) by replacing @xmath149 with @xmath150 . let us finally discuss the contributions due to the many - body part of the hamiltonian : carrier - carrier and carrier - phonon interactions ( the phonon - phonon one is not explicitly considered here ) . as discussed in @xcite , for both interaction mechanisms one can derive a hierarchy of equations involving higher - order density matrices . in order to close such equations with respect to our set of kinetic variables , approximations are needed . the lowest - order contributions to our equations of motion are given by first - order terms in the many - body hamiltonian : hartree - fock level . since we will neglect coherent - phonon states , the only hartree - fock contributions will come from carrier - carrier interaction . they simply result in a renormalization @xmath151 of the single - particle energy matrices @xmath152 as well as in a renormalization @xmath153 of the external field @xmath154 . we stress that the hartree - fock approximation , which consists in factorizing average values of four - point operators into products of two density matrices , is independent from the quantum - mechanical picture . it is then clear that the above kinetic equations are valid in any quantum - mechanical representation . all the contributions to the system dynamics discussed so far describe a fully coherent dynamics , i.e. no scattering processes . in order to treat incoherent phenomena , e.g. energy relaxation and dephasing , one has to go one step further in the perturbation expansion taking into account also second - order contributions ( in the perturbation hamiltonian @xmath155 ) . the derivation of these higher - order contributions , discussed in @xcite , will not be repeated here . again , in order to obtain a closed set of equations ( with respect to our set of kinetic variables ( [ eq24]-[eq25 ] ) ) additional approximations are needed , namely the mean - field and markov approximations . as for the hartree - fock case , the mean - field approximation allows to write the various higher - order density matrices as products of single - particle ones . the markov approximation allows to eliminate the additional higher - order kinetic variables , e.g. phonon - assisted density matrices , providing a closed set of equations still local in time , i.e. no memory effects @xcite . this last approximation is not performed in the quantum - kinetic theory discussed in @xcite where , in addition to our single - particle variables , one considers two - particle @xcite and phonon - assisted @xcite density matrices . while the mean - field approximation is representation - independent , this is unfortunately not the case for the markov limit . this clearly implies that the validity of the markov approximation is strictly related to the quantum - mechanical representation considered . the above kinetic description , based on intra- and interband density matrices , allows us to evaluate any single - particle quantity . in particular , for the analysis of the ultrafast carrier dynamics in photoexcited semiconductors two physical quantities play a central role : the intra- and interband total ( or macroscopic ) polarizations : @xmath156 where @xmath157 and @xmath158 denote , respectively , the intra- and interband dipole matrix elements in our @xmath120-representation . the time derivative of the intraband polarization @xmath159 describes the radiation field induced by the bloch - oscillation dynamics ( which for a superlattice structure is in the terahertz range ) while the fourier transform of the interband ( or optical ) polarization @xmath160 provides the linear and non - linear optical response of the system . in this section , we will discuss the dominant role played by optically - induced phase coherence on the ultrafast generation and relaxation of photoexcited carriers in bulk semiconductors . in particular , we will compare the description based on the theoretical approach discussed above with the more conventional picture based on the boltzmann theory . in order to study the ultrafast carrier dynamics in bulk systems , let us consider a two - band semiconductor model . for the case of a homogeneous system , the only relevant terms of the single - particle density matrix in @xmath2-space are the diagonal ones . this property , due to the translational symmetry of the hamiltonian , reduces the set of kinetic variables ( [ eq24]-[eq25 ] ) to the following electron and hole distribution functions ( intraband density - matrix elements ) @xmath161 together with the corresponding polarizations ( interband density - matrix elements ) has been applied to our plane - wave states . in particular , due to the charge - conjugation symmetry , the hole states are still labeled in terms of the corresponding valence - electron states , i.e. @xmath162 . ] @xmath163 the explicit form of the kinetic equations ( [ eq30 ] ) within the above two - band picture is given by @xmath164 \nonumber \\ \frac{d}{dt } f_{\bf k}^h = & & g_{\bf -k}(t ) - \sum_{\bf k ' } \left [ w^h_{\bf k'k } f_{\bf k}^h \left ( 1 - f_{\bf k'}^h \right ) - w^h_{\bf kk ' } f_{\bf k'}^h \left ( 1 - f_{\bf k}^h \right ) \right ] \nonumber \\ \frac{d}{dt } p_{\bf k}^ { } = & & \frac{1}{i\hbar}\left [ \left ( \epsilon_{\bf k}^e + \epsilon_{\bf -k}^h \right ) p_{\bf k}^ { } + u_{\bf k}\left ( 1 - f_{\bf k}^e - f_{\bf -k}^h \right ) \right ] \nonumber \\ & & { } - \sum_{\bf k ' } \left [ w^p_{\bf k'k } p_{\bf k}^ { } - w^p_{\bf kk ' } p_{\bf k'}^ { } \right ] \ , \end{aligned}\ ] ] with the generation rate @xmath165 \ .\ ] ] here , the incoherent contributions are treated within the usual markov limit as described in @xcite . within such approximation scheme , the incoherent contributions to the polarization dynamics exhibit the same structure as for the distribution functions in terms of the following in- and out - scattering rates @xmath166 \ , \ ] ] where @xmath167 are the usual scattering rates of the semiclassical boltzmann theory . this boltzmann - like structure of the scattering term in the polarization equation is the starting point of the generalized monte carlo method discussed in section [ s.int ] @xcite . the above kinetic equations are known as _ semiconductor bloch equations _ ( sbe ) ) reduce to the optical bloch equations introduced in section [ ss.moc ] , i.e. they describe a collection of independent ( non - interacting ) two - level systems . ] . as we can see , here the carrier photogeneration is a two - step process : the external field @xmath154 induces a coherent polarization @xmath53 which , in turn , generates electron - hole pairs via its coupling with the field according to equation ( [ eq4bulk ] ) . the generation rate is thus determined by the interband polarization @xmath168 which , in turn , is influenced by the various density - dependent scattering mechanisms , e.g. carrier - carrier processes . therefore , in contrast to the semiclassical case discussed below , the generation rate within the sbe model is clearly density dependent . the conventional boltzmann model is obtained from the above sbe by performing an adiabatic elimination of the interband polarization @xmath168 , i.e. by inserting a formal solution of the polarization equation into the generation rate ( [ eq4bulk ] ) and then performing a markov limit with respect to the electron and hole distribution functions @xcite . this corresponds to assuming that the carrier distributions are slowly - varying functions on the time - scale of the laser photoexcitation , i.e. the carrier generation and relaxation are treated as independent processes . within such approximation scheme , except for phase - space - filling effects , the generation rate entering the semiclassical boltzmann equations ( be ) for electrons and holes is fully determined by the temporal and spectral characteristics of the laser pulse . contrary to the sbe case , all effects related to the optically induced phase coherence and to its temporal decay are neglected . in order to compare the different models of carrier dynamics in photoexcited semiconductors , let us review some simulated experiments @xcite based on the generalized monte carlo solution of the sbe @xcite discussed in section [ s.int ] . the corresponding be has been solved , for comparison , by a standard ensemble monte carlo ( emc ) simulation @xcite . both be and sbe have been solved numerically for the case of gaas bulk excited by a 150 fs laser pulse . in figure [ fig1bulk ] the generation rate at different times as obtained from the sbe is plotted as a function of the carrier wave vector for three different densities . at the lowest density the behavior is essentially the same as in the case without carrier - carrier scattering : energy - time uncertainty leads to an initially very broad generation rate ; with increasing time the line narrows and , in the tails , exhibits negative parts due to a stimulated recombination of carriers initially generated off - resonance @xcite . after the pulse , the distribution function of the generated carriers is in good agreement with the be result . scattering processes destroy the coherence between electrons and holes which is necessary for the stimulated recombination processes . as a consequence , with increasing density the negative tails are strongly reduced and the generation remains broad for all times resulting in a much broader carrier distribution than in the be case . hot carrier luminescence has proven to be a powerful technique to study the ultrafast dynamics of photoexcited carriers in semiconductors @xcite . in band - to - acceptor ( ba ) luminescence experiments @xcite , due to their high sensitivity , carrier densities as low as several @xmath169 @xmath170 have been reached . thus , the transition between a dynamics dominated by carrier - phonon scattering to a dynamics dominated by carrier - carrier scattering ( with its consequences for the generation process discussed above ) is experimentally accessible . in the ba luminescence experiment reviewed here @xcite , a 3 @xmath1 m thick gaas layer doped with be acceptors of a concentration of @xmath171 @xmath170 is excited at a photon energy of 1.73 ev by transform limited 150 fs pulses from a mode - locked ti : sapphire laser . time - integrated luminescence spectra are recorded @xcite . the right column of figure [ fig2bulk ] shows measured ba luminescence spectra for three different values of the carrier density . at low density we observe an initial peak of the generated carriers ( marked bold ) and pronounced replicas due to the emission of an integer number of optical phonons . with increasing density the peaks become broader and at the highest density only a slight structure related to the phonons is still visible . these measured spectra are compared with corresponding simulated experiments based on the be and sbe models @xcite ( left and middle column in figure [ fig2bulk ] ) . we find several pronounced differences between the two models . in the semiclassical model the unrelaxed peak is clearly visible up to the highest density in contrast to the coherent model , where , in agreement with the experiment , this peak is strongly broadened . at the higher densities the semiclassical spectra exhibit an increase in the broadening of subsequent replicas which is not present in the bloch - model calculations as well as in the experimental results . to illustrate this difference quantitatively , in figure [ fig3bulk ] we have plotted the full width at half maximum ( fwhm ) of the three highest peaks in the spectra . from the semiclassical model we obtain a strong increase in the density dependence of subsequent peaks . this behavior can be easily understood : due to the emission time of an optical phonon of about @xmath172 fs , the carriers populate subsequent replicas at increasing times and , therefore , the efficiency of carrier - carrier scattering processes in broadening the peaks increases . however , the measured spectra exhibit a different behavior which is quantitatively reproduced by the sbe model : already the unrelaxed peak exhibits a strong increase with increasing density and the density dependence of all replicas is approximately the same . the reason is that the broadening of the generation process as discussed above obviously dominates over the broadening due to subsequent scattering processes . as discussed in @xcite , this is due to the fact that the generation process is strongly influenced by the decay of the interband polarization which , in turn , is due to electron plus hole scattering , while the broadening of the electron distribution during the relaxation process is due to electron scattering only . the above theoretical and experimental analysis constitutes a clear demonstration of the importance of dephasing processes for the analysis of luminescence spectra . the density dependence of the spectra in the interesting transition region between carrier - phonon and carrier - carrier dominated dynamics can only be explained by a coherent modeling of the carrier generation including the dynamics of the interband polarization . the main mechanism determining the width of the peaks is the broadening of the generation rate . it can be shown that for band - to - acceptor spectra the broadening of the recombination process , neglected in the present model , plays a minor role @xcite . however , for band - to - band spectra also this phenomenon should be taken into account . for the case discussed above the background due to band - to - band luminescence has been found to be of negligible importance @xcite . ever since the initial applications of quantum mechanics to the dynamics of electrons in solids , the analysis of bloch electrons moving in a homogeneous electric field has been of central importance . by employing semiclassical arguments , in 1928 bloch @xcite demonstrated that , a wave packet given by a superposition of single - band states peaked about some quasimomentum , @xmath173 , moves with a group velocity given by the gradient of the energy - band function with respect to the quasimomentum and that the rate of change of the quasimomentum is proportional to the applied field @xmath174 . this is often referred to as the `` acceleration theorem '' : @xmath175 thus , in the absence of interband tunneling and scattering processes , the quasimomentum of a bloch electron in a homogeneous and static electric field will be uniformly accelerated into the next brillouin zone in a repeated - zone scheme ( or equivalently undergoes an umklapp process back in to the first zone ) . the corresponding motion of the bloch electron through the periodic energy - band structure , shown in figure [ fig1sl ] , is called `` bloch oscillation '' ; it is characterized by an oscillation period @xmath176 , where @xmath177 denotes the lattice periodicity in the field direction . there are two mechanisms impeding a fully periodic motion : interband tunneling and scattering processes . interband tunneling is an intricate problem and still at the center of a continuing debate . early calculations of the tunneling probability into other bands in which the electric field is represented by a time - independent scalar potential were made by zener @xcite using a wentzel - kramers - brillouin generalization of bloch functions , by houston @xcite using accelerated bloch states ( houston states ) , and subsequently by kane @xcite and argyres @xcite who employed the crystal - momentum representation . their calculations lead to the conclusion that the tunneling rate per bloch period is much less than unity for electric fields up to @xmath178v / cm for typical band parameters corresponding to elemental or compound semiconductors . despite the apparent agreement among these calculations , the validity of employing the crystal - momentum representation or houston functions to describe electrons moving in a non - periodic ( crystal plus external field ) potential has been disputed . the starting point of the controversy was the original paper by wannier @xcite . he pointed out that , due to the translational symmetry of the crystal potential , if @xmath179 is an eigenfunction of the scalar - potential hamiltonian ( corresponding to the perfect crystal plus the external field ) with eigenvalue @xmath149 , then any @xmath180 is also an eigenfunction with eigenvalue @xmath181 , where @xmath182 is the so - called wannier - stark splitting ( @xmath183 being the primitive lattice vector along the field direction ) . he concluded that the translational symmetry of the crystal gives rise to a discrete energy spectrum , the so - called wannier - stark ladder . the states corresponding to these equidistantly spaced levels are localized states , as schematically shown in figure [ fig2sl ] for the case of a semiconductor superlattice . the existence of such energy quantization was disputed by zak @xcite , who pointed out that for the case of an infinite crystal the scalar potential @xmath184 is not bounded , which implies a continuous energy spectrum . thus , the main point of the controversy was related to the existence ( or absence ) of wannier - stark ladders . more precisely , the point was to decide if interband tunneling ( neglected in the original calculation by wannier @xcite ) is so strong to destroy the wannier - stark energy quantization ( and the corresponding bloch oscillations ) or not . it is only during the last decade that this controversy came to an end . from a theoretical point of view , most of the formal problems related to the non - periodic nature of the scalar potential ( superimposed to the periodic crystal potential ) were finally removed by using a vector - potential representation of the applied field @xcite . within such vector - potential picture , upper boundaries for the interband tunneling probability have been established at a rigorous level , which show that an electron may execute a number of bloch oscillations before tunneling out of the band @xcite , in qualitatively good agreement with the earlier predictions of zener and kane @xcite . the second mechanism impeding a fully periodic motion is scattering by phonons , impurities , etc . ( see figure [ fig1sl ] ) . this results in lifetimes shorter than the bloch period @xmath185 for all reasonable values of the electric field , so that bloch oscillations should not be observable in conventional solids . in superlattices , however , the situation is much more favourable because of the smaller bloch period @xmath185 resulting from the small width of the mini - brillouin zone in the field direction @xcite . indeed , the existence of wannier - stark ladders as well as bloch oscillations in superlattices has been confirmed by a number of recent experiments @xcite . the photoluminescence and photocurrent measurements of the biased gaas / gaalas superlattices performed by mendez and coworkers @xcite , together with the electroluminescence experiments by voisin and coworkers @xcite , provided the earliest evidence of the field - induced wannier - stark ladders in superlattices . a few years later , feldmann and coworkers @xcite were able to measure bloch oscillations in the time domain through a four - wave - mixing experiment originally suggested by von plessen and thomas @xcite . a detailed analysis of the bloch oscillations in the four - wave - mixing signal ( which reflects the interband dynamics ) has been also performed by leo and coworkers @xcite . in addition to the above interband - polarization analysis , bloch oscillations have been also detected by monitoring the intraband polarization which , in turn , is reflected by anisotropic changes in the refractive index @xcite . measurements based on transmittive electrooptic sampling ( teos ) have been performed by dekorsy and coworkers @xcite . finally bloch oscillations were recently measured through a direct detection of the terahertz ( thz ) radiation in semiconductor superlattices @xcite . let us now apply the theoretical approach presented in section [ s.tb ] to the case of a semiconductor superlattice in the presence of an uniform ( space - independent ) electric field . the non - interacting carriers within the superlattice crystal will then be described by the hamiltonian @xmath186 in equation ( [ eq9 ] ) , where now the electrodynamic potentials @xmath187 and @xmath188 ( in the following simply denoted with @xmath80 and @xmath81 ) correspond to a homogeneous electric field @xmath189 . as pointed out in section [ ss.ps ] , the natural quantum - mechanical representation is given by the eigenstates of this hamiltonian : @xmath190 \phi_n({\bf r } ) = \epsilon_n \phi_n({\bf r})\ .\ ] ] however , due to the gauge freedom discussed in section [ ss.ps ] , there is an infinite number of possible combinations of @xmath80 and @xmath81 and therefore of possible hamiltonians which describe the same homogeneous electric field @xmath191 . in particular , one can identify two independent choices : the vector - potential gauge @xmath192 and the scalar - potential gauge @xmath193 as shown in @xcite , the two independent choices correspond , respectively , to the well known bloch - oscillation and wannier - stark pictures . they simply reflect two equivalent quantum - mechanical representations and , therefore , any physical phenomenon can be described in both pictures . more specifically , within the vector - potential picture ( [ eq3sl ] ) , the eigenfunctions @xmath104 in ( [ eq2sl ] ) are the so - called accelerated bloch states ( or houston states ) @xcite . as discussed in @xcite , such time dependent representation constitutes the natural basis for the description of bloch oscillations , i.e. it provides a rigorous quantum - mechanical derivation of the acceleration theorem ( [ eq1sl ] ) , thus showing that this is not a simple semiclassical result . ) and the corresponding bloch - oscillation dynamics are usually regarded as a semiclassical result compared to the wannier - stark picture . on the contrary , they correspond to two different fully quantum - mechanical pictures . ] within such representation , bloch oscillations are fully described by the diagonal terms of the intraband density matrix ( [ eq24 ] ) ( semiclassical distribution functions ) . therefore , non - diagonal elements , describing phase - coherence between different bloch states , do not contribute to the intraminiband dynamics . however , they are of crucial importance for the description of interminiband dynamics , i.e. field - induced zener tunneling , which in this bloch - state representation originates from the time variation of our basis states ( see equation ( [ eq31 ] ) ) . on the contrary , within the scalar - potential picture ( [ eq4sl ] ) , the eigenfunctions @xmath104 in ( [ eq2sl ] ) are the well - known wannier - stark states @xcite . contrary to the previous bloch picture , within such representation the intraminiband bloch dynamics originates from a quantum interference between different wannier - stark states , thus involving non - diagonal elements of the intraband density matrix ( [ eq24 ] ) . in this section , we will review recent simulated experiments of the ultrafast carrier dynamics in semiconductor superlattices @xcite . they are based on a generalized monte carlo solution @xcite of the set of kinetic equations ( so - called semiconductor bloch equations ) derived in section [ ss.kd ] . in this case , the bloch representation discussed in section [ ss.qmp ] has been employed limiting the set of interband density - matrix elements in ( [ eq25 ] ) to the diagonal ones , i.e. @xmath194 . in addition , incoherent scattering processes have been treated within the usual markov limit as discussed in @xcite . within such approximation scheme , the explicit form of the kinetic equations ( [ eq30 ] ) coincides with the sbe ( [ eq3 ] ) ( obtained for the bulk case ) @xcite , provided we replace the wavevector @xmath2 with @xmath195 , @xmath105 being the superlattice miniband index . in the simulated experiments reviewed here the following superlattice model has been employed : the energy dispersion and the corresponding wavefunctions along the growth direction ( @xmath196 ) are computed within the well known kronig - penney model @xcite , while for the in - plane direction ( @xmath197 ) an effective - mass model has been used . starting from these three - dimensional wavefunctions @xmath198 , the various carrier - carrier as well as carrier - phonon matrix elements are numerically computed ( see equations ( [ eq21 ] ) and ( [ eq23 ] ) ) . they are , in general , functions of the various miniband indices and depend separately on @xmath196 and @xmath197 , thus fully reflecting the anisotropic nature of the superlattice structure . only coupling to gaas bulk phonons has been considered . this , of course , is a simplifying approximation which neglects any superlattice effect on the phonon dispersion , such as confinement of optical modes in the wells and in the barriers , and the presence of interface modes @xcite . however , while these modifications have important consequences for phonon spectroscopies ( like raman scattering ) , they are far less decisive for transport phenomena . we will start discussing the scattering - induced damping of bloch oscillations . in particular , we will show that in the low - density limit this damping is mainly determined by optical - phonon scattering @xcite while at high densities the main mechanism responsible for the suppression of bloch oscillations is found to be carrier - carrier scattering @xcite . this bloch - oscillation analysis in the time domain is also confirmed by its counterpart in the frequency domain . as pointed out in section [ ss.qmp ] , the presence of bloch oscillations , due to a negligible scattering dynamics , should correspond to wannier - stark energy quantization . this is confirmed by the simulated optical - absorption spectra , which clearly show the presence of the field - induced wannier - stark ladders@xcite . all the simulated experiments presented in this section refer to the superlattice structure considered in ref . @xcite : @xmath199 gaas wells and @xmath200 al@xmath201ga@xmath202as barriers . for such a structure there has been experimental evidence for a thz - emission from bloch oscillations @xcite . in the first set of simulated experiments an initial distribution of photoexcited carriers ( electron - hole pairs ) is generated by a @xmath203 fs gaussian laser pulse in resonance with the first - miniband exciton ( @xmath204 mev ) . the strength of the applied electric field is assumed to be @xmath205 kv / cm , which corresponds to a bloch period @xmath176 of about @xmath206 fs . in the low - density limit ( corresponding to a weak laser excitation ) , incoherent scattering processes do not alter the bloch - oscillation dynamics . this is due to the following reasons : in agreement with recent experimental @xcite and theoretical @xcite investigations , for superlattices characterized by a miniband width smaller than the lo - phonon energy as for the structure considered here and for laser excitations close to the band gap , at low temperature carrier - phonon scattering is not permitted . moreover , in this low - density regime carrier - carrier scattering plays no role : due to the quasi - elastic nature of coulomb collisions , in the low - density limit the majority of the scattering processes is characterized by a very small momentum transfer ; as a consequence , the momentum relaxation along the growth direction is negligible . as a result , on this picosecond time - scale the carrier system exhibits a coherent bloch - oscillation dynamics , i.e. a negligible scattering - induced dephasing . this can be clearly seen from the time evolution of the carrier distribution as a function of @xmath196 ( i.e. averaged over @xmath207 ) shown in figure [ fig3sl ] . during the laser photoexcitation ( @xmath34 ) the carriers are generated around @xmath208 , where the transitions are close to resonance with the laser excitation . according to the acceleration theorem ( [ eq1sl ] ) , the electrons are then shifted in @xmath2-space . when the carriers reach the border of the first brillouin zone they are bragg reflected . after about @xmath206 fs , corresponding to the bloch period @xmath185 , the carriers have completed one oscillation in @xmath2-space . as expected , the carriers execute bloch oscillations without loosing the synchronism of their motion by scattering . this is again shown in figure [ fig3sl ] , where we have plotted : ( b ) the mean kinetic energy , ( c ) the current , and ( d ) its time derivative which is proportional to the emitted far field , i.e. the thz - radiation . ( it can be shown that , by neglecting zener tunneling , the intraband polarization @xmath159 in equation ( [ eq37 ] ) is proportional to the current . ) all these three quantities exhibit oscillations characterized by the same bloch period @xmath185 . due to the finite width of the carrier distribution in @xmath2-space ( see figure [ fig3sl](a ) ) , the amplitude of the oscillations of the kinetic energy is somewhat smaller than the miniband width . since for this excitation condition the scattering - induced dephasing is negligible , the oscillations of the current are symmetric around zero , which implies that the time average of the current is equal to zero , i.e. no dissipation . as already pointed out , this ideal bloch - oscillation regime is typical of a laser excitation close to gap in the low - density limit . let us now discuss , still at low densities , the case of a laser photoexcitation high in the band . figure [ fig4sl](a ) shows the thz - signal as obtained from a set of simulated experiments corresponding to different laser excitations @xcite . the different traces correspond to the emitted thz - signal for increasing excitation energies . we clearly notice the presence of bloch oscillations in all cases . however , the oscillation amplitude and decay ( effective damping ) is excitation - dependent . for the case of a laser excitation resonant with the first - miniband exciton considered above ( see figure [ fig3sl ] ) , we have a strong thz - signal . the amplitude of the signal decreases when the excitation energy is increased . additionally , there are also some small changes in the phase of the oscillations , which are induced by the electron - lo phonon scattering . when the laser energy comes into resonance with the transitions between the second electron and hole minibands ( @xmath209 mev ) , the amplitude of the thz - signal increases again . the corresponding thz - transients show an initial part , which is strongly damped and some oscillations for longer times that are much less damped . for a better understanding of these results , we show in figure [ fig4sl](b ) the individual thz - signals , originating from the two electron and two heavy - hole minibands for the excitation with @xmath210 mev . the bloch oscillations performed by the electrons within the second miniband are strongly damped due to intra- and interminiband lo - phonon scattering processes @xcite . since the width of this second miniband ( @xmath211 mev ) is somewhat larger than the lo - phonon energy , also intraminiband scattering is possible , whenever the electrons are accelerated into the high - energy region of the miniband . the thz - signal originating from electrons within the first miniband shows an oscillatory behavior with a small amplitude and a phase which is determined by the time the electrons need to relax down to the bottom of the band . at the same time , the holes in both minibands exhibit undamped bloch oscillations , since the minibands are so close in energy that for these excitation conditions no lo - phonon emission can occur . the analysis shows that at early times the thz - signal is mainly determined by the electrons within the second miniband . at later times the observed signal is due to heavy holes and electrons within the first miniband . the above theoretical analysis closely resembles experimental observations obtained for a superlattice structure very similar to the one modelled here @xcite . in these experiments , evidence for thz - emission from bloch oscillations has been reported . for some excitation conditions these oscillations are associated with resonant excitation of the second miniband . the general behavior of the magnitude of the signals , the oscillations and the damping are close to the results shown in figure [ fig4sl ] . finally , in order to study the density dependence of the bloch - oscillation damping , let us go back to the case of laser excitations close to gap . figure [ fig5sl](a ) shows the total ( electrons plus holes ) thz - radiation as a function of time for three different carrier densities . with increasing carrier density , carrier - carrier scattering becomes more and more important : due to coulomb screening , the momentum transfer in a carrier - carrier scattering increases ( its typical value being comparable with the screening wavevector ) . this can be seen in figure [ fig5sl](a ) , where for increasing carrier densities we realize an increasing damping of the thz - signal . however , also for the highest carrier density considered here we deal with a damping time of the order of @xmath212 fs , which is much larger than the typical dephasing time , i.e. the decay time of the interband polarization , associated with carrier - carrier scattering . the dephasing time is typically investigated by means of four - wave - mixing ( fwm ) measurements and such multi - pulse experiments can be simulated as well @xcite . from a theoretical point of view , a qualitative estimate of the dephasing time is given by the decay time of the `` incoherently summed '' polarization ( isp ) @xcite . figure [ fig5sl](b ) shows such isp as a function of time for the same three carrier densities of figure [ fig5sl](a ) . as expected , the decay times are always much smaller than the corresponding damping times of the thz - signals ( note the different time - scale in figures [ fig5sl](a ) and ( b ) ) . this difference , discussed in more detail in @xcite , can be understood as follows : the fast decay times of figure [ fig5sl](b ) reflect the interband dephasing , i.e. the sum of the electron and hole scattering rates . in particular , for the coulomb interaction this means the sum of electron - electron , electron - hole , and hole - hole scattering . as for the case of bulk gaas discussed in section [ s.bulk ] , this last contribution is known to dominate and determines the dephasing time - scale . on the other hand , the total thz - radiation in figure [ fig5sl](a ) is the sum of the electron and hole contributions . however , due to the small value of the hole miniband width compared to the electron one , the electron contribution will dominate . this means that the thz damping in figure [ fig5sl](a ) mainly reflects the damping of the electron contribution . this decay , in turn , reflects the intraband dephasing of electrons which is due to electron - electron and electron - hole scattering only , i.e. no hole - hole contributions . from the above analysis we can conclude that the decay time of the thz - radiation due to carrier - carrier scattering differs considerably from the corresponding dephasing times obtained from a fwm experiment : the first one is a measurement of the intraband dephasing while the second one reflects the interband dephasing . for the case of bulk semiconductors , where the broadening of the photoexcited carrier distribution is mainly determined by the decay of the interband polarization ( interband dephasing ) while the subsequent energy broadening of the electron distribution is due to electron scattering only ( intraband dephasing ) ( see figure [ fig2bulk ] ) . ] let us now discuss the frequency - domain counterpart of the bloch - oscillation picture considered so far . similar to what happens in the time domain , for sufficiently high electric fields , i.e. when the bloch period @xmath176 becomes smaller than the dephasing time , the optical spectra of the superlattice are expected to exhibit the frequency - domain counterpart of the bloch oscillations , i.e. the wannier - stark energy quantization discussed in section [ ss.qmp ] . in the absence of coulomb interaction , the wannier - stark ladder absorption increases as a function of the photon energy in a step - like fashion . these steps are equidistantly spaced . their spacing , named wannier - stark splitting , is proportional to the applied electric field . the simulated linear - absorption spectra corresponding to a superlattice structure with @xmath213 gaas wells and @xmath214 al@xmath201ga@xmath202as barriers are shown in figure [ fig6sl ] @xcite . as we can see , the coulomb interaction gives rise to excitonic peaks in the absorption spectra and introduces couplings between these wannier - stark states . such exciton peaks , which are no longer equidistantly spaced , are often referred to as excitonic wannier - stark ladders @xcite of the superlattice . since for the superlattice structure considered in this simulated experiment @xcite the combined miniband width is larger than the typical two- and three - dimensional exciton binding energies , it is possible to investigate the quasi - three dimensional absorption behavior of the delocalized miniband states as well as localization effects induced by the electric field . for the free - field case , the electron and hole states are completely delocalized in our three - dimensional @xmath2-space . the perturbation induced by the application of a low field ( here @xmath215 kv / cm ) , couples the states along the field direction and in the spectra the franz - keldysh effect , well known from bulk materials @xcite , appears : one clearly notices oscillations which increase in amplitude with the field and shift with @xmath216 from the @xmath217 and @xmath218 levels toward the center of the combined miniband . for increasing field the potential drop over the distance of a few quantum wells eventually exceeds the miniband width and the electronic states become more and more localized . despite the field - induced energy difference @xmath219 , the superlattice potential is equal for quantum wells separated by @xmath220 . therefore , the spectra decouple into a series of peaks corresponding to the excitonic ground states of the individual electron - hole wannier - stark levels . each wannier - stark transition contributes to the absorption with a pronounced @xmath221 exciton peak , plus higher bound exciton and continuum states . the oscillator strength of a transition @xmath39 is proportional to the overlap between electron and hole wavefunctions centered at quantum wells @xmath41 and @xmath222 , respectively . the analysis shows that this oscillator strength is almost exclusively determined by the amplitude of the electron wavefunction in quantum well @xmath223 since for fields in the wannier - stark regime the hole wavefunctions are almost completely localized over one quantum well due to their high effective mass ( see figure [ fig2sl ] ) . thus , the oscillator strengths of transitions to higher @xmath224 become smaller with increasing @xmath224 and field . at high fields ( here @xmath225 kv / cm ) the separation between the peaks is almost equal to @xmath226 . for example , the peak of the @xmath227 transition which is shifted by the wannier - stark exciton binding energy with respect to the center of the combined miniband , demonstrates that the increasing localization also increases the exciton binding energy . this increased excitonic binding reflects the gradual transition from a three- to a two - dimensional behavior , discussed in detail in @xcite . for intermediate fields there is an interplay between the wannier - stark and the franz - keldysh effect . coming from high fields , first the wannier - stark peaks are modulated by the franz - keldysh oscillations . however , as soon as the separation @xmath228 between neighboring peaks becomes smaller than their spectral widths , the peaks can no longer be resolved individually so that only the franz - keldysh structure remains . the importance of coulomb - correlation effects in optical spectra of semiconductors and their dependence on dimensionality has now been long recognized @xcite . more recently , increasing interest has been devoted to one - dimensional ( 1d ) systems @xcite , prompted by promising advances in quantum - wire fabrication and application , e.g. quantum - wire lasers . the main goal is to achieve structures with improved optical efficiency as compared to two - dimensional ( 2d ) and three - dimensional ( 3d ) ones . a common argument in favour of this effort is based on the well known van hove divergence in the 1d joint density - of - states ( dos ) , which is expected to give rise to very sharp peaks in the optical spectra of 1d structures . such prediction is however based on free - carrier properties of ideal 1d systems and it ignores coulomb - correlation effects . early theoretical investigations by ogawa and coworkers @xcite , based on a 1d single - subband model of the wires , showed that the inverse - square - root singularity in the 1d dos at the band edge is smoothed when excitonic effects are taken into account . in this section , we will review recent theoretical results @xcite based on a full 3d description of realistic quantum - wire structures made available by the present technology , such as structures obtained by epitaxial growth on non - planar substrates ( v - shaped wires ) @xcite or by cleaved - edge quantum well overgrowth ( t - shaped wires ) @xcite . on the one hand , this approach allows an accurate determination of 1d - exciton binding energies ; on the other hand , it allows us to investigate also the non linear ( gain ) regime . in both cases , one finds a strong suppression of the 1d band - edge singularity , in agreement with previous results based on simplified 1d models @xcite . this theoretical approach @xcite is based on the general kinetic theory presented in section [ s.tb ] , which allows a full 3d description of coulomb interaction within a multiminiband scheme . in particular , we focus on the quasi - equilibrium regime where the solution of the coupled kinetic equations ( [ eq30 ] ) simply reduces to the solution of the interband - polarization equation . this is performed by direct numerical evaluation of the polarization eigenvalues and eigenvectors @xcite , which fully determine the absorption spectrum as well as the exciton wavefunctions . the main ingredients entering this calculation are the single - particle energies and wavefunctions , obtained numerically for the 2d confinement potential deduced , e.g. , from tem as in @xcite . the above theoretical scheme has been applied to realistic v- and t - shaped wire structures . in particular , here we show results for the gaas / algaas v - wire structure of ref . @xcite , whose cross - section is shown in figure [ fig1qwr ] . let us start considering the optical response of the system in the low - density limit . in figure [ fig2qwr ] we show the linear - absorption spectra obtained when taking into account the lowest wire transition only . results of our coulomb - correlated ( cc ) approach are compared to those of the free - carrier ( fc ) model . as we can see , electron - hole correlation introduces two important effects : first , the excitonic peak arises below the onset of the continuum , with a binding energy of about @xmath229 mev , in excellent agreement with recent experiments @xcite . second , one finds a strong suppression of the 1d dos singularity . a detailed analysis of the physical origin of such suppression @xcite has shown that the quantity which is mainly modified by cc is the oscillator strength ( os ) . in figure [ fig3qwr](a ) the ratio between the cc and fc os is plotted as a function of excess energy ( solid line ) . this ratio is always less than one and goes to zero at the band edge , and reflects a sort of hole in the electron - hole correlation function @xmath230 as shown in figure [ fig3qwr](b ) . such vanishing behaviour is found to dominate the 1d dos singularity and , as a result , the absorption spectrum exhibits a regular behaviour at the band edge ( solid line in figure [ fig2qwr ] ) . the above analysis shows that also for realistic quantum - wire structures electron - hole correlation leads to a strong suppression of the 1d band - edge singularity in the linear - absorption spectrum , contrary to the 2d and 3d cases . most of the potential quantum - wire applications , i.e. 1d lasers and modulators , operate in strongly non - linear - response regimes @xcite . in such conditions , the above linear - response analysis has to be generalized taking into account additional factors as : ( i ) screening effects , ( ii ) band renormalization , ( iii ) space - phase filling . figure [ fig4qwr ] reports quantitative results for non - linear absorption spectra of realistic v - shaped wire structures at different carrier densities at room temperature . as a reference we also show the results obtained by including the lowest subband only [ figure [ fig4qwr](a ) ] . in the low - density limit ( case a : @xmath231 @xmath232 ) we clearly recognize the exciton peak . with increasing carrier density , the strength of the excitonic absorption decreases due to phase - space filling and screening of the attractive electron - hole interaction , and moreover the band renormalization leads to a red - shift of the continuum . at a density of @xmath233 @xmath232 ( case d ) the spectrum already exhibits a negative region corresponding to stimulated emission , i.e. gain regime . as desired , the well pronounced gain spectrum extends over a limited energy region ( smaller than the thermal energy ) ; however , its shape differs considerably from the ideal fc one ( curve marked with diamonds in the same figure ) . in particular , the band - edge singularity in the ideal fc gain spectrum is clearly smeared out by electron - hole correlation . the overall effect is a broader and less pronounced gain region . finally , figure [ fig4qwr](b ) shows the non - linear spectra corresponding to the realistic case of a 12-subband v - shaped wire . in comparison with the single - subband case [ figure [ fig4qwr](a ) ] , the multisubband nature is found to play an important role in modifying the typical shape of the gain spectra , which for both cc and fc models result to extend over a range much larger than that of the single - subband case for the present wire geometry . in addition , the coulomb - induced suppression of the single - subband singularities , here also due to intersubband - coupling effects , tends to reduce the residual structures in the gain profile . therefore , even in the ideal case of a quantum wire with negligible disorder and scattering - induced broadening , our analysis indicates that , for the typical structure considered , the shape of the absorption spectra over the whole density range differs considerably from the sharp fc spectrum of figure [ fig2qwr ] . this tells us that , in order to obtain sharp gain profiles , one of the basic steps in quantum - wire technology is to produce structures with increased subband splitting . however , the disorder - induced inhomogeneous broadening , not considered here , is known to increase significantly the spectral broadening and this effect is expected to increase with increasing subband splitting . therefore , extremely - high - quality structures ( e.g. single - monolayer control ) seem to be the only possible candidates for successful quantum - wire applications . a review of coherent phenomena in optically - excited semiconductors has been presented . our analysis has allowed the identification of two classes of phenomena : _ optically - induced _ and _ coulomb - induced _ coherent phenomena . we have shown that both classes can be described in terms of a unique theoretical framework based on the density - matrix formalism . due to its generality , such quantum - kinetic approach allows a realistic description of coherent as well as incoherent , i.e. phase - breaking , processes , thus providing quantitative information on the coupled coherent vs. incoherent carrier dynamics in photoexcited semiconductors . the primary goal of the paper was to discuss the concept of quantum - mechanical phase coherence as well as its relevance and implications on semiconductor physics and technology . in particular , we have analyzed the dominant role played by optically induced phase coherence on the process of ultrafast carrier photogeneration . we have then discussed typical field - induced coherent phenomena in semiconductor superlattices , e.g. bloch oscillations and the corresponding thz radiation , as well as their dephasing dynamics . finally , we have analyzed the dominant role played by coulomb correlation on the linear and non - linear optical spectra of realistic quantum - wire structures , namely the coulomb - induced suppression of ideal 1d - band - edge singularities . our analysis shows that the conventional separation between coherent and incoherent regimes is no longer valid for most of the recent ultrafast optical experiments in semiconductors . the reason is that on such extremely short time - scales the carrier dynamics is the result of a strong interplay between coherence and relaxation , thus preventing a time - scale separation between photogeneration and relaxation dynamics . similar considerations apply to the various coulomb - induced phenomena : the strong modifications induced by coulomb correlation on the optical response of low - dimensional semiconductors play a significant role on the phase - breaking relaxation dynamics as well . it is thus evident that any theoretical analysis of the optical response of semiconductor heterostructures must provide a proper description of both coherent and incoherent phenomena as well as of their mutual coupling on the same kinetic level . i am particularly grateful to tilmann kuhn for his essential contribution in understanding most of the ideas and concepts discussed in this topical review . i wish to thank stephan w. koch , torsten meier , and peter thomas , as well as elisa molinari , for their relevant contributions to the research activity reviewed in the paper . i am also grateful to roberto cingolani , thomas elsaesser , alfred leitenstorfer , and peter e. selbmann for stimulating and fruitful discussions . j. shah , _ ultrafast spectroscopy of semiconductors and semiconductor nanostructures _ ( springer , berlin , 1996 ) . l. allen , j.h . eberly , _ optical resonance and two - level atoms _ ( interscience , new york 1975 ) . m. sargent iii , m.o . scully , w.e . lamb jr . , _ laser physics _ ( addison - wesley , new york 1977 ) . shen , _ the principles of nonlinear optics _ ( wiley , new york 1984 ) . levenson , s.s . kano , _ introduction to nonlinear laser spectroscopy _ ( academic , boston 1988 ) . p. meystre , m. sargent iii , _ elements of quantum optics _ , 2nd edn . ( springer , berlin , heidelberg 1991 ) . mills , _ nonlinear optics _ ( springer , berlin , heidelberg 1991 ) . h. haug and s.w . koch , _ quantum theory of the optical and electronic properties of semiconductors _ , 3rd edn . ( world scientific , singapore 1994 ) . w. demtroeder , _ laser spectroscopy _ , ( springer , berlin , heidelberg 1996 ) . f. henneberger , s. schmitt - rink , e.o . gbel ( eds . ) , _ optics of semiconductor nanostructures _ ( akademie verlag , berlin 1993 ) . phillips ( ed . ) _ coherent optical processes in semiconductors _ , nato asi series b : physics , vol . 30 ( plenum , new york 1994 ) . hahn , phys . rev . * 80 * , 580 ( 1950 ) . kurnit , i.d . abella , and s.r . hartmann , phys . * 13 * , 567 ( 1964 ) . abella , n.a . kurnit , and s.r . hartmann , phys . * 141 * , 391 ( 1966 ) . j. shah and r.c . c. leite , phys . rev . lett . * 22 * , 1304 ( 1969 ) . shank , r.l . fork , r.f . leheny , and j. shah , phys . * 42 * , 112 ( 1979 ) . j. shah , b. deveaud , t.c . damen , w.t . tsang , a.c . gossard , and p. lugli , phys . 59 * , 2222 ( 1987 ) . t. elsaesser , j. shah , l. rota , and p. lugli , phys . lett . * 66 * , 1757 ( 1991 ) . ulbrich , j.a . kash , and j.c . tsang , phys . lett . * 62 * , 949 ( 1989 ) . peterson and s.a . lyon , phys . 65 * , 760 ( 1990 ) . snoke , w.w . rhle , y .- c . lu , and e. bauser , phys . . lett . * 68 * , 990 ( 1992 ) . oudar , a. migus , d. hulin , g. grillon , j. etchepare , and a. antonetti , phys . rev . lett . * 53 * , 384 ( 1984 ) . foing , d. hulin , m. joffre , m.k . jackson , j .- l . oudar , c. tanguy , and m. combescot , phys . lett . * 68 * , 110 ( 1992 ) . a. leitenstorfer , c. frst , a. laubereau , w. kaiser , g. trnkle , and g. weimann , phys . . lett . * 76 * , 1545 ( 1996 ) . s. lutgen , r. kaindl , m. woerner , t. elsaesser , a. hase , h. knzel , m. gulia , d. meglio , and p. lugli , phys . lett . * 77 * , 3657 ( 1996 ) . m. joschko , m. woerner , t. elsaesser , e. binder , t. kuhn , r. hey , h. kostial , and k. ploog , phys . rev . * 78 * , 737 ( 1997 ) . c. jacoboni and p. lugli , _ the monte carlo method for semiconductor device simulations _ ( springer , wien 1989 ) . s.m . goodnick and p. lugli , in _ hot carriers in semiconductor nanostructures : physics and applications _ , edited by j. shah ( academic , san diego 1992 ) , p. 191 . f. rossi , p. poli , and c. jacoboni , semicond . * 7 * , 1017 ( 1992 ) . a. mysyrowicz , d. hulin , a. antonetti , a. migus , w.t . masselink , and h. morko , phys . 56 * , 2748 ( 1986 ) . n. peyghambarian , s.w . koch , m. lindberg , b.d . flgel , m. joffre , phys . * 62 * , 1185 ( 1989 ) . knox , d.s . chemla , d.a.b . miller , j.b . stark , and s. schmitt - rink , phys . * 1989 * , 1189 ( 1989 ) . kundiff , a. knorr , j. feldmann , s.w . koch , e.o . gbel , h. nichel , phys . 73 * , 1178 ( 1994 ) . becker , h.l . fragnito , c.h . brito cruz , r.l . fork , j.e . cunningham , j.e . henry , and c.v . shank , phys . lett . * 61 * , 1647 ( 1988 ) . y. masumoto , s. shionoya , and t. takagahara , phys . lett . * 51 * , 923 ( 1983 ) . l. schultheis , j. kuhl , a. honold , and c.w . tu , phys . 57 * , 1635 ( 1986 ) ; phys . * 57 * , 1797 ( 1986 ) . g. noll , u. siegner , s.g . shevel , and e.o . gbel , phys . lett . * 64 * , 792 ( 1990 ) . v. langer , h. stolz , and w. von der osten , phys . lett . * 64 * , 854 ( 1990 ) . gbel , k. leo , t.c . damen , j. shah , s. schmitt - rink , w. schfer , j.f . mller , and k. khler , phys . rev . lett . * 64 * , 1801 ( 1990 ) . d. frhlich , a. kulik , b. uebbing , a. mysyrowicz , v. langer , h. stolz , and w. von der osten , phys . * 67 * , 2343 ( 1991 ) . h. stolz , v. langer , e. schreiber , s.a . permogorov , w. von der osten , phys . lett . * 67 * , 679 ( 1991 ) . k. leo , j. shah , e.o . gbel , t.c . damen , s. schmitt - rink , w. schfer , and k. khler , phys . rev . lett . * 66 * , 201 ( 1991 ) . j. feldmann , t. meier , g. von plessen , m. koch , e.o . gbel , p. thomas , g. bacher , c. hartmann , h. schweizer , w. schfer , and h. nickel , phys . * 70 * , 3027 ( 1993 ) . c. waschke , h.g . roskos , r. schwedler , k. leo , h. kurz , and k. khler , phys . lett . * 70 * , 3319 ( 1993 ) . k. leo , m. wegener , j. shah , d.s . chemla , e.o . gbel , t.c . damen , s. schmitt - rink , and w. schfer , phys . 65 * , 1340 ( 1990 ) . s. weiss , m .- a . mycek , j .- y . bigot , s. schmitt - rink , and d.s . chemla , phys . * 69 * , 2685 ( 1992 ) . d .- s . kim , j. shah , j.e . cunningham , t.c . damen , w. schfer , m. hartmann , and s. schmitt - rink , phys . * 68 * , 1006 ( 1992 ) . d .- s . kim , j. shah , t.c . damen , w. schfer , f. jahnke , s. schmitt - rink , and k. khler , phys . * 69 * , 2725 ( 1992 ) . roskos , m.c . nuss , j. shah , k. leo , d.a.b . miller , a.m. fox , s. schmitt - rink , and k. khler , phys . lett . * 68 * , 2216 ( 1992 ) . planken , m.c . nuss , i. brener , k.w . goossen , m.s.c . luo , s.l . chuang , and l. pfeiffer , phys . lett . * 69 * , 3800 ( 1992 ) . c. comte and g. mahler , phys . rev.b * 34 * , 7164 ( 1986 ) . s. schmitt - rink , c. ell , and h. haug , phys . rev . b*33 * , 1183 ( 1986 ) . s. schmitt - rink , d.s . chemla , and h. haug , phys . b*37 * , 941 ( 1988 ) . k. henneberger and h. haug , phys . b*38 * , 9759 ( 1988 ) . kuznetsov , phys . rev . b*44 * , 8721 ( 1991 ) . h. haug and c. ell , phys . b*46 * , 2126 ( 1992 ) . i. balslev , r. zimmermann , and a. stahl , phys . b*40 * , 4095 ( 1989 ) . s. schmitt - rink and d.s . chemla , phys . * 57 * , 2752 ( 1986 ) . m. lindberg and s.w . koch , phys . b*38 * , 3342 ( 1988 ) . m. wegener , d.s . chemla , s. schmitt - rink , and w. schfer , phys . rev . a*42 * , 5675 ( 1990 ) . w. schfer , f. jahnke , and s. schmitt - rink , phys . rev . b*47 * , 1217 ( 1993 ) . e. binder , t. kuhn , and g. mahler , phys . rev . * b50 * , 18319 ( 1994 ) . o. hess and t. kuhn , phys . a * 54 * , 3347 ( 1996 ) ; ibid . 3360 ( 1996 ) . t. kuhn , in _ theory of transport properties of semiconductor nanostructures _ , edited by e. schll ( chapman & hall , london 1997 ) . r.l . fork , c.h . brito cruz , p.c . becker , and c.v . shank , phys . * 12 * , 483 ( 1987 ) . t. kuhn and f. rossi , phys . lett . * 69 * , 977 ( 1992 ) . t. kuhn and f. rossi , phys . b * 46 * , 7496 ( 1992 ) . f. rossi , s. haas , and t. kuhn , semicond . * 9 * , 411 ( 1994 ) . f. rossi , s. haas , and t. kuhn , phys . lett . * 72 * , 152 ( 1994 ) . s. haas , f. rossi , and t. kuhn , phys . b * 53 * , 12855 ( 1996 ) . a. lohner , k. rick , p. leisching , a. leitenstorfer , t. elsaesser , t. kuhn , f. rossi , and w. stolz , phys . lett . * 71 * , 77 ( 1993 ) . a. leitenstorfer , a. lohner , k. rick , p. leisching , t. elsaesser , t. kuhn , f. rossi , w. stolz , and k. ploog , phys . b * 49 * , 16372 ( 1994 ) . a. leitenstorfer , a. lohner , t. elsaesser , s. haas , f. rossi , t. kuhn , w. klein , g. boehm , g. traenkle , and g. weimann , phys . lett . * 73 * , 1687 ( 1994 ) . t. kuhn , f. rossi , a. leitenstorfer , a. lohner , t. elsaesser , w. klein , g. boehm , g. traenkle , and g.w . weimann , in _ hot carriers in semiconductors _ , edited by k. hess , j .- leburton , and u. ravaioli ( plenum press , new york , 1996 ) , p. 199 . a. leitenstorfer , t. elsaesser , f. rossi , t. kuhn , w. klein , g. boehm , g. traenkle , and g.w . weimann , phys . b * 53 * , 9876 ( 1996 ) . f. bloch , phys . rev . * 70 * , 460 ( 1946 ) . kash and j.c . tsang , in _ light scattering in solids vi _ , edited by m. cardona and g.gntherodt , springer , berlin ( 1989 ) , p. 423 . w. ptz and p. kocevar , phys . b * 28 * , 7040 ( 1983 ) . w. quade , e. schll , f. rossi , and c. jacoboni , phys . b * 50 * , 7398 ( 1994 ) . f. rossi , r. brunetti , and c. jacoboni , in _ hot carriers in semiconductor nanostructures : physics and applications _ , edited by j. shah ( academic press inc . , boston , 1992 ) , p. 153 . r. brunetti , c. jacoboni , and f. rossi , phys . b * 39 * , 10781 ( 1989 ) tran thoai and h. haug , phys . b * 47 * , 3574 ( 1993 ) . sayed , l. bnyai , and h. haug , phys . rev . * b50 * , 1541 ( 1994 ) . j. schilp , t. kuhn , and g. mahler , phys . b * 50 * , 5435 ( 1994 ) . v. meden , c. whler , j. fricke , and k. schnhammer , phys . rev . * b52 * , 5624 ( 1995 ) . kane , k.w . sun , and s.a . lyon , semicond . * 9 * , 697 ( 1994 ) kash , phys . rev . * b51 * , 4680 ( 1995 ) . f. bloch , z. phys . * 52 * , 555 ( 1928 ) . c. zener , proc . * 145 * , 523 ( 1934 ) . houston , phys . rev . * 57 * , 184 ( 1940 ) . e.o . kane , j. phys . solids * 12 * , 181 ( 1959 ) p.n . argyres , phys . rev . * 126 * , 1386 ( 1962 ) . g.h . wannier , phys . rev . * 117 * , 432 ( 1960 ) . j. zak , phys . lett . * 20 * , 1477 ( 1968 ) . c. kittel , _ quantum theory of solids _ ( wiley , new york , 1963 ) , p. 190 . j.b . krieger and g.j . iafrate , phys . b * 33 * , 5494 ( 1986 ) . g. nenciu , rev . phys . * 63 * , 91 ( 1991 ) . g. bastard , _ wave mechanics of semiconductor heterostructures _ , les editions de physique ( les ulis , france , 1989 ) . mendez , f. agullo - rueda , and j.m . hong , phys . lett . * 60 * , 2426 ( 1988 ) . p. voisin , j. bleuse , c. bouche , s. gaillard , c. alibert , and a , regreny , phys . lett . * 61 * , 1639 ( 1988 ) . j. feldmann , k. leo , j. shah , d.a.b . miller , j.e . cunningham , t. meier , g. von plessen , a. schulze , p. thomas , and s. schmitt - rink , phys . b * 46 * , 7252 ( 1992 ) . g. von plessen and p. thomas , phys . rev . b * 45 * , 9185 ( 1992 ) . k. leo , p.h . bolivar , f. brggemann , r. schwedler , solid state commun . * 84 * , 943 ( 1992 ) . p. leisching , p. haring bolivar , w. beck , y. dhaibi , f. bruggemann , r. schwedler , h. kurz , k. leo , and k. khler , phys . b * 50 * , 14389 ( 1994 ) . t. dekorsy , p. leisching , k. kohler , and h. kurz , phys . b * 50 * , 8106 ( 1994 ) . t. dekorsy , r. ott , h. kurz , and k. kohler , phys . b * 51 * , 17275 ( 1995 ) . roskos , c. waschke , r. schwedler , p. leisching , y. dhaibi , h. kurz , and k. khler , superlattices and microstructures * 15 * , 281 ( 1994 ) . f. rossi , in _ theory of transport properties of semiconductor nanostructures _ , edited by e. schll ( chapman & hall , london 1997 ) . f. rossi , t. meier , p. thomas , s.w . koch , p.e . selbmann , and e. molinari , phys . b * 51 * , 16943 ( 1995 ) . f. rossi , t. meier , p. thomas , s.w . koch , p.e . selbmann , and e. molinari , in _ hot carriers in semiconductors _ , edited by k. hess , j .- leburton , and u. ravaioli ( plenum press , new york , 1996 ) , p. 157 . t. meier , f. rossi , p. thomas , and s.w . koch , phys . lett . * 75 * , 2558 ( 1995 ) . je , t. meier , f. rossi , and s.w . koch , appl . lett . * 67 * , 2978 ( 1995 ) . koch , t. meier , t. stroucken , a. knorr , j. hader , f. rossi , and p. thomas , in _ microscopic theory of semiconductors : quantum kinetics , confinement and lasers _ , edited by s.w . koch ( world scientific , singapore , 1995 ) , p. 81 . f. rossi , m. gulia , p.e . selbmann , e. molinari , t. meier , p. thomas , and s.w . koch , in _ proc . 23rd icps , berlin , germany _ , edited by m. scheffler and r. zimmermann ( world scientific , singapore , 1996 ) , p. 1775 . h. rcker , e. molinari , and p. lugli , phys . b * 45 * , 6747 ( 1992 ) . e. molinari , in _ confined electrons and photons : new physics and applications _ , edited by e. burstein and c. weisbuch ( plenum , new york , 1994 ) . g. von plessen , t. meier , j. feldmann , e.o . gbel , p. thomas , k.w . goossen , j.m . kuo , and r.f . kopf , phys . b * 49 * , 14058 ( 1994 ) . m.m . dignam and j.e . sipe , phys . * 64 * , 1797 ( 1990 ) . see e.g. _ proc . 4th internat . conf . on optics of excitons in confined systems _ , il nuovo cimento d vol . 17 ( 1995 ) . for a review see : r. cingolani and r. rinaldi , rivista nuovo cimento * 16 * , 1 ( 1993 ) . t. ogawa and t. takagahara , phys . rev . * b43 * , 14325 ( 1991 ) . t. ogawa and t. takagahara , phys . rev . * b44 * , 8138 ( 1991 ) . f. rossi and e. molinari , phys . 76 * , 3642 ( 1996 ) . f. rossi and e. molinari , phys . rev . * b53 * , 16462 ( 1996 ) . f. rossi and e. molinari , in _ proc . 23rd icps , berlin , germany _ , edited by m. scheffler and r. zimmermann ( world scientific , singapore , 1996 ) , p. 1161 . r. rinaldi , r. cingolani , m. lepore , m. ferrara , i.m . catalano , f. rossi , l. rota , e. molinari , p. lugli , u. marti , d. martin , f. morrier - gemoud , p. ruterana , and f.k . reinhart , phys . lett . * 73 * , 2899 ( 1994 ) . t. someya , h. akiyama , and h. sakaki , phys . 76 * , 2965 ( 1996 ) .

it is generally expected that in a full description of quantum gravity the geometry of space - time at small scales will not resemble that of a manifold . the nature of this `` quantum geometry '' is one of the fundamental issues in the search for a quantum theory of gravity . one argument for quantum geometry is that quantum field theories give our best descriptions of microscopic physics , yet they must be regularized in order to yield meaningful predictions . the trouble being that quantum field theory suffers from ultra - violet divergences due to physical processes occurring at arbitrarily large momenta , or equivalently , arbitrarily small distances . some fundamental regularization that fixes these divergences can probably be interpreted as a modification of geometry at extremely short distances , perhaps even such that the concept of arbitrarily small distances is meaningless . another argument extends heisenberg s classic _ gedankenexperiment _ in support of the uncertainty relations . the observation of structures at very small distances requires radiation of very short wavelength and correspondingly large energy . attempting to observe a sufficiently small structure would thus require such a high concentration of energy that a black hole would be formed and no observation could be made . if this is so , then distances below about the plank scale are unobservable and thus operationally meaningless . if short distances are meaningless , then perhaps precise locations are as well . this suggests the possibility of uncertainty relations between position and position , analogous to the standard ones between position and momentum ; this has been argued on physical grounds ( see @xcite , or @xcite for a review ) and from string theory ( see @xcite ) . an uncertainty relation between , say , @xmath4-position and @xmath5-position , would mean that the @xmath4 and @xmath5 coordinates do not commute . since coordinates are just functions on space(-time ) this suggests that the algebra of functions on space might not be commutative ( see @xcite ) . that is the fundamental idea of noncommutative geometry . in noncommutative geometry ( see @xcite ) , familiar geometric concepts ( metric , measure , bundle , etc . ) are reformulated in an entirely algebraic way . this allows the generalization of geometry by replacing the algebra of functions on space with a noncommutative algebra . it may be that such a noncommutative generalization of ordinary geometry can describe the true quantum geometry of space - time . as i will explain , noncommutativity is no guarantor of regularization ( see also @xcite ) . in this paper , i will discuss a specific class of noncommutative geometries which do have the requisite regularization property . these models are not physically realistic ; they generalize euclidean ( space ) rather than lorentzian ( space - time ) geometry ( a sin shared by lattice models ) and they are not gauge theories . however , it is plausible that more realistic models may share some of the characteristics of these ones . although regularization in these geometries is quite manifest , a toolkit for coaxing actual predictions from field theory there has been lacking . my aim here is to present an approximation technique for field theory calculations in this regularization . this will show the leading order effects of noncommutativity on quantum field theory . the existing theory of noncommutative geometry ( see @xcite ) is largely inspired by the so - called gelfand theorem . according to this theorem , there is an exact correspondence between commutative c@xmath6-algebras and locally compact topological spaces . for any locally compact topological space , the algebra of continuous functions vanishing at @xmath7 is a commutative c@xmath6-algebra , and any commutative c@xmath6-algebra can be realized in this way . this suggests that c@xmath6-algebras in general should be considered as noncommutative algebras of `` continuous functions '' , and thus that the category of c@xmath6-algebras is the category of noncommutative topologies . the next step is to go from noncommutative topology to noncommutative geometry ; the most versatile noncommutative version of a riemannian metric is given by a dirac operator . on an ordinary manifold , taking the commutator of the dirac operator , @xmath8 , with a differentiable function gives @xmath9_- = i\gamma^jf_j \mbox.\ ] ] taking the norm of this gives @xmath10_-\rvert } = { \lvert\nabla f\rvert}$ ] ; thus , the dirac operator can detect the maximum slope of a function . from this , a construction for the distance between two points can be obtained ( see @xcite ) . this shows that the dirac operator contains all information of the riemannian metric . it also knows which functions are differentiable , smooth , lipschitz , etc . a c@xmath6-algebra with a dirac operator thus constitutes a noncommutative riemannian geometry . using a couple of additional structures , it is possible to characterize noncommutative riemannian manifolds axiomatically ( see @xcite ) . unfortunately , it is only known how to do this for what amounts to metrics of _ positive definite _ signature . this does not allow for the noncommutative generalization of space - time . another problem with these noncommutative manifolds is that they _ do not _ tend to regularize quantum field theory . a variation of noncommutative geometry that does have this property is matrix geometry ( see @xcite ) . in matrix geometry , every structure has only finite degrees of freedom . unfortunately , there is no axiomatic characterization of matrix geometries , as for noncommutative manifolds . the problem of characterizing noncommutative space - time is daunting ( see @xcite ) . the trouble is that the analytic properties of the dirac operator which are essential in the positive - definite case are not there in space - time . with everything finite , all analytic considerations evaporate in the case of matrix geometry , suggesting that the problem of noncommutative space - time might be solvable once matrix geometry is better understood . most existing applications of noncommutative geometry to physics have concerned connes - lott models . in these ( with the question of space - time deferred ) a simple type of noncommutative manifold provides an interesting interpretation of the standard model of particle physics . in particular , the higgs field and gauge bosons are unified . here , i am pursuing a different way of applying noncommutative geometry to physics . i am following @xcite and exploring the regularization effects of matrix geometry . i will begin in sec . [ regularization ] by discussing what it takes to regularize quantum field theory and describing the geometric quantization construction that is the basis of my approach . in sec . [ regularized.action ] , i discuss how to construct the action functional in this regularization , in slightly greater generality than has previously been discussed explicitly . this is followed by a brief speculation on convergence when the regularization is removed . perturbation theory in this regularization has not been described in detail before ; i present this in sec . [ feynman ] . in sec . [ scales ] , i explain ( heuristically ) how the infra - red and ultra - violet cutoff scales are balanced around the noncommutativity scale . in particular , the ultra - violet cutoff only exists when there is an infra - red cutoff . in sec . [ flat ] , i describe the weyl quantization of flat space and the effect of this on quantum field theory , giving a geometric algorithm for the modification . this is a prelude to the main result of this paper . i present , in sec . [ deformation ] , an approximation technique for perturbative calculations in this regularization . i illustrate exact and approximate calculations with a few examples in sec . [ examples ] , and in sec . [ divergences ] , i discuss the effect on noncommutativity on degrees of divergence . a euclidean quantum field theory can be defined by a path integral over the space of classical field configurations . given an action functional @xmath11 $ ] , the vacuum expectation value of some functional , @xmath12 $ ] , of the quantum fields is defined by z0f [ ] 0:= _ f [ ] e^-s [ ] , [ pathint]where the partition function , @xmath13 , is a normalizing factor such that @xmath14 . the celebrated divergences which plague quantum field theory are primarily due to the fact that the space of classical field configurations , @xmath15 , is infinite - dimensional ; this leaves the functional integral measure @xmath16 formal and awkwardly ill - defined . in the usual treatment of quantum field theory , perturbative feynman rules are derived from the formal path integral . unfortunately , these feynman rules typically lead to infinite results . in order to get meaningful results from computations , the feynman rules are usually regularized _ ad hoc_. this is quite effective for perturbative calculations since the details are independent of the choice of regularization . however , reality is not a perturbation . physical phenomena such as quark confinement are not reflected in strictly perturbative theories . a complete description of reality will presumably involve a nonperturbative regularization . in a euclidean quantum field theory , a nonperturbative regularization is implemented at the level of the path - integral rather than perturbation theory . one approach to regularization is to replace the original space of field configurations with some finite - dimensional approximation ; this essentially guarantees a finite theory . if @xmath17 is a single scalar field on a compact manifold , @xmath18 , then the space of field configurations is the algebra of smooth functions , @xmath19 . where algebra goes , other structures will surely follow ; for this reason , and simplicity , i shall largely restrict attention to a scalar field . to regularize , we would like to approximate @xmath20 by a finite dimensional algebra . the standard approach is to use the algebra of functions on some finite set of points a lattice which approximates @xmath18 . unfortunately , a lattice is symmetry s mortal enemy . if the space @xmath18 possesses a nontrivial group of isometries , it would be desirable to preserve these as symmetries in the regularized theory ; but for instance , in a lattice approximation to @xmath1 , the best possible approximation to the @xmath21-symmetry is the 60-element icosahedral group . we can maintain much greater symmetry with noncommutative approximating algebras . geometric quantization provides a method of constructing noncommutative approximations . as the name suggests , geometric quantization was originally intended as a systematic mathematical procedure for constructing quantum mechanics from classical mechanics . geometric quantization applies to a symplectic manifold ( originally , phase space ) with some additional structure ( a `` polarization '' ) . the terminology of quantization is unfortunate here , since i am concerned with quantum field theory . insofar as geometric quantization goes here , the manifold is not to be interpreted as phase space , the hilbert spaces are not to be interpreted as spaces of quantum - mechanical states , and the algebras do not consist of observables . in this paper , `` quantization '' and `` quantum '' have nothing to do with each other . here i shall use geometric quantization with a `` complex polarization '' . for a compact khler manifold , @xmath18 , this generates a sequence of finite - dimensional matrix algebras @xmath22 which approximate the algebra @xmath20 in a sense that i shall explain below . a khler manifold is simultaneously a riemannian , symplectic , and complex manifold . these structures are compatible with each other , such that raising one index of the symplectic 2-form , @xmath23 , with the riemannian metric gives the complex structure . ] i shall assume , for a length @xmath25 ( roughly , the size of @xmath18 ) , that the integral of @xmath26 over any closed 2-surface is an integer ; this implies the existence of a line bundle ( 1-dimensional , complex vector bundle ) , @xmath27 , with a connection @xmath28 and curvature @xmath29 . if @xmath18 is simply connected , then @xmath27 is unique . there is also a unique fiberwise , hermitian inner product on @xmath27 compatible with the connection . given two sections @xmath30 , their inner product is a function @xmath31 . compatibility with the connection means that for smooth sections , @xmath32 . using the fiberwise inner product , we can construct a global inner product , : = _ | .[inner.product]it is an elementary property of any khler manifold that the riemannian volume form can also be written in terms of the symplectic form as @xmath33 , where @xmath34 . a holomorphic section of @xmath27 is one satisfying the differential equation @xmath35 , or in index notation @xmath36 . the space of holomorphic sections @xmath37 is finite dimensional , and the inner product , , makes it a hilbert space . with this notation and structure , i can now present the geometric quantization construction . the tensor power bundle @xmath38 is much like @xmath27 ; it is a line bundle with an inner product , but with curvature @xmath39 . as @xmath40 increases , the spaces of holomorphic sections are increasingly large , finite - dimensional hilbert spaces , @xmath41 the dimension of @xmath42 grows as a polynomial in @xmath40 ( given by the riemann - roch formula , eq . ) . the algebra @xmath22 is now defined as @xmath43 that is , the space of @xmath44-linear maps from @xmath42 to itself in other words , matrices over @xmath42 . the inner product on @xmath42 gives @xmath22 an involution ( hermitian adjoint ) @xmath45 ; this makes @xmath22 a finite - dimensional c@xmath6-algebra . the collection of algebras @xmath46 alone knows nothing of @xmath18 . in order to connect @xmath22 with @xmath18 , we will need a structure such as the toeplitz quantization map @xmath47 . for any function @xmath48 , the matrix @xmath49 is defined by giving its action on any @xmath50 . since @xmath51 is a section of @xmath52 , the product @xmath53 is also a ( not necessarily holomorphic ) section of @xmath38 . using the inner product we can project @xmath53 orthogonally back to @xmath42 and call this @xmath54 . this implicitly defines @xmath55 . if @xmath18 were a phase space then @xmath49 would be interpreted as the quantum observable , @xmath56 , corresponding to @xmath57 . the toeplitz maps have the important property of being _ approximately _ multiplicative ; that is , for any two continuous functions on @xmath18 , @xmath58 according to rieffel @xcite , quantization should be expressed in terms of a continuous field of c@xmath6-algebras . a continuous field @xmath59 of c@xmath6-algebras over a compact topological space @xmath60 is essentially a vector bundle whose fibers are c@xmath6-algebras ( see @xcite ) . the space of continuous sections @xmath61 is a c@xmath6-algebra . for every point @xmath62 , the evaluation map from @xmath61 onto the fiber over @xmath63 is a @xmath64-homomorphism ( a c@xmath6-algebra map ) . the product of a continuous function on @xmath60 with a continuous section of @xmath59 is again a continuous section of @xmath59 . a continuous field of c@xmath6-algebras is completely specified by describing its base space , its fibers , and which sections are continuous . the geometric quantization of @xmath18 leads to a natural continuous field of c@xmath6-algebras , @xmath59 . the fibers of @xmath59 are each of the algebras @xmath22 and @xmath65 . this is one algebra for every natural number @xmath66 , plus one extra . the notion is that the sequence of algebras @xmath67 tends toward @xmath65 , so the natural topology for the base space is the one - point compactification @xmath68 , in which @xmath7 is the limit of any increasing sequence . the field @xmath59 is implicitly defined by the requirement that for every @xmath48 , there exists a section @xmath69 whose evaluation at @xmath40 is @xmath49 and at @xmath7 is @xmath57 . the existence of this continuous field is the sense in which the algebra @xmath22 approximates @xmath65 , but we can do better than this . there is a sense in which @xmath22 approximates the algebra of smooth functions @xmath19 . as i discussed in @xcite , the field @xmath59 has a further structure as a sort of _ smooth _ field of c@xmath6-algebras . obviously , our base space , @xmath70 is not a manifold . however , there is a reasonable notion of smooth functions on @xmath70 . we can identify @xmath70 with the homeomorphic set @xmath71 , and define the smooth functions @xmath72 as those which are restrictions of smooth functions on @xmath73 . equivalently , a smooth function on @xmath70 is one which can be approximated to arbitrary order by a power series in @xmath74 . the smooth structure of @xmath59 is given by a dense @xmath64-subalgebra @xmath75 of `` smooth '' sections . the product of a smooth function on @xmath70 with a smooth section of @xmath59 is again a smooth section of @xmath59 . the evaluation of any smooth section at @xmath7 is a smooth function on @xmath18 . the algebra @xmath76 is essentially defined by the condition that any smooth function @xmath77 gives a smooth section @xmath78 . some of the simplest spaces to work with are homogeneous spaces those such that any point of @xmath18 is equivalent to any other point under some isometry . in a homogeneous space with semisimple symmetry group , @xmath79 , we can do many calculations by simply using group representation theory . the homogeneous khler manifolds with semisimple symmetry group are the _ coadjoint orbits_. a coadjoint orbit of a lie group , @xmath79 , is simply a homogeneous space that is naturally embedded in the dual space , @xmath80 , of the lie algebra , @xmath81 , of @xmath79 . the coadjoint orbits of @xmath79 are classified by positive weight vectors of @xmath79 . weight vectors are naturally embedded in @xmath80 , and corresponding to the weight vector @xmath82 is its @xmath79-orbit @xmath83 , the set of images of @xmath82 under the action of elements of @xmath79 . in the case of @xmath84 , the space @xmath85 is 3-dimensional , and the coadjoint orbits are the concentric 2-spheres around the origin . the weight , @xmath82 , is simply a positive number , the radius . the geometric quantization of a coadjoint orbit is especially simple and respects the action of the symmetry group , @xmath79 . the hilbert space , @xmath42 , constructed in the geometric quantization of @xmath83 carries an irreducible representation of @xmath79 . namely , the representation with `` highest weight '' @xmath86 . in the case of @xmath87 , the hilbert space , @xmath42 , carries the representation of spin @xmath88 . the algebras are again defined by @xmath89 . again , we need to tie these together into a smooth field , @xmath59 , of c@xmath6-algebras . in the case of a coadjoint orbit , this can be done by using the lie algebra structure rather than the toeplitz quantization maps . let s briefly consider what can be said about @xmath59 , given only the collection of algebras @xmath46 . the restriction of @xmath59 to @xmath90 is a rather trivial continuous field ; since @xmath91 is discrete , any section is continuous . a section of @xmath59 over @xmath91 is nothing more than a sequence of matrices , one taken from each @xmath22 . the bounded sections of @xmath59 over @xmath91 ( norm - bounded sequences ) form a c@xmath6-algebra . so , you see , we already knew the restriction of @xmath59 to @xmath91 . since @xmath42 carries a @xmath79-representation , there is a linear map @xmath92 , taking lie brackets to commutators . because the representation is irreducible , the image of @xmath81 is enough to generate the entire associative algebra @xmath22 . in fact , @xmath22 can be expressed in terms of generators and relations based on this . let @xmath93 be a basis of self - adjoint generators of the complexified lie algebra , @xmath94 . the @xmath95 s can be thought of as sections of @xmath59 over @xmath91 , but they are unbounded sections , because their norms diverge linearly with @xmath40 . we can get bounded sections by dividing by @xmath40 . the operators @xmath96 are noncommutative embedding coordinates for the coadjoint orbit , @xmath83 , of radius @xmath97 . i would like to work with a coadjoint orbit of radius @xmath25 , so i define x_i:=j_i .[x ] we can construct @xmath98 as the c@xmath6-subalgebra of bounded sections of @xmath59 over @xmath91 that is generated by the @xmath99 s . this implicitly defines @xmath59 as a continuous field . the algebra @xmath22 can be expressed in terms of the generators @xmath95 or @xmath99 and three types of relations . in the @xmath87 case , the @xmath95 s are the three standard angular momentum operators , and @xmath100 . the first relations are the commutation relations that define the lie algebra . in the @xmath87-case , these are @xmath101_-=ij_3 $ ] and cyclic permutations thereof . in terms of the @xmath99 s these relations have a factor of @xmath74 on the right hand side , as @xmath102_-=\frac{i}2rn^{-1}x_3 $ ] . in the limit of @xmath103 , the relations simply become that the @xmath99 s commute . the second relations are casimir relations . these enforce that the various casimir operators have the correct eigenvalues . in the @xmath87-case , there is only one independent casimir , the quadratic one . the relation is @xmath104 , or in terms of the @xmath99 s , @xmath105 . the third relations are the serre relations . these enforce finite - dimensionality . in the @xmath87-case , this can be expressed as @xmath106 . the serre relations are equivalent to the requirement that @xmath22 is a c@xmath6-algebra , and become redundant in the @xmath103 limit . heuristically at least , in the case of @xmath87 , as @xmath103 , the relations become that the @xmath99 s commute and satisfy @xmath107 . clearly , this does generate the algebra of functions on @xmath1 of radius @xmath25 . in general , as @xmath103 , the @xmath99 s commute and satisfy polynomial relations which determine the relevant coadjoint orbit . only a few facts about the smooth structure of @xmath59 will be needed later . specifically , the @xmath99 s and @xmath74 are smooth sections , and any smooth section vanishing at @xmath108 is a multiple of @xmath74 . for a single real scalar field , @xmath109 , the general , unregularized action functional is s [ ] : = _ , [ general.action ] where @xmath110 is the volume form on @xmath18 , and @xmath111 is a lower - bounded polynomial self - coupling . complex conjugation on @xmath19 corresponds to the adjoint on @xmath22 , so our approximation to the space of real functions on @xmath18 will be the subspace @xmath112 of self - adjoint elements . to construct the regularized theory , we need a regularized action defined on @xmath113 that approximates . let s be precise about what it means to approximate an action functional in this way . we need a sequence of action functionals , @xmath114 , which converge to the unregularized action @xmath115 . this is nontrivial to define because these are functionals on different spaces . let @xmath116 be an arbitrary , self - adjoint , smooth section , and denote its evaluations as @xmath117 ; the sequence @xmath118 can be considered to converge well to the smooth function @xmath119 . my definition for convergence of @xmath120 is simply that for any such @xmath17 , @xmath121 \mbox.\ ] ] several ingredients are needed to construct a regularized action . the simplest is the product . it is the most elementary property of geometric quantization that multiplication in @xmath19 is approximated by multiplication in @xmath22 . the normalized trace on @xmath22 approximates the normalized integral on @xmath18 . that is , for any @xmath122 , @xmath123 the unregularized kinetic term can be written in terms of the laplacian , @xmath124 , using the elementary identity , @xmath125 mimicking this , we can safely write the regularized kinetic term as @xmath126 \mbox,\ ] ] since in fact , _ any _ quadratic functional can be written in this form . in the coadjoint orbit case , which i will mainly discuss , the laplacian is simply a multiple of the quadratic casimir operator . however , the discussion in this section will apply equally to any khler manifold for which a suitable approximate laplacian can be constructed . for this reason , i will write the approximate laplacian as @xmath127 until a more explicit form is required for examples . the regularized action approximating is ( using @xmath128 ) s_n ( ) = .[regact]this regularized action was originally formulated by grosse , klimk , and prenajder @xcite in the special case of @xmath1 ; however , the corresponding perturbation theory has not previously been discussed in detail . using the action , we can now define the regularized theory using a path integral formula , z0f ( ) 0:= _ _ n^ f ( ) e^-s_n ( ) d.[pathint2]this is formally identical to eq . , but it differs in that it is not merely a formal expression . the measure @xmath129 is simply the lebesgue measure on the finite - dimensional vector space @xmath113 . because @xmath130 increases at least quadratically in all directions , eq . is finite for all polynomial functionals @xmath131 . in the standard lattice regularization , it is necessary to verify that a theory is sufficiently well behaved in the `` continuum limit '' as the regularization is removed . the limit of removing the regularization in the present case is the commutative limit , @xmath103 . at this stage , it is not entirely clear what the correct definition of convergence in the commutative limit should be . certainly , renormalization is necessary . that is , the bare mass , @xmath132 , and coupling constants ( the coefficients in @xmath111 ) must depend on @xmath40 , and the field @xmath17 must also be renormalized by an @xmath40-dependent factor . some condition of convergence is then applied to the sequence of renormalized , regularized theories . a plausible form of the convergence condition is in terms of the one - particle irreducible generating functionals , @xmath133 . these are functions @xmath134 , derived from the path - integral . if the generating functionals are renormalized so that @xmath135 , then the condition may be that for any smooth , self - adjoint section @xmath116 , with evaluations @xmath117 , the sequence @xmath136 is convergent . the derivation of feynman rules from the path integral can now proceed in essentially the standard , heuristic way ( see , e. g. , @xcite ) , except that now it is not just a formal calculation . before we can discuss perturbation theory on a noncommutative space , we must first understand what green s functions are from an algebraic perspective . i begin in greater generality than just a scalar field . in general , the space of classical field configurations , @xmath15 , need not be a vector space . such is the case for nonlinear @xmath137-models . however , for a free field theory , @xmath15 is always a vector space . since we are going to apply perturbation theory about a free field theory , we must assume that @xmath15 is a vector space here . a green s function is the vacuum expectation value of a product of fields . an expectation value of a quantum field @xmath138 is an element of the space of classical field configurations , @xmath15 , since @xmath138 is thought of as a quantum field valued in @xmath15 . an expectation value of a product of two fields is an element of the ( real ) tensor product @xmath139 . carrying on like this , we see that a @xmath140-point green s function is an element of the real tensor power @xmath141 . the use of real tensor products here is actually not restrictive ; for a complex field , @xmath17 , a real tensor product will include all possible combinations of @xmath17 and its conjugate field . actually , i am overoptimistically allowing a great deal to fall into the ambiguity in a tensor product of infinite dimensional spaces . in the case of a real scalar field , @xmath142 , so the tensor product @xmath139 should intuitively be a space of functions of two points on @xmath18 . indeed , a 2-point green s function for a scalar field is a function of two points ; however , it has a singularity where the points coincide . this shows that the tensor product needs to be interpreted liberally . fortunately this issue is irrelevant to the case at hand . once @xmath15 is finite - dimensional , there is no ambiguity in the tensor product . in discussing divergences , one deals primarily with one - particle - irreducible ( 1p.08emi ) green s functions . these can be constructed as derivatives of an @xmath73-valued generating functional on @xmath15 . this shows immediately that a @xmath140-point , 1p.08emi greens function is a linear map from @xmath141 to @xmath73 . one - particle - irreducible green s functions thus live in the dual space of the corresponding ordinary green s functions . because of the coefficient in front of the action , there are powers of ( in the case of ) @xmath143 coming from the vertices and propagators . if , instead of setting @xmath144 , we had kept explicit factors of @xmath145 , then the combination @xmath146 multiplying the action would be the only appearance of @xmath145 in the functional integral . this means that the overall power of @xmath147 for a feynman diagram is the same as the overall power of @xmath145 namely , the number of loops plus @xmath148 . actually , implicit in most action functionals is an inner product on @xmath15 . this means that @xmath15 can be more or less naturally identified with a subspace of @xmath149 in general , and with all of @xmath149 when it is finite - dimensional . this inner product on @xmath15 tends to suffer an ambiguity of normalization . the most natural inner product of two functions is given by multiplying them and integrating . the most natural inner product on @xmath22 is given by multiplying matrices and taking the trace . unfortunately , these inner products disagree when we pair @xmath148 with itself . in @xmath65 that gives @xmath150 ; in @xmath22 it gives @xmath151 . we must correct the inner product on @xmath22 by a factor of @xmath147 , for consistency with that on functions . the natural form of the 1p.08emi green s functions is not actually the most useful normalization for comparison with the results of standard perturbation theory . in practice , we deal with ordinary feynman diagrams in momentum space . the quantities we usually deal with are not the momentum - space green s functions themselves , but have an overall , momentum - conserving @xmath152-function divided out . consider 2-point green s functions in a scalar theory . in momentum space , with the @xmath152-function divided out , these are simply functions of a single momentum . multiplying two such functions in momentum space corresponds to convolution of green s functions . in this normalization , a 2-point green s function is a convolution operator ; that is a linear map @xmath153 . in general , a @xmath140-point 1p.08emi green s function in this normalization is a linear map @xmath154 . in the regularized theory , this change simply amounts to dividing our amplitudes by @xmath147 . this simplifies the feynman rules slightly , so that the overall power of @xmath147 is now simply the number of loops . i ll now specialize to the theory given by the action , although most of the considerations are more general . as usual , the action splits into a free part ( the first two terms ) and an interaction part ( the @xmath111 term ) . the free part gives the propagator @xmath155 ; each monomial of @xmath156 gives a vertex whose valence is the degree of the monomial . a vertex of valence @xmath157 essentially represents the trace of a product of @xmath157 matrices . the term corresponding to a standard feynman diagram is a sum of terms with different orderings of the products . it is convenient to represent these subterms diagrammatically . for each vertex , the multiplicands correspond to incoming edges . because only the cyclic order matters in a trace , we only need to indicate a cyclic order to the edges . this is easily done graphically by drawing the vertex in the plane . the order of multiplication is indicated by the counterclockwise order of the attached edges . the distinct terms are thus labeled by `` framings '' of the feynman diagram in the plane . in ordinary quantum field theory involving a real field , there are symmetry factors to deal with . the contribution of a given diagram is divided by the number of its symmetries . in this case there is an additional type of combinatorial factor present . since a given ordinary feynman diagram corresponds to several framed diagrams , these framed diagrams are weighted by coefficients adding up to @xmath148 . determining these coefficients is a matter of enumerating the cyclic orientations of all vertices , and sorting the resultant framings into equivalence classes . to express the exact feynman rules , it is convenient to adapt the notation invented by t hooft for discussing the large @xmath40 limit of @xmath158-gauge theories ( see @xcite ) . in that diagrammar , the gluon propagator is represented by a double line ( two directed lines in opposite directions ) . an outgoing arrow indicates an upper index and an ingoing arrow indicates a lower index . the way lines are connected indicates how indices are contracted . in these diagrams , the two lines of a gluon propagator do not touch . this is appropriate , since they really have nothing to do with each other . the propagator is not just invariant under @xmath158 , but under 2 separate actions of @xmath158 corresponding to the 2 separate edges . in the present context , however , the notation needs to be modified . the propagator is not invariant under arbitrary unitary transformations ; it is only invariant under the isometries of @xmath18 ( if there are any ) . the two lines of the propagator are thus no longer independent , and i indicate this by linked double lines as shown in fig [ propagator ] . ] the upper indices ( outgoing arrows ) are factors of @xmath42 , while the lower indices ( ingoing arrows ) are factors of @xmath159 . each factor of @xmath160 thus gives an upper and a lower index , or an incoming and an outgoing arrow . ] figure [ doubled ] shows a more complicated doubled diagram . note that no distinction is made between overcrossings and undercrossings . a reader might reasonably question that this is truly a regularization of _ real _ scalar field theory . after all , the subspace @xmath112 is not closed under multiplication , so why should these feynman rules respect this subspace ? the issue is whether the green s functions are self - adjoint , in the obvious sense for elements of a tensor power of @xmath22 . in fact , a given ( framed ) feynman diagram may not be self - adjoint . however , its adjoint is the mirror - image diagram , which is another framing of the same diagram and therefore contributes to the same green s function . this makes the green s functions themselves self - adjoint , order - by - order in perturbation theory . this is a convenient point at which to introduce some further notation . there are three different length scales that are pertinent to quantum field theory on a quantized , compact space , and there are parameters characterizing each of these scales . i have already introduced the length @xmath25 . this characterizes the overall size of the manifold @xmath18 . it is relevant to quantum field theory as the infra - red cutoff scale ; that is , there do not exist modes of wavelength more than about @xmath25 . the second parameter , @xmath161 , characterizes the scale of noncommutativity . as i have mentioned ( sec . [ gq ] ) , geometric quantization was originally intended as a tool for deriving quantum mechanics from classical mechanics , so there is an analogy between some constructions here and in that problem . in this analogy , @xmath162 corresponds to @xmath145 . like a classical phase space , the khler manifold @xmath18 has a poisson bracket . the poisson bracket of two differentiable functions on @xmath18 is defined as @xmath163 , where the poisson bivector , @xmath164 , is in turn the inverse of the symplectic form in the sense that @xmath165 . the analogy continues with commutation relations , @xmath166_- = -i{\kappa}t_n(\{f , g\ } ) + { \mathcal{o}}^{2}({\kappa } ) \mbox,\ ] ] or heuristically , @xmath167_-\approx -i{\kappa}\{f , g\ } \mbox.\ ] ] note that @xmath162 has the dimensions of an area ; this balances the two derivatives occurring in the poisson bracket . in quantum mechanics , noncommutativity of observables leads to uncertainty relations . likewise , noncommutativity here should intuitively lead to uncertainty relations between coordinates ( _ roughly _ , something like @xmath168 ) . this suggests that the best that we can specify a point is to an uncertainty of the order @xmath169 in all directions . naively , we might conclude from this that @xmath18 is broken up into `` cells '' of this size and that the noncommutativity effects an ultraviolet cutoff at the mass scale @xmath170 , but this is not so . actually , if we divide @xmath18 into cells of size about @xmath169 , then the number of cells will be about @xmath151 . using the fact that the todd class , @xmath171 is equal to @xmath148 plus higher degree cohomology classes , the riemann - roch formula shows , _ n= _ te^ = + ^1-n ( ) .[riemann - roch]this shows that these cells correspond to the degrees of freedom of @xmath42 rather than of the scalar field . but what is the ultra - violet cutoff scale really ? the number of degrees of freedom associated to a scalar field is @xmath172 , so the volume belonging to each degree of freedom is @xmath173 the length scale of this cutoff is thus of the order @xmath174 , or as a mass ( inverse length ) , @xmath175 the noncommutativity scale set by @xmath162 is the geometric mean between the infra - red and ultra - violet scales . if we attempt to remove the infra - red cutoff ( decompactify the space ) by sending @xmath176 without @xmath162 diverging , then we must let @xmath177 . in other words , if the infra - red cutoff is taken away , then the ultra - violet cutoff goes away . this implies that noncommutativity only achieves an ultra - violet cutoff in the presence of an infra - red cutoff ! that will be proven in sec . [ flatqft ] . this is in marked contrast to lattice regularization . for one thing , an unbounded lattice can certainly achieve an ultra - violet cutoff . in a lattice , there is a sharply defined minimum separation between points ( the lattice spacing ) ; this is also the scale of the ultra - violet cutoff . on a quantized space , there is a fuzzy minimum observable distance , but this is much larger than the length scale of the ultra - violet cutoff . just because a computation is well - defined and possible in principle , does nt necessarily mean it is easy or practicable . doing exact calculations on coadjoint orbits requires working with representations of the symmetry group . while this is better than working on a space with no symmetry at all , it is not as easy as working on flat space . for flat space , calculations in quantum field theory are considerably simplified by the fact that momentum space is a vector space . locally , any manifold looks like flat space . topologically , this is the very definition of a manifold . geometrically , this is the content of einstein s principle of equivalence . if we are concerned with issues of small - scale physics ( like renormalization ) then it would be nice to do calculations in the simplified setting of flat space and not worry about the global structure of our manifold . the heuristic arguments of sec . [ scales ] have already indicated that things are not so simple . we can not ignore global structure , because ultra - violet regularization is dependent upon the global property of compactness . nevertheless , i will introduce in sec . [ deformation ] an approximation technique which takes advantage of the local resemblance to flat space . before i can describe this approximation i must discuss flat space itself . specifically , i will discuss quantized flat space , which does not have the effect of regularizing quantum field theory , but does modify it in a relevant way . if we `` zoom in '' around any point of a symplectic ( or even poisson ) manifold , then it will resemble a flat , affine space , with a poisson bracket determined by a constant poisson bivector , @xmath164 , as @xmath178 although the symplectic case is what we are really interested in , there is no need to assume that @xmath164 is nondegenerate in this section . a tensor product of two functions on the flat space @xmath179 is naturally regarded as a function on @xmath180 . the multiplication map @xmath181 is equivalent to the diagonal evaluation map , @xmath182 so that @xmath183 . if we regard @xmath164 as a second order differential operator on @xmath180 , then the poisson bracket can be expressed as @xmath184 with this notation , the weyl product corresponding to @xmath164 is defined as @xmath185 if we treat this as simply a formal power series in @xmath162 , then @xmath186 is an associative product . the space @xmath187 $ ] is defined to consist of formal power series in @xmath162 whose coefficients are smooth functions on @xmath179 . the formal deformation quantization algebra @xmath188 is @xmath187 $ ] with the product @xmath186 . unfortunately , if we insert an arbitrary pair of smooth functions into eq . , the series will typically diverge . fortunately , there is a sufficiently large space of functions for which @xmath186 does converge , that we can construct a sensible , concrete quantization of @xmath179 from this . the archetypal functions for which eq . is convergent are plane - wave functions . the weyl product of two plane - wave functions is simply e^-ipx * _ e^-iqx = e^i2\{p , q}e^-i(p+q ) x , [ expprod]using the shorthand @xmath189 since this combination will occur frequently . this notation is justified by the fact that if we think of @xmath190 and @xmath191 as linear functions on @xmath179 , then @xmath192 really is their poisson bracket . the `` good '' subalgebra @xmath193 consists of those functions whose dependence on @xmath162 is entire , and whose fourier transforms ( on @xmath179 ) are compactly supported . @xmath194 is algebraically closed , and consists of _ convergent _ power series in @xmath162 , so we can actually assign @xmath162 a concrete value . the mathematically sanctioned way to assign @xmath162 a concrete value , @xmath195 , is to quotient @xmath194 by the ideal generated by @xmath196 . the quotient algebra @xmath197 can then be completed to a c@xmath6-algebra . this is the concrete weyl quantization of @xmath179 at @xmath195 . this is the same algebra that would be obtained by geometric quantization . if we tried to construct the concrete weyl algebra directly from @xmath198 , we would have failed because the ideal generated in @xmath198 by @xmath196 is all of @xmath198 . the quotient @xmath199 is thus trivial . we can construct perturbative quantum field theory on quantized @xmath179 . the derivation of feynman rules is formally the same as on a quantized compact space ; because of noncommutativity , we still have to distinguish cyclic orderings of edges around vertices . however , there will again be divergent feynman diagrams which demand regularization ; thus justifying the claim that ultra - violet regularization is contingent upon infra - red regularization . in terms of momentum space , the feynman rules for vertices are modified by momenta dependent phase factors . these can be understood geometrically . in momentum space , which is dual to the position space @xmath179 , the bivector @xmath164 becomes a 2-form . the @xmath200 in eq . is precisely the flux of @xmath201 through the triangle formed by @xmath190 and @xmath191 ( see fig . [ triangle ] ) . ] for a valence @xmath157 vertex draw an @xmath157-sided polygon in momentum space such that the difference of the ends of a side is equal to the ingoing momentum of the corresponding propagator line . momentum conservation requires that the momenta add up to @xmath202 , which ensures that the polygon is closed . decomposing the polygon into triangles shows that the phase associated to the vertex is @xmath162 times the flux of @xmath201 through the polygon . note that this polygon is only fixed modulo an overall translation . in conventional euclidean real scalar field theory , the amplitudes are real ; the same is true here . we must again remember to sum over all framings of a feynman diagram . the phases lead to cosines of products of momenta . if the evaluation of a feynman diagram involves phases depending on internal momenta , then the resulting oscillatory integral may give a finite result where there once was a divergence . however , consistent with the heuristic argument about @xmath177 in sec . [ scales ] , this will not eliminate all divergences . ] planar diagrams remain just as divergent as those for commutative @xmath179 . the following proof is equivalent to that already given by filk @xcite ; however , i interpret it geometrically rather than in terms of `` special graph - topological properties of cocycles '' . consider a planar feynman diagram @xmath203 , such as that shown in fig . we can construct a dual polygonalization ( 2-dimensional cw - complex ) @xmath204 . this has a vertex for each open space in the planar rendering of @xmath203 , including both spaces enclosed by internal edges and spaces between external edges . the edges of @xmath204 are in one - to - one correspondence with the edges of @xmath203 . the 2-cells ( polygons ) of @xmath204 correspond to the vertices of @xmath203 . just as for a single vertex , we can embed @xmath204 in momentum space so that the separation between adjacent vertices of @xmath204 is the momentum associated to the connecting edge . equivalently , we can construct a polygon for each vertex of @xmath203 ( as above ) and fit these together . the total phase associated to @xmath203 is then given by the total flux of @xmath205 through all polygons of @xmath204 . however , @xmath164 is a constant and thus closed differential form , so the the flux only depends on the shape of the boundary of @xmath204 in momentum space . thus , the phase only depends on the momenta of the external lines . indeed the phase is the same as if the external lines entered a single vertex . a generalization of this construction to non - planar diagrams will be employed in sec . [ divergences ] . the deformation quantization of @xmath179 in sec . [ qflat ] is the archetype of a more general construction ( see @xcite for overview ) . in the formal deformation quantization of a manifold @xmath18 , the product @xmath186 is a formal power series in @xmath162 whose terms are bidifferential operators on @xmath18 ( as in eq . ) . just as for flat space , the algebra @xmath206 is equivalent as a vector space to the space , @xmath207 $ ] , of formal power series in @xmath162 with smooth functions as coefficients . in general , a deformation quantization can always be constructed to fit @xmath167_-=-i{\kappa}\{f , g\ } \mod { \kappa}^2 \mbox,\ ] ] for any poisson bracket on @xmath18 ; see @xcite . a deformation quantization can also be derived from a geometric quantization . geometric quantization can be loosely thought of as making the product of functions on @xmath18 variable ( dependent on @xmath162 ) ; the corresponding deformation quantization is the result of asymptotically expanding the product as a power series in @xmath162 . as i explained in @xcite , this algebra can be constructed from the smooth field of c@xmath6-algebras , @xmath59 , given by geometric quantization . the parameter @xmath162 is itself a smooth function on @xmath70 and vanishes at the point @xmath108 . in fact , any smooth section of @xmath59 which vanishes at @xmath7 is a multiple of @xmath162 . the space of smooth sections which vanish to order @xmath208 at @xmath7 is thus @xmath209 ; this is a 2-sided ideal . in the quotient algebra @xmath210 ( the algebra of `` jets '' about @xmath7 ) , the variability of the product is preserved to order @xmath211 . these quotient algebras naturally form an algebraic inverse system . taking the algebraic inverse limit gives the deformation quantization algebra corresponding to @xmath59 , @xmath212 we have just seen how to construct a deformation quantization from the geometric quantization . in sec . [ qflat ] , i described the opposite process reconstructing the geometric quantization from a deformation quantization in the case of flat space . this involved a `` good '' subalgebra @xmath193 . it may or may not be possible to make an analogous construction in all cases , but it can be done for coadjoint orbits . in this section i will describe a good subalgebra of @xmath213 . let @xmath59 be the smooth field given by geometric quantization of the coadjoint orbit @xmath83 . the sections defined in eq . are smooth sections , @xmath214 . since smooth sections form an algebra , any product of @xmath99 s , or linear combination thereof , is also a smooth section . such sections form a subalgebra i denote @xmath215 , the polynomial sections . since this is a subalgebra of the smooth sections , there is , for each @xmath208 , a natural homomorphism _ ( , ) ^(,)/^j+1 .[inclusion.quotient]the kernel of this is just @xmath216 , which vanishes as @xmath217 . when the @xmath217 limit is taken , becomes a natural , injective homomorphism , @xmath218 thus , @xmath215 is a subalgebra of both @xmath98 and @xmath198 . in the identification of @xmath213 with @xmath219 $ ] as a vector space , @xmath215 corresponds to @xmath220 $ ] , the space of polynomials in @xmath162 whose coefficients are polynomial functions on @xmath83 . this is the good subalgebra i wanted . in @xmath198 , @xmath221 is invertible , so the ideal it generates is all of @xmath198 ; thus we can not assign @xmath162 a concrete value in @xmath198 . on the other hand , the functions of @xmath162 in @xmath215 are all polynomials ; therefore , @xmath221 is not invertible and generates a nontrivial ideal in @xmath215 . we can meaningfully take the quotient algebra , @xmath222 in terms of the generators and relations representation of @xmath22 in sec . [ coadjoint ] , the quotient algebra @xmath223 is what we get by discarding the serre relations . there is therefore a natural surjective homomorphism e:_n^-14mu_n.[truncation]by characterizing the kernel of @xmath224 , we can effectively reconstruct the geometric quantization from the deformation quantization . as a @xmath79-representation , @xmath223 is indistinguishable from the space of polynomial functions on @xmath83 ; this is the direct sum of all irreducible @xmath79-representations appearing in @xmath225 . in @xmath22 , this lattice of irreducible representations is cut off ; @xmath22 is finite - dimensional . the kernel of @xmath224 is spanned by those @xmath79-representations which occur in @xmath225 , but not in @xmath22 . this is the key to doing approximate calculations . imagine that @xmath25 is very large . if we sit at some point of @xmath83 , then the region around us appears very close to the flat space @xmath226 ( where @xmath227 ) . we would like to describe the geometric quantization of @xmath83 in this approximation , but the geometric quantization of the noncompact space @xmath226 is very different from that of compact @xmath83 . their deformation quantizations , however , are similar , because the deformation quantization product is constructible locally from bidifferential operators . the geometric quantization of @xmath83 can be approximated using the weyl quantization of @xmath226 , and an approximation for the cutoff on representations . i refer to this as the deformation approximation if we are concerned with some field theory on a quantized coadjoint orbit then we would like to exploit the deformation approximation and the relationship with flat space in order to approximate the values of feynman diagrams for large @xmath40 . to accomplish this , we must characterize the kernel of the surjective homomorphism @xmath224 in in terms of momentum space . consider the case of @xmath1 . as an @xmath87-representation , @xmath228 contains all the irreducible representations of integer spin . on the other hand , @xmath22 is the direct sum of representations of integer spin @xmath229 . the kernel of @xmath224 consists of those representations of @xmath87 which are contained in @xmath228 but not @xmath22 . that means modes whose spin exceeds @xmath40 . the laplacian on @xmath1 with radius @xmath25 is @xmath230 , so the eigenvalue on a spherical harmonic with spin @xmath231 is @xmath232 . it is thus possible to characterize @xmath233 in terms of the laplacian : @xmath233 is spanned by the eigenfunctions of @xmath127 with eigenvalue greater than @xmath234 . now take the flat space approximation . in terms of momentum space , an eigenvalue of @xmath127 is simply the magnitude - squared of a momentum vector . modes with momentum greater than @xmath235 belong to the kernel of @xmath224 . we can approximate the algebra @xmath22 by the weyl product on @xmath236 , with modes of momentum greater than @xmath235 set to @xmath202 . as i have said , geometric quantization can be thought of as a modification of the product . in the feynman rules , products occur at the vertices . _ ab initio _ , one might expect that the feynman rules for the deformation approximation would differ from the flat space feynman rules only at the vertices . however , it is actually more convenient to shift some of the modification to the propagators . in the deformation approximation for @xmath1 , the product of a sequence of plane - wave functions is equal to their weyl product , _ if _ all the momenta have magnitude @xmath237 , otherwise the product is @xmath202 . in terms of the feynman rules , this translates into restricting ( `` cutting off '' ) momentum integration to the region in which all momenta have magnitude @xmath237 . in general , the cutoff can be more complicated . momenta are restricted to some region of size @xmath235 . this `` shape '' of the cutoff depends on the particular coadjoint orbit being considered , although it is significantly limited by symmetry . in our approximation of zooming in around one point of @xmath83 , there is still a symmetry group of rotations about that point ( isotropies ) . momentum space carries a representation of the isotropy group , and the cutoff must be invariant under this . this is a difference with lattice regularization ; that is much more arbitrary because features of the cutoff are not constrained by symmetry ; here , there is no arbitrariness . as examples , i shall consider the coadjoint orbits of dimension @xmath238 . there are really only three of these : @xmath1 , @xmath239 , and @xmath3 . for @xmath1 we have just seen that the cutoff shape is @xmath240 , a disc of radius @xmath235 , which is the only convex shape allowed by symmetry anyway . for @xmath2 , the cutoff shape derives from that of @xmath1 and is clearly @xmath241 . for @xmath3 , the cutoff can again be characterized by the laplacian ; it is @xmath242 , a ball of radius @xmath235 . again , this is the only convex shape allowed by symmetry . the feynman rules in the deformation approximation are a modification of those for quantized flat space . the cutoff is implemented by modifying the propagator . i summarize the feynman rules here . for each internal edge carrying momentum @xmath190 , there is a propagator s(p ) = , [ cutoff.propagator]where @xmath243 is a step function , equal to @xmath148 inside the cutoff and @xmath202 outside . to each independent closed loop in the diagram , there is an integration over the corresponding momentum and a factor of @xmath244 , where @xmath245 is the dimension of the space . a diagram may contain vertices of valence @xmath157 if the potential , @xmath111 , contains a monomial of degree @xmath157 . to each such vertex there is a coupling constant factor coming from the coefficient of the monomial . there is also a phase factor given by the flux of @xmath205 through a polygon in momentum space formed by the momenta entering the vertex . as always , the momenta entering a vertex add up to @xmath202 . as with the exact feynman rules , there are combinatorial factors coming from the symmetries of the diagram , and from the numbers of alternative framings . as examples , i shall consider ( in progressively diminishing detail ) the evaluation of some one - particle - irreducible feynman diagrams for the @xmath0 model on quantized coadjoint orbits of small dimension . in this case , the potential is @xmath246 , so all vertices are of valence 4 and carry a factor of @xmath247 . on a coadjoint orbit , the cutoff laplacian can be expressed in terms of the appropriately normalized quadratic casimir operator as @xmath248 . this has the insurpassable property that @xmath249 . the propagator is thus @xmath250 the first diagram to consider is the planar , 1-loop propagator correction , fig . [ diag1 ] . planar , one - loop propagator correction . ] ordinarily , this diagram would have a symmetry factor of @xmath251 . however , only 2 out of 6 framings are equivalent to this diagram . therefore , there is an overall combinatorial factor of @xmath252 . this is about the only diagram that can easily be evaluated exactly . the doubled diagram is shown in fig . [ ddiag1 ] . .[ddiag1 ] ] notice that the bottom line is detached from the rest of the diagram ; this shows that the diagram factorizes as the tensor product of the identity on @xmath159 with some linear map from @xmath42 to @xmath42 . this must be @xmath79-invariant , and since @xmath42 is an irreducible @xmath79-representation , this map must be proportional to the identity . in other words , the diagram must evaluate to a number . to determine this number , close up the upper part of fig . [ ddiag1 ] and divide by @xmath253 ( a circle ) to normalize . the factor of @xmath151 cancels the factor from the feynman rules . closing up the diagram amounts to taking a trace . the exact evaluation is thus @xmath254 now turn to the deformation approximation . because the diagram is planar , the phase factor can only depend on external momenta , but because there is only one external momentum , there is no phase factor . indeed , the diagram is independent of the external momentum it is simply a number , as in the exact evaluation . in the deformation approximation , @xmath255 for the exact evaluation , it remains to calculate the trace of the propagator @xmath256 . this is an @xmath87-invariant linear operator on the algebra @xmath22 . as an @xmath87 representation , @xmath22 is the direct sum of the irreducible representations of integer spin @xmath202 through @xmath40 . the spin @xmath231 subspace has dimension @xmath257 , and on that subspace the quadratic casimir reduces to @xmath258 . using @xmath259 , the exact evaluation is @xmath260 however , this is precisely the @xmath261-part midpoint approximation to the integral _ 0^n+1 dt = [ approxs2a ] . now evaluate fig . [ diag1 ] in the deformation approximation . the momentum cutoff is a disc of radius @xmath235 . this gives @xmath262 \label{approxs2b}\mbox.\end{aligned}\ ] ] in order for the deformation approximation to be valid , we must assume that @xmath263 . this means simply that the compton wavelength ( the distance , @xmath264 , determined by the bare mass ) should be much smaller than the universe and that @xmath132 should be much smaller than the cutoff mass . we need not assume that @xmath132 is smaller than the noncommutativity scale . using the formula for the leading correction to the midpoint approximation , we can find the leading order correction to ; this is @xmath265 so , in this case , the deformation approximation indeed converges if we take @xmath266 . as an @xmath267-representation , the algebra @xmath22 decomposes into a direct sum of irreducible subspaces numbered @xmath202 through @xmath40 . the @xmath231 subspace has dimension @xmath268 , and the quadratic casimir reduces to @xmath269 . for @xmath3 of circumference @xmath270 , the volume is @xmath271 . the exact evaluation of fig . [ diag1 ] is , @xmath272 as i have said , the momentum cutoff for @xmath3 is a ball of radius @xmath235 . so , in the deformation approximation , @xmath273 \label{planarcp } \mbox.\end{aligned}\ ] ] the resemblance to eq . is hopefully apparent . here , the exact evaluation is the double sum , @xmath274 the cutoff shape is the more complicated @xmath241 . this gives the deformation approximation as , @xmath275 \label{planarss } \mbox.\end{aligned}\ ] ] this demonstrates the effect of the cutoff shape . for @xmath276 , the results of eq.s and differ by a factor of @xmath277 . the only reason for this difference is the nontrivial cutoff shape for @xmath2 . the only nonplanar 1-loop propagator correction is shown in fig . [ diag2 ] . ] this is equivalent to all 4 out of 6 other framings of the original diagram ; so , the combinatorial factor is @xmath278 . the doubled diagram is shown in fig . [ ddiag2 ] . .[ddiag2 ] ] observe that fig . [ ddiag2 ] is really just the bare propagator ( fig . [ propagator ] on p. ) with the lines rearranged . the evaluation of fig . [ ddiag2 ] is @xmath279 times the propagator with the @xmath42 s exchanged . because this is a nonplanar diagram , there is a nontrivial phase factor . this makes the evaluation more interesting . with external momentum @xmath190 , it is ( again using the brace notation of eq . ) @xmath280 where @xmath281 is the fourier transform of the cutoff propagator and @xmath24 is the complex structure ( a @xmath282-rotation ) . equation is independent of any details of the propagator ; eq . used rotational invariance . it seems that rearranging lines in the doubled diagram corresponds to rotating by @xmath24 and taking a fourier transform . since the propagator is @xmath87-invariant , it can be written as a linear combination of projectors on irreducible representations . the rearranged version is also invariant and can also be so decomposed . calculating the rearrangement comes down to transforming between these two decompositions . the coefficients for this transformation are the famous 6-@xmath208 symbols ( see @xcite ) . since the value of this diagram is an invariant linear operator on @xmath22 , it can most conveniently be described by giving its eigenvalues on the irreducible subspaces of @xmath22 . the evaluation of fig . [ ddiag2 ] , acting on the spin @xmath231 subspace is @xmath283 this corresponds to evaluating with external momentum , @xmath190 , of magnitude @xmath284 . following eq . , the deformation approximation for fig . [ ddiag2 ] is , @xmath285 if we take @xmath177 while keeping @xmath162 fixed , then this becomes a hyperbolic bessel function , @xmath286 . this has a logarithmic singularity at @xmath287 , and falls off exponentially for large @xmath190 . aside from @xmath287 , it is finite . this means that this diagram , unlike the previous , planar diagram , is actually regularized by noncommutativity alone . the deformation approximation gives , @xmath288 if we take @xmath177 , this becomes @xmath289 . as in 2 dimensions , this falls off exponentially . the singularity is of the form @xmath290 , which is rather mild in 4 dimensions . there are several distinct framings of the one - loop vertex correction diagram . one of these is shown in fig . [ diag3 ] . ] the external momenta are all incoming ; the internal momentum @xmath191 is directed counterclockwise . by momentum conservation , @xmath291 . the phase associated to this diagram can be split into two factors . the first , @xmath292}$ ] is the same as for a bare vertex with the external edges oriented in this way . the second factor is @xmath293 . except for the combinatorial factors , the evaluation of fig . [ diag3 ] in the deformation approximation is e^2 s(q ) s(p_1+p_2-q ) e^i\{p_1,q } d^2nq .[diag3val]@xmath294 again denotes the cutoff propagator . note that this essentially amounts to a fourier transform of the product of propagators . in four dimensions , the 1-loop vertex correction is divergent , but if we take @xmath177 with @xmath162 fixed then remains finite except at @xmath295 let s consider how these amplitudes diverge when regularization is removed . actually , since there are both infra - red and ultra - violet cutoffs ( controlled by @xmath25 and @xmath235 ) , there are many possible ways to remove the cutoffs , one - by - one or simultaneously . as i have said , what has been described is far from a realistic physical model . despite this , we can optimistically juxtapose this model with reality and hope that some properties of the model are pertinent to reality . the kind of noncommutativity considered here has not yet been noticed in experiments . this means that the noncommutativity scale set by @xmath162 is , at best , about the smallest length scale explored by current experiments . on the other hand , @xmath25 should be something like the size of the universe a far larger scale , so @xmath40 must be _ very _ large . this shows that the cutoff mass , @xmath235 should be well beyond the scale of @xmath162 and thus far beyond the reach of experiments . the approximation relevant to physical predictions of noncommutativity is thus the limit of @xmath177 with @xmath162 fixed . note that this means simultaneously taking @xmath176 ; in other words , we remove the ultra - violet and infra - red cutoffs in unison . in this limit , the quantized compact space becomes , at least formally , quantized flat space . we should thus expect this limit of field theory to be described by field theory on quantized flat space . as i have claimed , and the above examples have corroborated , nonplanarity of a feynman diagram tends to decrease its degree of divergence . in the standard regularizations , the ultra - violet divergences are associated with the loops in the feynman diagram . it appears that the divergences when @xmath177 with @xmath162 fixed are associated more closely with the loops in the doubled diagram index loops , in the terminology of t hooft @xcite . the number of index loops is always less than or equal to the number of loops in the original diagram . although an arbitrary framed feynman diagram , @xmath203 , may not fit into the plane , it will always fit onto some oriented surface . such a surface , @xmath296 , can be constructed systematically by filling in a 2-cell for each line in the doubled diagram . it will have the topology of a riemann surface with at least one puncture ; each external leg of @xmath203 ends at a puncture . our first example , fig . [ diag1 ] , is planar ; thus @xmath296 is the plane ( equivalently , a sphere with 1 puncture ) . for both fig s [ diag2 ] and [ doubled ] , @xmath296 is a torus with 1 puncture . for fig . [ diag3 ] , @xmath296 is a sphere with 2 punctures . by construction , an index loop is always contractible in @xmath296 . we can make this more precise using homology . the group @xmath297 is trivial because @xmath296 is not closed ; the group @xmath298 is trivial because @xmath296 is obtained from @xmath203 by attaching 2-cells . inserting these facts into the long exact sequence for the pair @xmath299 , gives the short exact sequence , @xmath300 the group @xmath301 is generated by the 2-cells in @xmath296 that do not touch punctures ; these are in one - to - one correspondence with the index loops . the group @xmath302 classifies the loops of @xmath203 . the group @xmath303 classifies incontractible loops in @xmath296 . these are all free groups , so the sequence splits ( unnaturally ) . the loops of @xmath203 can therefore be divided into index loops and incontractible loops of @xmath296 . consider some loop @xmath304 in @xmath203 which is incontractible in @xmath296 , and examine what happens when we integrate over the momentum , @xmath190 , circulating around @xmath304 . if we leave the momenta in @xmath203 otherwise fixed and then add @xmath190 at each edge of @xmath304 . ] , then the part of the phase which depends on @xmath190 will be of the form @xmath305 , where @xmath191 is a linear combination of the other momenta in @xmath203 . in fact , @xmath191 is the momentum flowing across @xmath304 . the other part of the feynman integrand depending on @xmath190 is the product of propagators on the edges of @xmath304 ; each of these has the form @xmath306 ^ 2+m^2)^{-1}$ ] . taking the integral over @xmath190 effectively means taking the fourier transform of this product of propagators . the result may have an integrable singularity at @xmath307 ( an infra - red divergence because @xmath176 ) , but is otherwise finite , and falls of exponentially . this means that integration over @xmath190 does not contribute to the ultra - violet divergence ; moreover , if @xmath191 is also an internal momentum , then integration over @xmath191 will not contribute to the divergence either . the other important limit is the commutative limit . this is when @xmath25 is fixed , but @xmath308 and @xmath177 . in light of the deformation approximation , it appears that the behavior in this limit will be very much like that of the most elementary regularization , a simple momentum cutoff . for a single , complex scalar field , the general @xmath309-invariant action is s[]:=_,[complex.action]where @xmath310 is some real polynomial , lower bounded on the positive axis . the standard perturbation theory for this complex scalar field is not much different from that of the real field . every edge of a feynman diagram is now directed , and each vertex must have an equal number of ingoing and outgoing edges . the regularized action corresponding to this is a simple generalization of the real case . the only new ingredient is the fact that complex conjugation in @xmath20 corresponds to the hermitian adjoint in @xmath22 . the regularized version of is s_n ( ) : = .[complex.regact ] in every product inside the trace in eq . , @xmath17 alternates with @xmath311 . as a consequence , in the framed diagrams , ingoing and outgoing edges alternate around each vertex . except for this restriction , the feynman rules for the complex field are formally the same as for a real field . this restriction has an interesting interpretation in terms of the doubled diagram . there are really two types of lines in the doubled diagram , those on the left of a directed edge in the original diagram , and those on the right . the restriction is simply that left edges only connect to left edges and right to right . so far , i have only discussed topologically trivial scalar fields . a section of a nontrivial vector bundle is not a function , so sections of a vector bundle are not approximated by the algebra @xmath22 . as i have argued in @xcite , the noncommutative generalization of a vector bundle is a ( finitely generated , projective ) module of an algebra , so the geometric quantization of a vector bundle should be a module of @xmath22 . for the matrix algebra @xmath312 , any module is of the form , @xmath313 , a space of linear maps from some vector space to @xmath42 . i present a general construction in @xcite . in the simplest case a holomorphic vector bundle @xmath314 ; this generalizes @xmath315 . the simplest action for a nontrivial vector field is a trivial generalization of , although it requires a fiberwise , hermitian inner product on @xmath111 . the action is simply with @xmath316 and inner products understood between successive @xmath311 s and @xmath17 s . for the regularized fields , the inner product on @xmath111 leads to an inner product on @xmath317 ; because of this , for @xmath318 , we can define the product @xmath319 to be an element of @xmath22 . the regularized action for a nontrivial field is just with @xmath318 . in the doubled diagrams , the left and right lines are now truly distinct . left lines correspond to @xmath42 ; right lines correspond to @xmath317 . the complex scalar field really just wanted to be twisted . there is not much to be said about topologically nontrivial fields in the deformation approximation . from a local perspective , a nontrivial vector bundle is just a vector bundle with a fixed background gauge field . in the limit as @xmath176 , this gauge field vanishes . fermi statistics do not pose any particular obstacle in this regularization scheme . however , the kinetic terms for fermion action functionals may be more difficult to construct than that for bosons . whereas constructing a cutoff laplacian is elementary on coadjoint orbits , it is less obvious how to deal with the dirac operator . even on @xmath1 , there may be trouble . when the spinor bundle on @xmath1 is quantized by my prescription , the left and right chiral subspaces have different dimensions . this means that there does not exist a cutoff dirac operator which both anti - commutes with the chirality operator ( @xmath320 ) and has no kernel . in fact , the difference of the dimensions of the left an right subspaces grows linearly with @xmath40 ; so , if we require the dirac operator to anti - commute with the chirality operator , then the size of the kernel will diverge as @xmath103 . whether these problems can be circumvented , or whether they are really problems at all , remains to be seen . in all this i have largely dwelt on the regularization effects of noncommutativity . however , if there are effects of noncommutativity which are directly detectable by experiment , then these will probably be tree - level effects . in field theory on quantized flat space , the phase factors from different framings add up to a sum of cosines of products of momenta . this can , for instance , cause the amplitudes for processes with certain combinations of momenta , to vanish . this paper is essentially my thesis in physics at penn . state ; i would thus like to thank the members of my committee : lee smolin , abhay ashtekar , shyamoli chauduri , and paul baum . this material is based upon work supported in part by nsf grant phy95 - 14240 , by a gift from the jesse phillips foundation , and by the eberly research fund of the pennsylvania state university .

in this note , we review the connection between ads / cft duality and quark - gluon plasma ( qgp ) experiments at rhic ( see ref . @xcite for a review of qgp physics ) . rhic stands for relativistic heavy ion collider at brookhaven national laboratory . heavy ion " comes from the fact that it collides heavy ions such as gold nuclei @xmath1au instead of usual @xmath2 , @xmath3 or @xmath4 . the goal of the experiment is to realize the deconfinement transition and form the quark - gluon plasma . in principle , it should be possible to form qgp if one has high enough temperature or high enough density . however , it is not an easy job to confirm qgp formation because of the following problems : first , what one observes is not qgp itself but only the by - products after hadoronization , and one has to infer what had happened from the by - products . second , those secondary particles are mostly strongly - interacting , and the perturbative qcd is not very reliable for the current and near - future experimental temperatures . to resolve these problems , many attempts are made to identify the generic signatures of qgp . some of the generic signatures discussed to date are as follows : 1 . the elliptic flow which may be the consequence of very low viscosity of qgp 2 . the jet quenching 3 . @xmath5-suppression all of these signatures have been discussed in the ads / cft duality , so i review recent developments focusing on these phenomena . d - branes represent a @xmath6 gauge theory . ] since this review is aimed at non - experts , i first give a short course on string theory , in particular emphasizing on how gauge theories and black holes are described in string theory . figure [ fig : string ] shows main ingredients of string theory . there are two kinds of string : open strings with endpoints and closed strings with no endpoints . as i will show you shortly , an open string represents a gauge theory . a closed string represents a graviton . an open string propagates through spacetime just like a closed string , but an open string can also have its endpoints on an object , the so - called d - brane . the open string i consider in this lecture is always of this type . the easiest way to see that an open string represents a gauge theory is look at how the string oscillates . figure [ fig : level1 ] shows the simplest open string oscillation ( in 4 dimensions ) . as you can see , the string can oscillate in two directions . so , the open string has two degrees of freedom at this level . these degrees of freedom represent the two polarizations of the photon . in this sense , the open string represents a gauge theory . of course , our interest here is not qed , but rather qcd , so how can one describe a yang - mills theory ? an open string has endpoints on a d - brane , but if there are @xmath7 coincident d - branes , open strings can have endpoints in various ways ( fig . [ fig : sym ] ) . these new degrees of freedom precisely correspond to @xmath6 degrees of freedom . since these open strings are constrained to have their endpoints on the d - brane , the gauge theory described by the d - brane is localized on the d - brane . the d - branes arise with various dimensionalities . a d - brane with a @xmath8-dimensional spatial extension is called the d@xmath8-brane . namely , the d0-brane is point - like , the d1-brane is string - like , the d2-brane is membrane - like , and so on . thus , the d@xmath8-brane describes a @xmath9-dimensional yang - mills theory . we are interested in 4-dimensional gauge theories , so consider the d3-brane in order to mimic qcd . on the other hand , a closed string represents a graviton . again , the easiest way to see this is to look at how the string oscillates ( fig . [ fig : closed ] ) . in general , the oscillations on a string have two modes : the left - moving modes and right - moving modes . for an open string , these modes mix each other at endpoints , but these modes become independent for a closed string . so , one can oscillate the right - moving mode in one direction and the left - moving mode in the other direction . in a sense , a closed string oscillates in two directions simultaneously . this property explains the spin-2 nature of the graviton . in fact , a graviton also oscillates in two directions simultaneously . strictly speaking , a closed string represents a graviton and two undiscovered scalar particles , the dilaton and the axion . since each mode has two degrees of freedom , a closed string has 4 degrees of freedom at this level ( in 4 dimensions ) . the graviton has only 2 degrees of freedom , and two scalar fields cover the remaining degrees of freedom . we have seen that we can get a gauge theory from the d - brane , but it is not clear if the d - brane is simply described by a gauge theory . this is because string theory is more than a gauge theory , namely it has gravitons . at this point , it is not clear if the effect of gravity can be neglected . according to general relativity , any energy - momentum tensor curves spacetime . the d - brane of course has some energy , so how the d - branes curve spacetime ? since gravity is described by the newton potential _ newton ~ , one can measure the effect of curvature by @xmath10 . according to string theory , the newton constant @xmath11 and the mass @xmath12 of the d - brane are given by g & ~ & g_s^2 , + m & ~ & n / g_s , where @xmath13 is the string coupling constant which governs the strength of the interactions between strings . , which is consistent with the hypersurface picture of the d - brane , and the open strings attached represent the fluctuations of the d - brane . ] so , one gets @xmath14 . this means that as long as @xmath15 , one can neglect the effects of gravity and spacetime remains flat . in this case , the d - brane is simply described by a gauge theory . on the other hand , when @xmath16 , the d - brane starts to curve spacetime , and eventually it should become a black hole . so , in this case , the d - brane can be described by a black hole . the black hole here is not the usual schwarzschild - like black hole . we consider a d - brane , an object with infinite spatial extension . thus , the black hole formed from the d - brane has an horizon which extends indefinitely and is called the black brane . " to summarize , one can describe the d - brane both by a gauge theory and by a black hole ; which description is better depends on the value of @xmath17 . the string coupling and the yang - mills coupling @xmath18 are related by @xmath19 ( see sec . [ sec : details ] ) , so @xmath17 is nothing but the standard t hooft coupling , @xmath20 . this suggests that the black hole is the large t hooft coupling limit of a gauge theory . since t hooft coupling is the effective coupling of a gauge theory , the black hole is the strong coupling limit of the gauge theory ( fig . [ fig : largen ] ) . thus , the strategy here is to use black holes in order to compute gauge theory observables in the strong coupling regime . our argument here is very rough , but more refined version of the argument is known as the ads / cft duality @xcite ( see ref . @xcite for a review ) . the precise correspondence is as follows : + & & _ 5 s^5 . [ eq : dual ] here , @xmath21 means that the theory has 4 supercharges which are the maximum number of supercharges for a 4-dimensional theory . also , @xmath22 stands for the five - dimensional anti - desitter space . desitter was a dutch astronomer who found a solution of einstein equation with a constant positive curvature in 1917 . the space @xmath22 instead has a constant negative curvature ; this explains the prefix anti . " one can reach this correspondence by studying the d3-brane more carefully , but i will skip the argument . instead , i explain the correspondence from the symmetry point of view in appendix . we will use the finite temperature version of the duality and its cousins ; in this case , one needs to replace @xmath22 by a black hole in @xmath22 , which is known as the schwarzschild-@xmath22 black hole : + & & + & & _ 5 ) s^5 . according to the rhic experiment , qgp behaves like a liquid . the ads / cft then implies that a black hole also behaves like a liquid . then , plasma quantities should be calculable from black holes . in fact , black holes and hydrodynamic systems behave similarly . consider adding a perturbation to a black hole , _ e.g. _ , drop some object ( fig . [ fig : hydro ] ) . then , the shape of the black hole horizon becomes irregular , but such a perturbation decays quickly , and the black hole returns to the original symmetric shape . the no - hair theorem is one way to see this . according to the theorem , the stationary black hole is unique and symmetric . thus , the perturbed black hole can not be stable . if you regard this as a diffusion , the diffusion occurs since the perturbation is absorbed by the black hole . this behavior is very similar to a liquid . suppose that one drops a ball in a water pond . then , you generate surface waves , but they decay quickly , and the water pond returns to a state of stable equilibrium . in hydrodynamics , this is a consequence of viscosity . thus , one can consider the notion of viscosity for black holes as well . and the viscosity " for black holes should be calculable by considering the above process . let me remind you of freshman physics of viscosity . as a simple example , consider a fluid between two plates and move the upper plate with velocity @xmath23 ( fig . [ fig : viscosity ] ) . as the fluid is dragged , the lower plate experiences a force . this force is the manifestation of the viscosity . in this case , the force @xmath24 the lower plate experiences per unit area is given by = . the proportionality constant @xmath25 is called the ( shear ) viscosity . microscopically , the viscosity arises due to the momentum transfer between molecules . figure [ fig : viscosity ] shows a close - up view of the fluid and i put an artificial boundary to divide the fluid into two parts . the molecules collide with each other and are exchanged randomly through the boundary . but in the situation where you move the upper plate , the molecules in the upper - half part , on average , have more momentum in the @xmath26-direction than the ones in the lower - half part . these molecules are exchanged , which means that momentum in the @xmath26-direction is transported through the boundary . going back to the black hole , how can one calculate the plasma viscosity ? i will first give a brief explanation and its implications and then i will justify the claim more in detail . there are many ways to compute the viscosity and i will explain one simple method . in the gravity side , the diffusion occurs by black hole absorption . so , it is natural to associate the shear viscosity with the absorption cross section by black holes . ( i explain this point more in detail later . the detailed argument suggests that this is the cross section for the graviton of particular polarization . ) now , there is a general theorem on black holes @xcite that states that the cross section @xmath27 is equal to the horizon area @xmath28 for a broad range of black holes . ] : _ 0_bh = a . [ eq : absorption ] but the horizon area is the famous quantity , namely it represents the black hole entropy , so it must be the plasma entropy in this lecture . ] s_bh = k_b [ eq : entropy ] ( @xmath29 : boltzmann constant ) . then , the shear viscosity divided by the entropy becomes constant .. we consider black holes with infinite extension , so the entropy itself diverges . the area for the absorption cross section should be understood in a similar way . ] the constant can be determined from the argument later and the result is = . [ eq : viscosity_bound ] this value is very small . in comparison , @xmath30 for water is about @xmath31 under normal circumstances . now , the point is that all the relations we used ( cross section versus horizon area , black hole entropy versus horizon area ) are generic , so the result must be universal as well . namely , it does not depend on the details of black holes nor the details of gauge theories . so , the claim is @xcite _ gauge theory plasmas which have gravity duals have a universal low value of @xmath30 at large t hooft coupling . _ this is a rather indirect argument , but this claim has indeed been checked for many known gravity duals . this result is very important , so let me rephrase in a different way ( fig . [ fig : universality ] ) . gauge theories of which we can actually compute the shear viscosity are supersymmetric gauge theories , not the real qcd . we compute the shear viscosity from black holes , but the gravity dual of qcd is not known . so , one can not use ads / cft directly to compute qcd properties . however , as we saw , the quantity corresponding to the shear viscosity is universal on the black hole side , so one can immediately apply the @xmath21 result to the real qcd even though these two theories are completely different . in fact , rhic suggests that qgp has a very low viscosity and the estimated value @xcite ~o(0.1 ) ? is very close to the above ads / cft value . one important point is that the temperature in question is still order of @xmath32 . at this range of temperature , qcd is still strongly coupled and pqcd is not very reliable . in fact , the naive extrapolation of the weak coupling result gives a larger value for @xmath30 . thus , the ads / cft duality which predicts the strong coupling behavior may be useful to analyze qgp . i now describe the relation between the shear viscosity and the absorption cross section . to do so , one first has to understand the interactions of bulk and boundary fields . these fields can interact at the boundary . for example , the graviton produces the back - to - back scattering of gluons ( fig . [ fig : absorption ] ) . the relevant interactions are s_int ~d^4x f_^2 + h^ t_^ym + [ eq : interactions ] ( @xmath33 : dilaton , @xmath34 : graviton , @xmath35 and @xmath36 : the field strength and the energy - momentum tensor of the gauge theory ) . such interactions can be obtained by expanding the d - brane action around the expectation values of the bulk fields . the d - brane is described by the so - called dirac - born - infeld ( dbi ) action . the dbi - action with at most two derivatives contains the following term : ~e^- f_^2 + . if @xmath33 is constant , one gets the standard gauge theory lagrangian with @xmath37 , where @xmath38 . if @xmath33 fluctuates , the action is expanded as ~e^- ( f_^2 + f_^2 ) , where @xmath39 . the second term is nothing but the first term in eq . ( [ eq : interactions ] ) . this fact could be stated as _ bulk field fluctuations act as sources of boundary fields _ i just rephrase the same statement , but this leads to the so - called gkp - witten relation , " which is the definition of the ads / cft . given the interaction term , one can easily calculate the graviton decay rate ( with polarization @xmath40 , where @xmath26 and @xmath41 are directions along the brane ) from the standard field theory formula : _ qft & = & _ + & & ( 2)^4 ^4(p_f - p_i ) |m|^2 + & = & d^4 x e^it . the first equality is just the fermi s golden rule with matrix element @xmath42 . the matrix element is proportional to @xmath43 , so the formula is written by a correlator of the energy - momentum tensor . ( this is an optical theorem . ) the factor @xmath44 is a statistical factor for identical particles in the final state . in black hole description , it is natural to regard this decay rate as the absorption cross section of the graviton . this has been checked for the d3-brane at zero temperature @xcite , so ) is known to hold for the d3-brane at zero temperature but it is unclear if it also holds for generic ads black holes . the problem is that a black hole is the strong coupling limit of a gauge theory , so eq . ( [ eq : correlator ] ) is the strong coupling statement . but , of course , it is not easy to compute the right - hand side in gauge theory . one can compute it in the weak coupling , but weak coupling results differ from strong coupling results in general as we will see in sec . [ sec : others ] . in the special case of the zero - temperature d3-brane , such a comparison actually makes sense due to the nonrenormalization theorem , and this is the case where eq . ( [ eq : correlator ] ) has been checked . but it is an open question if this is also true for generic ads black holes . here , we simply assume that the relation holds at strong coupling . ] _ bh= d^4 x e^it . [ eq : correlator ] we can use this relation to compute the shear viscosity since the viscosity is given by a kubo formula microscopically : = _ 0 d^4 x e^it . this formula has the same form as the absorption cross section , so one can immediately write the viscosity in terms of the cross section : = . we have written plasma quantities in terms of black hole quantities . using eqs . ( [ eq : absorption ] ) and ( [ eq : entropy ] ) , one obtains = = . this is the previous formula ( [ eq : viscosity_bound ] ) . we saw that in gauge theories at strong coupling the ratio of shear viscosity to entropy density is universal . how far has the universality been shown ? recently , the universality has been extended to a variety of situations which are not covered in the original proofs . the shear viscosity was first computed for the d3-brane @xcite . this corresponds to the @xmath21 sym which is a scale - invariant theory . ( actually , the theory has a larger symmetry , a conformal symmetry . see appendix . ) but the universality has been shown even for theories with scales , _ i.e. _ , nonconformal theories . examples include d@xmath8-branes for @xmath45 , the klebanov - tseytlin geometry , the maldacena - nunez geometry , and the @xmath46 system ( these theories also have a reduced number of supersymmetries ) . in fact , various universality arguments were demonstrated for these cases @xcite . however , these proofs did not cover the cases with a chemical potential . so , the natural question is what happens to the universality at finite chemical potential . actually , it is not easy to realize the realistic finite density in ads / cft , _ i.e. _ , baryon number density . but there is a simple alternative . namely , consider charged ads black holes instead of neutral black holes . a black hole is known to obey thermodynamic - like laws and its first law is written as dm = tds_bh+dq ( @xmath47 : black hole temperature , @xmath48 : electromagnetic potential , @xmath49 : black hole charge ) . as one can see , the electromagnetic field @xmath48 plays a role of a chemical potential . in ads / cft , such a charge arises as follows . as in eq . ( [ eq : dual ] ) , the full geometry involves @xmath50 . one can add an angular momentum along @xmath50 , which is known as the spinning " d3-brane solutions @xcite . the angular momentum becomes a kaluza - klein charge after the @xmath50 reduction . what is the gauge theory interpretation of the charge ? the symmetry of @xmath50 corresponds to an internal symmetry of the sym , r - symmetry @xmath51 . this @xmath51 rotates adjoint scalars in the @xmath21 supermultiplet . ( see appendix . ) the r - symmetry group @xmath51 is rank 3 , so one can add at most three independent charges . the three - charge solution is known as the stu solution @xcite . when all charges are equal , the stu solution is the well - known reissner - nordstrm-@xmath22 black hole . thus , the charge in question is a @xmath52 charge . because the charge corresponds to the @xmath52 charge , this is by no means realistic . however , the theory has interesting features which are common to the real qcd . for instance , the phase diagram is qualitatively similar to the qcd diagram @xcite . this system does not represent a realistic chemical potential , but it may mimic the realistic case and one may learn an interesting lesson for gauge theory plasmas at finite density . the shear viscosity for charged ads black holes was computed by 4 groups , and the result turns out to be @xmath53 again @xcite-@xcite . so , the universality seems to hold even at nonzero chemical potential . if this is true for generic chemical potential , @xmath53 may be true even at finite baryon number density . in fact , when there is a chemical potential associated with a global symmetry , the universality has been proven in more general settings as well @xcite-@xcite . the theories described so far all have matter in the adjoint representation and not in the fundamental representation such as quarks . however , the universality has been proven even in the presence of fundamental matter @xcite . one way to include fundamental matter is to include d7-branes in addition to d3-branes . i will describe one simple way to realize fundamental matter in the next section , and the d3-d7 system is an extension of the method . this d3-d7 system is known as karch - katz model @xcite . gauge theory plasmas described so far are stationary ones . the real plasma at rhic is of course a rapidly changing system , and it is desirable to study such a plasma as well . in the gravity side , this corresponds to a time - dependent black hole . the universality has been claimed even for such a case @xcite . we saw that @xmath53 , but this is in the strong coupling limit or in the large t hooft coupling limit @xmath54 . real qcd of course has finite coupling , so how about @xmath30 at finite coupling ? in general , the shear viscosity is given by ~|vl_mfp ( @xmath55 : mass density , @xmath56 : mean velocity , @xmath57 : mean free path ) . at weak coupling , the interaction becomes weaker , so the mean free path becomes larger . the viscosity arises due to the momentum transfer , and the transfer is more effective at weak coupling . so , at weak coupling , one expects @xmath58 . and @xmath7 ( or @xmath18 and @xmath59 ) , so there are two kinds of finite - coupling corrections . first is the finite-@xmath59 correction described here . this corresponds to the so - called @xmath60-corrections in the gravity side . the effective actions of string theory contain higher curvature terms in addition to the einstein - hilbert action ; _ e.g. _ , @xmath61 . the finite-@xmath59 correction comes from such higher curvature terms . on the other hand , the finite-@xmath18 correction comes from the so - called string loop corrections since the string coupling constant is proportional to @xmath62 . such a correction to @xmath25 has never been evaluated . ] for the @xmath21 sym , the finite-@xmath59 correction can be computed , and the correction indeed raises the value of @xmath30 @xcite . the correction is about 9% if one uses @xmath63 ( or @xmath64 ) . thus , the strong coupling result seems a good approximation . the universality does not hold for finite-@xmath59 corrections though , so it is not clear how realistic the result is . figure [ fig : finite_coupling ] shows both the weak coupling result and the strong coupling result . they both increase at weak coupling . however , there is a difference . the viscosity is proportional to the mean free path , so is inversely proportional to the coupling constant . thus , the naive extrapolation of the weak coupling result suggests the perfect - fluid - like behavior in the strong coupling limit . instead , ads / cft tells that @xmath30 can not be small indefinitely and is saturated . incidentally , our problem is the strong coupling problem , so one might wonder how useful lattice computations are . a lattice computation of the viscosity for the pure yang - mills theory has been done in ref . unfortunately , the result has very large error bars , so one can not make a definite statement , but it would be very interesting to make the errors smaller . both from the weak coupling result and the strong coupling result . the red line represents the weak coupling result , _ i.e. _ , a naive extrapolation of the perturbative @xmath21 sym result ( based on ref . the blue line represents the strong coupling result , _ i.e. _ , the @xmath21 ads / cft result with the first order correction in @xmath59 ( based on ref . @xcite ) . ] recently , there has been much discussion on heavy quark dynamics in plasma medium . two applications have been discussed : issues related to @xmath0 suppression and issues related to jet quenching . in this section , i quickly summarize the discussion . since @xmath5 is heavy , charm pair production occurs only at the early stages of the nuclear collision . however , if the production occurs in the plasma medium , charmonium formation is suppressed due to the debye screening . one technical difficulty is that the @xmath65 pair is not produced at rest relative to the plasma . therefore , the screening length is expected to be velocity - dependent . such a computation has been done only for the abelian plasma @xcite . pair , the left configuration is not the lowest energy configuration . instead , the right configuration is energetically favorable . the energy difference is interpreted as a @xmath66 potential . at finite temperature , this is no longer true . for large enough separation of the pair , the left configuration becomes favorable . this phenomenon is the ads / cft description of the debye screening . ] first , one has to understand how to realize a heavy quark in ads / cft . to do so , let us go back to fig . [ fig : sym ] . when there are multiple number of d - branes , an open string can have endpoints in various ways ; there are @xmath67 possibilities . this means that the string transforms as the adjoint representation of @xmath6 gauge theory . now , consider an infinitely long string ( fig . [ fig : quark ] ) . in this case , the string can have endpoints in @xmath7 different ways . this means that the string transforms as the fundamental representation of @xmath6 gauge theory . in this sense , such a long string represents a quark . " such a string has an extension and tension , so the string has a large mass , which means that the long string represents a heavy quark . this kind of string has been widely studied in the past to measure the heavy quark potential @xcite . for a @xmath66 pair , two individual strings extending to the boundary is not the lowest energy configuration . instead , it is energetically favorable to have a single string that connects the pair ( fig . [ fig : wilson ] ) . the energy difference is interpreted as a @xmath66 potential , and one gets a coulomb - like potential for @xmath21 . at finite temperature @xcite , it is no longer true that a string connecting the @xmath66 pair is always the lowest energy configuration ; for large enough separation of the pair ( @xmath68 ) , isolated strings are favorable energetically . this phenomenon is the ads / cft description of the debye screening and @xmath68 may be interpreted as the screening length . references @xcite proposed how to compute the velocity dependence of the screening length . they computed the screening length in the @xmath66 rest frame , _ i.e. _ , they considered the plasma flowing at a velocity @xmath23 . such a plasma wind " is obtained by boosting a black hole . at the leading order in @xmath23 , the screening length is obtained as + & & ( 1-v^2)^1/4 + & ~ & ( ) ^-1/4 for the @xmath21 sym , where @xmath69 is the unboosted energy density . thus , the screening effect becomes stronger than the @xmath70 case . one can also compute the screening length for the other gauge theories @xcite . the leading behavior in @xmath23 seems universal . namely , if one writes the screening length as ( ) ( 1-v^2)^ , the exponent @xmath71 is determined by the speed of sound @xmath72 : 4= 1 - ( 1 - 3c_s^2 ) + [ eq : mu_vs_cs ] when the theory is nearly conformal , _ i.e. _ , @xmath73 . one can make a simple estimate of the exponent for qcd . according to the lattice results cited in ref . @xcite , all groups roughly predict @xmath74 around @xmath75 . bearing in mind that our results are valid to large-@xmath7 theories and not to qcd , eq . ( [ eq : mu_vs_cs ] ) gives @xmath76 . it would be interesting to compare this number with lattice calculations and experimental results . one can also study the screening length at finite chemical potential @xcite . at the leading order , the screening length at finite chemical potential is the same as the one at zero potential for a given energy density . another interesting qgp phenomenon is jet quenching . in the parton hard - scattering , jets are formed , but the jets have to travel in the qgp medium , so the jets are strongly suppressed . this phenomenon is known as jet quenching . so , the interesting quantity is the energy loss rate of partons . the ads / cft descriptions of jet quenching have been proposed recently @xcite-@xcite . the proposals have been quickly extended to the other gauge theories @xcite-@xcite . to discuss jet quenching , one now moves the fundamental string with a velocity @xmath23 along a brane direction . then , the momentum carried by the string flows towards the horizon and one interprets the flow as the energy loss rate . for example , the energy loss rate for the @xmath21 sym becomes = - t^2 . unfortunately , the result obtained in this way has some drawbacks . first , the result is not universal and is model - dependent . second , the black hole results become exact only in the @xmath54 limit , but the result does not have a finite large-@xmath59 limit . these two drawbacks are in contrast to the @xmath30 case . this may suggest that one has to be careful to apply ads / cft to qgp . namely , one should not always take ads / cft results at face value . hydrodynamic description of gauge theory plasmas using ads / cft is very powerful due to the universality . the ads / cft may be useful to analyze experiments . conversely , experiments or the other theoretical approaches ( such as lattice gauge theory ) may be useful to confirm ads / cft . this approach is also important since currently there are many loose ends on the ads / cft derivation . one has to clear up these loose ends , but it may be hard to make further progress within string theory alone . so , the inputs from the other areas may be useful to increase confidence in the ads / cft derivation . on the other hand , one has to be careful to apply ads / cft to qgp if the universality does not hold ( _ e.g. _ , jet quenching ) . in this talk , i emphasized the universality approach , but of course finding the gravity dual of qcd is desirable . the biggest assumption of the universality argument is that such a dual indeed exists , so finding the dual is necessary for the universality approach as well . one well - known model of qcd is the sakai - sugimoto model @xcite . this was a successful model in the confining phase , but unfortunately it is not a good model in the plasma phase . the sakai - sugimoto model involves the d4-brane compactified on a circle . above the deconfinement temperature , the model looks as a true five - dimensional theory . another problem of the current approach concerns the large-@xmath59 limit ; ads / cft and qgp are actually different limits . the large-@xmath59 limit of the ads / cft is @xmath77 and @xmath78 . but of course qgp has a large-@xmath59 since @xmath79 and @xmath80 . we are not taking the same limits , so it is not clear why both give such close results . i would like to thank elena cceres , kengo maeda , and takashi okamura for collaboration and for their comments on the manuscript . it is a pleasure to thank testuo hatsuda , tetsufumi hirano , kazunori itakura , tetsuo matsui , osamu morimatsu , berndt mller , and tadakatsu sakai for useful conversations . i would also like to thank the organizers of various workshops for the opportunity to give this lecture . the research of m.n . was supported in part by the grant - in - aid for scientific research ( 13135224 ) from the ministry of education , culture , sports , science and technology , japan . the precise correspondence between gauge theory and gravity is as follows : + & & _ 5 s^5 . [ eq : dual2 ] one can reach this correspondence by studying the d3-brane more carefully , but instead i explain the correspondence from the symmetry point of view in this appendix . to understand why the @xmath21 sym appears , let us go back to the open string oscillations on the d3-brane . superstring theory actually requires 10-dimensional spacetime for consistency . the open strings on a d - brane are bounded on the d - brane , so the d3-brane represents a 4-dimensional gauge theory , but these open strings still oscillate in the full 10-dimensional spacetime . thus , the simplest open string oscillations have 8 degrees of freedom instead of 2 discussed in the text . what are these degrees of freedom ? in other words , what kind of gauge theory the d3-brane represents ? to see this , note that there are two types of string oscillations . first are the oscillations in the brane and the other are the oscillations out of the brane . from the 4-dimensional point of view ( in terms of @xmath81 representations ) , the former represent a gauge field . the latter represent scalars . the spatial dimension is 9 and the brane dimension is 3 , so there are 6 scalar fields . thus , the gauge theory represented by the d3-brane inevitably comes with scalar fields . the @xmath21 sym is such a theory . the field contents of the @xmath21 sym include the gauge field @xmath82 and the scalars @xmath83 . ( the color indices are suppressed for simplicity . we will also suppress the spinor indices . ) in addition , there are 4 fermions @xmath84 due to supersymmetry ( which comes from the supersymmetry in superstring . ) the lorentz transformation properties are different , but they come from similar string oscillations , which means that all these fields transform as the adjoint representation . namely , the theory has no fundamental representation such as quarks . also , if we have only d3-branes , as in the present case , the directions transverse to the brane are all isotropic . these directions correspond to the scalar fields , so the isotropy means that there is a global @xmath51 symmetry for @xmath83 . such a global symmetry is known as r - symmetry . the action for the @xmath21 sym is given by & = & \ { - f_^2 - ( d__i)^2 - |_i ^d_^i + & & + o(^4 ) + o ( ) } . first 3 terms are standard kinetic terms for the gauge field , the scalar fields , and the fermions . in addition , the action contains interaction terms which are written only schematically ; @xmath85 interactions and yukawa - like interactions . the gauge field and the scalars have mass dimension 1 and the fermions have mass dimension @xmath86 , so all terms in the action have mass dimension 4 , which means that the theory has no dimensionful parameter and the theory is scale invariant . actually , it is often the case in a relativistic field theory that the scale invariance and poincar symmetry @xmath81 combine into a larger symmetry , the conformal symmetry @xmath87 . thus , the @xmath21 sym has the global @xmath88 symmetry . the @xmath22 space is a spacetime with constant negative curvature . we are familiar with spaces with positive curvature , like the sphere , and have no problem visualizing them , but it is more difficult to visualize a space with negative curvature . a space with constant negative curvature is known as a hyperbolic space . one way to visualize a hyperbolic space is illustrated in famous woodcuts by escher ( _ e.g. _ , circle limit iii " ) . it is not easy to visualize @xmath22 , but one can represent @xmath22 as a hypersurface in a flat spacetime , which is useful for our purpose . the @xmath22 space is described by x_0 ^ 2 + x_5 ^ 2 - x_1 ^ 2 - - x_4 ^ 2 = l^2 [ eq : embedding ] in a six - dimensional flat spaceds_6 ^ 2 = -dx_0 ^ 2 - dx_5 ^ 2 + dx_1 ^ 2 + + dx_4 ^ 2 . [ eq : flat ] here , @xmath90 gives the characteristic length scale of the @xmath22 space and it is related to the t hooft coupling from the discussion in the text . note that if all signs are positive in eq . ( [ eq : embedding ] ) , eq . ( [ eq : embedding ] ) defines a sphere . also , eq . ( [ eq : embedding ] ) is a higher dimensional generalization of the standard hyperbola @xmath91 . in this form , it is clear that @xmath22 has a @xmath87 symmetry just like the @xmath21 conformal symmetry ; to say that theories are dual to each other , they should have the same symmetries . in addition , the full geometry in eq . ( [ eq : dual2 ] ) involves @xmath50 which has a @xmath51 symmetry . this is the same symmetry as the @xmath21 r - symmetry . so , the gravity side also has the global @xmath92 symmetry . k. yagi , t. hatsuda and y. miake , _ quark - gluon plasma : from big bang to little bang _ ( cambridge univ . press , 2005 ) . a. strominger and c. vafa , `` microscopic origin of the bekenstein - hawking entropy , '' phys . b * 379 * ( 1996 ) 99 [ arxiv : hep - th/9601029 ] . j. m. maldacena , `` the large n limit of superconformal field theories and supergravity , '' adv . * 2 * ( 1998 ) 231 [ int . j. theor . phys . * 38 * ( 1999 ) 1113 ] [ arxiv : hep - th/9711200 ] . e. witten , `` anti - de sitter space and holography , '' adv . theor . math . * 2 * ( 1998 ) 253 [ arxiv : hep - th/9802150 ] . e. witten , `` anti - de sitter space , thermal phase transition , and confinement in gauge theories , '' adv . * 2 * ( 1998 ) 505 [ arxiv : hep - th/9803131 ] . s. s. gubser , i. r. klebanov and a. m. polyakov , `` gauge theory correlators from non - critical string theory , '' phys . b * 428 * ( 1998 ) 105 [ arxiv : hep - th/9802109 ] . o. aharony , s. s. gubser , j. m. maldacena , h. ooguri and y. oz , `` large n field theories , string theory and gravity , '' phys . rept . * 323 * ( 2000 ) 183 [ arxiv : hep - th/9905111 ] . s. r. das , g. w. gibbons and s. d. mathur , `` universality of low energy absorption cross sections for black holes , '' phys . * 78 * ( 1997 ) 417 [ arxiv : hep - th/9609052 ] . p. kovtun , d. t. son and a. o. starinets , `` viscosity in strongly interacting quantum field theories from black hole physics , '' phys . rev . lett . * 94 * ( 2005 ) 111601 [ arxiv : hep - th/0405231 ] . d. teaney , `` effect of shear viscosity on spectra , elliptic flow , and hanbury brown - twiss radii , '' phys . c * 68 * ( 2003 ) 034913 [ arxiv : nucl - th/0301099 ] . t. hirano and m. gyulassy , `` perfect fluidity of the quark gluon plasma core as seen through its dissipative hadronic corona , '' nucl . a * 769 * ( 2006 ) 71 [ arxiv : nucl - th/0506049 ] . s. gavin and m. abdel - aziz , `` measuring shear viscosity using transverse momentum correlations in relativistic nuclear collisions , '' phys . * 97 * ( 2006 ) 162302 [ arxiv : nucl - th/0606061 ] . a. adare [ phenix collaboration ] , `` energy loss and flow of heavy quarks in au + au collisions at s(nn)**(1/2 ) = 200-gev , '' arxiv : nucl - ex/0611018 . i. r. klebanov , `` world - volume approach to absorption by non - dilatonic branes , '' nucl . b * 496 * ( 1997 ) 231 [ arxiv : hep - th/9702076 ] . g. policastro , d. t. son and a. o. starinets , `` the shear viscosity of strongly coupled n = 4 supersymmetric yang - mills plasma , '' phys . * 87 * ( 2001 ) 081601 [ arxiv : hep - th/0104066 ] . kovtun , d. t. son and a. o. starinets , `` holography and hydrodynamics : diffusion on stretched horizons , '' jhep * 0310 * ( 2003 ) 064 [ arxiv : hep - th/0309213 ] . a. buchel and j. t. liu , `` universality of the shear viscosity in supergravity , '' phys . * 93 * ( 2004 ) 090602 [ arxiv : hep - th/0311175 ] . a. buchel , `` on universality of stress - energy tensor correlation functions in supergravity , '' phys . b * 609 * ( 2005 ) 392 [ arxiv : hep - th/0408095 ] . p. kraus , f. larsen and s. p. trivedi , `` the coulomb branch of gauge theory from rotating branes , '' jhep * 9903 * ( 1999 ) 003 [ arxiv : hep - th/9811120 ] . m. cvetic _ et al . _ , `` embedding ads black holes in ten and eleven dimensions , '' nucl . b * 558 * ( 1999 ) 96 [ arxiv : hep - th/9903214 ] . k. behrndt , m. cvetic and w. a. sabra , `` non - extreme black holes of five dimensional n = 2 ads supergravity , '' nucl . b * 553 * ( 1999 ) 317 [ arxiv : hep - th/9810227 ] . a. chamblin , r. emparan , c. v. johnson and r. c. myers , `` charged ads black holes and catastrophic holography , '' phys . d * 60 * ( 1999 ) 064018 [ arxiv : hep - th/9902170 ] . m. cvetic and s. s. gubser , `` phases of r - charged black holes , spinning branes and strongly coupled gauge theories , '' jhep * 9904 * ( 1999 ) 024 [ arxiv : hep - th/9902195 ] . j. mas , `` shear viscosity from r - charged ads black holes , '' jhep * 0603 * ( 2006 ) 016 [ arxiv : hep - th/0601144 ] . d. t. son and a. o. starinets , `` hydrodynamics of r - charged black holes , '' jhep * 0603 * ( 2006 ) 052 [ arxiv : hep - th/0601157 ] . o. saremi , `` the viscosity bound conjecture and hydrodynamics of m2-brane theory at finite chemical potential , '' arxiv : hep - th/0601159 . k. maeda , m. natsuume and t. okamura , `` viscosity of gauge theory plasma with a chemical potential from ads / cft correspondence , '' phys . d * 73 * ( 2006 ) 066013 [ arxiv : hep - th/0602010 ] . a. buchel and j. t. liu , `` gauged supergravity from type iib string theory on y(p , q ) manifolds , '' arxiv : hep - th/0608002 . p. benincasa , a. buchel and r. naryshkin , `` the shear viscosity of gauge theory plasma with chemical potentials , '' arxiv : hep - th/0610145 . j. p. gauntlett , e. o colgain and o. varela , `` properties of some conformal field theories with m - theory duals , '' arxiv : hep - th/0611219 . d. mateos , r. c. myers and r. m. thomson , `` holographic viscosity of fundamental matter , '' arxiv : hep - th/0610184 . a. karch and e. katz , jhep * 0206 * ( 2002 ) 043 [ arxiv : hep - th/0205236 ] . r. a. janik , `` viscosity from nonsingularity , '' arxiv : hep - th/0610144 . d. t. son and a. o. starinets , `` minkowski - space correlators in ads / cft correspondence : recipe and applications , '' jhep * 0209 * ( 2002 ) 042 [ arxiv : hep - th/0205051 ] . g. policastro , d. t. son and a. o. starinets , `` from ads / cft correspondence to hydrodynamics , '' jhep * 0209 * ( 2002 ) 043 [ arxiv : hep - th/0205052 ] . c. p. herzog , `` the hydrodynamics of m - theory , '' jhep * 0212 * ( 2002 ) 026 [ arxiv : hep - th/0210126 ] . g. policastro , d. t. son and a. o. starinets , `` from ads / cft correspondence to hydrodynamics . ii : sound waves , '' jhep * 0212 * ( 2002 ) 054 [ arxiv : hep - th/0210220 ] . a. nunez and a. o. starinets , `` ads / cft correspondence , quasinormal modes , and thermal correlators in n = 4 sym , '' phys . d * 67 * ( 2003 ) 124013 [ arxiv : hep - th/0302026 ] . c. p. herzog , `` the sound of m - theory , '' phys . d * 68 * ( 2003 ) 024013 [ arxiv : hep - th/0302086 ] . a. buchel , `` n = 2 * hydrodynamics , '' nucl . phys . b * 708 * ( 2005 ) 451 [ arxiv : hep - th/0406200 ] . k. kovtun and a. o. starinets , `` quasinormal modes and holography , '' phys . rev . d * 72 * ( 2005 ) 086009 [ arxiv : hep - th/0506184 ] . p. benincasa , a. buchel and a. o. starinets , `` sound waves in strongly coupled non - conformal gauge theory plasma , '' nucl . b * 733 * ( 2006 ) 160 [ arxiv : hep - th/0507026 ] . a. buchel , `` transport properties of cascading gauge theories , '' phys . d * 72 * ( 2005 ) 106002 [ arxiv : hep - th/0509083 ] . a. buchel , j. t. liu and a. o. starinets , `` coupling constant dependence of the shear viscosity in n = 4 supersymmetric yang - mills theory , '' nucl . b * 707 * ( 2005 ) 56 [ arxiv : hep - th/0406264 ] . s. c. huot , s. jeon and g. d. moore , `` shear viscosity in weakly coupled n = 4 super yang - mills theory compared to qcd , '' arxiv : hep - ph/0608062 . a. nakamura and s. sakai , `` transport coefficients of gluon plasma , '' phys . lett . * 94 * ( 2005 ) 072305 [ arxiv : hep - lat/0406009 ] . m. c. chu and t. matsui , `` dynamic debye screening for a heavy quark - antiquark pair traversing a quark - gluon plasma , '' phys . d * 39 * ( 1989 ) 1892 . j. m. maldacena , `` wilson loops in large n field theories , '' phys . * 80 * ( 1998 ) 4859 [ arxiv : hep - th/9803002 ] . s. j. rey and j. t. yee , `` macroscopic strings as heavy quarks in large n gauge theory and anti - de sitter supergravity , '' eur . j. c * 22 * ( 2001 ) 379 [ arxiv : hep - th/9803001 ] . s. j. rey , s. theisen and j. t. yee , `` wilson - polyakov loop at finite temperature in large n gauge theory and anti - de sitter supergravity , '' nucl . b * 527 * ( 1998 ) 171 [ arxiv : hep - th/9803135 ] . a. brandhuber , n. itzhaki , j. sonnenschein and s. yankielowicz , `` wilson loops in the large n limit at finite temperature , '' phys . b * 434 * ( 1998 ) 36 [ arxiv : hep - th/9803137 ] . h. liu , k. rajagopal and u. a. wiedemann , `` an ads / cft calculation of screening in a hot wind , '' arxiv : hep - ph/0607062 . m. chernicoff , j. a. garcia and a. guijosa , `` the energy of a moving quark - antiquark pair in an n=4 sym plasma , '' arxiv : hep - th/0607089 . k. peeters , j. sonnenschein and m. zamaklar , `` holographic melting and related properties of mesons in a quark gluon plasma , '' arxiv : hep - th/0606195 . e. caceres , m. natsuume and t. okamura , `` screening length in plasma winds , '' arxiv : hep - th/0607233 . s. d. avramis , k. sfetsos and d. zoakos , `` on the velocity and chemical - potential dependence of the heavy - quark interaction in n = 4 sym plasmas , '' arxiv : hep - th/0609079 . h. liu , k. rajagopal and u. a. wiedemann , `` wilson loops in heavy ion collisions and their calculation in ads / cft , '' arxiv : hep - ph/0612168 . m. chernicoff and a. guijosa , `` energy loss of gluons , baryons and k - quarks in an n = 4 sym plasma , '' arxiv : hep - th/0611155 . c. hoyos , k. landsteiner and s. montero , `` holographic meson melting , '' arxiv : hep - th/0612169 . f. karsch , `` lattice qcd at high temperature and the qgp , '' arxiv : hep - lat/0601013 . h. liu , k. rajagopal and u. a. wiedemann , `` calculating the jet quenching parameter from ads / cft , '' arxiv : hep - ph/0605178 . c. p. herzog , a. karch , p. kovtun , c. kozcaz and l. g. yaffe , `` energy loss of a heavy quark moving through n = 4 supersymmetric yang - mills plasma , '' arxiv : hep - th/0605158 . j. casalderrey - solana and d. teaney , `` heavy quark diffusion in strongly coupled n = 4 yang mills , '' arxiv : hep - ph/0605199 . s. s. gubser , `` drag force in ads / cft , '' arxiv : hep - th/0605182 . a. buchel , `` on jet quenching parameters in strongly coupled non - conformal gauge theories , '' arxiv : hep - th/0605178 . c. p. herzog , `` energy loss of heavy quarks from asymptotically ads geometries , '' arxiv : hep - th/0605191 . e. caceres and a. guijosa , `` drag force in charged n=4 sym plasma , '' arxiv : hep - th/0605235 . j. f. vazquez - poritz , `` enhancing the jet quenching parameter from marginal deformations , '' arxiv : hep - th/0605296 . e. caceres and a. guijosa , `` on drag forces and jet quenching in strongly coupled plasmas , '' arxiv : hep - th/0606134 . f. l. lin and t. matsuo , `` jet quenching parameter in medium with chemical potential from ads / cft , '' arxiv : hep - th/0606136 . s. d. avramis and k. sfetsos , `` supergravity and the jet quenching parameter in the presence of r - charge densities , '' arxiv : hep - th/0606190 . n. armesto , j. d. edelstein and j. mas , jet quenching at finite ` t hooft coupling and chemical potential from ads / cft,' arxiv : hep - ph/0606245 . e. nakano , s. teraguchi and w. y. wen , `` drag force , jet quenching , and ads / qcd , '' arxiv : hep - ph/0608274 . j. j. friess , s. s. gubser and g. michalogiorgakis , `` dissipation from a heavy quark moving through n = 4 super - yang - mills plasma , '' jhep * 0609 * ( 2006 ) 072 [ arxiv : hep - th/0605292 ] . j. j. friess , s. s. gubser , g. michalogiorgakis and s. s. pufu , `` the stress tensor of a quark moving through n = 4 thermal plasma , '' arxiv : hep - th/0607022 . j. j. friess , s. s. gubser , g. michalogiorgakis and s. s. pufu , `` stability of strings binding heavy - quark mesons , '' arxiv : hep - th/0609137 . s. s. gubser , `` comparing the drag force on heavy quarks in n=4 super - yang - mills theory and qcd , '' arxiv : hep - th/0611272 . s. s. gubser , `` jet - quenching and momentum correlators from the gauge - string duality , '' arxiv : hep - th/0612143 . p. c. argyres , m. edalati and j. f. vazquez - poritz , `` no - drag string configurations for steadily moving quark - antiquark pairs in a thermal bath , '' arxiv : hep - th/0608118 . p. c. argyres , m. edalati and j. f. vazquez - poritz , `` spacelike strings and jet quenching from a wilson loop , '' arxiv : hep - th/0612157 . s. j. sin and i. zahed , `` holography of radiation and jet quenching , '' phys . b * 608 * ( 2005 ) 265 [ arxiv : hep - th/0407215 ] . t. sakai and s. sugimoto , `` low energy hadron physics in holographic qcd , '' prog . * 113 * ( 2005 ) 843 [ arxiv : hep - th/0412141 ] .

inflation offers a simple solution to the shortcomings of the standard hot big bang model . in addition , its predictions about the initial density perturbations , leading to the observed structure formation , are in a remarkable agreement with measurements of the cosmic microwave background ( cmb ) temperature anisotropies . among the various inflationary models , one should select the ones which lead to a better agreement with data , with the additional requirement that such models should be naturally built within a fundamental theoretical framework . in spite of its enormous success , inflation remains still a paradigm in search of a model . despite the elegance of chaotic inflation @xcite , this simple model faces a fine tuning problem . consistency , between predictions for the amplitude of cmb temperature anisotropies and measurements , requires a tiny coupling constant . to avoid this problem , linde @xcite has proposed hybrid inflation , a model of inflation based on einstein gravity , but driven by a false vacuum . in this model , the inflaton field rolls while another scalar field remains trapped in a false vacuum state . the false vacuum becomes unstable when the magnitude of the inflaton field falls below some critical value , leading to a phase transition to the true vacuum . the energy density is dominated by the false vacuum energy density so that the phase transition signals the end of hybrid inflation . the phase transition at the end of inflation leads to topological defect formation @xcite . theoretically motivated inflationary models can be built in a context of supersymmetry ( susy ) and supergravity ( sugra ) theories . @xmath1 supersymmetric models contain complex scalar fields , which often have flat directions in their potential , thus offering natural candidates for inflationary models . in this framework , hybrid inflation ( driven by the f - terms or the d - terms ) are the most standard models . such inflationary models lead generically to cosmic string formation at the end of the inflationary era . cosmic strings in supersymmetric theories may have new properties , as compared to their non - supersymmetric counterparts ; we do not address this issue here . a gauge symmetry can be broken spontaneously in @xmath1 globally supersymmetric theories , either by adding f - terms to the superpotential or , in the abelian case , by introducing fayet - iliopoulos ( fi ) d - terms . the higgs mechanism leads generically @xcite to abrikosov - nielsen - olesen ( ano ) strings . depending whether they were formed at the end of f- or d - term inflation they are called f - term or d - term strings , respectively . f - term inflation is potentially plagued with the @xmath0-problem , while d - term inflation avoids it . this problem arises from the presence of large corrections ( of the order of the hubble parameter during inflation ) to the inflaton mass , which spoil the required flatness of the inflaton potential . d - term inflation can be successfully implemented in the framework of sugra , while in addition , it can be easily accommodated within string theory models . in the simplest models of d - term inflation within sugra , in which the constant fi term gets compensated by a single complex scalar field at the end of the inflationary era , the d - term strings formed at the end of inflation are topologically stable , since @xmath2 , with @xmath3 the vacuum manifold of the broken u(1 ) symmetry . our study concerns such models . however , they have been proposed @xcite models where d - term strings can become unstable . for example , one can introduce additional matter multiplets so as to obtain a non - trivial global symmetry such as su(2 ) , leading to a simply connected vacuum manifold and the production of semi - local strings . alternatively , it has been suggested @xcite that the waterfall higgs fields are non - trivially charged under some other gauge symmetries @xmath4 , such that the vacuum manifold , @xmath5/ u(1)$ ] , is simply connected , leading to the formation of semi - local strings . d - term inflation requires the existence of a non - zero constant fi term , which can be added to the lagrangian only in the presence of a u(1 ) gauge symmetry . this extra u(1 ) symmetry can be of a different origin . some models have been suggested for field - dependent fi terms , arising in the presence of a chiral superfield @xmath6 shifting under u(1 ) . the imaginary part of the scalar part of the chiral superfield plays the rle of an axion , and cancels the chiral anomaly by shifting under the u(1 ) symmetry . such a u(1 ) symmetry is called _ anomalous _ , or _ pseudo - anomalous _ since the total anomaly vanishes . here the fi term depends on the real part of the chiral superfield . in supersymmetry , models with anomalous fi terms have been developed @xcite within heterotic string theory . in a cosmological setup one has first to assume the stabilisation of the chiral superfield @xmath6 ; the rle of @xmath6 may be played by any modulus ( a dilaton or any volume modulus ) . only if this assumption holds the dilaton - dependent d - term can be considered as a constant fi term . however , the issue of dilaton and moduli stabilisation in the heterotic string theory is far from being resolved . it is still not clear how to derive constant fi terms from string or m - theory and only field - dependent d - terms have been identified . as we will discuss , in absence of constant fi terms , local supersymmetry requires the superpotential to be invariant under the u(1)-gauged symmetry . in the context of theories with large extra dimensions , brane inflation occurs in a similar way as hybrid inflation within supergravity , leading to cosmic string - like objects . in string theories , d - brane @xmath7-anti - brane annihilation leads generically to the production of lower dimensional d - branes , with d3- and d1-branes , which are d - strings , being predominant @xcite in iib string theories . to illustrate brane inflation let us consider @xcite a d@xmath8-@xmath9 system in iib string theory . six of the spatial dimensions are compactified on a torus , while the branes move relatively to each other in some directions . as the two branes approach , the open string modes between the branes develop a tachyon , thus an instability . brane inflation @xcite ends by a phase transition mediated by open string tachyons . since the tachyonic vacuum has a non - trivial @xmath10 homotopy group , one concludes that there must exist stable tachyonic string solutions with @xmath11 co - dimensions ; they are stable bps ( bogomolnyi - prasad - sommerfield ) d@xmath12-branes . since all dimensions are compact these daughter branes are seen as one - dimensional objects for a four - dimensional observer ; they are the d - strings . the d - strings ( d1-branes ) have been identified @xcite in the low - energy supergravity with the d - term strings . the justification for this conjecture is that only d - term strings remain bps states in @xmath13 supergravity . in supergravity , f - term strings break all the supersymmetries , whereas the d - term strings preserve half of it . an interesting and successful brane inflationary model is the d3/d7 one @xcite , which has also an effective description as a d - term inflationary model . the flat direction of the inflaton potential is associated with the shift symmetry , which protects the inflaton field from acquiring a large mass that would spoil the required flatness of the potential . in its original version this model leads to the formation of topologically stable ano bps strings . in a later developed version @xcite of the d3/d7 model , where in terms of an effective gauge theory the model has a local u(1 ) gauge symmetry and a global su(2 ) symmetry , semi - local strings are formed . such strings can unwind without any cost of potential energy . in what follows we concentrate on inflationary models leading to the formation of topologically stable strings . strings formed at the end of an inflationary era , contribute @xcite in the spectrum of temperature anisotropies ; their contribution is heavily constrained from cmb data . compatibility between cmb measurements and theoretical predictions constrain @xcite the parameters space ( mass scales and couplings ) of the inflationary models . these constraints have been obtained @xcite for d - term inflation within minimal sugra . here , we would like to investigate whether the constraints on the parameters space are a result of our choice of minimal supergravity , or whether they are a generic outcome of d - term hybrid inflation . we will therefore examine d - term inflation originated by different choices of the khler potential . the main motivation being that a minimal khler potential can be considered as a peculiar and unmotivated choice @xcite . we base our study in a formulation of supergravity constructed from superconformal theory , since the standard formulation may be insufficient in the presence of constant fayet - iliopoulos terms @xcite . to be able to constrain the parameters space of the models we should know the power spectrum of a cosmic strings network and the allowed upper limit on the cosmic string contribution to the measured temperature anisotropy spectrum . the upper limit imposed on the cosmic string contribution to the cmb data depends on the numerical simulation employed in order to calculate the cosmic string power spectrum . the upper limit found in the literature @xcite is @xmath14 or @xmath15 , depending on the simulation , with @xmath16 confidence level . there are some uncertainties in these results due to the cosmic string evolution codes they are based on was found using the results of ref . @xcite , based on the velocity - dependent one - scale model for the string evolution . this approach is certainly not the best . the upper limit of @xmath15 was found using the results of ref . @xcite ; however recently the authors have corrected @xcite a mistake in their code , which has not been taken into account in ref . note however that the corrected version of the code @xcite seems to give similar results . we believe that the recent publication given in ref . @xcite is the best , at present , approach . ] and therefore in our calculations we use an average value for the upper bound equal to @xmath17 . note that none of the existing simulations take into account a non trivial microstructure of the cosmic strings formed : however they have been shown to be superconducting in supersymmetric abelian symmetry breaking through f- or d - terms @xcite . we plan the rest of the paper as follows : in section ii we address hybrid d - term inflation , first within the standard formulation of supergravity , and subsequently within the effective supergravity theory built upon superconformal field theory . in the rest of the paper we focus on the second approach . we briefly review the effective supergravity formulation because of its consequences for d - term inflation , often not taken into consideration . in section iii we review d - term inflation in minimal supergravity . in section iv we discuss inflation in a supergravity theory with shift symmetry . in section v we consider d - term inflation in supergravity models with higher order terms , including all corrections up to order @xmath18 . we round up our conclusions in section vi . we first review hybrid d - term inflation model in its standard supergravity formulation . following the standard formulation of supergravity @xcite , the general sugra lagrangian for chiral superfields @xmath19 , and a vector superfield , depends on three generic functions : the khler potential @xmath20 , the superpotential @xmath21 , and the kinetic function @xmath22 for the vector multiplets . it can be expressed as an integral over the grassmann variables @xmath23 and @xmath24 ( over superspace ) , as @xcite : @xmath25\nonumber\\ & & + \int { \mathrm{d}}^2\theta [ f_{ab}(\phi_i)w_a^{\ \alpha } w_{\alpha b } + { \rm h.c.}]~ , \label{lagrsugra}\end{aligned}\ ] ] where _ h.c . _ stands for hermitian conjugate . it turns out that the effective lagrangian obtained from eq . ( [ lagrsugra ] ) depends only on one combination of the khler potential @xmath20 and the superpotential @xmath21 , the following : @xmath26 a priori , the only restriction for the choice of the khler potential @xmath27 and the superpotential @xmath28 is that @xmath27 must be a real function and @xmath28 must be a holomorphic function . however , there exists a degeneracy in the choices of @xmath27 and @xmath28 , while the lagrangian is invariant under the _ khler transformation _ @xmath29 assuming that the lagrangian must be invariant also under a gauge symmetry , @xmath28 and @xmath27 must be invariant , at least up to a khler transformation . the gauge kinetic function @xmath22 must be holomorphic and covariant under gauge symmetry . choosing for example the _ minimal khler potential _ @xmath30 the kinetic terms for the scalar parts of the @xmath19 s are simply @xmath31 in minimal sugra , we also set the gauge kinetic function @xmath32 equal to @xmath33 . within this framework of supergravity , and assuming that there is a gauge symmetry , the scalar potential @xmath34 for the scalar components @xmath35 of the superfields @xmath19 reads @xmath36 + \frac{1}{2}\left[{\rm re } f_{ab}(\phi_i)\right]^{-1}\sum_a g^2_a d^2_a~;\ ] ] the first term in the _ r.h.s _ of eq . ( [ formpotenscal ] ) is called _ f - term _ and the second one _ d - term_. while the f - term has in general positive and negative contributions , the d - term is positive definite . @xmath37 is given by @xmath38 where the fayet - iliopoulos term , @xmath39 , is non - zero only if the gauge symmetry is abelian ; @xmath40 is the gauge coupling of the symmetry generated by @xmath41 . assuming a constant fi term , d - term potential may lead to de sitter type of solutions , which is particularly interested for building an inflationary model within supergravity . the standard d - term hybrid inflation model is based on the superpotential @xcite @xmath42 where @xmath43 are three chiral superfields and @xmath44 is the superpotential coupling . this model assumes an invariance under an abelian gauge group @xmath45 , under which the three superfields , @xmath43 , have charges @xmath46 , @xmath47 and @xmath48 , respectively . this model also assumes the existence of a constant fi term . however , the above summarised supergravity formulation ( called hereafter _ standard _ ) is inappropriate to describe d - term inflation @xcite . indeed , in d - term inflation the superpotential vanishes at the unstable de sitter vacuum , as it also vanishes in the absolute minkowski vacuum ; anywhere else the superpotential is non - zero . thus , the standard formulation of supergravity , where the lagrangian depends on @xmath27 and @xmath28 only through the combination given in eq . ( [ kwcomb ] ) is inappropriate , since the theory is ill defined at @xmath49 . in conclusion , d - term inflation in supergravity must be described with a non - singular formulation of sugra when the superpotential vanishes . various formulations of effective supergravity can be constructed from the superconformal field theory ( see , ref . @xcite for a recent review ) . the main idea is , in a first stage , to build a lagrangian with full superconformal theory . in a second stage , the gauge symmetries that are absent in poincar supergravity ( e.g. , local dilatations , local chiral u(1)-symmetry and local s - supersymmetry ) are gauge fixed . starting from this framework in order to derive the effective supergravity theory , one can construct a non - singular theory at @xmath49 , where the action depends on all three functions , @xmath50 and @xmath32 . the superconformal field theory ( scft ) is based on the @xmath51 symmetry . the superconformal lagrangian contains three parts , each of them being separately conformally invariant . for @xmath52 chiral multiplets @xmath53 ( with @xmath54 and some vector multiplets @xmath55 superconformally coupled to supergravity , the superconformal lagrangian reads @xcite @xmath56_d + \left[{\mathcal{w}}(x ) \right]_f + \left[f_{\alpha\beta}(x ) \bar{\lambda}^\alpha \lambda^\beta\right]_f~.\ ] ] the homogeneous function @xmath57 and the holomorphic functions @xmath58 , @xmath59 encode the khler potential , the superpotential and gauge kinetic function , respectively , once the extra gauge symmetries have been gauge - fixed . after constructing the lagrangian , we fix the extra symmetries . fixing local dilatation , the number of chiral scalars is decreased by one : there will be only @xmath60 physical scalar fields in supergravity . fixing s - supersymmetry makes a free fermion field to be removed . one thus makes a transformation from the @xmath52 variables @xmath53 to the _ conformon _ scalar @xmath61 and @xmath60 physical scalars @xmath62 , the @xmath60 chiral superfields of standard supergravity . more precisely , @xmath63 where @xmath64 are a set of holomorphic functions . the khler potential @xmath65 is related to the function @xmath57 , appearing in the superconformal lagrangian , and the conformon superfield @xmath61 through @xmath66 chiral and dilatation symmetry imply that the holomorphic function @xmath67 is @xmath68 apart the local @xmath51 symmetry , the lagrangian may also have some yang - mills gauge symmetries , which commute with local superconformal symmetries . the superconformal functions @xmath69 are invariant , while the superconformal function @xmath32 is covariant under the yang - mills gauge symmetries . the yang - mills transformations of all chiral superfields in the superconformal action are @xmath70 @xmath71 and @xmath72 are @xmath52 holomorphic functions for every symmetry . note that the splitting given in eq . ( [ splitting]a ) is not unique ; the action invariance under khler transformation , eq . ( [ defkahltransf ] ) , has its origin exactly in this remark . the meaning of the value of @xmath73 is the transformation of the conformon field @xmath61 . the invariance of @xmath74 leads to @xcite @xmath75\right\}~,\ ] ] implying that @xmath71 describes the non - invariance of the khler potential @xmath76 given by @xmath77 . note that derivatives @xmath78 ( and @xmath79 ) stand for derivatives w.r.t . @xmath62 ( and @xmath80 , respectively ) . assuming that the transformation of the conformon superfield @xmath61 is given by imaginary constants , then @xmath81 @xmath82 is the gauge coupling constant . if @xmath83 , the conformon superfield @xmath61 transforms under a @xmath84 . only if @xmath85 the symmetry is preserved without corrections of the superconformal u(1 ) . equation ( [ transfofi]a ) shows the superconformal origin of fi terms via the gauge transformations of the conformon field . equation ( [ ralpha ] ) implies that the khler potential is invariant @xcite @xmath86 the function @xmath67 should be invariant under yang - mills transformations , @xmath87 implying @xmath88 since the _ r.h.s . _ of the above equation , eq . ( [ noninv ] ) , is non - zero when fi terms are present , one concludes that the superpotential @xmath28 can not be gauge invariant , if the conformon multiplet @xmath61 transforms under gauge transformations . assuming that @xmath73 is given by eq . ( [ transfofi]a ) , which corresponds to a constant fi term in supergravity , one gets @xcite @xmath89 to construct a formulation of supergravity with constant fi terms from superconformal theory , one finds @xcite that under u(1 ) gauge transformations in the directions in which there are constant fi terms @xmath90 , the superpotential @xmath28 must transform as in eq . ( [ transfow ] ) . as a consequence , we can not keep any longer the charge assignment described in the case of standard supergravity . as it was explicitly shown in ref . @xcite , the gauge fixing of the local dilatational invariance introduces the mass scale @xmath91 , which we set @xmath92 , and fixes the value of the conformon superfield in terms of the khler potential @xmath76 , which depends only on the physical scalars @xmath93 and @xmath94 , namely @xmath95 the value @xmath96 being fixed , we then fix the phase of @xmath61 . the superconformal lagrangian , eq . ( [ lagrangsuperconf ] ) , is invariant under the redefinition @xmath97 for an arbitrary holomorphic function @xmath98 , since this redefinition can be absorbed in a redefinition of @xmath62 . we can check that the effect of this redefinition on @xmath65 and @xmath28 is precisely the khler transformation introduced in eq . ( [ defkahltransf ] ) . two choices have been proposed @xcite in order to fix the phase of @xmath61 . either @xmath99 leading to the standard formulation of sugra , described previously , or alternatively @xmath100 the first choice , eq . ( [ choice1 ] ) , makes sense only for @xmath101 , while the second one , eq . ( [ choice2 ] ) , is appropriate for cases where @xmath67 can become zero . since in what follows we focus on d - term inflation , where the superpotential vanishes during inflation as well as at the end of the inflationary era , we adopt the second choice of gauge fixing . in this case , the form of the lagrangian is explicitly given in refs . we give below several quantities which are of interest for the purpose of our study . to switch to dimensional khler potential and fields , we redefine @xmath62 and @xmath65 , as @xcite @xmath102 @xmath103 is regular at @xmath104 . we denote the scalar part of the superfield @xmath19 by @xmath35 . from now on derivatives @xmath78 stand for derivatives w.r.t . @xmath35 ; we thus use the fields @xmath35 and their complex conjugates @xmath105 to indicate scalar fields . the previous choice of conformon transformation , eq . ( [ transfofi ] ) , now reads @xmath106 . the supergravity lagrangian depends on @xmath107 and @xmath108 . constant fi terms can be introduced for some of the u(1 ) gauge groups . we write below the bosonic and fermionic parts of the lagrangian for the scalar fields @xmath35 , which are relevant for d - term inflation in the context of supergravity formulated using superconformal theory . the part of the bosonic sector which interests us , is @xmath109 with @xmath110\nonumber\\ v_d&=&\frac{1}{2}\left[({\rm re}f)^{-1}\right]^{\alpha\beta } { \mathcal{p}}_\alpha { \mathcal{p}}_\beta~,\end{aligned}\ ] ] where @xmath111~,\ ] ] and @xmath112 the fermionic mass terms for fermions @xmath113 of chiral supermultiplets @xmath35 are given by the fermionic sector of the lagrangian , namely @xcite @xmath114 -m^{ij}\bar{\chi}_i\chi_j - m_{ij}\bar{\chi}^i\chi^j + e^{-1}{\mathcal{l}}_{\rm mix}~,\ ] ] where @xmath115 and @xmath116 and where the khler metric and khler connection are defined by @xmath117 in addition to the standard mass terms proportional to @xmath118 , some mixing mass terms between the chiral fermions @xmath113 and the gaugino or the gravitino are contained in @xmath119 . they can potentially contribute to the fermionic mass matrix @xcite : @xmath120~,\end{aligned}\ ] ] where @xmath121\nonumber\\ v_l^1&=&\frac{1}{2}i{\mathcal{p}}_\alpha\lambda_l^\alpha+m^i\chi_i~.\end{aligned}\ ] ] in the case of non vanishing @xmath119 , the masses of are obtained by diagonalising the fermionic mass matrix . let us investigate the consequences of a formulation of supergravity constructed from superconformal theory for d - term inflation . first , the transformation of the sugra superpotential given by the eq . ( [ transfow ] ) imposes a modification of the charge assignment of section [ sec : stdsugraformul ] . this charge assignment holds for the global supersymmetric limit . since the charge of the superpotential is given by the sum of the charges of the fields @xmath122 , @xmath123 , we must impose @xcite @xmath124 note that as a consequence , we can relate the fi amplitude to the anomaly of the symmetry u(1 ) @xmath125 similarly to the generation of fi terms with anomalous u(1 ) symmetry within weakly coupled string theories . in ref . @xcite the authors suggest a way to solve the anomalous u(1 ) symmetry . more precisely , the authors introduce new matter superfields , which are not involved in inflation , with appropriate charge assignments so that the anomaly gets cancelled four - dimensional supergravity with fayet - iliopoulos couplings has been recently addressed in refs . @xcite . ] . in order to avoid an @xmath126problem we set @xmath127 . this modification of the charge assignment induces a modification of the d - terms of the scalar potential . since the function @xmath27 is real and @xmath28 is holomorphic , we can check that the expression of the f - terms , eq . ( [ potenfterms ] ) , is identical to the standard formulation of supergravity . concerning the radiative corrections , the modification of the charges induces a modification of the mass terms for the scalar components of the chiral superfields , since part of their mass terms comes from the d - terms . the fermionic lagrangian , eq . ( [ lagrangfermio ] ) , is identical to the expression obtained in the context of standard supergravity given in @xcite . note that in the case of interest , there are no additional contributions to the chiral fermion mass terms from @xmath119 given in eq . ( [ lagrangfermiomix ] ) . the last consequence of the superconformal origin of supergravity concerns the form of the superpotential . we first remark that the transformation of the superpotential , eq . ( [ transfow ] ) , combined with the requirement that the inflaton field should remain uncharged in order to avoid the @xmath0-problem , protect the form of the superpotential @xmath128 all non - renormalisable terms of the form @xmath129 are forbidden . secondly , if we assume in addition that the superfields and the superpotential transform under an r - symmetry as @xmath130 then all terms proportional to @xmath131 , with @xmath132 , and all terms of the form @xmath133 or @xmath134 are forbidden . we end up with a d - term inflation that is precisely described by the minimal superpotential , eq . ( [ superpot ] ) , even if in supergravity non - renormalisable terms can be present . this motivates our choice to consider in this work , the standard form of superpotential . the minimal supergravity description is based on the minimal khler potential @xmath135 and the minimal structure for @xmath22 , namely @xmath136 . the tree level scalar potential reads @xcite @xmath137\nonumber \\ & & + \frac{g^2}{2}\left(q_+|\phi_+|^2+q_-|\phi_-|^2+\xi\right)^2~.\end{aligned}\ ] ] the next step is the calculation of the masses of the components of the superfields @xmath123 . the scalar mass squared can be read directly from the scalar potential @xcite , namely @xmath138 strictly speaking , one must use the charge assignment given by eq . ( [ chargassign ] ) , for example , @xmath139 hereafter we assume @xmath140 and therefore the expression for the scalar masses , eq . ( [ scalmass ] ) , can be approximated as @xmath141 in the inflationary valley , where @xmath142 , the associated dirac fermions have a mass squared @xmath143 unchanged compared to the case of standard minimal supergravity @xcite . using the coleman - weinberg formula , we can then compute the effective inflationary potential , taking into account the tree - level and the one - loop radiative corrections due to the mass splitting between components of the chiral superfields @xmath123 . the effective potential reads @xcite @xmath144 \right\ } ~,\ ] ] where @xmath145 and @xmath146 d - term inflation leads to cosmic strings formation at the end of the inflationary era . consistency between cmb measurements and theoretical predictions constrain the parameters space . the constrains are imposed on the couplings and mass scales . they are shown in fig . [ fig : dtermstandard ] and can be summarised as @xcite @xmath147 the above constraints , eq . ( [ constrdmin ] ) , can be expressed @xcite as a single constraint on the fayet - iliopoulos term @xmath148 , namely , @xmath149 we can therefore see that our assumption that @xmath150 in eq . ( [ approxmass ] ) is justified @xcite , which implies that for the purpose of our study and for minimal d - term inflation , we can neglect the corrections introduced by the superconformal origin of supergravity . moreover , since the tree level potential is given by @xmath151 , the limit @xmath152 should hold since otherwise the energy density would become transplankian and the quantum gravity corrections , which have been so far neglected @xcite , would become important @xcite . + , as a function of the superpotential coupling constant @xmath44 . ] in what follows we address whether the restrictions found in the framework of minimal sugra , are still qualitatively valid for a non - minimal sugra theory . we will thus study d - term hybrid inflation with a superpotential @xmath28 defined by eq . ( [ superpoteninflad ] ) and different choices for the form of the khler potential . we first study hybrid inflation in the context of supergravity , with a khler potential obeying a shift symmetry . this symmetry can be used for model building in the framework of supergravity , while it is also motivated from string theory @xcite . more precisely , the shift symmetry , @xmath153 with @xmath154 a real constant , has been used in order to obtain flat potentials from f - terms in sugra , so that the @xmath0-problem is cured , or to make compatible chaotic inflation , which requires an inflaton field larger than the planck scale , and supergravity @xcite . thus , we consider the following form for the khler potential @xmath155 and minimal structure for the kinetic function . we note that compared to the minimal khler potential , eq . ( [ kmin ] ) , @xmath156 has two additional terms , , @xmath157 and @xmath158 , which do not affect the kinetic terms . comparing with the scalar potential obtained in minimal sugra , we can identify two differences . the first one is that there are several new terms proportional to @xmath164 . these terms will not affect the effective mass terms of the @xmath123 fields , thus they will not affect the inflationary potential , even when one - loop radiative corrections have been taken into account . the second one is that the exponential factor @xmath165 in minimal sugra , has been here replaced by @xmath166 . writing @xmath167 , we get @xmath168 . if we identify the inflaton field with the real part of @xmath122 , then we obtain exactly the same potential as in the minimal case . however , with the present choice for the khler potential , @xmath156 , it would be more logical to identify the inflaton field with the imaginary part of @xmath122 , namely @xmath169 . the reason being that with this choice the exponential term is constant during inflation , thus it can not spoil the slow roll conditions . then we can check that the inflationary potential we get is identical to the usual d - term inflation within the global susy framework . this model has been studied in ref . the result obtained in ref . @xcite is reminded in fig . [ dtermsusy ] . , in the case of the d - term inflation model in the framework of global susy . figure taken from ref . @xcite . ] clearly , in such a model the cosmic string contribution to the cmb anisotropies is dominant , in contradiction with the cmb measurements , unless the superpotential coupling is constrained to be @xmath170 concluding , we state that simply imposing a shift symmetry to the khler potential is not enough to _ escape _ the _ cosmic string problem _ of d - term inflation . in this section , we consider another choice of khler potential that contains non renormalisable terms . more precisely , we consider the following form @xmath171 where @xmath172 are arbitrary functions of @xmath173 . contrary to the previous section , this new form should modify the kinetic terms . the motivation for this choice is that it contains the next - to - minimal khler potential , with all terms up to order @xmath174 . we consider in a first place a reduced version of this potential and then we proceed with the full expression . we remind to the reader that our primary goal is not to build a new model for d - term inflation in the context of supergravity , but to study the robustness of our prediction that there is a dominant contribution of cosmic strings to the cmb temperature anisotropies , unless we fine tune the superpotential coupling constant @xmath44 . we first consider the case of the previous khler potential , eq . ( [ gen ] ) , setting @xmath175 , implying latexmath:[\[k_3=|s|^2+|\phi_+|^2+|\phi_-|^2+f_+\bigg(\frac{|s|^2}{m_{\rm pl}^2}\bigg)|\phi_+|^2+f_-\bigg(\frac{|s|^2}{m_{\rm pl}^2}\bigg ) arbitrary functions of @xmath173 , while the superpotential is given in eq . ( [ superpoteninflad ] ) . one can argue that the khler potential given in eq . ( [ kahlerseto1 ] ) is quite general , since during inflation the higgs fields are small . however , one can also criticise this choice for the following reason : even though @xmath177-terms are indeed negligible as compared to the @xmath178-ones ( the @xmath179-terms are small during the inflationary era ) there is no reason for neglecting @xmath180-terms . they will be taken into account in next section . note that the expression for the effective scalar potential is identical to that of minimal supergravity , except for the expression of @xmath93 . the first derivative of the potential is equal to @xmath194 where @xmath195 @xmath196 ~;\ ] ] @xmath197 denote the first derivative of @xmath172 with respect to @xmath198 . a choice which we will later consider for @xmath199 is @xmath200 in the large @xmath201-limit the effective potential and its first derivative with respect to @xmath201 reduce to @xmath202 \label{limitz1}\ ] ] and @xmath203 respectively . at this point , we would like to note that d - term inflation can be realised with the last 60 e - folds being very close to the critical value @xmath204 , implying that the above reduced formulae , eqs . ( [ limitz1 ] ) , ( [ limitz2 ] ) , _ can not _ be used to compute the predictions of the model regarding the cmb temperature anisotropies . therefore , we disagree with the approach of ref . @xcite , where the authors have used the reduced formulae for the effective potential and its first derivative , to compute the inflaton contribution to the cmb temperature anisotropies . to illustrate the above remark , we represent in fig . [ fig : error ] the error made by this assumption . the cosmic string contribution is drawn as a function of the superpotential coupling constant @xmath44 , considering either the partial or the full scalar potential . one can realise that the error on the cosmic string contribution can reach @xmath205 , leading to a considerable error in the induced constraints on the parameters space . with partial ( dashed line ) or full ( continuous line ) scalar potential . ] the number of e - foldings of inflation between the initial and final value of the inflaton field is @xmath206 f_{v'}(z)}~,\ ] ] where we note that the index @xmath207 denotes the scale responsible for the quadrupole anisotropy in the cmb . for the integral appearing in the above expression for @xmath208 to be correctly defined , one must assume that the function @xmath209 can be inverted . since it is a monotonic and increasing function , one can verify , at least numerically , that the function @xmath210 can be inverted . we denote the inverted function by @xmath211 . the above expression for @xmath208 , eq . ( [ efoldings ] ) , will enable us to write a relation between @xmath148 and @xmath212 . setting @xmath213 , we can fix the value of the inflaton field @xmath214 , responsible for the quadrupole anisotropy of the cmb . d - term inflation leads generically to cosmic strings formation , which also contribute to the cmb temperature anisotropies . the total contribution , from the inflaton field as well as from the cosmic strings , has to be normalised to the value measured by the cobe - dmr experiment . this normalisation will fix the mass scale @xmath215 as a function of the superpotential coupling @xmath44 and the gauge coupling @xmath82 , which are considered here as parameters . thus , we have to calculate both , the cosmic strings contribution and the inflaton field one , and normalise their _ sum _ to cobe - dmr . here we emphasise that different approaches can be found in the literature where the authors do not normalise the sum of the two contributions to the data . clearly not to normalise the sum to the data is a simplification , which may lead to an important error for the result . normalising only the inflationary contribution to the quadrupole absolute value can be considered as an assumption where the cosmic string contribution is neglected . this assumption could be _ allowed _ only if one wants to argue that the cosmic strings are sub - dominant . however , since we already know that cosmic strings are indeed sub - dominant and our aim is to constrain the parameters space and calculate exactly how sub - dominant the strings are , this simplification in the normalisation implies important errors in the calculated cosmic string contribution . from the scalar potential given in eq . ( [ dpotensugranonmin ] ) , one can see that the spontaneous symmetry breaking is generated when @xmath216 takes a vacuum expectation value @xmath215 . thus , the quadrupole contribution to the cmb temperature anisotropies from the cosmic strings formed at the end of hybrid inflation , which is @xmath217 where @xmath218 is the vacuum expectation value of the higgs field responsible for the formation of cosmic strings , is approximately equal to @xmath219 for the contribution from the inflaton field , we evaluate the sachs - wolfe term splitted into the scalar and tensor parts , using @xmath220 and @xmath221 from eqs . ( [ treelevelsc ] ) , ( [ veff ] ) , ( [ inflsc ] ) and ( [ inflsctens ] ) we get @xmath222 and @xmath223 we note that the ratio of the tensor part of the inflaton field contribution to the cosmic strings one is constant , and that the tensor part contribution can be neglected . the tensor over scalar ratio , @xmath224 , is less straightforward and it is given by @xmath225 we then proceed with the cosmic string contribution to the quadruple cmb temperature anisotropy . this contribution in computed as a function of the superpotential coupling @xmath44 , for various values of @xmath82 and @xmath226 , which are considered as parameters . the results are plotted in fig . [ csdelambda ] . from the top to the bottom , respectively . this is for @xmath227 . ] to show the dependence of the cosmic string contribution on the gauge coupling @xmath82 , we draw in fig . [ csdelambdag1 ] the cosmic string contribution for @xmath228 and @xmath229 . clearly , this case ( @xmath228 ) is excluded since the cosmic string contribution is above the allowed one . we note that we do not take @xmath230 higher than 5 , since positivity condition @xmath231 requires @xmath232 @xcite . from the top to the bottom , respectively . this is for @xmath228 . ] we quantify below the constraints on the parameters space imposed from the three - year wilkinson microwave anisotropy probe ( wmap ) @xcite measurements . we want to set precise constraints on the free parameters since this can be of importance for concluding of whether d - term inflation remains in agreement with cmb data once we also include constraints imposed from the allowed value of the spectral index @xcite . for @xmath233 , there is no parameters space in agreement with measurements . in the range @xmath234 $ ] , the parameters space is extremely small , around @xmath235 ; the presence of @xmath230 enlarges slightly the allowed window . for @xmath227 , the @xmath17 upper limit on the allowed cosmic string contribution to the cmb imposes , at @xmath236 of confidence level , the following constraints : @xmath237 or , equivalently , @xmath238 clearly the new degree of freedom , namely @xmath226 , allows a slightly higher upper bound on the coupling , which is however at best higher by only a factor of 2 , thus concluding that fine tuning is still required . all constraints are equivalent to a single constraint on @xmath239 , or @xmath240 , as already stated in refs . therefore , there is a bijection between the cosmic string contribution and the mass scale of inflation , which nevertheless does not mean that one can normalise only the inflaton contribution to the cmb data . clearly , the error made by this assumption is big when cosmic strings have an important weight , while it is small if the cosmic string contribution is small , of the order the one found here . with the current upper limit of @xmath17 on the allowed cosmic string contribution to the temperature anisotropies , the relative error made on the string contribution , by not normalising the sum , is of the order of @xmath241 . as one can see from figs . [ csdelambda ] , [ csdelambdag1 ] the curve which showing the cosmic string contribution becomes singular at its minimum , for large values of @xmath242 . the reason for this behaviour is that the function @xmath243 , denoting the sum of the cosmic strings and inflaton contributions as a function of @xmath148 becomes non - bijective for certain values of @xmath44 . this implies that there are more than one solutions for the function @xmath244 normalised to the cobe data , for a tiny window of the parameter @xmath44 around the value @xmath245 . we illustrate this in fig . [ multval ] . . in terms of the cosmic string contributions these solutions correspond to @xmath246 and @xmath247 from left to right , equivalently . ] we have checked that this _ degeneracy _ does not influence the validity of our results since the different solutions for the normalisation to cobe are of the same order of magnitude and this behaviour is observed for a tiny parameters space . as a final remark , we would like to note that there is also another point where we disagree with ref . @xcite . in their model specified by the choice of the khler potential and with the simplifications made for the scalar potential , the authors obtain @xcite a value for @xmath239 , required to generate the appropriate magnitude of density perturbations . the authors argue that this value is consistent with the upper limit imposed on @xmath148 so that the cosmic string contribution is within the allowed window , following the results of ref . @xcite . however , the analysis of ref . @xcite was done employing the curvaton mechanism ; a contribution which the authors of ref . @xcite have not considered at all . we proceed with the general case , namely we include all terms of the next to leading order latexmath:[\[k_2=|s|^2+|\phi_+|^2+|\phi_-|^2+f_+\bigg(\frac{|s|^2}{m_{\rm pl}^2}\bigg)|\phi_+|^2+f_-\bigg(\frac{|s|^2}{m_{\rm pl}^2}\bigg ) @xmath172 is just @xmath249 the motivation for such a choice is that beyond the minimal supergravity part of the khler potential , the leading order is @xmath174 , implying that one should consider all corrections up to this order . during inflation , the charged fields @xmath123 vanish , meaning that terms of the form @xmath250 or @xmath251 can be neglected compared to @xmath252-terms , which are in @xmath253 . thus , the next - to - minimal d - term inflationary model should be constructed with khler potential @xmath254 . we again consider the superpotential @xmath255 involving the superfields @xmath122 and @xmath123 with charges @xmath46 and @xmath256 , respectively , under the @xmath45 symmetry of the lagrangian . the charges @xmath256 must satisfy the constraint induced by the superconformal origin of the fi term , as discussed in section [ csqforsugra ] . the d - term part of the scalar potential is thus unchanged as compared the previous section and given by eq . ( [ vdnonmin ] ) . equation ( [ fnew ] ) implies that during inflation @xmath182 is minimised for @xmath267 , thus @xmath268 and inflation is driven by the d - term . the total scalar potential @xmath34 is constant at tree level and equal to @xmath269 . having the scalar potential we calculate the masses of the canonically normalised scalar components of the superfields @xmath123 : @xmath270 whereas from eq . ( [ lagrangfermio ] ) we get the mass squared for their fermionic partners : @xmath271 following the same procedure as in the previous sections , we obtain the effective potential @xmath190 \right\ } ~,\ ] ] where @xmath272 is the same as in section [ secpotenseto ] , eq . ( [ deffv ] ) ; the only difference is the expression for @xmath93 , namely here @xmath273 the first derivative of the scalar potential with respect to @xmath201 reads @xmath274 where @xmath275 is the same as in section [ secpotenseto ] , eq . ( [ deffv ] ) , but @xmath276 is in this case given by @xmath277~. \label{newfz}\ ] ] the number of e - folds and the inflationary contribution to the cmb quadrupole anisotropy are still calculated using eqs . ( [ efoldings ] ) and ( [ dtsurt ] ) , respectively , with @xmath276 given by eq . ( [ newfz ] ) above . the cosmic strings contribution to the cmb temperature anisotropies is the same as in the minimal d - term sugra case . the contribution of cosmic strings to the cmb anisotropies in the whole parameters space is represented in figs . [ csdelambdanonminextc0g2 ] , [ csdelambdanonminextc0b1 ] , [ csdelambdanonminextb1c1 ] . , in the next - to - minimal d - term inflationary model . the value of the gauge coupling constant is fixed and equal to @xmath227 ; there are no cross terms between @xmath122 and @xmath123 , i.e. , @xmath278 . the minimal sugra for @xmath175 is represented by the dashed line , while the different plain lines are calculated for @xmath279 , going from the bottom to the top . a fine tuning of the superpotential coupling @xmath44 is still required , in order to avoid a dominant contribution of cosmic strings . ] , in the next - to - minimal d - term inflation model . the parameter @xmath280 is set equal to @xmath281 ; there are no cross terms between @xmath122 and @xmath123 , ie @xmath278 . the different plain lines are calculated for @xmath282 , going from the bottom to the top . a fine tuning of the superpotential coupling @xmath44 is still is required to avoid a dominant contribution of cosmic strings . ] , in the next - to - minimal d - term inflation model . the parameter @xmath280 is set equal to @xmath281 ; there are no cross terms between @xmath122 and @xmath123 , i.e. , @xmath278 . the different plain lines are calculated for @xmath282 , going from the bottom to the top . a fine tuning of the superpotential coupling @xmath44 is still required , in order to avoid a dominant contribution of cosmic strings . ] studying the above figures one easily concludes that considering a more general form for the khler potential does not solve the fine tuning problem of d - term inflation . the new term in the khler potential , whose weight is given by the parameter @xmath280 , induces an enhancement of the cosmic string contribution at low @xmath44 . we still observe a dominant contribution of cosmic strings if the superpotential coupling @xmath44 is close to unity . therefore , the constraints on @xmath44 found on the previous section remain unchanged as given in eq . ( [ contraintelambdanonmin ] ) . d - term hybrid inflation is a successful and interesting model . in the context of supergravity , d - term inflation avoids the _ hubble - induced mass _ problem , which plagues f - term hybrid inflation , while it can easily be implemented in string theory . in a standard formulation of d - term inflation , where the constant fi term gets compensated by a single complex scalar field , we do not add an additional discrete symmetry , and do not consider non - renormalisable terms in the potential , d - term strings are formed at the end of the phase transitions which signals the end of inflation . these strings are analogous to the d - strings formed at the end of brane inflation , which is the result of brane collisions . d - term inflation can not be studied in the standard formulation of supergravity , which is ill - defined whenever the superpotential vanishes and there are present constant fayet iliopoulos terms . following an effective supergravity formulation based on superconformal theory , the superpotential transforms under the u(1 ) symmetry along the directions where the fi terms are constant . this transformation defines the charge assignments of the superfields . d - term inflation has to be studied within this new formulation of supergravity , which is well defined when the superpotential vanishes . cosmic strings contribute to the cosmic microwave background temperature anisotropies . all current measurements put severe constraints on the allowed cosmic string contribution . to achieve a compatibility between measurements and theoretical predictions one should fine tune the couplings . this was already found in the case of minimal supergravity . it was therefore natural to ask the question of whether this result still holds in non - minimal supergravity . it was previously claimed in the literature that higher order khler potentials suppress the cosmic strings contribution . studying a case of non - minimal supergravity , where we include higher order corrections in the khler potential , as well as supergravity with shift symmetry we conclude that cosmic string contribution will be dominant unless the couplings are fine - tuned . we also find that , as in the minimal case @xcite , the @xmath17 constraint on the cosmic string contribution is equivalent to the constraint @xmath283 we would also like to emphasise that if @xmath215 is higher by a factor of 2 , the cosmic string contributison is of the order of @xmath284 , as show in fig . 7 of ref . @xcite . in conclusion , we definitely disagree with the statement that non - minimal khler potentials avoid the cosmic strings problem , which should imply that fine tuning is not necessary . even though we have not studied a large number of khler potentials , our current findings indicate that the problem of fine tuning is unavoidable unless one considers a more complicated model , for example where strings become topologically unstable , namely semi - local strings . we have not yet studied @xcite the spectral index in our models to check the consistency with the limits imposed by the recent three - year wmap data , which however have been given only for purely adiabatic models . if the spectral index is higher than the one preferred from the measurements and since , in addition , the requirement for very small couplings seems more difficult to be satisfied in string theory , one should then look for mechanisms to suppress the rle of strings , for example by making them unstable , along the lines of refs . @xcite , by adding new terms in the superpotential @xcite , or by considering the curvaton mechanism @xcite . as we were completing this work , ref . @xcite came to our attention . in that study , the authors find , in the context of minimal supergravity , slightly higher upper bounds for the parameters @xmath44 and @xmath82 , whereas the obtain the same with us upper bound on the fi term . this small discrepancy originates from the different analysis followed in ref . @xcite ; the authors perform a full markov - chain - monte - carlo analysis and consider one more parameter , namely the spectral index . it is a pleasure to thank n. chatillon , r. kallosh , p. greene , j. gray , j. martin , a. van proeyen and j. yokoyama for stimulating discussions . jr acknowledges support by the national science foundation under grant no . phy-0455649 and by the university of texas at san antonio , texas 78249 , usa . 10 a. linde , phys . b * 129 * , 177 ( 1983 ) .

the study of galaxy formation and evolution requires detailed information about statistically significant samples of dim objects . this , in turn , requires deep imaging and spectroscopy over wide areas of the sky . in pursuit of these data , several wide - deep surveys are now underway . those covering several square degrees or more either lack spectroscopic follow - up ( e.g. noao dwfs , @xcite , and odts , @xcite ) or are restricted to the study of objects at @xmath9 ( except for quasars ) by their imaging depth e.g. the sloan digital sky survey ( @xcite and @xcite ) , the virmos - vlt deep survey ( @xcite , @xcite ) and deep2 @xcite . other surveys target the high - redshift universe with deep hst imaging over fractions of a square degree i.e. the hubble deep fields @xcite , the hubble ultra deep field ( hudf ) , goods @xcite , and gems @xcite . spectroscopic coverage is feasible over these areas but is presently unable to probe deeper than @xmath10 , making the added imaging depth useful only for morphological studies and photometric redshifts . the multiwavelength survey by yale - chile ( musyc ) probes the intermediate regime of a square degree of sky to the spectroscopic limit of @xmath10 . section 2 describes the design of our survey . 3 reports our imaging observations for the extended hubble deep field south . 4 describes our imaging reduction , and 5 covers our photometric calibration and photometry . 6 gives our results for @xmath4-band number counts and the sky density and angular clustering of @xmath6-selected lyman break galaxies . 7 concludes . our analyses assume a standard @xmath11cdm cosmology with @xmath12=0.27,@xmath13 , and @xmath14=70km / s / mpc . the multiwavelength survey by yale - chile ( see http://www.astro.yale.edu/musyc ) is designed to provide a fair sample of the universe for the study of the formation and evolution of galaxies and their central black holes . the core of the survey is a deep imaging campaign in optical and near - infrared passbands of four carefully selected @xmath15 fields . musyc is unique for its combination of depth and total area , for additional coverage at x - ray , uv , mid- and far - infrared wavelengths and for providing the @xmath16 photometry needed for high - quality photometric redshifts over a square - degree of sky . the primary goal is to study the properties and interrelations of galaxies at a single epoch corresponding to redshift @xmath17 , using a range of selection techniques . we chose to use the @xmath0 filter set in the optical in order to obtain six nearly - independent flux measurements with the broadest possible wavelength coverage . lyman break galaxies at @xmath5 are selected through their dropout in @xmath18-band images combined with blue continuua in @xmath19 ( @xmath20 in the rest - frame ) typical of recent star formation @xcite . imaging depths of @xmath18,@xmath21,@xmath22,@xmath23 were chosen to detect the lbgs , whose luminosity function has a characteristic magnitude of @xmath24 in @xmath25 , and to find their lyman break decrement in the @xmath18 filter via colors @xmath26 . lyman @xmath27 emitters at @xmath5 are selected through additional deep narrow - band imaging using a 50 fwhm filter centered at 5000 . these objects can be detected in narrow - band imaging and spectroscopy by their emission lines , allowing us to probe to much dimmer continuum magnitudes than possible for lyman break galaxies . it has recently become clear that optical selection methods do not provide a full census of the galaxy population at @xmath28 , as they miss objects which are faint in the rest - frame ultraviolet @xcite . with this in mind , musyc has a comprehensive near - infrared imaging campaign . the nir imaging comprises two components : a wide survey covering the full square degree and a deep survey of the central @xmath29 of each field . this division between deep and wide was chosen because of the @xmath30 field - of - view of the ispi near - infrared camera on the ctio 4 m telescope . the @xmath31 point source sensitivities of the wide and deep components are @xmath32 and @xmath33 respectively . nir imaging over the full survey area provides a critical complement to optical imaging for breaking degeneracies in photometric redshifts and modeling star formation histories . deeper @xmath34 imaging over @xmath29 subfields opens up an additional window into the @xmath35 universe as the @xmath36 selection technique @xcite will be used to find evolved optically - red galaxies at @xmath37 through their rest - frame balmer/4000 break . extensive follow - up spectroscopy is being conducted over the square degree . a subset of the color - selected lyman break galaxy candidates will turn out to be agn based on broad- or narrow - line emission features seen in follow - up spectroscopy @xcite . damped lyman @xmath27 absorption systems @xcite at @xmath38 , which comprise the neutral gas reservoir needed to form most of the stars in the universe @xcite , will be searched for in the spectra of the brightest color - selected lbg / agn candidates ( typically quasars at @xmath28 ) . in addition to the optical and near - infrared , imaging campaigns at other wavelengths and follow - up spectroscopy are integral parts of musyc . x - ray selection will be used to study agn demographics over the full range of accessible redshifts , @xmath39 , ( see * ? ? ? * ) with spitzer imaging used to detect optically- and x - ray - obscured agn @xcite . this also allows a census of accreting black holes at @xmath5 in the same fields to study correlations between black hole accretion and galaxy properties at this epoch . future epochs of optical imaging will be used to conduct a proper motion survey to find white dwarfs and brown dwarfs in order to study galactic structure and the local initial mass function ( see * ? ? ? * ) ; the additional epochs will also enable a variability study of agn . the four survey fields ( see table [ tab : fields ] ) were chosen to have extremely low reddening , hi column density @xcite , and 100@xmath40 m dust emission @xcite in order to facilitate satellite coverage with spitzer , hst , chandra , and xmm , to take advantage of existing multiwavelength data and to enable flexible scheduling of observing time during the year . additionally , each field satisfies all of the following criteria : minimal bright foreground sources in the optical and radio , high galactic latitude ( @xmath41 ) to reduce stellar density , and accessibility from observatories located in chile . the survey fields will be a natural choice for future observations with alma . the remainder of this paper describes our optical images and catalog of e - hdfs . the techniques used for data reduction and photometry are the same as those used for the analysis of the other three fields . optical imaging from the full survey will be reported in e. gawiser et al . ( 2005 , in preparation ) . the near - infrared data will be discussed in r. quadri et al . ( 2005 , in preparation ) . the e - hdfs has deep public space - based observations at uv , optical , near - infrared , and far - infrared wavelengths . the hdfs itself covers a small @xmath42 central region with wfpc2 plus stis and nicmos regions , with deep ground - based jhk coverage of the wfpc2 region available from the fires survey @xcite . spitzer irac and mips coverage of the central @xmath43 is being performed in gto time . the extended area around hdfs has previously been imaged by @xcite and @xcite to a depth sufficient for the study of galaxies at @xmath44 . these images were made public and combined with deep h - band images by the las campanas infrared survey ( lcirs , @xcite ) to study red galaxies out to @xmath45 . our survey goes about one magnitude deeper in ubvri@xmath46 to probe the @xmath2 universe . our extended hubble deep field south ( e - hdfs ) field center ( see table [ tab : fields ] ) was chosen to keep a bright star ( m=6.8 ) which lies just north of the wfpc2 field off of the ctio+mosaic detectors . lcirs covered an h - shape centered on wfpc2 and thereby provides public @xmath47-band coverage of roughly half of our e - hdfs field . optical images of e - hdfs were taken on the nights of 2002 october 6,8,10,12 and 2003 may 26,27,28 using the 8192@xmath488192 pixel mosaic ii camera consisting of 8 2048@xmath484096 ccds , each with 2 amplifiers , on the blanco 4 m telescope at ctio . afternoon calibrations were obtained , including zero - second exposures to trace the readout bias pattern , and domeflats in each filter to be observed except for @xmath18 for which the counts from the dome lamps were insufficient . twilight flats were therefore obtained for the @xmath18 filter . dark exposures of comparable length to the imaging observations were obtained , but due to the negligible dark current they were not used for data reduction . a standard dither pattern was used to fill in gaps between the 8 ccds . the pixel scale is 0.267@xmath49/pixel , leading to coverage spanning @xmath50 of sky with each pointing . figure [ fig : filters ] shows the filter response curves and their multiplication with the ccd quantum efficiency and atmospheric transmission at one airmass . table [ tab : obs ] gives details of the exposure times in each filter for each run along with the approximate average seeing measured in the raw images during observations . the seven nights were mostly cloudless , but moderate clouds affected some of the @xmath22 imaging of e - hdfs on 2002 october 10,12 and some of the @xmath4 imaging on 2003 may 28 ; our reduction methods described below allow these images to be used without biasing the photometry . photometric standard fields from @xcite large enough to cover the full mosaic ii field of view were observed each night , and the nights of 2002 october 6 and 2003 may 26 proved photometric . @xmath0 images from ctio4m+mosaic ii were reduced using the mscred and mscdb packages in iraf v.2.12 following the noao deep wide field survey cookbook v.7.02 . we used custom software to work around a few difficulties in these packages , as described below . a composite zero image is subtracted from each raw image to remove the amplifier bias level and pattern . the resulting image is then `` flat - fielded '' by dividing by a composite domeflat , or a composite twilight flat in the case of @xmath18-band . superskyflat _ for each filter is made by combining all of the flat - fielded , unregistered images taken in each filter each night with rejection used to remove sources . our 2@xmath51 amplitude dither pattern was designed to eliminate the wings of bright sources from the _ superskyflat_. each flat - fielded image is then divided by the appropriate _ superskyflat _ , which offers an estimate of the pixel - by - pixel response to the spectrum of the night sky with sufficient counts to achieve 1% precision per pixel . because the _ superskyflat _ was produced using flat - fielded images , dividing by it serves to remove the original domeflat ( twilight flat ) from the reduction process . the real influence of the original flat - fielding is to remove the illumination pattern and gross pixel - to - pixel variations before looking for cosmic rays and bright objects to reject in making the _ superskyflat_. using the _ superskyflat _ to correct for the pixel response is considered preferable to using the domeflat because the ccd response to the spectrum of the dome lamp ( or twilight sky ) can have significant systematic differences from its response to the spectrum of the night sky . for background - limited photometry , this is an important effect . we then find an astrometric solution for each image , starting with fiducial wcs headers provided for each fits extension of the raw images and comparing the claimed positions with the known positions of stars in the usno - b catalog using the mscred routine _ msccmatch_. we found _ msccmatch _ to be finicky ; for some runs the fiducial headers were too inaccurate to be corrected with the maximum second - order terms used by the mscred package . an iterative , non - interactive procedure of calling _ msccmatch _ multiple times proved sufficient . the final rms astrometric errors are between 0.2@xmath49 and 0.3@xmath49 in each image , which is consistent with the uncertainties on individual usno - b stars despite having fit many more stars than free parameters , and the actual solution should be better than 0.2@xmath49 rms . to perform aperture photometry later on , our final images are transformed to have a common pixel scale and tangent plane projection point . this is accomplished by projecting each processed image ( after bias - subtraction , flat - fielding , _ superskyflat_-fielding , and correction of wcs header information ) onto the tangent plane of a common reference image using _ mscimage_. the reprojection performed by _ mscimage _ should be performed with flux conservation set to _ no _ the first time but then set to _ yes _ for any further reprojections . this is because the pixel scales in raw mosaic ii images are a function of radius from the field center with several percent variation from center to corner of the field . the process of flat - fielding removes the true illumination pattern caused by more photons falling in larger pixels leading to an illusion of flat background counts despite the variation in pixel scales . mscimage _ is used for astrometric projection into a uniformly sized grid of pixels , flux conservation would re - introduce the illumination pattern , thereby causing a variable photometric zeropoint across the image . turning off flux conservation instead produces an average of the values of initial pixels neighboring each final pixel location that offsets the error introduced by flat - fielding . we make three different stacked `` final '' images for each filter . unweighted versions are made first in order to determine the weights for the other two versions . the versions denoted by _ xs _ were combined using weights optimized for surface brightness , which is the method used by the ndwfs cookbook . the versions denoted by _ ps _ were made using weights optimized for point sources ; details can be found in the appendix . we synthesize a @xmath52__ps _ image to use as our photometric detection image by adding the @xmath21 , @xmath22 , and @xmath4 _ ps _ stacks using point - source - optimized weights derived from the signal and noise statistics of these stacks . we decided not to add @xmath18 to @xmath52 although it is also very deep because some low - redshift objects have different morphologies in @xmath18 , the seeing is typically worse in @xmath18 , and objects at @xmath5 will be nearly invisible in @xmath18 so adding it to @xmath52 would just add noise for these objects . note that for similar reasons objects that drop out in @xmath21 and @xmath22 are somewhat less likely to be detected in our @xmath52 image than in a @xmath53 or @xmath4 image alone . we prefer the @xmath52 approach to the creation of a @xmath54 image advocated by @xcite because it allows us to measure object morphological parameters directly from the same image used for object detection , which is not recommended in the latter approach . moreover , the optimally - weighted @xmath52 combination reduces the emphasis on single - filter outliers employed by the @xmath54 approach and avoids its flaw of treating both negative and positive sky fluctuations as evidence of an object . we trim the final @xmath52__ps _ image to the maximum size region that has nearly uniform signal - to - noise ratio in all three input images @xmath21 , @xmath22 , and @xmath4 . for e - hdfs , the trimmed image is 7395@xmath487749 pixels or @xmath55 . then the iraf task _ imalign _ is used to shift and trim the other images to match , giving our images the common origin and size required for aperture photometry with sextractor . the trimmed images are normalized to an effective exposure time of one second . after the pipeline reduction , the background is flat to better than 1% in all filters . the remaining low level , large scale fluctuations were subtracted using sextractor @xcite with mesh size back_size=64 and median filter back_filtersize set to 6 mesh units . the resulting final @xmath52 image is shown in figure [ fig : bvr ] with a magnified image of a small section shown in fig . [ fig : zoom ] . new fits headers have been added to these final images indicating input information for sextractor photometry runs and other routines , specifically : + satur_level , the empirically determined saturation level in each image , which is usually a factor of a few less than the apparent saturation level of the brightest stars , + seeing_fwhm , the mode of the seeing for a set of bright unsaturated stars , + mag_zeropoint , the ab magnitude of an object that gives 1 count per second derived from photometric calibration as described below , + flux_zeropoint , the flux in @xmath40jy corresponding to 1 count per second , + tot_exptime , the total exposure time that each final image represents , and + gain_tot , the number of photoelectrons represented by each count in the final image . our calibration scheme is to take @xmath565 minute exposures of our fields ( referred to as `` calibration images '' ) in all optical filters on photometric nights . this was achieved for e - hdfs on oct . photometric calibration is performed using a version of the landolt catalog with all magnitudes and colors converted into the ab95 system of @xcite ( referred to throughout this paper as ab magnitudes ) . we included @xmath46 magnitudes for landolt standard stars in our catalog of standards using the formula @xmath57 which was generated from the formulae of @xcite and has been corrected for a 0.02 magnitude bias found when comparing this prediction with measurements of @xmath46 magnitudes of bright landolt stars by @xcite . the resulting 0.04 magnitude rms error was used to predict errors in these estimates . this procedure offers a significant reduction in observing time and a tremendous increase in the number of calibration stars versus the traditional method of using spectrophotometric standard stars for filters outside the johnson - cousins system . only a few stars per landolt field have been turned into sdss calibrators @xcite and they are typically too bright for the 4 m telescope to use without defocusing . table [ tab : photcal ] lists the airmass and seeing for each of our calibration images along with our photometric solution for the magntiude in terms of measured counts per second , @xmath58 where @xmath59 is the magnitude zeropoint , @xmath60 is the airmass coefficient , @xmath61 is airmass , and the color term is listed in table [ tab : photcal ] . our images of landolt standard fields occupied a range of airmasses bracketing the airmasses of these calibration images but insufficient to break the degeneracy between zeropoint and airmass coefficient . we therefore used fixed airmass coefficients , with the resulting uncertainty reflected in the uncertainty in the zeropoint fit , which is @xmath621.5% in each filter . since ab95 was carefully calibrated for the standard johnson - cousins @xmath63 filter set , the conversion of the landolt catalog is quite accurate . however , the filter system used at ctio is not a precise match to johnson - cousins , making the use of a color term helpful in photometric calibration . the relatively small color coefficients shown in table [ tab : photcal ] minimize the scatter in our photometric solution by providing a better estimate of the true ab magnitude of each landolt star in our observed filter system . we set the ab color coefficient to zero when determining photometry for our final object catalog ; this places the fluxes on our observed filter system rather than standard johnson - cousins filters . a color correction can be calculated later for applications that require fluxes extrapolated to johnson - cousins colors . because photometric redshift codes multiply spectral templates by the atmospheric transmission , filter transmission , and ccd quantum efficiency used for the observations , they already account for the origin of the color term and should be given fluxes in the observed filter system . the presence of non - power - law features such as lyman and balmer breaks in these template spectra ( and real objects ) makes it desirable to avoid color transformations calibrated with stars . we then used the calibration images to create standard stars in each filter for e - hdfs and used these new standard stars to calculate the zeropoint in our final images . this avoids attempting to explicitly correct the photometry for the airmass and cirrus extinction in each image which contributed to our final images , allowing us to use images taken in non - photometric conditions as part of the final stack . for e - hdfs we used a set of 10 stars , each from a different original amplifier , chosen to be bright enough to provide good signal in @xmath18 and yet to avoid saturation in @xmath4 , @xmath64 , and @xmath46 . care was taken to choose some stars near the field center and some near the sides and corners of the field to check for systematic effects . the photometry of the new standards was determined using the iraf routine _ phot_. we used the same @xmath65 diameter aperture for photometry of these standard stars in order to minimize aperture losses . we repeated this process by using _ phot _ on our final science images for the same stars . we checked for systematic patterns in the offsets but found none , and the rms scatter was 0.03 magnitudes , implying a smaller error in the mean offset . the total exposure time , seeing , photometric zeropoints , and 5 @xmath66 point source detection depths ( ab ) of our final images are given in table [ tab : final ] . flux zeropoints are also determined using the definition @xcite @xmath67 given the measured uncertainties in the initial zeropoints and our measured scatter in the solution for the zeropoints of the final images , we estimate a total systematic uncertainty of 3% in the zeropoint of each filter . derivations of optimal photometric techniques and their uncertainties typically assume poisson noise which is uncorrelated between image pixels . however , the astrometric reprojection used to place all images on a common grid correlates neighboring pixels , leaving the large - scale noise properties unchanged but making a pixel - by - pixel rms a poor measure of the typical noise due to sky fluctuations in a larger aperture . as this is the method used by sextractor to estimate photometric errors , we undertook a detailed investigation to test its accuracy , which turns out to be poor , and to find an improved method . our empirical method for estimating photometric uncertainties as a function of aperture size is also sensitive to large - scale noise correlations from unsubtracted nearby objects and from any ccd defects that survive flat - fielding . wings from bright objects appear to affect the noise statistics ; when we set all pixels belonging to detected objects to zero and displayed the resultant image , wings are clearly visible around the brightest objects . we estimate the error due to background fluctuations and noise correlations in each filter via a custom idl code which places @xmath68 random apertures of a given size on the sky - subtracted image which do not overlap with any of the pixels in the segmentation map of isophotal object regions produced by sextractor . we use circular apertures of area @xmath69 centered at integer pixels and describe them by an effective size @xmath70 . a gaussian is then fit to the histogram of aperture fluxes to yield the rms background fluctuation as shown in fig . [ fig : hist ] . the histograms appear well described by the best - fit gaussians . figure [ fig : sigmav ] shows the rms of aperture fluxes vs. n for raw , sky - subtracted raw , and final @xmath22 images . the solid curve gives our fit to the noise properties of the final @xmath22 image using the function @xmath71 where @xmath72 , @xmath73 for the measured rms pixel noise of @xmath74=0.014 . an alternative formula suggested by @xcite , @xmath75 yields an equally good fit , with @xmath76 , @xmath77 . this formula shows an explicit sum between contributions from poisson noise , which is independent from one pixel to another and hence has rms proportional to @xmath78 , and correlated noise from fluctuations in background level on scales larger than the aperture which yields an rms proportional to @xmath79 . these simple , extreme cases are shown as dotted lines starting from the value of @xmath80 in figure [ fig : sigmav ] and are seen to bracket the true behavior except at very small aperture sizes . apertures of just a few pixels in area are affected by the small - scale noise correlations introduced by re - pixelization performed during re - projection ; this effect should actually depress @xmath80 since each final pixel is the average over several nominally independent input pixels . we prefer the simplicity of eq . [ eq : power ] which reflects the reality that noise exists on a range of scales leading to an effective power - law behavior intermediate between these two extremes . table [ tab : noise ] lists our measured rms - per - pixel @xmath74 and the fit coefficients @xmath27 and @xmath81 for all of our filters ; the uncertainties in @xmath27 and @xmath81 are highly correlated but both have been determined to 5% precision , whereas the uncertainties in @xmath74 measurements are at the 1% level . a similar analysis of our raw images shown in figure [ fig : sigmav ] reveals that the noise in the raw images ( squares ) is fully correlated due to the domination of background fluctuations . once a sky - subtraction has been performed ( diamonds ) , the noise properties resemble those of the final image except for being noisier by a factor roughly equal to the square root of the number of exposures . while some of this improvement comes from flat - fielding , it appears that the correlated noise is due to background fluctuations from objects whose subtraction improves with the square root of observing time . this is inconsistent with confusion noise from undetected sources , which is not expected to be a significant factor even in optical images this deep . one possibility for the origin of this correlated noise is that the global sky subtraction performed by sextractor is insufficient . however , when we mimic the backphoto_type local mode of sextractor by estimating the local sky in a square annulus around each aperture and subtracting this off we find a slight increase in the noise correlations . hence it appears that the background contributions from sky gradients and nearby objects have been subtracted about as well as one could expect , and that the remaining contributions increase the noise significantly above the extrapolation from @xmath74 one would assume for the case of uncorrelated noise . we wish to determine accurate photometry for a wide range of object sizes and brightnesses , with particular attention to high - redshift galaxies with half - light radii 0.2 - 0.3@xmath49 @xcite ) which are dim and typically unresolved in our @xmath82 seeing and can therefore be approximated as point sources . indeed , @xcite showed that most galaxies at @xmath83 have half - light radii @xmath84 regardless of redshift . in traditional aperture photometry , the flux of an object is measured by giving constant weight to one region of the image and zero weight to the rest , with fractional weights used for pixels which fall only partially in the chosen aperture . the simplest case is a circular aperture centered at the object centroid . for the case of uncorrelated noise dominated by the sky background and a gaussian psf , the circular aperture with optimal signal - to - noise is easy to derive and has a diameter equal to 1.35 times the seeing fwhm . however , the actual psf in each filter has a near - gaussian core with broad wings i.e. more of the flux is found outside the fwhm than for a gaussian . to investigate the effect of the non - gaussian psf on photometry we measured the enclosed flux as a function of radius for a set of objects selected to be bright but unsaturated ( @xmath85 ) and unresolved according to sextractor ( @xmath86 , as discussed below ) . [ fig : psf ] shows the results for the @xmath18 , @xmath22 , and @xmath64 bands plus a gaussian of 0.96@xmath49 fwhm that represents an excellent fit to the core of the @xmath22-band psf but only encloses 80% of the total flux . a 14@xmath49 diameter aperture was used to define the total object flux , with rapid convergence seen in the bluer filters but only gradually in @xmath4,@xmath64 , and @xmath46 . while this could be due in part to contributions from neighboring objects , we used the median statistic to define these profiles to avoid sensitivity to the small fraction of objects with bright neighbors . the contribution of nearby objects should lead to a divergence with increasing aperture diameter which is not seen , so it appears that the psf wings do contain a higher fraction of the total flux in the redder bands , with perhaps a few percent missed by our 14@xmath49 diameter aperture . figure [ fig : sn ] plots the enclosed flux , the rms background fluctuation , and the resulting signal - to - noise for point sources as a function of aperture diameter . there are competing effects which serve to make the optimal aperture slightly smaller than the idealized result of 1.35 @xmath48 fwhm . the non - gaussian psf places more flux outside the `` optimal '' diameter , so a larger aperture would find more signal . however , the large - scale noise correlations mean that noise increases with aperture size faster than in the ideal poisson case . table [ tab : apertures ] lists various seeing measurements along with the optimal circular aperture diameters and fractional flux enclosed within these apertures for each filter which is used to correct the observed fluxes to estimate the total object flux in each filter . a systematic increase in the seeing occurs when the estimate is made using sextractor s _ fwhm _ parameter , which assumes a gaussian psf , instead of iraf s _ imexam _ routine , which allows a moffat fit . the non - gaussian psf shape is reflected in the further increase in the diameter containing half of the object light implied by the half - light radii @xmath87 output by sextractor ; a gaussian contains half of the flux within one fwhm . the optimal circular apertures for point sources turn out to be quite close to these values for the half - light radii . the image quality is measured to be nearly constant across the @xmath88 field , with systematic variations in fwhm and half - light radius seen at the 10% level in the @xmath64 band but at no more than the 5% level in the other filters . for extended sources , our apertures are no longer optimal , and the corrections for the fraction of enclosed flux are incorrect . these sources are barely affected by seeing smaller than their intrinsic size , leading to a flux better described by the aperture area than by the fractional enclosed point - source flux . the traditional solution is to use larger apertures ( @xmath89 diameter ) to measure colors of extended objects as well as point sources . our analysis shows that the signal - to - noise of point source photometry is reduced by up to 30% by using these larger apertures , with a bias in the @xmath90 color of point sources of 20% ( 0.2 magnitudes ) introduced by the different seeing in these images . this could be solved by convolving all of the images to the relatively poor seeing ( fwhm=1.3@xmath49 ) of the @xmath18 band ( see @xcite ) . however , the non - gaussian psf shape requires a non - gaussian convolution kernel ; simply matching the fwhm by gaussian convolution is not sufficient to fully match the psf . moreover , smoothing the other images to match the psf shape of the @xmath18 band would be costly in terms of decreasing the signal - to - noise for point source photometry , especially given the observed large - scale correlations in background noise . in choosing the `` optimal '' aperture sizes for each filter in table [ tab : apertures ] , we have chosen amongst the range of apertures with signal - to - noise@xmath9195% of the optimal value in a manner that reduces the range of aperture sizes in order to reduce the errors introduced for extended sources . this has the effect of choosing slightly larger apertures in the redder filters than would formally maximize the signal - to - noise . this provides additional benefits , as the approximations that high - redshift galaxies are point sources , image co - registration is perfect , and object positions have no error all become less valid as the apertures decrease in size . our simulations indicate that these combined effects will not affect object colors beyond the 10% level for the chosen aperture sizes . our primary solution to the challenge of obtaining decent colors for both unresolved and extended objects dimmer than the night sky is to create corrected aperture fluxes ( hereafter referred to as apcorr ) . we tried using the a_image and b_image parameters to model the flux distribution as an ellipsoid but encountered too many systematic problems with these parameters to prefer this approach . we therefore use the half - light radius ( flux_radius ) of each object determined by sextractor in the @xmath52 detection image and assume a two - dimensional gaussian light profile with twice this value as fwhm . the half - light radius measured in the @xmath52 detection image represents the object s intrinsic size convolved with the @xmath52 seeing , and this allows us to predict the observed size in each filter given the known seeing measured from the median half - light radii of bright point sources listed in table [ tab : apertures ] . for object @xmath92 in filter @xmath93 we expect @xmath94 \ ; \ ; \ ; , \label{eq : sigma_squared_ij}\ ] ] where the last term in parentheses modifies the quadrature sum of intrinsic object size and bvr seeing to the needed sum of intrinsic object size and seeing in filter @xmath93 and the prefactor converts from half - light radius ( which equals half - width half maximum for a gaussian ) to the rms of the light profile . the fractional flux of object @xmath92 falling within the optimal circular aperture radius @xmath95 for filter @xmath93 listed in table [ tab : apertures ] is then assumed to be @xmath96 where the first value in parentheses refers to the fractional flux for point sources listed in table [ tab : apertures ] and the second gives the fractional flux of the two - dimensional gaussian which approximates the object . taking the minimum of these two values prevents noisy data from causing an object to be inferred to be smaller than a point source . each object s flux measured within the optimal aperture for a given filter is then corrected by dividing by @xmath97 to infer the total flux in that filter . another solution for attempting to handle both extended and unresolved sources is sextractor s auto aperture , which uses a kron ellipse adapted to each object s light profile to find @xmath98 of the flux for extended sources and @xmath99 for point sources using the parameters we have adopted , according to @xcite . in the @xmath18 , @xmath21 , and @xmath22 filters , we find that auto indeed measures 96% of the point source flux found in a 14@xmath49 diameter aperture , with the value decreasing to 93% in @xmath4 , 92% in @xmath64 , and 89% in @xmath46 in keeping with the broader psf discussed above . because sextractor assumes poisson sky noise , it produces underestimated photometric uncertainties for objects dimmer than the background sky . detection image as a source of uncertainty in object detection and the background in the measurement image as a source of uncertainty in the photometry . ] the standard formula used by sextractor to predict photometric errors in flux as a function of aperture size is @xmath100 where the first term represents sky fluctuations assuming uncorrelated noise between pixels , @xmath101 is the object flux in counts within the aperture , and @xmath102 is the total effective gain used to turn counts in the normalized image into original photons detected by the ccd , with the fraction @xmath103 giving the poisson variance . sextractor measures @xmath74 using the median fluctuation in its internally computed background map , and this measurement agrees with ours to within a few percent . we modified this to make use of our fits for ( correlated ) background noise which yields @xmath104 where the first term now represents empirical sky fluctuations which include the subdominant contribution from electronic readout noise . both noise terms will be larger in regions of the image where reduced exposure time leads to a lower effective gain and hence higher background rms . sextractor accounts for this by using a variance map @xmath105 which is either input as a weight_image or generated internally from a map of background fluctuations in the image . the errors produced by sextractor as a function of position are then @xmath106 the ratio of the formulas in equations [ eq : uncertainty_se ] and [ eq : uncertainty ] allows us to make a simple correction to the flux uncertainties output by sextractor to yield @xmath107 for negative fluxes , @xmath101 is set to zero in all of these error terms , since the uncertainty is entirely dominated by the background fluctuations . for auto apertures , the area is given by that of the kron ellipse used i.e. @xmath108 ( the so - called kron_radius is really a scaling of the a_image and b_image parameters based on the second moment of the object light distribution ) . the resulting rms has units of counts per second and is converted to flux using the zeropoints in table [ tab : final ] . the uncertainty determined for optimized aperture fluxes is then multiplied by the same correction factor used to estimate the total object flux i.e. the reciprocal of @xmath97 defined in eq . [ eq : fractional_flux_ij ] . finally , the uncertainty in the corrected aperture flux is increased by a factor @xmath109 which serves to amplify the uncertainties for extended sources to account for the uncertainty in their correction factors . this reproduces the errors found in corrected aperture fluxes for extended objects in our simulations described in [ sec : simulations ] . the photometric errors derived in this manner do not include the 3% calibration uncertainty which is common to all sources in a given filter . we used the iraf package _ artdata _ to simulate stars and galaxies with known magnitudes and positions and to add them to our observed images to get realistic crowding effects and background noise . we ran sextractor on these `` simulated '' images in dual - image mode to simulate our full photometric pipeline . we found object centroiding errors to be @xmath110 pixels i.e. @xmath111 at @xmath112 . in the centroiding of individual objects we found auto photometry to be nearly unbiased for both single filter fluxes and colors . however , figure [ fig : stars ] shows that auto has larger errors for point sources than the corrected aperture fluxes due to the larger auto apertures including significantly more sky noise , so we recommend auto fluxes only for extended sources and give fluxes for both types of apertures in our catalog . the errors seen in the @xmath18 and @xmath22 filters appear to be entirely uncorrelated . figure [ fig : gals ] shows that auto performs similarly to corrected aperture fluxes for extended objects , although both show significant covariance between @xmath18 and @xmath22 due to misestimation of the true object light profile . figure [ fig : sigma_stars ] shows the flux errors divided by their uncertainties for simulated point sources . apcorr again appears unbiased but this plot reveals that the error estimates are roughly equally accurate for both cases . the median squared value is close to one in both cases , so the reported flux uncertainties in the catalog appear trustworthy for point sources . figure [ fig : sigma_gal ] shows the flux errors divided by their nominal uncertainties for simulated galaxies . this plot illustrates that the estimated uncertainties are typically too small for galaxies in both cases . figure [ fig : colors ] shows that @xmath90 colors of simulated objects appear slightly biased for auto but unbiased for apcorr , and the errors are a bit smaller for apcorr for point sources and of similar size for both types of fluxes for galaxies . the color errors are highly correlated , especially for galaxies , implying that these errors are primarily caused by errors in the isophotal object detection by sextractor on which both fluxes depend . these include problems caused by blending with nearby neighbors , although the median object is unaffected by neighbors in this uncrowded field . the auto and apcorr colors of real objects are discussed in [ sec : lbg_select ] . we performed aperture photometry using sextractor in dual - image mode with @xmath52 as the detection image and each final image as the measurement image , leading to a common catalog for all filters . detection was performed after filtering with a @xmath113 pixel gaussian convolution kernel with 4 pixel ( 1.07@xmath49 ) fwhm which is well - matched to the seeing in our @xmath52 image ( see appendix ) . we used a threshold of 1.2 for both detection and analysis ; a single pixel at the 5 @xmath66 level in the original @xmath52 image would barely make the 1.2@xmath66 threshold after filtering . using the identical parameters , we detected 318 objects in a negative of the @xmath52 image , so given the symmetrical nature of sky fluctuations and readnoise we estimate an equivalent number of spurious objects in our final catalog i.e. 0.5% . for each filter , we used the `` optimal '' apertures and corrections described above as well as the auto aperture determined for each object by sextractor centered on the @xmath52 barycenter . the photometry was corrected for galactic dust extinction of e(b - v)=0.03 @xcite . the 5@xmath66 point source detection depths in table [ tab : apertures ] were determined by multiplying the @xmath66 of the measured sky noise for the optimal point source aperture sizes by 5 and then correcting for the flux missed by these apertures to turn sky fluctuations into total point source magnitudes . many estimates of point source detection limits in the literature are based on extrapolating the rms pixel noise , @xmath74 , to the chosen aperture size assuming uncorrelated sky noise . for correlated noise like that in our images , which we expect is typical , this will overestimate the true depth in a 2@xmath49 diameter aperture by 0.4 magnitudes and in a 3@xmath49 diameter aperture by 0.6 magnitudes . our optical photometric catalog of 62968 objects in e - hdfs is available in full in the electronic version of the journal and from our website , with the first several lines shown in table [ tab : catalog ] . all positions and shape parameters are measured from the final @xmath52 image . the columns in the catalog offer the following information : + _ column 1 _ : object number starting with first object as 0 + _ column 2 _ : object name ( musyc- ) + _ column 3 _ : right ascension in decimal hours , double precision + _ column 4 _ : declination in decimal degrees , double precision + _ column 5 _ : x barycenter + _ column 6 _ : y barycenter + _ column 7 _ : stellarity classification + _ column 8 _ : half - light radius + _ column 9 _ : rms of flux distribution along major axis + _ column 10 _ : rms of flux distribution along minor axis + _ column 11 _ : object position angle counterclockwise from x - axis + _ column 12 _ : flags output by sextractor + _ column 13,15,17,19,21,23 _ : auto flux in @xmath18,@xmath21,@xmath22,@xmath4,@xmath64,@xmath46 + _ column 14,16,18,20,22,24 _ : auto flux uncertainty in @xmath18,@xmath21,@xmath22,@xmath4,@xmath64,@xmath46 from ( [ eq : correction ] ) + _ column 25,27,29,31,33,35 _ : apcorr flux in @xmath18,@xmath21,@xmath22,@xmath4,@xmath64,@xmath46 after division by correction factor based on half - light radius described in eq . [ eq : fractional_flux_ij ] + _ column 26,28,30,32,34,36 _ : apcorr flux uncertainty in @xmath18,@xmath21,@xmath22,@xmath4,@xmath64,@xmath46 from ( [ eq : correction ] ) after division by correction factors based on half - light radius described in eq . [ eq : fractional_flux_ij ] and multiplication by additional factor for extended sources described above + photometric measures in the catalog are given in units of flux density in @xmath40jy ( 1 @xmath40jy = @xmath114 w m@xmath7 hz@xmath115 ) . the conversion to ab magnitudes is simple using the formula @xmath116 @xcite . this avoids the loss of information that comes from sextractor representing objects with negative aperture fluxes as having @xmath117 and the confusion that results from flux errors consistent with a non - detection being turned into extremely large magnitudes . photometric errors are nearly symmetrical in flux but not in magnitudes , even for objects with low signal - to - noise , making it acceptable to represent the uncertainties with a single number . this makes flux density the natural unit to be used for calculating photometric redshifts if one wishes to use a @xmath54 function to compute the likelihood as is done by the publicly available codes . for color selection , one can either turn color criteria into desired flux ratios between filters or turn the measured fluxes into magnitudes and subtract to obtain colors . sextractor offers a neural network classification of objects into `` stars '' and galaxies ; @xmath86 implies an unresolved profile consistent with the psf , @xmath118 implies an extended profile , and @xmath119 represents uncertain objects which are typically too dim for this classification to work successfully . figure [ fig : class ] shows the stellarity results for our catalog . the classification starts to break down at @xmath120 , with brighter objects confidently split between stellar and extended profiles but many dimmer objects receiving an inconclusive value . stellarity becomes nearly useless at @xmath121 where the majority of values fall in the uncertain regime , which is appropriate given that @xmath122 for most galaxies at @xmath123 @xcite . however , the vast majority of objects at @xmath120 are galaxies due to the rapidly rising galaxy luminosity function and the plateau in the stellar equivalent . this means that for many scientific goals all objects with @xmath124 may be considered galaxies . figure [ fig : radec ] shows the locations of all objects in our catalog . there is an overabundance of objects along the image border which is not apparent to the eye but can be eliminated by removing the 574 objects with @xmath125 , which identifies objects truncated by the image border . the one visual blemish is a line of spurious objects detected along a heavily saturated column extending from the brightest star in the field . this line is removed by eliminating all 1226 objects with @xmath126 , which implies saturation in at least one filter . the sky density of objects in our catalog is greatly reduced near the brightest stars , leading us to create a mask to properly exclude these regions from analyses of sky density and clustering . our methodology is described in detail in l. infante et al . ( 2005 , in preparation ) . a careful analysis of the implied stellar and galaxy luminosity functions as a function of stellarity cut shows that stellar contamination is minimized with negligible loss of galaxy counts by requiring @xmath124 . the total @xmath4-band magnitude counts ( from auto ) are shown in figure [ fig : rcounts ] and appear 90% complete to @xmath127 and 50% complete to @xmath128 . part of the incompleteness comes from auto beginning to underestimate the flux of very dim objects , as noted by @xcite . the fit for the number of galaxies per magnitude per square degree is @xmath129 . our slope of 0.34@xmath1300.01 agrees well with previous measurements of 0.321@xmath1300.001 by @xcite , 0.361@xmath1300.004 by @xcite , 0.31@xmath1300.01 and 0.34@xmath1300.01 in two different fields by @xcite , and 0.39 by @xcite . our differential counts at @xmath128 of 1.3@xmath131 agree well with the values of 1.2@xmath132 reported by @xcite . -band galaxy counts per unit magnitude per square degree with poisson error bars and solid fit of @xmath133 based on counts at total ab magnitudes @xmath134 . incompleteness is minimal below total magnitude @xmath127 and reaches 50% around @xmath128 . [ fig : rcounts ] ] various filter sets have yielded success at finding lbgs , including @xmath135 @xcite , sdss @xmath136 @xcite , @xmath137 @xcite , @xmath138 @xcite , and @xmath139 ( @xcite , @xcite ) . figure [ fig : uvr_stars ] shows star colors expected and measured in our survey versus our adopted @xmath6 color selection criteria of @xmath140 where all magnitudes are ab . the color criteria shown as dashed lines in the figure result from the transformation of the @xmath141 @xmath142 @xmath143 criteria described by @xcite into @xmath6 . and @xmath144 . ] although the upper branch of the stellar distribution at red @xmath145 colors represents giants which are rare at our survey depth and high galactic latitude , we shifted our criteria to avoid this region as it also contains dim dwarf stars with correspondingly large errors in color , which are the primary expected source of interlopers . the region of the steidel et al . criteria avoided by this shift had the highest fraction of interlopers and also shows a high interloper fraction in the more liberal @xmath6-selected subset of the @xcite lbg sample . the lbg template curve falls within our selection region for @xmath146 . at higher redshifts , the lbgs rapidly become too red in @xmath145 to distinguish from lower - redshift objects and are better selected as dropouts in @xmath147 . we allow the color selection region to extend far to the blue in @xmath145 to account for the effect of lyman @xmath27 emission lines , which add flux in @xmath22 . for spectroscopic selection and clustering analysis , objects with @xmath18 fluxes , @xmath148 , less than their @xmath149 flux uncertainty , @xmath150 , were assigned an upper limit of @xmath151 to make it less likely that negative sky fluctuations were responsible for turning a dim object in @xmath18 into a formal dropout in @xmath90 . this correction avoids lower - redshift interlopers at the cost of some incompleteness . the vast majority of our lbg candidates are formal dropouts in the @xmath18 filter , with @xmath152 before being set to these limit values . the limit values were used to generate the @xmath90 colors plotted in fig . [ fig : uvr ] , which shows our full catalog including @xmath35 lyman break galaxy candidates selected by their @xmath6 colors . our selection criteria yield 1607 candidates in 1137 square arcminutes , or 1.4 arcmin@xmath7 . @xcite found 1.2 arcmin@xmath7 at @xmath153 using stricter criteria than ours , @xcite found 1.7 arcmin@xmath7 using the extended @xmath141 @xmath142 @xmath143 criteria shown translated into @xmath6 in figure [ fig : uvr_stars ] , @xcite found 1.5 arcmin@xmath7 using ubr , and @xcite found 1.5 arcmin@xmath7 using @xmath154 criteria . @xcite found a redshift distribution @xmath155 . our @xmath6 criteria are expected to start selecting lbgs at redshifts higher by 0.16 because the red cutoff of the @xmath18-band filter and its effective wavelength after accounting for the mosaic ii ccd qe are both @xmath156 redder than for @xmath141 . our criteria are expected to stop selecting lbgs at redshifts higher by up to 0.4 because the lyman alpha forest causes red @xmath145 colors at a higher redshift than in @xmath157 due to the longer effective wavelength of @xmath22 . hence we expect @xmath158 , and this will be calibrated via spectroscopy . figure [ fig : lbg_radec ] shows the spatial distribution of these lbg candidates . we estimate the angular correlation function of these objects using the @xcite estimator @xmath159 where @xmath160 represents the number of pairs at separation @xmath161 in the object catalog , @xmath162 represents the appropriately - normalized number of pairs at separation @xmath161 in a poisson - distributed mock catalog , and @xmath163 represents the appropriately - normalized number of pairs at separation @xmath161 with one in the object catalog and the other in the mock catalog . this estimator has variance @xmath164 \simeq { \frac{(1 + w(\theta))^2}{rr } } \ ; \ ; \ ; , \label{eq : ls_var}\ ] ] where the denominator gives the poisson error contribution i.e. the number of pairs expected to be found in the bin centered on @xmath161 given the survey geometry and the total number of lbgs in the survey . from @xcite ] the observed values of @xmath165 must then be corrected for the integral constraint caused by estimating the true sky density of lbgs from our survey to find a power - law fit @xcite , @xmath166 in practice , this correction is performed by using a modified estimator @xmath167 where the correction factor @xmath168 is given by @xmath169 a is estimated iteratively and usually converges in a few iterations . the angular correlation function found for lyman break galaxy candidates in figure [ fig : w_lbg ] is @xmath170 , where @xmath161 is measured in arcsec , intermediate between previous results for fixed @xmath171 from @xcite of @xmath172 and @xcite of @xmath173 . this implies a correlation length for our sample of @xmath174@xmath175mpc and an lbg bias factor of @xmath176 for the @xmath11cdm cosmology . these results will be updated after spectroscopic calibration of our @xmath6 lbg selection and by including our full square degree survey . ] musyc is unique for its combination of depth and total area , for the coverage in x - ray , uv , mid- and far - infrared wavelengths and for providing the @xmath0 plus near - infrared photometry needed to produce high - quality photometric redshifts over a square - degree of sky . this multiwavelength coverage will enable comparison of selection effects which have previously complicated the study of galaxies in the high - redshift universe . the large area ensures good statistics for studies of clustering , luminosity functions , and surface densities . we expect that the careful attention paid to optimized photometry methods and the estimation of accurate photometric uncertainties will allow our public data to make a significant impact for a wide range of scientific studies . in particular , this high - quality @xmath0 data , combined with forthcoming near - infrared photometry , should yield photometric redshifts with good accuracy . it is interesting to note that some of the standard techniques perform well despite making a number of idealized assumptions . for instance , we found that the optimal point source apertures are remarkably close in diameter to the idealized case of 1.35 fwhm due to a partial cancellation of the non - gaussian psf preferring a larger aperture and the correlated noise preferring a smaller aperture . the more traditional approach of using @xmath177 apertures produces biases in point - source colors . alleviating those biases by `` psf - matching '' i.e. smoothing the images to match the fwhm of the image with the worst seeing must be done carefully given the non - gaussian psf shapes and will result in degraded signal - to - noise for point sources . estimates of point source detection depth that assume uncorrelated noise and extrapolate from the rms pixel noise will overestimate the depth by about 0.5 magnitudes for 2@xmath49 or 3@xmath49 apertures . the apcorr fluxes introduced in this paper perform extremely well , as seen in fig . [ fig : uvr ] . the nearly unbiased performance of apcorr fluxes for both point sources and extended objects is impressive , and the correlated errors between filters for extended objects are acceptable if the flux is misestimated by a constant fraction in all filters . however , one should use caution when calculating photometric redshifts on extended objects in photometric catalogs generated using apertures which might miss a different fraction of the object flux in each filter . the best results should be obtained by using the apcorr fluxes in our catalog for unresolved and barely resolved sources , with further investigation needed to determine whether apcorr outperforms auto for extended sources . our @xmath0 images and catalogs of the @xmath88 extended hubble deep field south are available to the public at http://www.astro.yale.edu/musyc . we thank buell januzzi and frank valdes for their detailed answers and trouble - shooting with questions on the mscred package and benne w. holwerda whose guide , `` source extractor for dummies , '' provided a valuable supplement to the official sextractor user s manual written by e. bertin . we acknowledge helpful conversations with anton koekemoer , david spergel , and nick suntzeff . we thank the additional members of the musyc collaboration , including danilo marchesini , ken rines , ned taylor , and bill van altena for their valuable input . we are grateful for support from fundacin andes , the fondap centro de astrofsica , and the yale perspectives in science program . this material is based upon work supported by the national science foundation under grant . ast-0201667 , an nsf astronomy and astrophysics postdoctoral fellowship ( aapf ) awarded to e.g. s.t . acknowledges the support of the danish natural research council . we thank the staff of cerro tololo inter - american observatory for their invaluable assistance with our observations . this research has made use of nasa s astrophysics data system . we wish to combine a series of processed individual images which have signal @xmath178 and noise @xmath179 . the exact signal obviously depends on which star is measured , but if a common set of stars are considered the variations in signal will be due to differences in exposure time , cirrus extinction , and atmospheric extinction . the noise should be sky - dominated away from bright objects but measuring the rms in `` blank '' regions of sky properly accounts for the readnoise as well . for surface - brightness - optimization we want to average the total signal from a set of stars , as measured by the iraf routine _ phot _ or the inverse of the _ mscscale _ header output by _ mscimatch_. for point - source - optimization we want to determine the average signal that falls within our eventual optimal aperture , which will have a diameter of roughly 1.35 times the fwhm of the final image . the seeing in the final image can be estimated or measured from an initially unweighted stack , or an iterative procedure can be used . alternatively , a correction to the total star counts @xmath180 can be calculated by integrating an assumed gaussian psf out to the diameter of the expected optimal aperture of radius r ; if the seeing in each image is @xmath181 fwhm@xmath182 , we obtain @xmath183 \ ; \ ; \ ; . \label{eq : s_i}\ ] ] in either type of optimization , the noise value used should be the background rms for the expected aperture size , which depends on the level of noise correlation between pixels . assuming that the noise correlations have the same behavior in each individual image , it is sufficient to use the pixel - by - pixel rms i.e. @xmath74 measured in each image since this will produce the right relative weights . one can derive the weights that optimize the @xmath184 of the combined image by noting that the combined image will have @xmath185 and @xmath186 ( the weighted sum is just an unweighted sum of new images having signal @xmath187 and noise @xmath188 ) . maximizing the function @xmath184 produces the result @xmath189 , where multiplying all the weights by a constant @xmath190 preserves the final @xmath184 . in astronomy , it is typical to first scale the images to have equal signal levels ; this accounts for differences in exposure time and extinction to provide constant photometry and makes it easier to delete outlying pixels due to cosmic rays , satellite trails , etc . the iraf-_mscred _ routine _ mscstack _ performs this scaling by multiplying by the _ mscscale _ header value and then uses a weightfile to weight the * scaled * images . so the weights used must be for the scaled images , which leads to the formula for surface - brightness - optimized weighting of @xmath191 which is equivalent to @xmath192 in the original image because the image being weighted is different than the one for which @xmath178 and @xmath179 were measured . the output image is the same since instead of multiplying @xmath178 by @xmath193 and adding we are equating the signal levels and then multiplying 1 by @xmath192 , but performing the scaling first makes it easier to remove outlying pixels from this sum as mentioned above . the weights being used correspond to @xmath194 for the scaled images ; they reduce to the familiar case of inverse - variance weighting since the signal levels have already been equalized . for point - source optimized weighting we use @xmath195 where @xmath196 is given by @xmath197 where we have re - expressed the formula for @xmath178 from eq . [ eq : s_i ] in terms of the measured seeing in an unweighted ( or surface - brightness optimized ) stack , @xmath198 , and that in each individual image @xmath199 . again , these weights are only valid when used on the post - scaled images , and they differ by a factor of @xmath178 from the weights appropriate for the original images . due to the large number of individual exposures taken in e.g. @xmath18-band and multiple observing runs , it is sometimes necessary to perform stacking as an iterative procedure . to do this , the signal , noise , and seeing in the intermediate stacked images should be measured empirically as before . then _ mscstack _ or the equivalent can be used to stack these intermediate images just as if they were individual images , scaling by the new _ mscscale _ values and calculating a new weights file using the above equations . as long as the final seeing is estimated well , no loss of signal - to - noise should occur from the iteration . a corollary question is when to discard an image with poor @xmath184 rather than to include it in the stack . the formal answer appears to be never as @xmath200 for all @xmath92 . the magic of the optimal weighting formula can be seen : @xmath201 hence the optimal weights cause @xmath184 to add in quadrature so it will never formally decrease no matter how poor an input image is . note that having a magnitude of extinction or twice the seeing still leaves an image with significant @xmath184 ; a @xmath184 threshold of 40% would require discarding images with about 2 times worse seeing or 1 magnitude of extinction ( if the sky background is unaffected ) . the @xmath184 of the combination of the image with a given @xmath184 and an image with 40% of that @xmath184 is 8% higher than that of the better image alone so this is somewhat useful . a conservative approach would be to cut entirely any images with @xmath202% that of the median image as a way of reducing systematic effects not accounted for by these idealized formulae . @xcite showed using simulations that the best performance for point source detection was achieved by filtering the image with the psf itself . this result is derived in the sextractor manual using fourier transforms . in practice due to the constraints of computation time one chooses to cut off the psf at a finite radius , as is standard in the convolution kernels offered as part of the sextractor package . the general case for an optimized weighted sum of pixel values @xmath203 , where each pixel has signal @xmath204 and noise @xmath205 , can be derived in analogy to the result for adding images found above , yielding @xmath206 . the constant background noise for sky - dominated objects thus leads to a filter of the precise shape of the psf . for objects significantly brighter than the sky , @xmath207 giving constant weight i.e. a tophat which gradually morphs into the roughly gaussian shape of the psf as you move far enough away from the object center for the sky to dominate the noise . sextractor only allows for a fixed convolution kernel so we optimized the detection for sky - dominated objects by using the psf ( truncated to 7@xmath487 pixels ) as our filter . by the same argument , the optimum signal - to - noise measurement of the photometry of a point source will be obtained by a weighted sum of pixel fluxes weighted using the psf shape itself centered at the barycenter of the object . this is typically referred to as `` psf photometry '' but it is not a supported feature of sextractor photometry , just of object detection . note that this is no longer strictly optimal when the noise is correlated between pixels as we have found in our images , since the derivation of the optimal weights for each pixel assumed uncorrelated noise . we decided against using psf photometry for this reason , and also because many of our science objects are slightly resolved and this would overweight the fluxes in their core , increase the risk of color biases , and complicate the task of correcting the aperture fluxes of extended objects . llrllccc e - hdfs & 22:32:35.6 & -60:47:12 & ( 328,-49 ) & ( 311,-47)&0.03 & 1.37 & 1.6e+20 + e - cdfs & 03:32:29.0 & -27:48:47 & ( 224,-54)&(41,-45)&0.01 & 0.40 & 9.0e+19 + sdss1030 + 05 & 10:30:27.1&05:24:55 & ( 239,50)&(157,-4)&0.02&1.01&2.3e+20 + cw1255 + 01 & 12:55:40&01:07:00 & ( 306,64 ) & ( 192,7 ) & 0.02 & 0.81 & 1.6e+20 + lcrrc 2002 oct 6,8 & @xmath18 & 47 & 28200 & 1.40 + 2002 oct 6,8,10 & @xmath21 & 13 & 7500 & 1.35 + 2002 oct 6,10,12 & @xmath22 & 30 & 10440 & 0.90 + 2002 oct 6,10,12 & @xmath4 & 21 & 6300 & 0.85 + 2002 oct 6,12 & @xmath64 & 26 & 6300 & 0.85 + 2002 oct 6,12 & @xmath46 & 11 & 2700 & 0.90 + 2003 may 26,27 & @xmath18 & 25 & 15000 & 1.20 + 2003 may 27 & @xmath21 & 10 & 6000 & 1.10 + 2003 may 27,28 & @xmath4 & 20 & 6000 & 1.10 + 2003 may 26 & @xmath46 & 15 & 3600 & 1.30 + ccclcr u & 1.34 & 2.1 & 24.15@xmath1300.012 & 0.55 & ( -0.08@xmath1300.01)@xmath208 + b & 1.38 & 1.8 & 25.24@xmath1300.009 & 0.24 & ( -0.03@xmath1300.004)@xmath209 + v & 1.40 & 1.6 & 25.64@xmath1300.004 & 0.15 & ( 0.03@xmath1300.003)@xmath210 + r & 1.42 & 1.5 & 25.95@xmath1300.005 & 0.09 & ( -0.009@xmath1300.005)@xmath211 + i & 1.44 & 1.5 & 25.40@xmath1300.006 & 0.05 & ( 0.006@xmath1300.017)@xmath212 + @xmath46 & 1.46 & 1.5 & 24.61@xmath1300.014 & 0.03 & ( -0.06@xmath1300.054)@xmath212 + ccccccc bvr & 0.95 & 0.99 & 0.59 & 1.2 & 0.48 & 26.3 + u & 1.30 & 1.48 & 0.78 & 1.6 & 0.50 & 26.0 + b & 1.25 & 1.29 & 0.71 & 1.4 & 0.47 & 26.1 + v & 0.90 & 0.96 & 0.56 & 1.2 & 0.52 & 26.0 + r & 0.90 & 0.97 & 0.58 & 1.2 & 0.49 & 25.8 + i & 0.80 & 0.99 & 0.56 & 1.2 & 0.49 & 24.7 + @xmath46 & 0.90 & 1.06 & 0.62 & 1.2 & 0.43 & 23.6 + lllllllll 0&j223023.68 - 604817.4&22.506580&-60.804850&7397.04&3613.83&0.490000&0.377004&0.0771630 + 0.0771630&45.0000&16 & 0.00599998&0.0123559&0.0268263&0.0133826&0.0721889&0.0189753 + 0.0535715&0.0192685&0.0584936&0.0536170&0.105203&0.142492 & 0.0153194&0.0299205&0.0582284 + 0.0300672&0.129481&0.0322477&0.0994542&0.0346933&0.0963358&0.0964758&0.129852&0.292286 + 1&j223024.28 - 604008.9&22.506747&-60.669150&7395.69&5443.56&0.380000&0.381543&0.0771630 + 0.0771630&45.0000&16 & 0.0545032&0.0134836&0.0528900&0.0133148&0.0600472&0.0181627 + 0.0531777&0.0189421&0.0346594&0.0537312&-0.0605122&0.0999073 & 0.123143&0.0319542&0.111140 + 0.0292123&0.0983857&0.0301251&0.0985063&0.0331430&0.0733320&0.0938256&-0.113113&0.287106 + 2&j223023.85 - 604558.3&22.506627&-60.766210&7397.14&4134.88&0.490000&0.555360&0.216003 + 0.133500&-0.200000&24 & 0.0909909&0.0243018&0.0549541&0.0255182&0.0440691&0.0353969 + 0.130563&0.0386372&0.159939&0.108677&0.0605380&0.286208 & 0.100151&0.0285476&0.0699377 + 0.0270144&0.0861040&0.0280836&0.125653&0.0311628&0.113170&0.0863858&0.0282574&0.261025 + 3&j223024.98 - 603027.9&22.506940&-60.507770&7394.68&7619.57&0.940000&0.328410&0.199716 + 0.120951&18.8000&24 & -0.0217973&0.0159112&0.000543353&0.0123818&0.115021&0.0169463 + 300.840&0.0511371&577.641&0.128364&-0.00643368&0.124763 & -0.0458961&0.0422158&-0.000965204 + 0.0289711&0.222445&0.0300914&601.662&0.125434&1191.94&0.301193&0.00430643&0.280111 + 4&j223023.35 - 605242.2&22.506487&-60.878400&7397.96&2622.20&0.610000&0.613032&0.369528 + 0.202653&-5.40000&24 & -0.00917560&0.0288420&0.0469781&0.0383380&0.134234&0.0542395 + 0.155328&0.0622344&-0.104166&0.130648&0.110060&0.458585 & -0.0224367&0.0274903&0.100572 + 0.0275703&0.155969&0.0290837&0.143652&0.0332477&0.0475046&0.0869854&0.677608&0.276257 +

future @xmath8 colliders will likely have the option of operating in @xmath9 or @xmath10 collision modes @xcite . these modes are reached by compton scattering laser light off one or more of the incoming fermion beams , and then colliding the resulting high energy photons with the remaining fermion beam or with each other . there is a large potential for @xmath11 and @xmath12 collisions to elucidate possible physics beyond the standard model ; previous investigations have focused on anomalous couplings @xcite , searches for extra dimensions @xcite , properties of supersymmetry @xcite , and a broad host of other topics . one subject that deserves further study is @xmath3-violating gauge boson self - couplings . @xmath3 violation is one of the most poorly understood aspects of the standard model ( sm ) . present data merely fixes the value of the @xmath3 violating phase in the ckm matrix , and can not test if this phase constitutes the only source for @xmath3 violation . in fact , studies of baryogenesis within the sm suggest that additional @xmath3 violating terms are required in order to generate the observed baryon asymmetry in the universe @xcite . most discussions of @xmath3 violation at photon colliders work within the context of models such as supersymmetry , and focus upon either higgs or top quark production @xcite . however , since almost every extension of the sm contains new @xmath3 violating phases , it is desirable to eliminate any possible theoretical bias by studying @xmath3 violation within the generic context of an effective lagrangian approach , without assuming any underlying mechanisms . a handful of such works have been performed in the past @xcite , concentrating on @xmath13 , @xmath14 , or top quark production . here , we extend these studies by examining possible @xmath3 violating quartic gauge boson couplings . @xmath12 colliders are particularly suited to studying such couplings ; they can be probed in @xmath5 scattering processes , unlike at @xmath8 colliders . the sm contributions to the processes @xmath0 , @xmath15 , and @xmath16 , including electroweak contributions , were first computed in @xcite . recently , they have been reexamined @xcite , and shown to exhibit several interesting features that motivate our analysis . the sm amplitudes vanish at tree level , and may therefore be quite sensitive to the effects of new physics . at one loop they acquire large imaginary parts , completely dominated at high energies by the helicity amplitudes @xmath17 , @xmath18 , and @xmath19 , where @xmath20 denotes the direction of transverse polarization relative to the beam direction . such amplitudes can interfere strongly with @xmath3 violating contributions from new sources , yielding greatly enhanced sensitivity to these new effects . in addition , the ability to polarize both the initial laser light and fermion beams allows the construction of observables which are sensitive only to these interference effects . these properties , together with the experimental cleanliness of @xmath0 , @xmath1 , and @xmath2 scattering , suggest that these processes might provide powerful tools in searches for new physics . in this paper we will show that @xmath0 , @xmath1 , and @xmath2 at a photon collider can provide sensitive tests of @xmath3 violation in the gauge boson sector . we limit our study to genuinely quartic gauge boson operators , as contributions to the two point functions are strongly excluded by current data and the three point functions are likely better tested elsewhere . we consider manifestly su(2 ) @xmath4 u(1 ) invariant operators constructed from the appropriate field strength tensors , making no assumptions as to the origin of electroweak symmetry breaking . as a result , our operators are of dimension eight . throughout the paper we will identify features of our analysis that might be useful for other new physics searches using these processes . the paper is organized as follows . we first present the density matrix formalism for photon - photon collisions in some detail , both for completeness and to motivate the expression for the ensemble average of the scattering amplitude , which will be used throughout our analysis . we then briefly discuss the construction of the anomalous @xmath3 violating operators we have studied . our results for the three processes considered are presented next , including discussions of the various initial laser and fermion beam polarizations . finally , we present our conclusions , including a comparison of our resulting sensitivity to these effects with others in the literature . in this section we present the density matrix formalism for photon - photon collisions . our discussion is similar to that found in the literature ( see , for example , @xcite ) , but is included here to provide motivation for eq . ( 9 ) , which is used throughout our analysis . consider a photon moving along the @xmath22 axis ; its polarization vectors for positive and negative helicities are given by @xmath23 we will denote the photons corresponding to states with these polarizations by @xmath24 . the most general photon state can be written in this basis as @xmath25 where @xmath26 and @xmath27 . writing this state in 4-vector form demonstrates that the choices @xmath28 and @xmath29 lead to circularly polarized states , while @xmath30 leads to a linearly polarized state . the angle @xmath31 describes the direction of linear polarization in the @xmath32 plane . the experimental setup in @xmath9 colliders has been described in detail elsewhere @xcite ; briefly , one compton scatters laser photons off fermion beams , and then collides the backscattered photons . when observing the scattering process @xmath33 , the energies or helicities of the incoming photons are not known ; they will be statistically distributed , with the specific distribution being determined by the details of the compton scattering process . when measuring the differential cross section , one no longer determines @xmath34 , but rather the ensemble average of @xmath35 : @xmath36 where the sum over @xmath37 sums over all possible initial states weighted by @xmath38 , the probability of their occurrence . we can write this in the form @xmath39 where we have introduced the density matrix for the incoming photons in the basis defined by @xmath40 : @xmath41 let us compute the density matrix for the laser photon of eq . ( 2 ) ; introducing the notation @xmath42 and @xmath43 , we have @xmath44 @xmath45 and @xmath46 measure the amounts of circular and linear polarization , respectively . this is consistent with the polarizations for the values @xmath47 and @xmath30 noted before . after the compton scattering process , the degrees of linear and circular polarization are described by distributions which are dependent upon the fraction of the fermion beam energy the laser photons acquire . denoting this fraction by @xmath48 , and the helicity distribution functions by @xmath49 , the density matrix becomes @xmath50 the eigenvalues of this matrix are @xmath51 we no longer have a pure state unless @xmath52 . as the processes considered in this paper involve two compton scattered photons , the complete density matrix will be the tensor product of the density matrix for each photon . also , since one of the initial laser photons will be moving along the @xmath53 axis , we must take @xmath54 in its density matrix . referring back to eqs . ( 4 ) and ( 7 ) , we can write the ensemble average for the process @xmath33 as @xmath55 \nonumber \\ & & + \xi_c ( x_2 ) \left [ \mid { \cal m}_{++ } \mid^2 - \mid { \cal m}_{-- } \mid^2 - \mid { \cal m}_{+- } \mid^2 + \mid { \cal m}_{-+ } \mid^2 \right ] \nonumber \\ & & + \xi_c ( x_1 ) \xi_c ( x_2 ) \left [ \mid { \cal m}_{++ } \mid^2 + \mid { \cal m}_{-- } \mid^2 - \mid { \cal m}_{+- } \mid^2 - \mid { \cal m}_{-+ } \mid^2 \right ] \nonumber \\ & & + 2 \xi_{t } ( x_1 ) \xi_{t } ( x_2 ) \ , { \rm re } \left [ { \cal m}_{+- } { \cal m}^{*}_{-+ } e^{2i ( \phi_1 + \phi_2 ) } + { \cal m}_{++ } { \cal m}^{*}_{-- } e^{2i ( \phi_1 - \phi_2 ) } \right ] \nonumber \\ & & -2 \xi_{t } ( x_1 ) \ , { \rm re } \left [ { \cal m}_{++ } { \cal m}^{*}_{-+ } e^{2i \phi_1 } + { \cal m}_{+- } { \cal m}^{*}_{-- } e^{2i \phi_1 } \right ] \nonumber \\ & & -2 \xi_{t } ( x_2 ) \ , { \rm re } \left [ { \cal m}_{+- } { \cal m}^{*}_{++ } e^{2i \phi_2 } + { \cal m}_{-- } { \cal m}^{*}_{-+ } e^{2i \phi_2 } \right ] \nonumber \\ & & + 2 \xi_{t } ( x_1 ) \xi_c ( x_2 ) \ , { \rm re } \left [ { \cal m}_{+- } { \cal m}^{*}_{-- } e^{2i \phi_1 } - { \cal m}_{++ } { \cal m}^{*}_{-+ } e^{2i \phi_1 } \right ] \nonumber \\ & & + 2 \xi_{c } ( x_1 ) \xi_t ( x_2 ) \ , { \rm re } \left [ { \cal m}_{-- } { \cal m}^{*}_{-+ } e^{2i \phi_2 } - { \cal m}_{+- } { \cal m}^{*}_{++ } e^{2i \phi_2 } \right ] \bigg\ } \,\ , , \end{aligned}\ ] ] where , for example , @xmath56 denotes the helicity amplitude for incoming photons with helicities @xmath57 and @xmath58 . we have suppressed the final state @xmath59 ; depending upon the process considered the observable final state might require a sum over various helicities . this expression will be used to construct observables for all the processes considered in this paper . the physical cross section involves a convolution of this scattering amplitude with the photon number density function of each photon . the explicit form of this function , as well as the forms for the linear and circular helicity distribution functions , can be found in the appendix . here we construct the most general set of operators that contribute to neutral gauge boson self - interactions , subject to the following constraints . we consider only su(2 ) @xmath4 u(1 ) invariant operators , and further restrict these to @xmath3-odd terms . as the effects of @xmath3-odd trilinear operators have been extensivley studied @xcite , we consider only those terms that lead to quartic or higher gauge boson interactions . we also make no assumption as to the mechanism of electroweak symmetry breaking . these restrictions lead us to the following set of seven dimension eight operators constructed from the @xmath7 and @xmath60 field strength tensors : @xmath61 here @xmath62 is the su(2 ) field strength tensor and @xmath63 the u(1 ) field strength tensor . we have introduced the notation @xmath64 and our convention for the @xmath65 tensor is @xmath66 . a brief comment on the completeness of these operators is in order . @xmath3 violating operators containing three @xmath67 tensors are reducible to those containing only one such tensor , as the product of two @xmath67 tensors can be written as a determinant of metric tensors . operators such as @xmath68 , where the @xmath67 tensor contracts an index on each field strength tensor , are reducible to those where the @xmath67 tensor fully contracts one of the field strength tensors through use of the schouten identity ( see the first reference in @xcite , @xcite ) , @xmath69 which states that no tensor antisymmetric in five indices exists in four dimensions . similarly , the operators @xmath70 can be reduced to those listed in eq . ( 10 ) through use of the identity @xmath71 this is true for @xmath72 antisymmetric in four - dimensional spaces ( an analog holds in arbitrary dimensions @xcite ) . finally , since the processes considered here only involve @xmath73 , the operators @xmath74 become equivalent to the second operator in eq . ( 10 ) through use of the above identity , while the operators @xmath75 only contribute to processes involving @xmath76 . we thus see that the seven operators delineated in eq . ( 10 ) form a complete and independent set . each of these operators will give rise to a number of different @xmath77 , @xmath78 , and @xmath79 operators , as well as other quartic operators involving three or more @xmath80 bosons and various quintic terms . in this paper we will concentrate on those quartic operators relevant for the scattering processes @xmath0 , @xmath1 , and @xmath2 ; the other terms can not be probed in @xmath5 scattering , and the resulting constraints obtainable on the operators of eq . ( 10 ) would be weakened . denoting the operators to be studied by @xmath81 , @xmath82 , and @xmath83 , each su(2 ) @xmath4 u(1 ) operator will have an expansion of the form @xmath84 where the @xmath85 , @xmath86 , and @xmath87 are functions of the weak mixing angle and the ellipsis denotes the neglected quartic and quintic terms . to determine the independent @xmath12 , @xmath1 , and @xmath2 structures , we set @xmath88 where @xmath89 doing so , and making use of the identity in eq . ( 14 ) , leads to the following forms for the @xmath90 : @xmath91 we will present the coefficients @xmath85 , @xmath86 , and @xmath87 when we discuss the relevant process . we will organize our study as follows . a section will be devoted to each of the processes @xmath0 , @xmath15 , and @xmath16 . in each section we will discuss the properties of the relevant @xmath90 , focusing on the observable asymmetries associated with each operator and the role of varying the initial polarization states . we will then discuss the measurement of the various @xmath92 , and estimate the sensitivity to these operators that can be obtained at a photon collider . we will bound each of these operators separately ; each is multiplied by an arbitrary coefficient that could cause contributions from different operators to cancel , but we will ignore this possibility . to preview what we will observe , it will turn out that in certain @xmath92 the structures of eq . ( 20 ) will interfere destructively , while in others they will combine constructively , leading to widely varying sensitivity to the @xmath92 . in this section we consider the process @xmath93 . we also present and motivate the various parameterizations and approximations used throughout the paper . a detailed presentation of the standard model amplitudes can be found in @xcite ; we focus here only upon those features relevant to our analysis . let us first discuss the properties of the @xmath3 violating amplitudes we are considering . the detailed expressions for these amplitudes , and those for the other processes considered in this paper , can be found in the appendix . bose symmetry implies the relations @xmath94 where @xmath95 , @xmath96 , and @xmath97 are the usual mandelstam invariants : @xmath98 , @xmath99 , and @xmath100 . we have denoted the @xmath3 violating amplitudes with the superscript @xmath3 . the operators considered in this paper are odd under both @xmath101 and @xmath102 , which implies @xmath103 in addition , the standard crossing symmetries hold . these relations immediately imply @xmath104 the only non - vanishing amplitudes are @xmath105 . the standard model amplitudes are dominated at high energies by the amplitudes @xmath106 , @xmath107 , and @xmath108 . these amplitudes are primarily imaginary , although the real parts are non - negligible ; features of all @xmath109 amplitudes are discussed in @xcite . all of the @xmath3 odd asymmetries that can be constructed involve the interference of @xmath105 with a standard model amplitude ; to increase our sensitivity to the anomalous interaction we should attempt to define an observable with interference between @xmath105 and one of the dominant standard model terms . unfortunately , no observable exists which contains such an interference with one of the large imaginary sm amplitudes . we can , however , construct the following observable which contains an interference with one of the real pieces of the sm amplitudes : @xmath110}{\int_{0}^{2 \pi } \int_{0}^{2 \pi } \ , d \phi_1 \ , d \phi_2 \ , \left [ \left ( \frac{d \sigma}{d \omega}\right ) \delta(\phi_1 - \phi_2 -\pi/4 ) + \left ( \frac{d \sigma}{d \omega}\right)\delta(\phi_1 - \phi_2 + \pi/4 ) \right ] } \ , , \ ] ] where @xmath111 , @xmath112 denote the angles of linear polarization introduced in our discussion of the density matrix formalism . the experimental determination of this asymmetry requires several measurements , as indicated in eq . ( 24 ) by the integration over linear polarization directions . the delta functions in @xmath113 denote the need to maintain a fixed angle between the linear polarization directions of the initial beams . this observable has been considered previously in @xcite . referring to eq . ( 9 ) , and implementing the relations in eqs . ( 21 - 22 ) , we can show that the numerator of this asymmetry contains the term @xmath114 as shown in the appendix , the @xmath3 violating amplitudes are purely imaginary , leading to a non - vanishing result . as this observable is sensitive only to the degree of linear polarization , we will use the following set of initial parameters : @xmath115 we note that with no circular polarization the fermion beam polarizations do not enter any expression . we can simultaneously measure another independent quantity , the denominator of @xmath113 . this is the unpolarized cross section @xmath116 the form of which can also be obtained from eq . . we will now consider the operator relevant to @xmath9 final states in eq . we write it in the form @xmath117 where @xmath118 is the energy scale of the physics leading to these operators , and @xmath119 the associated coupling constant . we will hereafter set @xmath120 , effectively absorbing this coupling constant into the definition of @xmath118 . although we will quote our sensitivity in terms of the energy scale that can be probed , the reader should realize that this is not exactly the scale associated with the new physics leading to these anomalous couplings . we have also ignored a possible @xmath121 appearing in front ; the unpolarized cross section is unaffected by this sign , and @xmath113 only acquires an overall sign change . in our analysis , we have used the approximate sm amplitudes found in @xcite , valid for @xmath122 , which holds for the collider energies considered here . we will use similar approximate expressions when considering @xmath15 and @xmath16 . the features of the sm amplitudes noted earlier occur when the energy of the process becomes greater than several hundred gev . in accordance with this approximation we employ the cuts @xmath123 throughout this paper , we take the @xmath124 center of mass energy to be @xmath125 gev when obtaining our search reaches . in this case , at the endpoints of this region , where @xmath126 and @xmath127 , we have @xmath128 ; this value becomes much smaller as we approach the center of the parameter region , thus validating our approximation . although a more detailed analysis taking into account detector properties and experimental cuts would need to use the complete sm expressions , our study captures the salient points . we show the total cross section and asymmetry versus @xmath129 for this operator in fig . ( 1 ) for a `` typical '' value of @xmath130 for purposes of demonstration . in presenting these results we have assumed an @xmath124 integrated luminosity @xmath131 , which is the quoted yearly integrated luminosity for planned linear colliders @xcite . however , a determination of @xmath113 requires several separate measurements , and it is certainly not expected that colliders will operate primarily with purely linearly polarized beams , or even primarily in a photon collision mode . these numbers should be interpreted as results coming from several years of operation and are intended for purposes of illustration . to provide a more conservative outlook , our quoted sensitivity will be given as a function of integrated luminosity , beginning at the modest value @xmath132 . we have also assumed the value @xmath133 in this and all other numerical results presented . at the luminosity @xmath131 , the asymmetry is much smaller than the associated errors , and only a change in the total counting rate is statistically observable . such an effect can arise from a variety of sources , and the identification of @xmath3 violation requires higher luminosities to observe a non - vanishing asymmetry . the transcription of these results into statements about the su(2 ) @xmath4 u(1 ) operators is simplified by there being only one @xmath12 operator . the coefficients @xmath134 are @xmath135 where @xmath136 . the su(2 ) @xmath4 u(1 ) operators of eq . ( 10 ) take the form @xmath137 to estimate the value of @xmath138 that can be probed at a @xmath12 collider we have performed a combined least - squares fit to the total cross section and asymmetry . we have assumed standard statistical errors and an additional 1 % luminosity error in the integrated cross section . the fit was performed with @xmath139 gev ; approximate results for other energies can be obtained by scaling these numbers . as can be seen from eq . ( 30 ) there are only four different results for this process ; these are presented in fig . although this asymmetry is not particularly sensitive to @xmath3 violation as parameterized by the operators we consider , we note that it is a good test of any @xmath3 violating amplitude for which any of the real amplitudes @xmath140 , @xmath141 , or @xmath142 are non - vanishing . in these cases terms of the form @xmath143 appear in the fifth line of eq . ( 9 ) , leading to a large @xmath113 . in this section we study the effects of anomalous @xmath3 violating operators in @xmath145 . a detailed study of the sm amplitudes for this process can be found in @xcite . as in @xmath0 , the @xmath146 amplitudes satisfy certain relations dictated by bose symmetry ; the anomalous couplings considered satisfy additional relations because of their transformations under @xmath101 . let @xmath147 denote either longitudinal or transverse z polarizations . bose symmetry implies @xmath148 and oddness under @xmath101 requires @xmath149 since the initial and final states are different in this process , the transformation properties under @xmath102 do not imply any relations between the amplitudes . the explicit computation of the amplitudes is presented in the appendix . unlike in @xmath0 , there are several surviving amplitudes . the structure of the sm amplitudes for @xmath144 is similar to that for @xmath0 , and even more pronounced . at high energies the process is dominated by the imaginary parts of the amplitudes @xmath106 , @xmath107 , and @xmath108 ; all other amplitudes are completely negligible @xcite . as before , we must construct an observable that contains an interference of a @xmath3 odd amplitude with one of the dominant sm terms . the asymmetry considered in the previous section contains only interference with the real parts of the sm amplitudes , and is therefore unacceptable . for interference effects between imaginary @xmath3 odd and @xmath3 even amplitudes , we must consider the following asymmetry which is measurable with circularly polarized beams : @xmath150 this asymmetry has been considered previously in @xcite . the subscripts @xmath151 denote the initial polarization states of the laser and fermion beams , which we will now discuss . setting @xmath152 , we are left with four parameters describing the inital state polarization : @xmath153 , @xmath154 , @xmath155 , and @xmath156 . we will set @xmath157 and @xmath158 , consistent with the expected capabilities of future facilities @xcite , and label our inital states as @xmath159 . in the sm , there are six independent states : @xmath160 and @xmath161 , where , for example , @xmath162 means @xmath163 , @xmath164 , @xmath165 , and @xmath166 . states obtained by an overall sign flip are identical ; for example , @xmath162 and @xmath167 lead to the same observables . this is not true when @xmath3 violating interactions are present . the asymmetry of eq . ( 35 ) , where the subscript @xmath168 refers to a given inital state and @xmath169 to the state obtained by flipping the signs of the polarizations , will vanish in the sm and be non - zero in the presence of the anomalous couplings . in addition to this asymmetry we can simultaneously measure another independent quantity , the denominator of the asymmetry . referring to the ensemble average in eq . ( 9 ) and using the relations in eqs . ( 33 - 34 ) , we see that this denominator is just twice the polarized cross section , @xmath170 \ , .\end{aligned}\ ] ] the operators relevant to @xmath146 final states can be found in eq . ( 20 ) ; we will write them as follows : @xmath171 our philosophy in parameterizing these operators is identical to that discussed in the @xmath0 section ; there is again a coupling constant @xmath119 associated with each operator that we have set equal to 1 , and a possible factor of @xmath172 appearing in each operator that we have ignored . the detailed amplitudes are presented in the appendix ; here we will discuss several properties relevant to our analysis . the first is that the amplitudes for longitudinally polarized @xmath80 states are suppressed relative to those with transverse @xmath80 states by a factor of @xmath173 , as dictated by the goldstone boson equivalence theorem . this , and the fact that the tagging efficiency for longitudinal states is at best 10% , means that we will not explore the effect of selecting final state polarizations in our analysis . before presenting our results , let us examine @xmath174 in more detail . our discussion will illustrate the complementary information that can be obtained from examining all of the possible initial polarizations . concentrating on the numerator of @xmath174 , we see that @xmath175 \nonumber \\ & & + \left ( \xi_{c}(x_{1 } ) - \xi_{c}(x_{2 } ) \right ) \left [ \mid { \cal m}_{+- } \mid^2 - \mid { \cal m}_{-+ } \mid^2 \right ] \,\ , , \end{aligned}\ ] ] where a sum over final state helicities is implied . the first term is symmetric under the interchange of the two initial photons , which implies symmetry under the transformation @xmath176 and under the interchange of the initial polarization states , @xmath177 . the second term is antisymmetric under both of these exchanges . by selecting an appropriate initial polarization state , we can isolate each term . for example , the choice @xmath178 is symmetric under the interchange of the two initial polarizations ; with this choice only the first term of @xmath174 will contribute , whereas the choice @xmath179 is antisymmetric under interchange of the two initial polarizations , and with this selection only the second term contributes . a choice such as @xmath161 is of mixed symmetry , and is sensitive to both terms in @xmath174 . such a variation of the inital states allows us to isolate the various anomalous amplitudes which contribute to @xmath174 is sensitive ; with symmetric initial polarizations it is sensitive to @xmath180 , while with antisymmetric polarizations it is sensitive to @xmath141 and @xmath142 . the four largest asymmetries for @xmath181 are presented in fig . ( 3 ) as a function of @xmath182 , employing the same cuts as used in @xmath0 . at @xmath183 and @xmath184 tev , the `` symmetric '' asymmetries are statistically significant throughout the entire angular region , while the `` antisymmetric '' asymmetries are significant in the outer regions . in fig . ( 4 ) we present the total cross section versus @xmath129 for each inital polarization state , and the integrated asymmetry versus @xmath129 for the four symmetric initial states . although a deviation in the cross section is not seen in any polarization state until high energies , the integrated asymmetry becomes quite large at low energies , suggesting that it provides a sensitive test of the anomalous couplings under consideration . so far we have only considered the operator @xmath185 ; the distributions for the other two operators differ only in relative sign and overall magnitude at high energies . the asymmetries for polarization states @xmath162 and @xmath179 for all three operators are shown in fig . ( 5 ) ; those for @xmath185 and @xmath186 are of similar magnitude but opposite sign . this can be understood in a simple way . at high energies , a transversely polarized @xmath80 effectively looks like a photon . when computing the amplitudes for the various operators in eq . ( 37 ) , we must include all possible permutations of the three photons , as they are identical particles ( for example , denoting the field strength tensors of the three photons by 1,2 , and 3 , the six permutations required to compute @xmath185 become @xmath187 , @xmath188 , @xmath189 , @xmath190 , @xmath191 , and @xmath192 ) . the permutations required for @xmath185 and @xmath186 are identical to the permutations necessary to compute the amplitudes for the four photon operator of eq . ( 28 ) ; we can verify using the explicit forms of the amplitudes in the appendix that the sum of @xmath193 and @xmath194 gives @xmath195 , up to an overall constant . since @xmath102 violation leads to vanishing of the amplitudes @xmath180 , @xmath141 and @xmath142 in @xmath0 , the asymmetries for @xmath185 and @xmath186 must therefore be of opposite sign . su(2 ) @xmath196 u(1 ) operators whose coefficients @xmath197 , @xmath198 are of the same sign will see destructive interference between @xmath185 and @xmath186 , while those whose coefficients have opposite signs will see constructive interference . we will see similar behavior when investigating @xmath16 . table ( 1 ) presents the coefficients @xmath199 relating the @xmath200 to the @xmath82 . .coefficients @xmath199 relating the @xmath92 to the @xmath82 . [ cols="<,<,<,<",options="header " , ] unlike the situation in @xmath15 , we see that all of the interference effects are destructive , given the relative signs of the @xmath201 for each operator . however , since @xmath202 and @xmath203 , a complete cancellation does not occur except for the three operators @xmath204 , @xmath205 , and @xmath206 . we have performed a combined least - squares fit to the normalized binned cross section , binned asymmetry , and total cross section , with @xmath207 gev , for the polarization state @xmath162 to estimate the value of @xmath138 that can be probed in this process at a photon collider . the search reaches obtainable from the other symmetric polarization states are similar . the results are presented in fig . we only display results for the operators @xmath204 , @xmath206 , @xmath208 , and @xmath209 ; the sensitivity to @xmath204 and @xmath205 is identical at high energies . similarly , @xmath208 and @xmath210 differ only in the sign of their asymmetries , as do @xmath209 and @xmath211 , and hence yield identical search reaches . we see that overall , this process is not quite as sensitive to the anomalous operators as @xmath15 due to the destructive interference arising from the su(2 ) @xmath4 u(1 ) embedding of the @xmath79 vertex structures . however , those operators for which the asymmetry does not vanish can still be constrained quite stringently . in conclusion , the sensitivity of @xmath212 to interference with the large sm amplitude @xmath17 renders this process a sensitive probe of @xmath3 violation appearing in gauge boson self couplings . in this paper we have examined the possibility of using the processes @xmath0 , @xmath1 , and @xmath2 to search for sources of @xmath3 violation that manifest themselves at energy scales above those of future colliders . we have parameterized these effects by an effective lagrangian containing the complete and independent set of @xmath3-odd , su(2 ) @xmath4 u(1 ) invariant quartic gauge boson operators relevant to these interactions . the considered processes vanish at tree level in the sm , and at one loop are completely dominated by the imaginary parts of the amplitudes @xmath17 , @xmath18 , and @xmath19 . the fact that they might be powerful tools in indirect searches for new physics was pointed out in @xcite , where their sensitivity to supersymmetric loop effects was discussed . our results show that they are also of utility in searches for sources of @xmath3 violation that contribute to self interactions of gauge bosons . to demonstrate that these processes are indeed sensitive probes of the gauge boson sector , we compare our results with four others found in the literature . we first note that while our operators are of dimension eight , those with which we compare are of dimension six . although our comparisons are for the search reaches for the new physics scale @xmath213 , the reader should be aware that a given limit on @xmath213 translates into a greater sensitivity to a dimension eight operator , which scales as @xmath214 , than a dimension six operator , which scales as @xmath215 , for @xmath216 . we also note that the constraints on the anomalous su(2 ) @xmath4 u(1 ) operators obtained in this paper are stronger than those implied by unitarity . we first contrast our search reaches with the direct bounds on quartic gauge boson couplings given recently by the lep experiments @xcite . they consider operators of the form @xmath217 , and obtain bounds of @xmath218 on @xmath3-conserving @xmath219 , @xmath220 , and @xmath221 vertices . the operators considered in this paper are of the form @xmath222 ( with @xmath120 ) , and therefore our search reach for @xmath223 are to be contrasted with their bounds on @xmath224 . as expected , our sensitivity at a higher energy and higher luminosity collider are very significantly better . we next compare our results with those of a study of these same operators at a @xmath225 gev @xmath124 collider with an integrated luminosity of @xmath226 @xcite . in this study the operators are parameterized in the form @xmath227 , and bounds of @xmath228 obtained on vertices such as the @xmath229 vertex . setting @xmath230 , this results in @xmath231 gev , at the @xmath232 level . the sensitivities obtained in this paper , normalized to the appropriate luminosity and energy are @xmath233 gev , where the better constraints are for those operators contributing to one of the considered asymmetries . the search reaches obtained here on @xmath213 , after adjusting the statistical significances , are still stronger , even though the operators we consider are of dimension eight . the sensitivity of future hadron colliders such as the lhc to these anomalous @xmath3-even @xmath79 and @xmath234 has been studied in @xcite . this paper obtains search reaches of @xmath235 at the 95 % cl , which are again somewhat weaker than those we have presented . finally , the authors of @xcite consider @xmath3-odd dimension six operators contributing to the process @xmath236 at @xmath237 tev assuming @xmath238 of integrated luminosity . they cast their operators in the form @xmath239 , and obtain bounds of @xmath240 . note the we have improved their results by a factor of 2.5 to simulate the search reach assuming @xmath241 of integrated luminosity . these translate into limits of @xmath242 gev in our notation , at the @xmath243 level . the relevant limits in this paper to compare to are those assuming @xmath241 of integrated luminosity , which range from @xmath244 gev to @xmath245 gev at the 95 % cl . our constraints on the new physics scale @xmath213 are not quite as strong , which is not too surprising considering the large cross section for @xmath236 and the fact that their operators are of dimension six while ours are of dimension eight , although the differences are lessened after accounting for the greater statistical significance of our results . however , the processes considered are complementary ; they study the charged gauge boson production process @xmath236 while we examine the neutral gauge boson processes @xmath0 , @xmath1 , and @xmath2 . in summary , the processes @xmath0 , @xmath1 , and @xmath2 are sensitive probes of @xmath3 violation in the gauge boson sector . the sensitivities obtained here to the anomalous @xmath3-odd operators affecting these interactions compare favorably with similar limits on anomalous gauge boson couplings found in the literature . in addition , the examination of these processes nicely complements previous studies that have focused primarily on @xmath7 boson , top quark , or higgs production . their utility in other indirect physics searches at future photon colliders should certainly be studied in more detail . it is a pleasure to thank hooman davoudiasl for several useful conversations , and tom rizzo for suggesting this project and for many helpful discussions . the work of f. p. was supported in part by an nsf graduate research fellowship . we collect here formulae describing the various photon distribution functions ; derivations are given in @xcite . we first define the auxiliary functions @xmath246 where @xmath247 $ ] , and @xmath248\ ] ] @xmath249.\ ] ] here @xmath22 is a variable describing the laser photon energy , and is given by @xmath250 , where @xmath251 is the electron beam energy and @xmath252 the laser energy . varying @xmath22 changes the maximum value of the backscattered photon beam energy @xmath253 , where @xmath254 . we will set @xmath255 , which maximizes @xmath253 while preventing interactions between the backscattered photons and laser beam . in terms of these functions and variables the photon number and helicity distributions take the form @xmath256 - p_l \ , ( 2r - 1)\left(1 - x + \frac{1}{1 - x}\right)\right\}\ , . \xi_{t}(x , p_e , p_l , p_t ; z ) & = & \frac{2 r^2 p_t } { c(x ) } \,\,.\end{aligned}\ ] ] @xmath257 is the photon number density function , while @xmath258 and @xmath259 are respectively the circular and linear helicity distribution functions . we can now write the observable differential cross section as @xmath260 where @xmath261 and @xmath262 accounts for any final state bose symmetries . here we present the kinematics and amplitudes used for the processes in this paper . the generic interaction considered is @xmath263 , where @xmath264 , @xmath265 are the incoming photons and @xmath266 , @xmath267 the outgoing gauge bosons . the momenta and polarization vectors take the following form in the c.m.s . frame : @xmath268 with @xmath269 , @xmath270 , @xmath271 , and @xmath272 . given these expressions , it is straightforward to compute the amplitudes for the operators of eq . ( 20 ) . as mentioned in the text we neglect terms of @xmath273 , as the sm amplitudes we use also omit terms of this order . we find that this approximation is numerically valid for the energies considered here . the amplitude for the anomalous @xmath12 operator of eq . ( 20 ) takes the form @xmath274 \left(3+c^{2}_{\theta } \right ) \ , .\ ] ] this amplitude is exact , as @xmath275 and @xmath276 appear only in the sm amplitudes . note that this amplitude is odd under the interchange @xmath277 ; this is a consequence of @xmath102 violation and the fact that the two final state particles are the same as the initial states . the anomalous amplitudes for the process @xmath15 ( with @xmath278 ) are , neglecting terms of @xmath279 , @xmath280 \nonumber \\ { \cal m}_{1}^{\gamma z}(\lambda_1,\lambda_2;\lambda_3,0 ) & = & \frac{ie^2 m_z \hat{s}}{2(\lambda_{1}^{\gamma z})^4 } \ , s_{\theta } \ , \sqrt{\frac{\hat{s}}{2 } } \ , \bigg[\lambda_3 ( \lambda_1 - \lambda_2 ) + \bigg(2- \lambda_3 ( \lambda_1 + \lambda_2)\bigg ) c_{\theta } \bigg ] \nonumber \\ { \cal m}_{2}^{\gamma z}(\lambda_1,\lambda_2;\lambda_3,\lambda_4 ) & = & \frac{ie^2 \hat{s}^2}{8(\lambda_{2}^{\gamma z})^4 } \ , \bigg [ -5(\lambda_1 + \lambda_2)(1+\lambda_3 \lambda_4 ) + 2(\lambda_3 + \lambda_1 \lambda_2 \lambda_4 ) \nonumber \\ & & -2(\lambda_2 - \lambda_1)(1-\lambda_3 \lambda_4 ) c_{\theta } + \bigg(2 ( \lambda_3 + \lambda_1 \lambda_2 \lambda_4 ) - ( \lambda_1 + \lambda_2 ) ( 1 + \lambda_3 \lambda_4 ) \bigg ) c^{2}_{\theta } \bigg ] \nonumber \\ { \cal m}_{2}^{\gamma z}(\lambda_1,\lambda_2;\lambda_3,0 ) & = & \frac{ie^2 m_z \hat{s}}{2(\lambda_{2}^{\gamma z})^4 } \ , s_{\theta } \ , \sqrt{\frac{\hat{s}}{2 } } \ , \bigg [ \lambda_3 ( \lambda_2 - \lambda_1 ) + \bigg(2 \lambda_1 \lambda_2 - \lambda_3 ( \lambda_1 + \lambda_2 ) \bigg ) c_{\theta } \bigg ] \nonumber \\ { \cal m}_{3}^{\gamma z}(\lambda_1,\lambda_2;\lambda_3,\lambda_4 ) & = & \frac{ie^2 \hat{s}^2}{32(\lambda_{3}^{\gamma z})^4 } \ , \bigg [ 3(\lambda_1 + \lambda_2)(1+\lambda_3 \lambda_4 ) + 12(\lambda_4 + \lambda_1 \lambda_2 \lambda_3 ) + 6(\lambda_3 + \lambda_1 \lambda_2 \lambda_4 ) \nonumber \\ & & + 6(\lambda_2 - \lambda_1)(1-\lambda_3 \lambda_4 ) c_{\theta } + \bigg(2 ( \lambda_4 + \lambda_1 \lambda_2 \lambda_3 ) - ( \lambda_1 + \lambda_2 ) ( 1 + \lambda_3 \lambda_4 ) \nonumber \\ & & + ( \lambda_4 - \lambda_3 ) ( 1- \lambda_1 \lambda_2 ) \bigg ) c^{2}_{\theta } \bigg ] \nonumber \\ { \cal m}_{3}^{\gamma z}(\lambda_1,\lambda_2;\lambda_3,0 ) & = & \frac{ie^2 m_z \hat{s}}{8(\lambda_{3}^{\gamma z})^4 } \ , s_{\theta } \ , \sqrt{\frac{\hat{s}}{2 } } \ , \bigg [ \lambda_3 ( \lambda_1 - \lambda_2 ) + \bigg ( 4 -2 \lambda_1 \lambda_2 - \lambda_3 ( \lambda_1 + \lambda_2 ) \bigg ) c_{\theta } \bigg ] \ , .\end{aligned}\ ] ] again neglecting terms of @xmath279 , the @xmath2 amplitudes become @xmath281 \left(1+c^{2}_{\theta } \right ) \nonumber \\ { \cal m}_{3}^{z z}(\lambda_1,\lambda_2;\lambda_3,0 ) & = & \frac{ie^2 m_z \hat{s}}{2(\lambda_{3}^{z z})^4 } \ , s_{\theta } \ , \sqrt{\frac{\hat{s}}{2 } } \ , \bigg [ ( 1 + \lambda_1 \lambda_2 ) - \lambda_3 ( \lambda_1 + \lambda_2 ) \bigg ] c_{\theta } \nonumber \\ { \cal m}_{4}^{z z}(\lambda_1,\lambda_2;\lambda_3,\lambda_4 ) & = & \frac{ie^2 \hat{s}^2}{32(\lambda_{4}^{z z})^4 } \ , \bigg [ 6 ( \lambda_1 + \lambda_2 ) ( 1 + \lambda_3 \lambda_4 ) + 2 ( \lambda_3 + \lambda_4 ) ( 1 + \lambda_1 \lambda_2 ) \nonumber \\ & & + 2 \bigg ( - ( \lambda_1 + \lambda_2 ) ( 1 + \lambda_3 \lambda_4 ) + ( \lambda_3 + \lambda_4 ) ( 1 + \lambda_1 \lambda_2 ) \bigg ) c^{2}_{\theta } \bigg ] \nonumber \\ { \cal m}_{4}^{z z}(\lambda_1,\lambda_2;\lambda_3,0 ) & = & \frac{ie^2 m_z \hat{s}}{4(\lambda_{4}^{z z})^4 } \ , s_{\theta } \ , \sqrt{\frac{\hat{s}}{2 } } \ , \bigg [ ( 1 + \lambda_1 \lambda_2 ) - \lambda_3 ( \lambda_1 + \lambda_2 ) c_{\theta } \bigg ] \nonumber \\ { \cal m}_{5}^{z z}(\lambda_1,\lambda_2;\lambda_3,\lambda_4 ) & = & \frac{ie^2 \hat{s}^2}{32(\lambda_{5}^{z z})^4 } \ , \bigg [ -6 ( \lambda_3 + \lambda_4 ) ( 1 + \lambda_1 \lambda_2 ) - 2 ( \lambda_1 + \lambda_2 ) ( 1 + \lambda_3 \lambda_4 ) \nonumber \\ & & + 2 \bigg ( ( \lambda_3 + \lambda_4 ) ( 1 + \lambda_1 \lambda_2 ) - ( \lambda_1 + \lambda_2 ) ( 1 + \lambda_3 \lambda_4 ) \bigg ) c^{2}_{\theta } \bigg ] \nonumber \\ { \cal m}_{5}^{z z}(\lambda_1,\lambda_2;\lambda_3,0 ) & = & \frac{ie^2 m_z \hat{s}}{4(\lambda_{5}^{z z})^4 } \ , s_{\theta } \ , \sqrt{\frac{\hat{s}}{2 } } \ , \bigg [ ( 1 + \lambda_1 \lambda_2 ) - \lambda_3 ( \lambda_1 + \lambda_2 ) c_{\theta } \bigg]\end{aligned}\ ] ] amplitudes with two longitudinal @xmath80 bosons are of @xmath279 , and hence are negligible in our approximation . i. f. ginzburg , g. l. kotkin , v. g. serbo and v. i. telnov , nucl . methods * 205 * ( 1983 ) 47 ; i. f. ginzburg , g. l. kotkin , s. l. panfil , v. g. serbo and v. i. telnov , nucl . instr . methods * 219 * ( 1984 ) 5 . g. blanger and f. boudjema , phys . lett . * b288 * ( 1992 ) 210 ; k. a. peterson and s. godfrey , hep - ph/9302297 ; o. j. p. obli , m. c. gonzalz - garca and s. f. novaes , nucl * b411 * ( 1994 ) 381 , hep - ph/9306306 ; o. j. p. obli , m. b. magro , p. g. mercandante and s. f. novaes , phys.rev . * d52 * ( 1995 ) 15 , hep - ph/9503432 ; s. godfrey , hep - ph/9505252 ; g. j. gounaris , f. m. renard , z. phys . * c69 * ( 1996 ) 513 , hep - ph/9505429 ; g. j. gounaris , j. layssac , f. m. renard , z. phys . * c69 * ( 1996 ) 505 , hep - ph/9505430 ; m. baillargeon , g. blanger and f. boudjema , nucl.phys . * b500 * ( 1997 ) 224 , hep - ph/9701372 ; e. boos . hep - ph/0103090 . t. g. rizzo , phys . rev . * d60 * ( 1999 ) 115010 , hep - ph/9904380 ; k. cheung , phys . rev . * d61 * ( 2000 ) 015005 , hep - ph/9904266 ; h. davoudiasl , phys . rev . * d60 * ( 1999 ) 084022 , hep - ph/9904425 ; phys . rev . * d61 * ( 2000 ) 044018 , hep - ph/9907347 . s. hesselbach and h. fraas , phys . * d55 * ( 1997 ) 1343 , hep - ph/9604439 ; v. barger , t. han , and j. kelly , phys . b419 * ( 1998 ) 233 , hep - ph/9709366 ; s. berge , m. klasen and y. umeda , phys.rev . * d63 * ( 2001 ) 035003 , hep - ph/0008081 ; k. hagiwara , hep - ph/0011360 ; m. m. mhlleitner , m. krmer , m. spira and p. m. zerwas , hep - ph/0101083 ; m. m. mhlleitner , hep - ph/0101263 . g. r. farrar and m. e. shaposhnikov , phys . * 70 * ( 1993 ) 2833 ; erratum - ibid . * 71 * ( 1993 ) 210 ; phys . rev . * d50 * ( 1994 ) 774 ; hep - ph/9406387 ; m.b . gavela , p.hernndez , j. orloff , o.pne , mod . a9 * ( 1994 ) 795 ; hep - ph/9407403 ; p. huet and e. sather , phys . rev . * d51 * ( 1995 ) 379 , hep - ph/9404302 . h. anlauf , w. bernreuther and a. brandenburg , phys.rev . * d52 * ( 1995 ) 3803 , erratum - ibid . * d53 * ( 1996 ) 1725 , hep - ph/9504424 ; p. poulose ans s. d. rindani , phys.rev . * d57 * ( 1998 ) 5444 , erratum - ibid . * d61 * ( 2000 ) 119902 , hep - ph/9709225 ; s. y. choi and j. s. lee , phys.rev . * d62 * ( 2000 ) 036005 , hep - ph/9912330 ; i. f. ginzburg and i. p. ivanov , hep - ph/0004069 ; d. atwood , s. bar - shalom , g. eilam and a.soni , hep - ph/0006032 . g. blanger and g. couture , phys . * d49 * ( 1994 ) 5720 ; s. choi and k. hagiwara , phys . lett . * b359 * ( 1995 ) 369 , hep - ph/9506430 ; g. j. gounaris and g. p. tsirigoti , phys.rev . * d56 * ( 1997 ) 3030 , erratum - ibid . * d58 * ( 1998 ) 059901 , hep - ph/9703446 ; m.s . baek , s.y . choi and c.s . kim , phys . rev . * d56 * ( 1997 ) 6835 , hep - ph/9704312 ; i. f. ginzburg and i. p. ivanov , hep - ph/0004069 ; e. asakawa , s.y . choi , k. hagiwara and j. s. lee , phys . rev . * d62 * ( 2000 ) 115005 , hep - ph/0005313 . s. y. choi , k. hagiwara and m. s. baek , phys.rev . * d54 * ( 1996 ) 6703 , hep - ph/9605334 . g. jikia and a. tkabladze , phys . lett . * b323 * ( 1994 ) 453 ; phys . lett . * b332 * ( 1994 ) 441 ; g. jikia , nucl . phys . * b405 * ( 1993 ) 24 ; phys . lett . * b298 * ( 1993 ) 224 . g. j. gounaris , p. i. porfyriadis , and f. m. renard , phys . lett . * b452 * ( 1999 ) 76 , hep - ph/9812378 . g. j. gounaris , p. i. porfyriadis , and f. m. renard , eur . j. * c9 * ( 1999 ) 673 , hep - ph/9902230 . g. j. gounaris , j. layssac , p. i. porfyriadis , and f. m. renard , eur . j. * c10 * ( 1999 ) 499 , hep - ph/9904450 . g. j. gounaris , j. layssac , p. i. porfyriadis , and f. m. renard , eur . j. * c13 * ( 2000 ) 79 , hep - ph/9909243 . k. hagiwara , r. d. peccei , d. zeppenfeld , and k. hikasa , nucl . b282 * ( 1987 ) 253 ; m. baillargeon , g. blanger , and f. boudjema , nucl . phys . * b500 * ( 1997 ) 224 , hep - ph/9701372 ; f. berends et . al . , j. phys . * g24 * ( 1998 ) 405 , hep - ph/9709413 ; g. j. gounaris , j. layssac , and f. m. renard , phys . rev . * d61 * ( 2000 ) 073013 . a. denner , s. dittmaier , m. roth , and d. wackeroth , hep - ph/0104057 . d. lovelock and h. rund , _ tensors , differential forms , and variational principles _ , new york : dover publications , 1989 . b. badelek et . al . , _ the photon collider at tesla _ , http://www.desy.de/lcnotes/tdr/ ; j. bagger et . al . , hep - ex/0007022 . the opal collaboration , g. abbiendi et al , phys . lett . * b471 * ( 1999 ) 293 , hep - ex/9910069 ; the l3 collaboration , m. acciarri et . al . , phys . lett . * b490 * ( 2000 ) 187 , hep - ex/0008022 ; the l3 collaboration , m. acciarri et . al . , hep - ex/0102024 . g. blanger , f. boudjema , y. kurihara , d. perret - gallix and a. semenov , eur . phys . j. * c13 * ( 2000 ) 283 , hep - ph/9908254 . o. j. p. obli , m. c. gonzalz - garca , s. m. lietti , and s. f. novaes , phys . rev . * d63 * ( 2001 ) 075008 , hep - ph/0009262 .

there is presently a strong interest in computation and information processing based on fundamental principles of quantum mechanics @xcite . quantum information technology has the potential both to address problems that can not be solved by standard , classical information technology as well as to radically improve the performance of existing classical schemes . the prospect of scalability and integrability with conventional electronics makes solid state systems a likely future arena for quantum information processing . of particular interest is the entanglement between the elementary charge carriers , quasiparticles , in meso- or nanoscopic solid state conductors . entanglement , or quantum mechanical correlations , constitutes a resource for any quantum information process . moreover , due to controllable system properties and coherent transport conditions , conductors on the meso and nano scale constitute ideal systems for the generation and detection of quasiparticle entanglement . this opens up for quantum bits based on the spin or orbital quantum states of individual electrons , the ultimate building blocks for solid state quantum information processing . to date quasiparticle entanglement has however remained experimentally elusive . in particular , there is no unambiguous experimental demonstration of entanglement between two spatially separated quasiparticles . a class of mesoscopic systems that appear promising for a successful entanglement experiment are conductors without direct interactions between the quasiparticles . it was shown by beenakker _ et al _ @xcite that fermions emitted from a thermal source can , in contrast to bosons , be entangled by scattering at a beam - splitter . this was originally discussed for electron - hole pairs @xcite and shortly afterward for pairs of electrons @xcite . since then there has been a large number of works on entanglement of non - interacting particles , see e.g. @xcite for a number of representative papers and also @xcite for a review . several of the entanglement proposals have been based on electrical analogs of optical interferometers and beam - splitter geometries . such electronic systems are conveniently implemented in conductors in the quantum hall regime , where electrons propagate along chiral edge states @xcite and quantum point contacts constitute reflectionless beam - splitters @xcite with controllable transparency , see e.g. @xcite . recent experimental progress on electronic mach - zehnder @xcite and hanbury brown twiss @xcite interferometers has provided further motivation for a theoretical investigation of entanglement in such systems . in addition , the experimental realization @xcite of time - controlled single - electron emitters @xcite in quantum hall systems has opened up the possibility for a dynamical generation of entangled quasiparticles , entanglement on demand @xcite . in this work we will focus on the electronic two - particle , or hanbury brown twiss , interferometer . a theoretical proposal for an implementation of this two - particle interferometer ( 2pi ) in a conductor in the quantum hall regime was proposed by two of us , p.s and m.b . , together with e. v. sukhorukov in ref . recently , the heiblum group , including one of us , i.n . , was able to realize the 2pi in a versatile system which could be electrically tuned between with two independent mach - zehnder interferometers and a 2pi . in perfect agreement with the theoretical predictions @xcite , the two - particle interference pattern was visible in the current correlations but not in the average current . as discussed in ref . @xcite , there is an intimate relation between two - particle interference and entanglement in the fermionic 2pi . under ideal conditions , i.e. zero temperature and perfect coherence , two - particle interference implies that the two particle wave function is on the form @xmath0 . \label{introsing}\ ] ] here @xmath1 denote the sources and @xmath2 the sites of detection , as shown in fig . [ system ] . the wavefunction @xmath3 is maximally entangled , it is a singlet in the orbital , or pseudo spin , space @xmath4 . however , in the experiment @xcite , @xmath5 visibility of the current correlation oscillations was observed . this indicates that both decoherence and finite temperature is important . dephasing can qualitatively be accounted for @xcite by a suppression of the off - diagonal components of the density matrix @xmath6 . it was shown that at zero temperature the entanglement survives for arbitrary strong dephasing . the effect of finite temperature was not investigated at the time of the experiment . the experimental findings thus raised two important questions : are the electrons reaching the detectors at a and b entangled and if so , can this two - particle entanglement be unambiguously detected by measurements of currents and current correlators , the standard quantities accessible in electronic transport measurements ? in our recent work @xcite we provided a positive answer to both these questions . we first calculated the entanglement of the emitted two - particle state and found that the state was clearly entangled . thereafter we showed that under very general conditions the entanglement of the reduced two - particle density matrix provides a lower bound for the entanglement of the emitted two - particle state . since the reduced density matrix is possible to reconstruct tomographically by current and current correlation measurements @xcite , this provides an unambiguous way to detect the entanglement of the emitted state . in the present paper we discuss these findings in more detail . interference is most often investigated in structures that lead to a superposition of amplitudes of a single particle . however , in 1956 , hanbury brown and twiss ( hbt ) invented an optical interferometer based on correlations of light intensities @xcite , an optical 2pi , see fig . [ system ] . the intensity interferometer allowed hbt to determine the angular diameter of a number of visual stars , not possible with available single particle , or michelson , interferometers . the hbt intensity interferometer displays two distinct but fundamentally interrelated features : + @xmath7 first , there is a direct statistical effect since photons from a thermal light source tend to bunch , whereas fermions would anti - bunch . this effect has been used in a large number of experiments in different fields of physics such as elementary particles @xcite , solid state @xcite and free @xcite electrons and recently cold atoms @xcite . + @xmath7 second , light from two different , completely uncorrelated sources gives rise to an interference effect in intensity correlations but not in the intensities themselves . this is the two - particle interference effect . in optics , various aspects of two - particle interference have been investigated extensively since the hbt - experiment , see e.g. @xcite for a short review , and is still a subject of interest @xcite . in electronics , only very recently was a fermionic two - particle interferometer realized @xcite , the subject of this work . + fundamentally both of these effects are related to the symmetry of the multiparticle wave function under exchange of two particles . we note that albeit the hbt - experiment could be explained by a classical electro - magnetic theory , a compelling quantum mechanical picture based on individual photons was put forth soon after the experiment @xcite . importantly , for fermions no classical theory exists . to obtain a qualitative understanding of the physics of two - particle interferometers it is rewarding to compare the properties of optical , bosonic interferometers and electronic , fermionic interferometers . in fig . [ system ] a schematic of a two - particle interferometer , topologically equivalent to the hbt - interferometer , is shown . a natural measure of the correlations between the particles at @xmath8 and @xmath9 is the probability to jointly detect one particle at @xmath8 and one at @xmath9 . an expression for this joint detection probability for photons was derived by glauber @xcite . in ref . @xcite this was adapted to detection of electrons . here we consider the probability to detect one photon / electron in detector @xmath10 , @xmath11 , at time @xmath12 and one in detector @xmath13 , @xmath14 at a time @xmath15 , given by @xmath16 the photon / electron creation operators at a are @xmath17 , with @xmath18 creating a particle in @xmath10 at energy @xmath19 and similarly at b. for photons we consider thermal sources in @xmath20 and @xmath21 while @xmath22 and @xmath23 are left empty . a detector frequency window of size @xmath24 is assumed , over which the distribution functions of the sources are constant , i.e. @xmath25 . for electrons we assume zero temperature and the sources @xmath20 and @xmath21 biased at @xmath26 while sources @xmath22 and @xmath23 are grounded . only quasiparticle excitations , @xmath27 are considered . the probabilities are normalized such that @xmath28 . following the scattering theory for intensity / current correlations for bosons / fermions emitted from thermal sources @xcite , we get @xmath29+|s_{a\alpha 2}|^2|s_{b\beta 2}|^2\left[1\pm g(\tau)\right ] \nonumber \\ & + & |s_{a\alpha 1}|^2|s_{b\beta 2}|^2+|s_{a\alpha 2}|^2|s_{b\beta 1}|^2 \nonumber \\ & \pm & g(\tau)\left[s_{a\alpha 1}^*s_{b\beta 2}^*s_{b\beta 1}s_{a\alpha 2}+s_{a\alpha 1}s_{b\beta 2}s_{b\beta 1}^ * s_{a\alpha 2}^*\right ] \label{jdpt}\end{aligned}\ ] ] where @xmath30 contains the time dependence , with @xmath31 the coherence time for electrons and @xmath32 for photons . here @xmath33 is the amplitude to scatter from source 2 to detector @xmath10 etc . the upper / lower signs @xmath34 correspond to electrons / photons . several interesting conclusions can be drawn directly from eq . ( [ jdpt ] ) : \1 ) for @xmath35 , @xmath36 approaches zero and @xmath37 is just proportional to the product of the two mean currents / intensities . the fermionic versus bosonic statistics of the particle plays no role . \2 ) for shorter times , @xmath38 , @xmath36 is finite and the statistics is important . note that , as pointed out above , that the statistics of the particles enter in two different ways . + i ) the first two terms in eq . ( [ jdpt ] ) describe a direct bunching ( + ) or anti - bunching ( - ) effect for two particles emitted from the same reservoir within a time @xmath38 . this effect would still be present if one of the sources @xmath20 or @xmath21 is removed . + ii ) the last two terms describe the two - particle , or exchange @xcite , interference , where the @xmath34 sign explicitly follows from the interchange of the two detected particles . this two particle interference is only present when both sources are active . for semitransparent beam - splitters @xmath39 and @xmath40 and coincident detection @xmath41 we have @xmath42 & \mbox{fermions } \\ \frac{1}{4}\left[1+\frac{\alpha\beta}{2}\cos \phi\right ] & \mbox{bosons } \end{array}\right . \label{jdpt2}\end{aligned}\ ] ] where @xmath43 is a scattering phase . from this expression a very important difference between bosonic and fermionic thermal sources is apparent : the visibility @xmath44 of the oscillations is @xmath20 for fermions but only @xmath45 for bosons . this is directly related to the fact that while the emitted fermionic two - particle state is maximally entangled , the bosonic state is unentangled @xcite . in ref . @xcite we proposed an implementation of an electronic 2pi in a conductor in the quantum hall regime , with electrons propagating along single , spin polarized edge states ( see fig . [ hbtferm ] ) . two electronic reservoirs @xmath1 biased at @xmath26 act as sources for electrons while the reservoirs @xmath46 as well as the detector reservoirs are grounded . all reservoirs are kept at the same temperature @xmath47 . moreover , we consider here only the linear regime in voltage where electron - electron interactions can be neglected . this regime is relevant for the experiment @xcite . the qpc s at @xmath39 and @xmath40 act as beamsplitters with transparencies @xmath48 and @xmath49 respectively . the scattering amplitude @xmath50 , where @xmath51 and @xmath52 is the scattering phase picked up by the electron up when traveling from c to a. similar relations hold for the other scattering amplitudes . note that the total phase @xmath53 is , up to a constant term , given by @xmath54 where @xmath55 is the magnetic flux threading the 2pi and @xmath56 , the single particle flux quanta . importantly , the corbino geometry in fig . [ hbtferm ] with unidirectional edge states and reflectionless beam - splitters is topologically equivalent to the 2pi shown in fig . [ system ] . the standard tools for investigating transport properties in mesoscopic electronic systems are average electrical current and current correlation measurements @xcite . a scattering theory calculation @xcite gives the average current at contact @xmath10 @xmath57 \nonumber \\ \label{curr}\end{aligned}\ ] ] and similar at @xmath58 . here @xmath59 and @xmath60 are the fermi distributions of the biased , @xmath1 and the grounded , @xmath46 reservoirs respectively . the irreducible zero frequency correlator @xmath61 between currents @xmath62 and @xmath63 @xcite becomes @xmath64 ^ 2 \label{noise}\ ] ] these expressions are valid for arbitrary temperature but for the rest of the discussion in this section we only consider the zero temperature case . in particular , for the simplest possible case , with all beam - splitters semitransparent and energy - independent scattering amplitudes , we have @xmath65 \label{currnoise}\end{aligned}\ ] ] while the average current is a function of qpc - transparencies only , the current cross correlator depends also on the phase @xmath43 . since this phase is proportional to the magnetic flux @xmath55 threading the 2pi , we call this a two - particle aharonov - bohm ( ab ) effect . interestingly , we can directly relate the coincident detection probability in eq . ( [ jdpt ] ) at times @xmath66 with the currents in eq . ( [ curr ] ) and the zero frequency noise correlators in eq . ( [ noise ] ) as @xmath67 $ ] @xmath68 this is a direct consequence of fermionic anti - bunching , leading to a filled stream of electrons emitted from the source reservoirs and hence making long time observables an effective average of many individual , short time , single and two - particle events . the connection between this two - particle aharonov - bohm effect and entanglement can be seen by considering the many - body ground state @xmath69 of the electrons injected into the 2pi . electrons at different energies are independent and the many - body state at zero temperature is thus a product state in energy @xmath70 where @xmath71 is the filled fermi sea and @xmath72 creates an electron at energy @xmath19 , incident from reservoir @xmath20 . adopting the formalism of ref . @xcite we first define @xmath73 the injected state at energy @xmath19 . we have the scattering relations at the two source beam splitters , suppressing energy notation @xmath74 for incoming ( a s ) and outgoing ( b s ) electrons . the primed scattering amplitudes thus describes particles incoming from the unbiased sources . this gives the emitted state for the electrons at energy @xmath19 , after beam - splitters @xmath75 but before impinging on the detector beam splitters @xmath2 , as @xmath76 since we are interested in entanglement between particles in the two , spatially separated detector regions a and b we project out the part of the wave function with one particle in a and one in b yielding the normalized wavefunction @xmath77 with @xmath78 the normalization constant . here we introduced the transmission and reflection probabilities of the source beam splitters as @xmath79 and @xmath80 for c and similarly for d. to make this more transparent we can , since the two particles live in well separated hilbert spaces , introduce the dirac notation @xmath81 etc , and write @xmath82 \label{pureab}\ ] ] which for semi - transparent beam splitters ( and scattering phase @xmath83 ) reduces to the singlet state @xmath3 in eq . ( [ introsing ] ) . the orbital states are shown in fig . [ hbtferm ] the entanglement of the state @xmath84 can conveniently be quantified in terms of the concurrence @xmath85 @xcite , which ranges from zero for an unentangled state to unity for a maximally entangled state . working in the computational basis @xmath86 , for the pure state @xmath87 in eq . ( [ pureab ] ) we have @xmath88 where @xmath89 is @xmath87 with all coefficients complex conjugated , @xmath90 a pauli matrix and @xmath91 the direct , tensor product . we thus find for @xmath87 the concurrence @xmath92 which reaches unity for semitransparent beam splitters , i.e. for the singlet state in eq . ( [ introsing ] ) . note that the normalization factor @xmath93 is maximal , equal to @xmath45 , for semitransparent beam splitters . this demonstrates that at most only half of the particles injected from @xmath20 and @xmath21 lead to split pairs , with one particle emitted towards @xmath8 and one towards @xmath9 , i.e. a maximal pair emission rate of @xmath45 . for a measurement during a time @xmath94 the maximum concurrence production @xcite is thus @xmath95 , where @xmath96 the number of pairs injected from @xmath20 and @xmath21 in the time @xmath94 and energy interval @xmath97 there are several microscopic mechanisms that can lead to dephasing , typically suppressing the two - particle interference . for low temperatures it is commonly believed that the dominatinating mechanism for dephasing is electron - electron interactions , but this is still a topic of ongoing research and goes beyond the scope of the present work . here we consider no specific mechanism but model dephasing qualitatively by coupling one of the interferometer arms to a dephasing voltage probe @xcite . in this context we point out a recent experiment @xcite : a voltage probe was coupled , via a tunable quantum point contact , to one arm of a mach zehnder interferometer in the quantum hall regime , demonstrating controllable dephasing . considering semitransparent beam splitters , the dephasing probe coupled with a strength @xmath98 lead to a modification of the current correlator in eq . ( [ currnoise ] ) to @xcite @xmath99 \end{aligned}\ ] ] from this expression it is clear that @xmath100 enters as a decoherence parameter ; decreasing @xmath100 from @xmath20 to @xmath101 leads to a suppression the phase dependence of the current correlator . in the presence of dephasing the emitted state is no longer a pure state , it is instead a mixed state described by a density matrix @xmath102 . considering zero temperature , working in the computational basis the result for @xmath103 corresponds to a suppression of the off - diagonal components of @xmath104 as @xmath105 the concurrence for a mixed state is @xcite @xmath106 where @xmath107 , @xmath108 , are the eigenvalues in decreasing order of @xmath109 . we then have @xmath110 this means that the entanglement persists even for very strong dephasing @xcite . this is a consequence of the 2pi - geometry , where scattering between the arms , i.e. pseudo spin - flip scattering , is prohibited . very recently the electronic 2pi was realized experimentally by neder _ et al_. in the experiment , in the quantum hall regime , it was possible to electrically tune the system between two individual mach zehnder interferometers and a 2pi , as shown schematically in fig . [ expfig ] . the authors first tuned the system to two mach - zehnder interferometers and measured the single particle interference in the average current for each interferometer . they found a very large visibility in both interferometers , around @xmath111 . they also determined the periods of the single particle ab - oscillations as a function of both the area and the magnetic flux enclosed by the interferometers . thereafter the system was tuned to a single 2pi . as predicted by theory @xcite no single - particle ab - oscillations in the average current were observed but the current cross correlations displayed clear two - particle ab - oscillations , with an amplitude @xmath112 of the predicted coherent , zero temperature value . by measuring also the period of the two - particle oscillations as a function of interferometer area and enclosed flux and comparing to the sum of the periods for the two mach zehnder interferometers , the two - particle nature of the ab - oscillations could be established beyond doubt . in the experiment semitransparent beam splitters were used , @xmath113 . for the current cross correlations , theory for finite temperature and dephasing @xcite predicts , for @xmath114 , @xmath115 . \label{noise2}\ ] ] the temperature dependence is fully contained in @xmath116 varying from unity for @xmath117 to zero for @xmath118 . the effect of finite temperature is thus to suppress the overall amplitude of the current cross correlation oscillations . in the experiment , the applied bias was @xmath119 . the electron temperature was estimated from independent auto - correlation measurements to be @xmath120 . this yields the temperature suppression factor @xmath121 . a direct comparison to eq . ( [ noise2 ] ) then gives the oscillation amplitude @xmath122 , i.e. @xmath123 , a substantial dephasing . our main aim of this work is to theoretically investigate the effects of finite temperature on the entanglement of the state emitted out from the source , towards the detectors . a prerequisite is to obtain both a qualitative and a quantitative description of the emitted many - body state at finite temperature . we consider the experimentally relevant situation with all source and detector reservoirs kept at the same temperature @xmath47 . due to the finite temperature , not only the electrons emitted from the source in the energy range @xmath124 are of interest , we must in principle take into account particles emitted from all reservoirs at all possible energies . however , due to the chiral geometry of the 2pi in fig . [ hbtferm ] , particles emitted from the detectors can never scatter back to the detectors , i.e. detector cross talk is topologically prohibited . the particles arriving at the detectors thus all originate from the source reservoirs and we can focus on the many body state emitted by source @xmath20 to @xmath23 . we note that in the slightly different geometry realized experimentally @xcite , there is the possibility for scattering between the detectors . it can however be shown @xcite that this does not influence the entanglement of the emitted state . at finite temperature the state injected from the sources is mixed and described by a density matrix @xcite @xmath125|0\rangle\langle 0|+f_{\kappa}(e)a^{\dagger}_{\kappa}(e)|0\rangle\langle 0|a_{\kappa}(e)\right ] \label{rhoin}\end{aligned}\ ] ] where @xmath126 is the fermi distribution of source reservoir @xmath127 . the outgoing state is then obtained by inserting the scattering relations of eq . ( [ scatrel ] ) int eq . ( [ rhoin ] ) . one can see from eq . ( [ rhoin ] ) that the effect of finite temperature is to give rise to states with 0 to 4 particles emitted at a given energy . for the terms of interest , i.e. with at least one particle at both a and b , there is at finite temperature the possibility for e.g. two particles at a and one at b etc . these terms are of central importance in the discussion below . a theory for entanglement production in non - interacting @xcite conductors at finite temperature was presented by beenakker @xcite and along similar lines in closed condensed matter systems by dowling , doherty and wiseman @xcite . at a given energy , only the component of the emitted many - body state with one particle in detector region a and one in b has nonzero entanglement . moreover , as emphasized in ref . @xcite , only this term describes two particles which each live in a well defined @xmath128 hilbert spaces at a and b respectively , i.e. two coupled orbital qubits . we point out that this definition does not take into account occupation - number , or fock - space entanglement . the first step is thus to project out the two - particle component from the many - body wave function , which is accomplished by the projection operator @xmath129 where @xmath130 with @xmath131 etc is the number operator ( suppressing energy notation ) . this yields the projected density matrix @xmath132 the elements of the density matrix @xmath133 are conveniently calculated from the relation @xcite @xmath134_{ij , kl}=\langle \pi b_{ai}^{\dagger}b_{bj}^{\dagger}b_{bk}b_{al } \pi \rangle\ ] ] where , for any operator x , @xmath135 $ ] is the standard quantum - statistical average . some algebra gives the projected density matrix , formally equivalent to the density matrix calculated in @xcite , eqs . ( b9 ) - ( b13 ) , @xmath136 where @xmath137 and @xmath138 and @xmath139 the fermi distribution functions of the grounded and biased source reservoirs respectively . the coefficients @xmath140+\chi)(t_d[1-\chi]+\chi ) , \nonumber \\ c_{21}^{21}&=&(t_c[1-\chi]+\chi)(r_d[1-\chi]+\chi ) , \nonumber \\ c_{12}^{21}&=&(c_{21}^{12})^*=-\gamma \sqrt{r_ct_cr_dt_d}e^{i\phi_0}(1-\chi)^2\end{aligned}\ ] ] with @xmath141 an overall scattering phase of the beam splitters c and d. thus , only the prefactor @xmath142 depends on energy . as for the zero temperature case we have introduced dephasing as a suppression of the off - diagonal components of the density matrix . it follows from eq . ( [ projected ] ) that finite temperature leads to + i ) an overall modification of the energy - dependent probability for two - particle emission via the prefactor @xmath143 . + ii ) a suppression @xmath144 of the off - diagonal components , equivalent to the effect of dephasing . + iii ) a finite amplitude for the diagonal density matrix elements @xmath145_{11,11}$ ] and @xmath145_{22,22}$ ] , i.e for two particles being emitted from either sources 1,3 or 2,4 . + additional insight follows from writing the projected density matrix as @xmath146 \label{projected2}\ ] ] where the diagonal density matrix @xmath147\ ] ] with the zero temperature single particle density matrices @xmath148 and @xmath149 . the density matrix @xmath150\end{aligned}\ ] ] results from the two - particle interference . here we used the shorthand notation @xmath151 with @xmath152 etc . note that the effect of decoherence enters as a suppression of the two - particle interference @xmath153 , where @xmath154 . writing @xmath155 in the form in eq . ( [ projected2 ] ) shows that , taken the energy dependent prefactor @xmath142 aside , the effects of finite temperature can be viewed as follows : first , the amplitude of the two - particle interference component @xmath156 is suppressed with increasing temperature as @xmath144 . second , the density matrix acquires a purely diagonal component @xmath157 with an amplitude @xmath158 ( note that @xmath159=4 $ ] , independent on temperature ) . for the entanglement , following @xcite we introduce @xmath160 and @xmath161 , the normalized density matrix and the emission probability of the emitted two - particle state respectively , defined from @xmath162=(1-f)^2f_v^2[(r_ct_d+t_cr_d)(1-\chi)^2 + 4\chi]\end{aligned}\ ] ] where we note that @xmath160 is independent on energy . the emission probability @xmath161 is thus the probability , per unit energy , that the ( normalized ) two - particle state @xmath160 is emitted . the concurrence production per unit energy is then @xmath163 and the total entanglement production during a time @xmath94 , @xmath164 , is then ( @xmath96 ) @xmath165 we denote this the projected entanglement . as shown in fig . [ fig2 ] , @xmath166 decreases monotonically as a function of @xmath47 . it reaches zero at a critical temperature @xmath167 given by @xmath168 for semi - transparent beam - splitters and zero dephasing , @xmath169 , the entanglement thus survives up to @xcite @xmath170 . inserting the parameter values from the experiment , we get @xmath171 and @xmath172 , i.e. _ the state emitted by the 2pi is clearly entangled_. importantly , the effect of finite temperature is essentially negligible , the reduction in entanglement comes from decoherence . the entanglement of the projected density matrix is the entanglement one could access , had one been able to do arbitrary local operations and classical communication between a and b , i.e. fully energy and particle resolved measurements . under realistic conditions this is not possible , the accessible physical quantities are currents and current cross correlators . is it possible to determine the projected entanglement with such measurements ? the answer to this question is no , for two main reasons : + i ) as discussed above , at nonzero temperatures it is not only the biased source reservoirs which emit particles but also the grounded source reservoirs do . as a consequence , there is a finite amplitude for emitted states with two - particles at a and/or at b. these unentangled states contribute to currents and current correlators , which results in a detectable state with suppressed entanglement . + ii ) the current and current correlators provide information on the energy integrated properties of the many - body state , not on the emitted state at each energy . this lack of energy - resolved information leads to a further suppression of the detectable entanglement . + clearly , these effects of the thermally excited fermi sea constitute generic problems when trying to detect entanglement in mesoscopic conductors . as a remedy for these finite temperature read - out problems it was suggested to work with detectors at very low temperatures @xcite . another idea was recently presented by hannes and titov @xcite . they investigated detection of entanglement at finite temperatures via a bell inequality and proposed to introduce energy filters at the drains . however , both schemes @xcite would lead to additional experimental complications in systems which already are experimentally very challenging . our idea is instead to investigate what information about the projected entanglement can actually be deduced from current and current correlation measurements . in this context we also mention the recent proposal by kindermann @xcite , to produce and detect entangled electron - hole pairs in graphene via a bell inequality formulated in terms of the transport part of the current cross correlators @xcite , i.e. by subtracting away the thermal equilibrium correlators from the finite bias ones . in our work @xcite we proposed a similar scheme for a general mesoscopic conductor . however , as was pointed out in @xcite and is further discussed below , it is important that one performs a detailed comparison of the projected entanglement and the entanglement obtained from current cross correlation measurements . without such a comparison , there is the possibility that one concludes , based on correlation measurements , finite entanglement where there is none , i.e. the projected entanglement is zero . we first consider the expression for the current and zero frequency current cross correlators at contacts @xmath173 and @xmath174 at finite temperatures . we have @xmath175 , \hspace{0.5 cm } i_{b+}=\frac{e}{h}\int de \left[\langle n_{b+}\rangle - f\right ] , \nonumber \\ s_{a+b+}&=&\frac{e^2}{h}\int de \langle \delta n_{a+ } \delta n_{b+}\rangle \label{currnoisered}\end{aligned}\ ] ] where @xmath176 is the irreducible correlator . as discussed above , the many - body state incident on the detectors originates from the sources . it is the properties of this state that determines the observables @xmath177 and @xmath178 and thus establishes a connection between the emitted state and the physical quantities accessible in a measurement . in order to better understand the readout problem discussed above , we first discuss the energy resolved properties of the emitted state . if one would have access to energy filters , as proposed in @xcite , or would be working at zero temperature , by combining current and current cross correlations it would be possible to get direct access to the energy resolved quantities @xmath179 and @xmath180 . as is discussed below , by a suitable set of measurements with different settings of the beam splitters at a and b one could then tomographically reconstruct the ( unnormalized ) density matrix of the state emitted out from the source beam splitters c and d , @xmath181 , with elements given by @xmath182_{ij , kl}=\langle b_{ai}^{\dagger}b_{bj}^{\dagger}b_{bk}b_{al } \rangle\ ] ] we denote @xmath181 the energy resolved reduced density matrix . by comparing @xmath181 with the expression for the projected density matrix in eq . ( [ projected ] ) we see that it differs by the projection operators . consequently , the reduced density matrix contains also the contributions from processes with more than one particle at a and/or at b. after some algebra we find the density matrix @xmath183 where we introduced @xmath184 $ ] and the coefficients @xmath185+\chi)(t_d[1-\chi]+\tilde \chi ) , \nonumber \\ \tilde c_{21}^{21}&=&(t_c[1-\chi]+\chi)(r_d[1-\chi]+\tilde \chi).\end{aligned}\ ] ] a comparison to the projected density matrix in eq . ( [ projected ] ) shows that @xmath181 only differs formally from @xmath155 by the change @xmath186 at a number of places . this has the consequence that the normalized density matrix @xmath187 , with @xmath188 $ ] depend on energy . that is , in contrast to @xmath189 both the normalized , emitted two - particle state as well as the emission probability depend on energy . qualitatively , as discussed above , the difference between @xmath181 and @xmath155 arises from the fact that also states with more than one particle at a and/or b contribute to @xmath181 but not to @xmath155 . writing @xmath181 on a form similar to eq . ( [ projected2 ] ) one sees that these three and four particle states contribute only to the diagonal part of @xmath181 . turning to the entanglement , the concurrence production @xmath190 at energy @xmath19 is then @xmath191 from the expression for the concurrence it becomes clear that the separable three and four - particle states are detrimental for the entanglement . hence , finite temperature leads to a stronger suppression of the reduced , energy resolved density matrix than of the projected one . this is illustrated in fig . [ redeconc ] where the corresponding concurrencies are plotted for semitransparent beam - splitters and different values of @xmath192 . as is clear from the figure , there is an energy @xmath193 above which the concurrence is finite ( up to @xmath194 ) . the energy @xmath193 is given by the condition @xmath195 , as @xmath196-\ln \left[(1-\chi)\sqrt{1 + 4\sqrt{r_ct_cr_dt_d}}-(1+\chi)\right]\right ) \label{encond}\ ] ] what is moreover clear from fig . [ redeconc ] is that , for all energies , @xmath197 . the difference is obvious for energies @xmath198 , where @xmath199 . at these energies the probability for emission of separable three and four particle states is thus large enough to completely suppress the entanglement of the reduced density matrix . importantly , the relation @xmath197 holds for all settings of the beam splitters @xmath200 and @xmath49 , as is clear by comparing eqs . ( [ conceproj ] ) and ( [ concered ] ) . the reason for this is that the reduced density matrix contains contributions from all individual particle density matrices @xmath201 with @xmath202 ( e.g. @xmath203 describes one particle at a and two at b ) while the projected density matrix only depends on @xmath204 . since all @xmath205 are separable and the concurrence is a convex quantity , i.e. @xmath206 for @xmath207 , the concurrence @xmath208 is always smaller than @xmath209 . we point out that this carries over to the total concurrence production found by integrating eq . ( [ concered ] ) over energy ( result not presented here ) . it follows from eq . ( [ encond ] ) that for a critical temperature @xmath210 the energy @xmath211 , i.e. the entanglement is zero for any energy . interestingly , this happens for the same temperature as for the projected concurrence , eq . ( [ crittc ] ) . importantly , at finite temperature , without any energy filters , we do not have access to the energy resolved quantities discussed above , only to the total currents and current correlators measured at contacts @xmath212 . in ref . @xcite it was discussed how to , at zero temperature , tomographically reconstruct the reduced density matrix using currents and current correlations . extending this scheme to nonzero temperatures it is natural to define the finite temperature reduced density matrix @xmath213 via the relation @xmath214\rho_r\right\}. \label{noisecurrrel}\end{aligned}\ ] ] we emphasize that @xmath213 is reconstructed from observables already integrated over energy and does hence not depend on energy . also note that @xmath213 is not given by integrating @xmath181 over energy , in fact the difference between the two density matrices is further discussed below . in eq . [ noisecurrrel ] the orbital current operators in the local basis @xmath215 , including the rotations at the detector splitters , are @xmath216 and @xmath217 , with @xmath218 and @xmath219 where @xmath220 $ ] a vector of pauli matrices and @xmath221 the scattering matrix of the beam splitter at a ( b ) . making use of the results for finite temperature current and current correlations in @xcite we obtain the reduced density matrix @xmath222 where @xmath223 . comparing @xmath213 to both @xmath155 and @xmath181 in eqs . ( [ projected ] ) and ( [ redendens ] ) it is clear that the qualitative effect of finite temperature is the same for the reduced density matrix . the quantitative effects are however different . first , the temperature dependence enters via @xmath224 rather than via @xmath225 , giving a much stronger effect of finite temperature . this is the effect of having access to energy integrated quantities only . second , in the expression for the average current in eq . ( [ currnoisered ] ) , in the integrand one subtracts @xmath138 which arises due to particles flowing out of the detector reservoirs . this yields smaller diagonal terms , to be further discussed below . it is illuminating , just as for @xmath155 , to write @xmath213 as a sum of a diagonal and an interference part , @xmath226+h\rho^{int}. \label{reduced2}\ ] ] from this we see that the effect of increasing temperature is to monotonically increase the amplitude for the separable product state @xmath227 , while the amplitude of the interference component is suppressed . we can thus conclude the following properties for all three density matrices @xmath228 and @xmath213 : + i ) at zero temperature they all reduce to the same expression , @xmath156 . + ii ) increasing temperature leads to a monotonic suppression of the two - particle interference component . + iii ) finite temperature introduces an additional diagonal component , different for the three density matrices . turning to entanglement , introducing the normalized reduced density matrix @xmath229 we can write @xmath230=[r_ct_c+r_dt_d](1-h)+r_ct_d+r_dt_c.\end{aligned}\ ] ] we then define the total entanglement production during a time @xmath94 as @xmath231 . it is @xmath232,0\ } \label{concred}\ ] ] here called the reduced entanglement . as @xmath166 , @xmath233 decreases monotonically with increasing @xmath47 . it reaches zero at a critical temperature @xmath234 given by the relation @xmath235 for perfect coherence , @xmath169 , we have @xmath236 , close to one half of @xmath237 . importantly , in contrast to @xmath167 , @xmath234 is independent on the setting of the beam splitters . by comparing the expressions for the two quantities of main interest , the projected and reduced concurrencies , @xmath166 in eq . ( [ concproj ] ) and @xmath233 in eq . ( [ concred ] ) , we can conclude the following : + i ) for both @xmath166 and @xmath233 the origin of the entanglement is the two - particle interference , in fact the component @xmath156 gives rise to the positive term @xmath238 , identical for @xmath166 and @xmath233 . + ii ) for both @xmath166 and @xmath233 finite temperature introduces a negative term , @xmath239 $ ] for @xmath166 and @xmath240 for @xmath233 , which leads to a suppression of the concurrence . these terms arise from the separable , diagonal components of the corresponding density matrices . comparing eqs . ( [ concproj ] ) and ( [ concred ] ) quantitatively we find that @xmath241 for @xmath242 independent on @xmath100 ( see fig . [ fig2 ] ) . consequently , for beam splitters away from the strongly asymmetrical ( tunneling ) limit , _ the reduced entanglement constitutes a lower bound for the projected entanglement_. in the tunneling limit , however , the reduced entanglement is larger than the projected one . thus , in contrast to the energy - resolved reduced density matrix @xmath181 , @xmath213 can be more entangled than @xmath189 . the origin of this difference is , as pointed out above , that when calculating ( and measuring ) @xmath213 the average currents flowing out from the detector reservoirs are subtracted , yielding a smaller diagonal component and hence a larger entanglement @xmath233 . importantly , since the transparencies @xmath200 and @xmath49 can be controlled and measured via average currents in the experiment , it is always possible to verify independently that the condition in eq . ( [ bondcond ] ) is satisfied . turning to the experiment @xcite , for the relevant parameters we have @xmath243 , showing the validity of the bound . however , @xmath244 and based on the measurement @xcite no conclusive statement can be made about @xmath233 and hence not about @xmath166 . in order to detect entanglement via measurements of currents and current correlations , one thus need to work at even lower temperature and further reduce the dephasing in the experiment . a more detailed understanding of this finite temperature readout problem can be obtained by comparing the properties of @xmath160 and @xmath229 . for perfect coherence @xmath169 and identical beam splitters @xmath245 one can ( up to a local phase rotation ) write @xmath246 a werner state @xcite , with singlet weight [ @xmath3 is the singlet in eq . ( [ introsing ] ) ] @xmath247 increasing @xmath192 from zero , @xmath248 becomes exponentially small while @xmath249 increases linearly . these qualitatively different behaviors , clearly illustrated in fig . [ fig2 ] , are a striking signature of how a small @xmath192 , having negligible effect on @xmath250 , leads to a large suppression of @xmath251 . from eqs . ( [ concproj ] ) and ( [ concred ] ) follows also a counter - intuitive result : _ finite amplitude of the ab - oscillations is no guarantee for finite two - particle entanglement_. this is apparent for @xmath229 in the limit of no decoherence @xmath169 and identical beam splitters @xmath252 , since a separable werner state , @xmath253 , can be decomposed @xcite as @xmath254 with the normalized states at @xmath8 and @xmath9 @xmath255/4}\sin \theta^{a / b}_n|2\rangle , \nonumber \\ \theta_1^{a / b}&=&\theta_3^{a / b}=\mbox{atan}[y^{a / b } ] , \hspace{0.5 cm } \theta_2^{a / b}=\theta_4^{a / b}=-\mbox{acot}[y^{a / b } ] \nonumber \\ y^{a / b}&=&\frac{\sqrt{2-\xi_r}+\sqrt{3\xi_r-2}}{\sqrt{\xi_r}\pm \sqrt{4 - 3\xi_r } } , \hspace{0.5 cm } + ( -)~ \mbox{for}~ a(b)\end{aligned}\ ] ] this _ classically _ correlated state gives , via eq . ( [ noisecurrrel ] ) , ab - oscillations with amplitude @xmath256 . moreover , the reduced local single particle states are completely featureless , @xmath257 which means that there is no single particle aharonov - bohm effect . the existence of classically correlated two - particle states giving rise to aharonov - bohm oscillations in the current cross correlations but not in the currents provides further motivation for a complete tomographic reconstruction of the reduced density matrix in order to provide an unambiguous experimental demonstration of entanglement . as pointed out at several places above , the reduced density matrix can be reconstructed by a suitable set of current and current correlations measurements with different settings of the beam splitters parameters , i.e. different @xmath258 . a detailed description of this scheme is given in @xcite . here we only emphasize that the necessary tools , controllable reflectionless electronic beams splitters and phase gates , are experimentally available , as demonstrated in e.g. @xcite another widely discussed @xcite approach to detect the entanglement in mesoscopic conductors is to use a bell inequality . violation of a chsh - bell inequality @xcite formulated in terms of currents and low - frequency current correlations demonstrates finite entanglement of @xmath213 . we point out that an optimal bell test , requiring control over all three components of @xmath259 and @xmath260 , demands the same number of measurement and level of experimental complexity as a tomographic reconstruction of @xmath213 . the chsh - bell inequality is @xmath261 where @xmath262 is the bell parameter for the projected / reduced state . the bell parameter is formally determined by the projected / reduced density matrix @xmath263 and different settings of the detector beam splitters , reaching its maximum value @xmath264 for an optimal setting of @xmath259 and @xmath265 . from @xmath160 and @xmath229 above , we can , using ref . @xcite , calculate the maximal bell parameters . for symmetric beam splitters , @xmath266 , we have the simple result @xmath267 where the singlet weights @xmath268 and @xmath269 are given in eq . ( [ singweights ] ) . this shows that the effects of decoherence and finite temperature enters separately in the bell parameter . moreover , as pointed out in refs . @xcite , at zero temperature a bell inequality can in principle be violated for arbitrary dephasing . we also point out that a detailed investigation of conditions for violation of a bell inequality in the presence of dephasing , in the solid state , was recently performed in ref . @xcite . the limiting value for violation @xmath270 for @xmath271 plotted in fig . it is clear that for the values @xmath192 and @xmath100 of the 2pi - experiment , while @xmath272 in principle can be violated , a detection of entanglement by violating @xmath273 is not possible . this demonstrates in a striking way the known fact @xcite that there are entangled states that do not give a violation of a bell inequality . in conclusion , we have investigate the effect of finite temperature on the entanglement production and detection in the fermionic two - particle interferometer , presenting an extended discussion of the results in ref . @xcite . a calculation of the entanglement of the two - particle state projected out from the emitted , finite temperature many body state shows that the state emitted in the two - particle interferometer in the experiment by neder et al @xcite is clearly entangled . by comparing the entanglement of the projected two - particle state with the entanglement of the reduced two - particle state , accessible via quantum state tomography based on current and current correlation measurements , we establish that the entanglement of the reduced state constitute a lower bound for the entanglement of the projected state . in the two - particle interferometer experiment the reduced state is however marginally entangled . moreover , a finite temperature bell inequality formulated in terms of currents and current correlators can not be violated in the experiment . this shows that an unambiguous demonstration of the entanglement via measurements of currents and current correlations requires a reduction of the dephasing and the temperature . the work was supported by the swedish vr , the israeli sf , the minerva foundation , the german israeli foundation ( gif ) and project cooperation ( dip ) , the us - israel binational sf , the swiss nsf and manep . beenakker , m. kindermann , c.m . marcus , and a. yacoby , in _ fundamental problems of mesoscopic physics _ , edited by i.v . lerner , b.l . altshuler , and y. gefen , nato science series ii . 154 ( kluwer , doordrect , 2004 ) .

the subject of superfluidity in ultracold trapped fermionic gases is an exciting field @xcite , not only for its implications in atomic physics , but also because there are important implications for condensed matter systems , including perhaps high temperature superconductors @xcite . there are two important aspects which are particularly notable about these systems , from the perspective of the present paper . they can be tuned in various ways which are not available to nature - made superconductors . thus , one can study the entire regime from bcs to bose - einstein condensation ( bec ) , simply by the application of a magnetic field in concert with a feshbach resonance . moreover , one can vary the concentrations of the two spin species arbitrarily @xcite ; in a fashion , this simulates the application of a magnetic zeeman field . this latter tuneability has important implications for other sub - disciplines in physics such as dense qcd and ( isospin asymmetric ) nuclear matter @xcite . equally important is the fact that there is a rich collection of experimental data from two different atomic physics groups @xcite on @xmath2li gases near unitarity , with which one can compare various theoretical results . the goal of this paper is to present an overarching theoretical framework for dealing with fermionic gases over the entire range of bcs to bec crossover and the entire range of temperature , as well as the entire range of population imbalance . the foundations of this theory lie with the initial observation of eagles @xcite and of leggett @xcite that the bcs - like wavefunction has a much greater generality than was originally recognized at the time of its proposal . it is capabable of describing both bcs and bec like systems , providing the pairing attraction is tuneable from arbitrarily weak ( bcs ) to arbitrarily strong ( bec ) and one self consistently solves for the fermionic chemical potential . this mean field - like ground state wave function is also readily generalized to include population balance . indeed , there are at least three well studied phases @xcite which have been proposed to accomodate a difference in the population of the two spin species . what we want to stress is that these same mean field theories have a natural extension to finite temperature @xcite . this extension will be a focus of the present paper . at strictly zero temperature , there is a rather extensive literature @xcite on these population - imbalanced superfluids and superconductors , including recently the effects of the crossover from bcs to bec . there have also been some studies at finite @xmath0 , which are at the mean field level and do not include the effect of the non - condensed pairs @xcite we consider here . the three most studied phases are the sarma "- like @xcite or breached pair states in which ( as in bcs theory ) the condensed pairs have zero net momentum , but polarization can , nevertheless , be introduced . at @xmath3 this state appears to be stable in the deep bec regime . additionally , two different phases @xcite have been proposed by larkin and ovchinnikov and by fulde and ferrell ( loff ) in which the condensate has a net momentum of a pre - determined @xmath4 or of @xmath5 . even more elaborate crystalline lattices of various @xmath6 have also been contemplated @xcite . it is believed that these states are more appropriate closer to the bcs side of resonance , although rather little has been determined about the two - plane wave " loff state near @xmath3 or in the presence of crossover effects . added to this complexity is the possiblity of heterogeneous or phase separated states @xcite . in this paper we will present the theoretical formalism for the sarma and one and two - plane wave loff states at zero and finite @xmath0 as one varies from bcs to bec . we note that because theories of population imbalanced superfluids are ( currently , without exception ) based on bcs - leggett - type @xcite ground states , it is important to determine their finite temperature implications within this broad class of ground states , as we do here . our premise is that the effects of finite @xmath0 , which necessarily must be accomodated in any comparison with experiment , must be compatible with the @xmath3 formalism . indeed , one of the most important effects of temperature is to stabilize the sarma - like phase . in this way one finds an intermediate temperature superfluid @xcite , one that exists at @xmath7 , but not at @xmath3 . our studies of the two - plane wave loff state present new results by extending the current literature away from the landau - ginzburg regime ( near @xmath1 ) . with this class of generalized mean field theories we will show that the effects of temperature enter in a fashion which is strongly reminiscent of bose gas condensation . except in the bcs limit , pairs form at higher temperatures , than the ( transition ) temperature @xmath1 at which they condense . thus we have to distinguish the excitation gap [ called @xmath8 from the order parameter [ called @xmath9 . alternatively , this means that there is an excitation gap ( or pseudogap @xmath10 ) for fermionic excitations even in the normal phase . it also implies that below @xmath1 there will be additional modes of exciting the condensate via pair excitations . these latter are non - condensed or incoherent pairs with finite center of mass momentum . we have found @xcite that in a trapped geometry they are particularly important for providing a mechanism of getting polarization into the gas . & & @xmath11 , @xmath12 + & & + & & + these effects of finite @xmath0 can be compared with an alternative class of theories in the literature based on work by nozieres and schmitt - rink ( nsr ) @xcite . this approach is known to lack self - consistency @xcite . we stress that the finite temperature nsr approach was not designed to be consistent with the standard ground state equations . this observation has also been made in ref . indeed , these latter authors have presented in considerable detail @xcite a more complete finite @xmath0 approach based on the @xmath1 calculations of ref . @xcite . a major concern about this class of theories remains to be addressed . because the gap and the number equations are not treated on an equivalent basis , it is possible that the superfluid density will not consistently vanish at @xmath1 . at this temperature there has to be a precise , but delicate cancellation of paramagnetic and diamagnetic current contributions , as found in the present theory @xcite . in a related fashion , pseudogap effects ( associated with the presence of non - condensed pairs ) appear , within an nsr - based approach , in the number equation but not in the gap equation . we begin at the more physical level by stressing the analogy between condensation in this composite boson or fermionic superfluid and condensation in a gas of ideal point bosons . our theory treats self - consistently two - particle and one - particle green s functions on an equal footing . because the physics is so simple and clear , we can fairly readily anticipate the form of the central equations of this bcs - bec generalization of bcs theory . it is important to stress , however , that these equations can be derived more rigorously from a truncated series of equations of motion for the appropriate green s functions @xcite . there are three principle equations which govern bose condensation : the vanishing of the bosonic chemical potential at all @xmath13 is the first . throughout this paper we will refer to this condition as the bec condition " . it is related to the usual thouless criterion , but the latter is generally associated only with the temperature @xmath1 . the second equation is the boson number equation . all bosons " must be accounted for as either condensed or non - condensed . the third equation is the number of non - condensed bosons " , which are created by thermal excitations . this is determined simply by inserting the known excitation spectrum of the excited pairs or bosons , into the bose distribution function . with this equation , and the first equation , one can then deduce the number of condensed bosons . these three central equations for bosons are indicated in table i , on the far right , for true point bosons , and in the second column for the composite bosons which appear in fermionic superfluids . for these composite bosons the quantity which provides a measure of the number " of bosons ( @xmath14 ) is given by @xmath15 ( up to a constant coefficient , @xmath16 ) . this is reasonably easy to see . in the fermionic regime , when the fermionic chemical potential is positive , @xmath15 represents the square of the excitation gap . this is the energy which must be supplied to break apart the pairs . thus , @xmath15 , in some sense then , reflects the number of pairs . how does one quantitatively establish the appropriate boson number " for the fermionic case ? this is determined via the self consistent gap equation for @xmath17 , which , in turn , is determined using the first condition : that the pair chemical potential is zero at and below @xmath1 . how does one compute the number of excited pairs ? once the gap equation is interpreted in terms of the appropriate non - condensed pair propagator , then one knows the related excitation spectrum @xmath18 of this propagator . the quantity @xmath16 which appears in the last equation of the table ( for the composite bosons ) gives the relation between the gap associated with non condensed pairs ( @xmath10 ) and the number of pairs ( @xmath19 ) . it can be readily calculated in this theory ; once one has the non - condensed pair propagator , @xmath16 appears as the inverse residue . ( deep in the bec regime , @xmath16 is relatively simple to compute , for here the boson number density approaches the asymptote @xmath20 , where @xmath21 is the fermion density ) . more precisely , the total number of bosons in the present case has to be determined self - consistently through the gap equation . it also involves the fermion number equation through the related fermionic chemical potential . in this last context , it should be stressed that there is one important aspect of the fermionic superfluids , which is not apparent in table i. for bcs - bec crossover , it is essential to derive the self consistent equation for the fermionic chemical potential ; in this problem the fermions are the fundamental statistical entity . this can be readily accomplished within the same framework used to arrive at the gap equation . the vanishing of the pair chemical potential is associated with a particular choice for the pair propagator involving dressed green s functions . these , in turn , determine the fermionic chemical potential through the fermion number equation . in the next two sections we turn to the gap and number equations , and show through a green s function formulation , how strongly these two equations are inter - connected . the rest of the paper is organized as follows . we conclude this section with a summary of the central equations associated with our @xmath0-matrix scheme . in sec . [ sec : ii ] , we present a mean - field theory for the sarma and one - plane - wave loff state for @xmath22 . due to the complexity of the two - plane - wave loff state , we dedicate an entire section ( sec.[sec : iii ] ) to its mean - field treatment . in sec . [ sec : iv ] we present a generalization of our @xmath0-matrix formalism to include pairing fluctuation effects . section [ sec : v ] recapitulates our simple physical picture and section [ sec : vi ] presents our conclusions . additional , more technical details are given in two appendices for the one and two plane wave loff states . to make contact with the general class of mean field theories ( including the sarma and loff states ) , we introduce a @xmath0-matrix approximation . this means that we consider the coupled equations between the particles ( with propagator @xmath23 ) and the pairs [ with propagator @xmath24 and drop all higher terms . this theory does not include direct boson - boson " interactions , although the pairs do interact indirectly via the fermions , in an averaged or mean field sense . throughout sections [ sec : ii ] and [ sec : iii ] of this paper we will be showing that the bec condition noted in the previous section will give the same gap equation we find using standard techniques , such as bogoliubov diagonalization applied to the linearized mean field hamiltonian . here , for all @xmath25 , the bec condition is interpreted as requiring that the pair chemical potential @xmath26 associated with the non - condensed pairs vanish . in sections [ sec : ii ] and [ sec : iii ] we will address only the first line of table i. the second two lines , or sets of equations will be discussed in section [ sec : iv ] . within a @xmath0-matrix scheme , the pair propagator is given by @xmath27 where @xmath28 is the spin _ symmetrized _ pair susceptibility , and @xmath29 is the pairing interaction strength . the function @xmath30 is , in many ways , the most fundamental quantity we introduce in this paper . it provides the basis for obtaining well known ( as well as new ) results of the zero temperature theory . moreover , it provides the basis for arriving at a finite temperature description , which appears in section [ sec : iv ] . the introduction of spin symmetrization is only important for the case of population imbalance . in earlier literature @xcite , this complexity did not arise . we will show that one obtains consistent answers between @xmath0-matrix based approaches and standard mean field theories , provided the components of the pair susceptibility in the presence of population imbalance , defined by @xmath31 \label{eq:2}\ ] ] is given by the product of one dressed and one bare green s function [ eq:3 ] @xmath32 where @xmath33 , and @xmath23 and @xmath34 are the full and bare green s functions respectively . we will discuss @xmath23 in more detail on a case by case basis . here @xmath35 , @xmath36 , @xmath37 is the kinetic energy of fermions , and @xmath38 is the fermionic chemical potential for spin @xmath39 . throughout this paper , we take @xmath40 , @xmath41 , and use the four - vector notation @xmath42 , @xmath43 , @xmath44 , etc , where @xmath45 and @xmath46 are the standard odd and even matsubara frequencies @xcite ( where @xmath21 and @xmath47 are intergers ) . for the mean field discussions in sections [ sec : ii ] and [ sec : iii ] of this paper we will not be considering general values of @xmath48 but only zero frequency limits with special values of @xmath49 associated with sarma ( @xmath50 ) and loff ( @xmath51 ) states however , when we include the contribution of non - condensed pairs ( or pseudogap effects ) the general values of @xmath48 become important . for the sarma phase we have the bec condition @xmath52 more generally , for loff - like states we have the bec condition at finite @xmath4 : @xmath53 we will discuss , in considerable detail , the nature of the mean field self energy which appears in the full green s function @xmath54 . not only does this determine the gap equation but it also leads to the number equations which can be written in terms of green s functions as @xmath55 . the number and gap equations then provide the underlying basis for the mean field approach . and we will see that the _ same _ propagator for non - condensed pairs [ @xmath24 enters into the beyond - mean - field corrections . it is important to stress that when we refer to `` mean field '' based approaches in sections [ sec : ii ] and [ sec : iii ] of this paper , we will not be distinguishing between the order parameter @xmath56 and the excitation gap @xmath57 . subsequently we will show that this distinction is actually an important one in all but the bcs limit . specifically , we note that the expressions we present in sections [ sec : ii ] and [ sec : iii ] within our green s function - based formulation , are more generally valid below @xmath1 , but for the excitation gap @xmath57 , not for the order parameter . we now want to study the sarma or breached pair state . on the fermionic side of resonance , the dispersion relation of one of the two quasi - particle bands ( @xmath58 and @xmath59 , defined below ) has two zero crossings , at momenta @xmath60 @xmath61 and @xmath62 @xmath63 . this is associated with gapless excitations . in a momentum space representation , pairing is confined to @xmath64 and @xmath65 . this pairing is `` breached '' by a normal component in the intermediate region @xmath66 . it is this normal component , then , which carries the bulk of the polarization . if we consider pairing between @xmath67 and @xmath68 states , the mean field hamitonian can be rewritten as @xmath69 where we have chosen @xmath70 to be real . using standard bogoliubov diagonalization techniques , we readily arrive at the mean field gap equation via the self - consistency condition @xmath71 which can be written as @xmath72\nonumber\\ & = & \frac{1}{u}+\sum_{\mathbf{k}}\frac{1 - 2\bar{f}(e_{k})}{2e_{k } } \ , . \label{eq:10}\end{aligned}\ ] ] here @xmath73 and @xmath74 , @xmath75 , @xmath76 and @xmath77 , where @xmath78 . in addition , we define the average @xmath79/2 $ ] , where @xmath80 is the fermi distribution function . the coupling constant @xmath81 can be replaced in favor of the dimensionless parameter , @xmath82 , via the relationship @xmath83 , where @xmath84 is the two - body @xmath85-wave scattering length , and @xmath86 is the noninteracting fermi wave vector for the same total number density in the absence of population imbalance . therefore the gap equation can be rewriten as @xmath87 \ , . \label{eq:11}\ ] ] this is a familiar gap equation which has appeared many times in the literature . the mean field number equations can be readily deduced @xmath88 \ , , \label{eq:12}\end{aligned}\ ] ] where @xmath89 and the coherence factors @xmath90 . equivalently , they can be rewritten as [ eq : neq ] @xmath91,\\ pn & = & \sum_\mathbf{k } [ f({e_{\mathbf{k}}}-h)-f({e_{\mathbf{k}}}+h ) ] \ , , \label{eq : neqb}\end{aligned}\ ] ] where @xmath92 is the total atomic density , @xmath93 is the number difference and @xmath94 is the polarization . we next show that the same results can be obtained from a @xmath0-matrix based approach . this discussion provides the link between the first line of table i and the previous subsection . the one particle green s function for particles with spin @xmath95 is @xmath96 the self - energy @xmath97 can be shown to be of the bcs - like form @xmath98 we will see later in section [ sec : iv ] how this form for the self energy very naturally arises ( below @xmath1 ) in a @xmath0-matrix approach . thus @xmath99 then , using the coherence factors we defined earlier , the green s functions become @xmath100 now we are in position to calculate the pair susceptibility at @xmath101 for the sarma phase based on eqs . ( [ eq:2 ] ) and ( [ eq:3 ] ) . @xmath102 substituting this expression into our bec condition eq . ( [ eq:5 ] ) , we obtain the same gap equation ( [ eq:10 ] ) , after carrying out the matsubara summation . in terms of green s functions , we readily arrive at the number equations : @xmath55 , which reduce to the number equations ( [ eq:12 ] ) we found earlier . if we now consider condensates in which momentum @xmath67 pairs with @xmath103 , for , as yet undermined @xmath4 , the mean field hamiltonian can be rewritten as @xmath104 upon bogoliubov diagonalization , the self - consistency condition @xmath105 readily leads to the mean - field gap equation @xmath106 with @xmath107 , @xmath108 , @xmath109 , @xmath110 . the regularized gap equation is @xmath111 \ , . \label{eq:22}\ ] ] finally , the mean field number equations for @xmath112 , are given by @xmath113 \ , . \label{eq:23}\end{aligned}\ ] ] here the coherence factors are @xmath114 and @xmath115 . there must be another equation which governs @xmath4 . this can be obtained by minimizing the thermodynamical potential or free energy with respect to @xmath4 . equivalently , we will derive @xmath4 from the @xmath0-matrix method described below . we now use the same @xmath0-matrix based approach ( as we did for the sarma case , and as indicated by table i ) to make contact with the one plane wave loff state . we take the self energy of the form @xmath116 so that @xmath117 then we have @xmath118 here @xmath119 and @xmath120 . note if @xmath121 , then @xmath122 and @xmath123 . the number equations are @xmath55 , which yields the same answer ( eq . [ eq:23 ] ) we found in the mean field approach . for @xmath25 , using the bec condition @xmath124 we thus arrive at the gap equation we found earlier in eq . ( [ eq:21 ] ) . we next investigate the pair susceptibility and its extremal value at @xmath125 . we define the contributions to the pair susceptibility in the presence of population imbalance as in eqs.([eq:1])-([eq:7 ] ) . for the one plane wave loff state , the bec condition is that @xmath126 diverges at a non - zero momentum @xmath127 . thus , at this momentum , the quantity @xmath128 should reach a _ maximum _ at @xmath127 . we determine @xmath4 by requiring @xmath129 where @xmath30 is explicitly shown in eq . ( [ eq : loff1chi ] ) . this yields @xmath130\frac{\xi_{kq}}{e_{kq}}\big ] \nonumber\\ & & { } + \left(\mathbf{k-}\frac{\mathbf{q}}{2}\right)[f(e_{1,\uparrow})-f(e_{1,\downarrow})]\bigg\}\ , . \label{eq:27}\end{aligned}\ ] ] it is important to stress , as we show in appendix [ app : loff1 ] , that this extremal condition on @xmath30 is equivalent to the condition that the net current is identically zero in this situation . @xmath131=0 \label{eq:28}\end{aligned}\ ] ] this key observation shows that the present way of computing @xmath4 directly from the pair susceptibility is consistent with the counterparts in the literature , which are based on the vanishing of the net current in equilibrium . at the mean field level we , then , have four unknowns @xmath57 , @xmath4 , @xmath132 and @xmath133 , and four equations : two number equations , the gap equation , and the condition on the vanishing of the first order derivative . the two plane wave loff state has not been studied in much detail away from @xmath1 and the associated tri - critical point @xcite . here we address this state at more general temperatures , starting with the very natural mean field hamiltonian in the same spirit as in the previous two cases . the hamiltonian is given by @xmath134 to solve for the value of the gap we generalize the bogoliubov diagonalization procedure . if one rewrites the effective hamiltonian in the basis set @xmath135 and @xmath136 , the hamiltonian can be written in two @xmath137 matrices . here we show the matrix for @xmath138 : @xmath139 we can equivalently write the @xmath140 component of the @xmath137 matrix in terms of the basis @xmath141 the mean field gap equation is derived from the averages obtained from a symmetric combination of the two basis sets as [ eq : loff2_gap ] @xmath142 which is equal to @xmath143 we choose this symmetrized form to write down the gap equation , which reflects the underlying symmetry of the hamiltonian . the resulting gap equation can be shown to be of the form @xmath144\frac{e_{1}+\mathcal{e}_{kq}}{(e_{1}-e_{2 } ) ( e_{1}-e_{3})}\right.\nonumber\\ & & { } + [ f(e_{2\uparrow})+f(e_{2\downarrow})]\frac{e_{2 } + \mathcal{e}_{kq}}{(e_{2}-e_{1})(e_{2}-e_{3})}\nonumber \\ & & { } + \left . [ f(e_{3\uparrow})+f(e_{3\downarrow})]\frac{e_{3}+\mathcal{e}_{kq } } { ( e_{3}-e_{1})(e_{3}-e_{2})}\right\}\ , . \label{eq:33}\end{aligned}\ ] ] here we define @xmath145 , @xmath146 , and @xmath147 as the solutions to the cubic equation which is related to the determinant of the 3x3 eigenvalue equation @xmath148x-\mathcal{e}_{kq}\big[\tilde{e } _ { kq}^{2}-\big(\frac{\mathbf{k\cdot q}}{m}\big)^{2}\big]=0 . \label{eq:34}\ ] ] where @xmath149 and @xmath150 . then @xmath151 and @xmath152 for @xmath153 . the mean field number equations are @xmath154 \label{eq:35}\end{aligned}\ ] ] where we define @xmath155 for the counterpart mean field theory , approached from a @xmath0-matrix scheme , we define the self energy as @xmath156 \nonumber \\ & = & \frac{1}{2}\big[\frac{\delta^{2}}{i\omega_{n}+\xi_{\mathbf{k - q}\bar{\sigma } } } + \frac{\delta^{2}}{i\omega_{n}+\xi_{\mathbf{k+q}\bar{\sigma } } } \big ] \label{eq:37}\end{aligned}\ ] ] here @xmath157 . the green s function is then given by @xmath158 again , following eqs . ( [ eq:1])-([eq:7 ] ) we can write out the form of the pair susceptibility . for the two plane wave loff state , this equation is presented as eq . ( [ eq : loff2symchi ] ) in appendix [ app : loff2 ] . this quantity , in turn , enters the gap equation , given by the bec condition , @xmath159 . we may write this gap equation in compact form as @xmath160 this will , in turn , reduce to the gap equation ( [ eq:33 ] ) we deduced directly from the @xmath137 matrix analysis . similarly , in our green s function formalism , we have @xmath55 , which reduces to the number equations we found in eq . ( [ eq:35 ] ) . next we determine the momentum @xmath49 that maximizes @xmath30 when @xmath161 , i.e. , we need to find a solution to @xmath162 . since the green s function is symmetric under @xmath163 , it can be shown that @xmath50 is a solution , which corresponds to the sarma phase . here , however , we are interested in a loff - like state @xmath164 . where both signs contribute in a symmetric fashion . thus we choose @xmath127 , and , thereby , arrive at the defining equation for the net momentum of the pairs . @xmath165}{(e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m } ) } \nonumber \\ & & { } + \frac{f(e_{1\uparrow})+f(e_{1\downarrow})}{(e_{1}-e_{2})(e_{1}-e_{3})}\,\frac{e_{1}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}+\frac{f(e_{2\uparrow})+f(e_{2\downarrow})}{(e_{2}-e_{1})(e_{2}-e_{3})}\,\frac{e_{2}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m } } \nonumber \\ & & { } + \frac{f(e_{3\uparrow})+f(e_{3\downarrow})}{(e_{3}-e_{1})(e_{3}-e_{1})}\,\frac{e_{3}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}\big\ } \ , . \label{eq:40}\end{aligned}\ ] ] this determines the magnitude @xmath4 ; for definiteness we take the direction of @xmath4 as along the @xmath166-axis . the case of the single plane wave loff state should be contrasted . there we showed that there was an intimate relation between the condition that there be no net equilibrium current and the extremal requirement on @xmath30 . for the two plane wave loff state the current can be shown to be identically zero . rather , the only condition one has to determine @xmath4 is the vanishing of the first derivative of the pair susceptibility . a similar condition was imposed in the original larkin - ovchinnikov ( lo ) paper @xcite . diagrams for sarma states or alternatively one plane wave loff states including ( a ) bare green s function @xmath167 . ( b ) vertices from pairing . the momenta are @xmath168 and @xmath169 , while for single plane - wave loff states the momenta are @xmath170 and @xmath171 . ( c ) green s function @xmath23 . ( d ) @xmath172 for sarma states and @xmath173 for single plane - wave loff states . here the thin and thick lines represent bare and full green s functions , respectively.,scaledwidth=45.0% ] diagrams representing the original ovchinnikov - larkin theory for ( a ) non - interacting green s function @xmath167 . ( b ) vertices from pairing with @xmath4 . ( c ) vertices from pairing with @xmath174 . ( d ) green s function @xmath23 . ( e ) @xmath175.,scaledwidth=45.0% ] we now want to compare the present approach for the two plane wave loff state with that proposed in the original lo paper @xcite . we will do this comparison within a diagrammatic framework . it is useful first to illustrate the diagrammatic scheme by referring to the simpler sarma and one plane wave loff states . these two states have a rather similar diagrammatic formulation . we write these diagrams in a consolidated form in fig . [ fig : bcsdia ] . the first line indicates the bare quantities , that is , the bare green s function @xmath34 in part ( a ) and the quantity @xmath176 in part ( b ) . this is written for the sarma state and readily generalized to the one plane wave loff phase . the second line indicates the ( diagonal ) dressed green s function , while the third line shows effectively the gorkov @xmath177 function . it should be noted for fig . [ fig : bcsdia]d that this latter anomalous " or off - diagonal green s function is explicitly seen to depend on a symmetrized sum of the product @xmath23 and @xmath34 . it is this combination which we have seen appear in the pair susceptibility @xmath30 . indeed , throughout this paper we have found that @xmath178 is essential for arriving at the standard mean field theoretic approach . we can , thus , conclude from the last line in the figure that the @xmath177 function is given by a spin symmetrized combination of @xmath179 . in fig . [ fig : lodia ] we show the two plane wave loff diagrams originally proposed by lo . in order to discuss their implications , we present the diagrams associated with the mean field matrix scheme of the previous section which are plotted in fig.[fig : qcdia ] . as in figure [ fig : bcsdia ] , one can see from the last line in part e of fig.[fig : qcdia ] that the off - diagonal green s function depends on a symmetrized sum of @xmath23 and @xmath34 . one can also see that there are similarities as well as differences in these latter two approaches . in the approach of section [ sec : iiia ] , just as in lo we restrict our calculations to diagrams that do not contain propagators with momentum beyond @xmath180 and @xmath181 . the differences are also apparent . in the lo paper , the ( diagonal component of the ) green s function diagrams were summed up to second order in @xmath57 . in the gap equation lo dropped diagrams containing propagators with higher momenta and truncated the series at third order in @xmath57 . diagrams associated with the 3x3 bogoliubov diagonalization theory for ( a ) non - interacting green s function @xmath167 . ( b ) vertices from pairing with @xmath4 . ( c ) vertices from pairing with @xmath174 . ( d ) green s function @xmath23 . ( e ) @xmath175.,scaledwidth=45.0% ] if one were to follow the original lo scheme , but sum the entire series , then we will arrive at a slightly modified gap equation . we would find instead the gap is determined by @xmath182 -\frac{1}{2}[g_{0\uparrow}(-k+q)g^{(1)}_{\downarrow}(k)+g_{0\downarrow}(-k+q ) g^{(1)}_{\uparrow}(k)]\big\ } \nonumber \\ & = & \sum_{\mathbf{k}}\big\{[f(e_{1\uparrow } ) + f(e_{1\downarrow})]\frac{e_{1}+\mathcal{e}_{kq}}{(e_{1}-e_{2 } ) ( e_{1}-e_{3})}+[f(e_{2\uparrow})+f(e_{2\downarrow})]\frac{e_{2 } + \mathcal{e}_{kq}}{(e_{2}-e_{1})(e_{2}-e_{3 } ) } \nonumber \\ & & { } + [ f(e_{3\uparrow})+f(e_{3\downarrow})]\frac{e_{3}+\mathcal{e}_{kq } } { ( e_{3}-e_{1})(e_{3}-e_{2})}\big\}-\sum_{\mathbf{k } } \frac{f(e^{(1)}_{1,\uparrow})+f(e^{(1)}_{1,\downarrow})-1}{2e^{(1)}_{kq } } \label{eq:41}\end{aligned}\ ] ] and thus the pair susceptibility derived from this method is @xmath183\nonumber\\ & & { } -\big[g_{0\uparrow}(p - k)g^{(1)}_{\downarrow}(k)+g_{0\downarrow}(p - k)g^{(1)}_{\uparrow}(k)\big]\big\}.\nonumber\\ \label{eq:42}\end{aligned}\ ] ] we could similarly use this modified pair susceptibility to arrive at the constraint on the value of @xmath4 . we discuss these contributions in appendix [ app : loff2 ] . for the above equation we define the green s function for the single plane - wave loff states as @xmath184 and @xmath185 . the energy spectrum of the single plane - wave loff states is @xmath186 , @xmath187 , @xmath188 , and @xmath189 , where @xmath190 and we define @xmath191 . in summary , to make contact with the original results of lo , one must subtract a second symmetrized term , which represents pairing with only one momentum . it would thus appear , that relative to lo , there is an overcounting in the summation of the two series from the diagrams shown in fig.[fig : qcdia ] , and associated with the 3x3 matrix or mean field representation . at this stage it is difficult to determine which of the two plane wave loff representations is the more appropriate . it will be essential in future to study them both numerically in the presence of bcs - bec crossover effects . for definiteness the equations which appear in this section apply to the sarma - like phase . we can readily generalize to include the two different loff states . this diagrammatic representation of our @xmath0-matrix scheme is shown in figs . [ fig:4 ] and [ fig:5 ] . the first of these indicates the propagator for non - condensed pairs which we refer to as @xmath192 , and the second of these the total self energy . one can see throughout the combination of one dressed and one bare green s function , as represented by the thick and thin lines . the self energy consists of two contributions from the non - condensed pairs or pseudogap ( @xmath193 ) and from the condensate ( @xmath194 ) . there are , analogously , two contributions in the full @xmath0-matrix @xmath195 where we write @xmath196 . similarly , we have for the fermion self energy @xmath197 we can see at once that @xmath198 the vanishing of the pair chemical potential implies that @xmath199 moreover , a vanishing chemical potential means that @xmath200 is strongly peaked around @xmath101 . thus , we may approximate @xcite eq . ( [ eq : sigma2 ] ) to yield @xmath201 where @xmath202 importantly , we are led to identify the quantity @xmath203 @xmath204 note that in the normal state ( where @xmath26 is non - zero ) eq . ( [ eq : sigma3 ] ) is no longer a good approximation . we now have a closed set of equations for addressing the ordered phase . we can similarly extend this approach to temperatures somewhat above @xmath1 , by self consistently including a non - zero pair chemical potential . this is a necessary step in addressing a trap as well @xcite . additionally , the propagator for non - condensed pairs can now be quantified , using the self consistently determined pair susceptibility . at small four - vector @xmath48 , we may expand the inverse of the @xmath0-matrix , after analytical continuation ( @xmath205 ) , to obtain @xmath206 where the imaginary part @xmath207 rapidly as @xmath208 below @xmath1 . because we are interested in the moderate and strong coupling cases , we drop the @xmath209 term in eq . ( [ omega_q : exp ] ) , and hence @xmath210 where we associate @xmath211 this establishes a quadratic dispersion and defines the effective pair mass , @xmath212 . -matrix diagrams . note that there is one dressed and full green s function . this represents the propagator for the non - condensed pairs.,width=288 ] -matrix scheme , indicating the condensed ( @xmath213 ) and non - condensed ( @xmath214 ) pairs . ] finally , one can rewrite eq . ( [ eq : delta_pg ] ) as @xmath215 where @xmath216 is the bose distribution function . analytical expressions for this mass are possible via a small @xmath217 expansion of @xmath28 . in this way we find @xmath218 the coefficients are @xmath219 and @xmath220 the theoretical framework in the previous section can now be implemented to address the physics at finite @xmath0 . we do this here only for the sarma state , thereby expanding on earlier papers @xcite . the counterpart results for the two loff phases are presented in the two appendices . we begin by determining the form of the pair dispersion . the pair susceptibility is given by @xmath221 } \nonumber \\ & & = \sum_{\mathbf{k}}\bigg\ { u_{k}^{2}\frac{\bar{f}(e_{k } ) + \bar{f}(\xi_{p - k})-1}{i\omega-\xi_{p - k}-e_{k } } + v_{k}^{2}\frac{\bar{f}(\xi_{p - k})- \bar{f}(e_{k})}{i\omega-\xi_{p - k}+e_{k}}\bigg\}.\nonumber\\\end{aligned}\ ] ] inserting this form into the @xmath0-matrix [ eq . ( [ eq:1 ] ) ] we readily find that the coefficent @xmath16 is @xmath222.\end{aligned}\ ] ] here @xmath223 is the total density . next we want to calculate @xmath224 ( defined above ) and its derivatives . @xmath225\bigg\ } \end{aligned}\ ] ] since @xmath226 , we have @xmath227 and @xmath228 . then @xmath229 \nonumber \\ & & { } -\bar{f}(e_{k\uparrow})\big[\frac{6\xi_{k}}{e_{k}}+ \frac{4(\xi_{k}+\mu)(e_{k}^{2}+\xi_{k}^{2})}{me_{k}\delta^{2}}\big ] \nonumber \\ & & { } + 4\bar{f}^{\prime}(\xi_{k})\frac{(\xi_{k}+\mu)}{m}-3\big(1-\frac{\xi_{k}}{e_{k}}\big ) \nonumber \\ & & { } + \frac{2e_{k}}{m\delta^{2}}\big(1-\frac{\xi_{k}}{e_{k}}\big)^{2 } ( \xi_{k}+\mu)\big\}.\end{aligned}\ ] ] here @xmath230 is the derivative of @xmath231 . with this dispersion , then one can compute the number of non - condensed pairs , @xmath232 . we then have @xmath233 where @xmath234 is the riemann zeta function . when @xmath235 , @xmath236 . physically , these non - condensed pairs , which appear at finite @xmath0 , have been shown to have important physical consequences . in the homogeneous situation , we have found that finite @xmath0 stabilizes the superfluid state leading to an intermediate temperature superfluid " @xcite . in addition , the presence of a finite excitation gap at @xmath1 [ @xmath237 makes the behavior of the superfluid transition temperature more complex than its mean field counterpart , @xmath238 . we frequently find a double - valued structure @xcite , with superfluidity existing only for temperature intermediate between the two @xmath1 s . in the trapped situation , we have found @xcite that these pairs enter explicitly as the mechanism for carrying polarization within the sarma phase . at low @xmath0 there is very little polarization carried by the superfluid core ; instead , the region of the trap where @xmath239 , but @xmath240 which can be called the mixed normal region" carries the bulk of the polarization , much as observed experimentally @xcite . we now return to the strong analogies between this bcs - based or leggett mean field theory @xcite and bose condensation of point bosons , as summarized in table i. we have three central equations . \1 . the pair chemical potential must vanish at and below @xmath1 @xmath241 importantly this condition leads to the mean field gap equations derived in sections [ sec : ii ] and [ sec : iii ] . these gap equations then provide a specific value for @xmath17 , according to the different phases being contemplated . there must be a conservation of the total number of ( composite ) bosons `` in the system . for this condition , our central equation is eq . ( [ eq : sum ] ) . here it is understood that the number of ' ' bosons " is effectively represented by the parameter @xmath242 . in the fermionic limit , this parameter reflects the number of bosons through the energy which is needed to create fermions , and thereby break the bosons apart . unlike the point boson case , here the total boson number " is temperature dependent and has to be self - consistently determined . as expected , in the deep bec regime , where the fermionic excitations are negligible , the pair density is given by @xmath243 . \3 . the number of non - condensed pairs is readily computed in terms of the pair dispersion , just as in conventional bec . for this condition our central equation is eq . ( [ eq:81 ] ) . then , just as in conventional bec , the number of condensed bosons ( proportional to @xmath244 ) is determined by the difference between @xmath15 and @xmath245 . this , in turn , determines the transition temperature @xmath1 as the lowest temperature(s ) in the normal state at which noncondensed pairs exhaust the total weight of @xmath246 so that @xmath247 . solving for the transition temperature " in the absence of pseudogap effects @xcite leads to the quantity @xmath238 . more precisely , @xmath238 should be thought of as the temperature at which the excitation gap @xmath17 vanishes . this provides a reasonable estimate , for the pairing onset temperature @xmath248 , ( when a stable superfluid phase exists ) . this is distinguished from the transition temperature . we note that @xmath248 represents a smooth crossover rather than an abrupt phase transition . it should be stressed that the dispersion relation for the non - condensed pairs is quadratic . this appears in the sarma and both one- and two - plane wave loff states . for the latter two states , it can be seen to be closely related to the fact that @xmath249 must reach a maximum at @xmath4 . while _ one will always find a linear dispersion in the collective mode spectrum @xcite _ , within the present class of bcs - bec crossover theories , the restriction to a @xmath0-matrix scheme means that there is no feedback from the collective modes onto the pair excitation spectrum . in effect , the @xmath0-matrix approximation does not incorporate pair - pair interactions at a level needed to arrive at this expected linear dispersion in the pair excitation spectrum . nevertheless , because essentially all theories which address population imbalance build on the simplest bcs - leggett mean field theory , there is good reason to first address this level of approximation when including finite temperature effects . it should be clear from the previous sections that the zero and finite temperature theories of population imbalanced superfluids can be consolidated into one general theory , based on the quantity which we call the pair susceptibility , @xmath30 . this has a very specific form in the class of mean field theories currently applied to address population imbalance . the quantity @xmath30 enters into the propagator for non - condensed pairs eq . ( [ t - matrix_pg ] ) which is just the @xmath0-matrix , @xmath250 . when non - condensed pairs are in equilibrium with a condensate , they must have zero chemical potential . this , in turn , yields the various gap equations for the sarma and the one- and two - plane - wave loff states , provided one take a special form for the pair susceptibility involving one dressed and one bare green s function . importantly , the same pair propagator characterizes the effective number of non - condensed pairs as seen in eq . ( [ eq : delta_pg ] ) . each of these population imbalanced superfluids has an associated @xmath251 pseudogap contribution , which is summarized for the sarma case in section [ sec : iii ] and for the two loff cases in appendices [ app : loff1 ] and [ app : loff2 ] . this pseudogap contribution serves to differentiate the order parameter from the gap parameter at all non - zero @xmath0 , and in all cases except strict bcs theory . at a more physical level , in the normal state there is an excitation gap for fermionic excitations associated with pre - formed " pairs . in the superfluid phase , there is a new form of condensate excitation , not found in bcs theory and associated with excited pair states . as the pairing attraction becomes stronger , it pays to excite pairs of atoms rather than create single fermion excitations which cost an energy gap . these pseudogap effects are an essential component of bcs - bec crossover theory , and they are necessary in order to smoothly involve from the fermionic statistics of bcs to the bosonic statistics of bec . interestingly , they are also widely observed in high temperature superconductors @xcite . thus far , we have applied the present theoretical formalism to the sarma state both in the homogeneous @xcite and trapped @xcite configurations . from the point of view of comparing with experiment , our trapped calculations have again underlined the importance of pseudogap effects . we find that the bulk of the polarization is carried in the pseudogap region of the trap : outside the condensate but in the region where both spin states ( and thus pairing ) are present . because we find that only modest polarizations are stable , it appears necessary in future to include loff - like condensates as well , although their stability must be demonstrated @xcite . this observation is also consistent with numerical calculations @xcite based on bogoliubov - de gennes theory . a key contribution of the present work is that it lays the groundwork for addressing the one and two plane wave loff phases , at general temperature . in summary , this paper has presented a theoretical formalism for the sarma and one- and two - plane wave loff states at zero and finite @xmath0 as one varies from bcs to bec , in the presence of an arbitrary population imbalance . our premise is that the effects of finite @xmath0 , which necessarily must be accomodated in any comparison with experiment , must be compatible with the @xmath3 formalism . the zero temperature formalism we use here reduces to the standard one in the literature @xcite for the sarma and one - plane wave loff states . however , our studies of the two - plane wave loff state present new results by extending the current literature away from the landau - ginzburg regime ( near @xmath1 ) . this work was supported by nsf phy-0555325 and nsf - mrsec grant no . dmr-0213745 . the following two appendices are dedicated to presenting additional details on the one- and two - plane - wave loff states . the results are presented in appendix form both for clarity and to avoid some of the more technical details in the main text . first , for the one - plane - wave loff state , we wish to explore the relationship between the zero current condition applied by fulde and ferrell @xcite and the extremal condition of @xmath252 . for the former , we determine momentum @xmath253 by requiring the total current to be zero . from @xmath254 $ ] we get @xmath255-\sum_{\mathbf{k}}\mathbf{(k - q)}[f(e_{1,\downarrow})u_{\mathbf{k}}^{2}+f(-e_{1,\uparrow})v_{\mathbf{k}}^{2}]\nonumber \\ & = & \sum_{\mathbf{k}}\big[\frac{\mathbf{q}}{2 } \big(1-\frac{\xi_{kq}}{e_{kq}}\big)+\frac{\mathbf{q}}{2}[f(e_{1,\uparrow})+f(e_{1,\downarrow})]\frac{\xi_{kq}}{e_{kq}}+(\mathbf{k-}\frac{\mathbf{q}}{2})[f(e_{1,\uparrow})-f(e_{1,\downarrow})]\big ] \nonumber \\ & = & 0\nonumber\end{aligned}\ ] ] we can take @xmath4 as the @xmath166 direction , and all the above equations need to be integrated over the angular variable @xmath256 . the pair susceptibility , even for the simpler loff state is reasonably complex so we present it here for completeness . @xmath257 $ ] is @xmath258.\ ] ] next , we characterize the pseudogap contributions which enter via @xmath259 . here @xmath16 and @xmath260 are determined as follows . @xmath261\big\}.\end{aligned}\ ] ] here @xmath21 is the total density . note that the linear derivative vanishes , as was established in section [ sec : iib ] . thus we turn next to the quadratic term . the coefficient of this second order term corresponds to the inverse pair mass , @xmath262 . this is given by @xmath263-[f(e_{1,\uparrow})+f(e_{1,\downarrow})]\big[\frac{3\xi_{kq}}{e_{kq } } + \frac{2[(\mathbf{k - q})^2+q^2](e_{kq}^{2}+\xi_{kq}^{2})}{me_{kq}\delta^{2}}\big ] \nonumber \\ & & { } + \frac{2[f(e_{1,\uparrow})-f(e_{1,\downarrow})]}{m\delta^2}[(\mathbf{k - q})^2-q^2]\xi_{kq } + \frac{8}{m}\bar{f}^{\prime}(\xi_{\mathbf{k - q}})(\mathbf{k - q})^2 - 3\big(1-\frac{\xi_{kq}}{e_{kq}}\big ) \nonumber \\ & & { } + \frac{2e_{kq}}{m\delta^{2}}\big(1-\frac{\xi_{kq}}{e_{kq}}\big)^{2}(\mathbf{k - q})^2\big\}\end{aligned}\ ] ] we begin with the results that relate to the simplest mean field - based scheme discussed in section [ sec : iiia ] . we then proceed to the slightly different scheme based on a resummation of the lo diagrams . @xmath264 \nonumber \\ & = & \frac{1}{2}t\sum_{n}\sum_{\mathbf{k},\sigma}\frac{1}{i\omega - i\omega_{n}- \xi_{\mathbf{p - k}\sigma}}\,\frac{(i\omega_{n}+ \xi_{\mathbf{k+q}\sigma})(i\omega_{n}+\xi_{\mathbf{k - q}\sigma } ) } { ( i\omega_{n}-e_{1\bar{\sigma}})(i\omega_{n}-e_{2\bar{\sigma } } ) ( i\omega_{n}-e_{3\bar{\sigma } } ) } \ ] ] the explicit forms of the coefficients @xmath16 and @xmath262 are @xmath265}{(e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})}+\frac{f(e_{1\uparrow})+f(e_{1\downarrow})}{(e_{1}-e_{2})(e_{1}-e_{3})}\,\frac{e_{1}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m } } \nonumber \\ & & { } + \frac{f(e_{2\uparrow})+f(e_{2\downarrow})}{(e_{2}-e_{1})(e_{2}-e_{3})}\,\frac{e_{2}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}+\frac{f(e_{3\uparrow})+f(e_{3\downarrow})}{(e_{3}-e_{1})(e_{3}-e_{1})}\,\frac{e_{3}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m } } \big\}\end{aligned}\ ] ] and @xmath266 \nonumber \\ & & { } -\frac{\frac{8}{m^{2}}(\mathbf{k\cdot q})(\xi_{\mathbf{k - q}}+\mu)[f^{\prime}(\xi_{\mathbf{k - q}\uparrow})+f^{\prime}(\xi_{\mathbf{k - q}\downarrow})]}{(e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m } ) } \nonumber \\ & & { } + \frac{f(e_{1\uparrow})+f(e_{1\downarrow})}{(e_{1}-e_{2})(e_{1}-e_{3})}\,\big[\frac{\frac{4}{m}(\xi_{\mathbf{k - q}}+\mu)(e_{1}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m})}{(e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})^{2}}-6\frac{e_{1}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}\big ] \nonumber \\ & & { } + \frac{f(e_{2\uparrow})+f(e_{2\downarrow})}{(e_{2}-e_{1})(e_{2}-e_{3})}\,\big[\frac{\frac{4}{m}(\xi_{\mathbf{k - q}}+\mu)(e_{2}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m})}{(e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})^{2}}-6\frac{e_{2}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}\big ] \nonumber \\ & & { } + \frac{f(e_{2\uparrow})+f(e_{2\downarrow})}{(e_{2}-e_{1})(e_{2}-e_{3})}\,\big[\frac{\frac{4}{m}(\xi_{\mathbf{k - q}}+\mu)(e_{3}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m})}{(e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})^{2}}-6\frac{e_{3}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}\big ] \big\}\end{aligned}\ ] ] for the lo - based pair susceptibility we have @xmath267-\frac{1}{2}\sum_{k}[g_{0\uparrow}(p - k ) g^{(1)}_{\downarrow}(k)+g_{0\downarrow}(p - k)g^{(1)}_{\uparrow}(k ) ] \nonumber \\ & = & t\sum_{n}\sum_{\mathbf{k},\sigma}\big\{\frac{1}{i\omega_l - i\omega_{n}- \xi_{\mathbf{p - k}\sigma}}\,\frac{(i\omega_{n}+ \xi_{\mathbf{k+q}\sigma})(i\omega_{n}+\xi_{\mathbf{k - q}\sigma } ) } { ( i\omega_{n}-e_{1\bar{\sigma}})(i\omega_{n}-e_{2\bar{\sigma } } ) ( i\omega_{n}-e_{3\bar{\sigma}})}\big\ } \nonumber \\ & & { } -\frac{1}{2}t\sum_{n}\sum_{\mathbf{k}}\big\{\frac{1}{i\omega_l - i\omega_{n } -\xi_{\mathbf{p - k}\uparrow}}\,\frac{i\omega_{n}+ \xi_{\mathbf{k - q}\uparrow}}{(i\omega_{n}-e^{(1)}_{2,\downarrow } ) ( i\omega_{n}+e^{(1)}_{2,\uparrow } ) } \nonumber \\ & & { } + \frac{1}{i\omega_l - i\omega_{n}-\xi_{\mathbf{p - k}\downarrow } } \,\frac{i\omega_{n}+\xi_{\mathbf{k - q}\downarrow } } { ( i\omega_{n}-e^{(1)}_{1,\uparrow})(i\omega_{n } + e^{(1)}_{1,\downarrow})}\big\}\end{aligned}\ ] ] the momentum @xmath4 is determined by minimizing @xmath128 , i.e. , @xmath268 at @xmath127 . this gives the * momentum equation * @xmath269 \nonumber \\ & = & \sum_{\mathbf{k}}\big(\frac{\mathbf{k - q}}{m}\big)\big\{\frac{\frac{2}{m}(\mathbf{k\cdot q})[f(\xi_{\mathbf{k - q}\uparrow})+f(\xi_{\mathbf{k - q}\downarrow})-2]}{(e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m } ) } \nonumber \\ & & { } + \frac{f(e_{1\uparrow})+f(e_{1\downarrow})}{(e_{1}-e_{2})(e_{1}-e_{3})}\,\frac{e_{1}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}+\frac{f(e_{2\uparrow})+f(e_{2\downarrow})}{(e_{2}-e_{1})(e_{2}-e_{3})}\,\frac{e_{2}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m } } \nonumber \\ & & { } + \frac{f(e_{3\uparrow})+f(e_{3\downarrow})}{(e_{3}-e_{1})(e_{3}-e_{1})}\,\frac{e_{3}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}\big\ } \nonumber \\ & & { } -\frac{1}{2\delta^{2}}\sum_{\mathbf{k}}\big\{\frac{\mathbf{q}}{2}\big(1-\frac{\xi^{(1)}_{\mathbf{kq}}}{e^{(1)}_{kq}}\big)+\frac{\mathbf{q}}{2}\frac{\xi^{(1)}_{\mathbf{kq}}}{e^{(1)}_{kq}}[f(e^{(1)}_{1,\uparrow})+f(e^{(1)}_{1,\downarrow})]+(\mathbf{k-}\frac{\mathbf{q}}{2})[f(e^{(1)}_{1,\uparrow})-f(e^{(1)}_{1,\downarrow})]\big\}.\end{aligned}\ ] ] the coefficients in the pseudogap dispersion are @xmath270}{(e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})}+\frac{f(e_{1\uparrow})+f(e_{1\downarrow})}{(e_{1}-e_{2})(e_{1}-e_{3})}\,\frac{e_{1}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}\right . \nonumber \\ & & { } + \frac{f(e_{2\uparrow})+f(e_{2\downarrow})}{(e_{2}-e_{1})(e_{2}-e_{3})}\,\frac{e_{2}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}+\frac{f(e_{3\uparrow})+f(e_{3\downarrow})}{(e_{3}-e_{1})(e_{3}-e_{1})}\,\frac{e_{3}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m } } \nonumber \\ & & { } \left . + \frac{1}{2\delta^{2}}\bigg[[1-f(\xi_{\mathbf{k - q}\uparrow})-f(\xi_{\mathbf{k - q}\downarrow } ) ] -\frac{\xi^{(1)}_{\mathbf{k\cdot q}}}{e^{(1)}_{kq}}[1-f(e^{(1)}_{1,\uparrow})-f(e^{(1)}_{1,\downarrow})]\bigg]\right\}\end{aligned}\ ] ] and @xmath271 \nonumber \\ & & { } -\frac{\frac{8}{m^{2}}(\mathbf{k\cdot q})(\xi_{\mathbf{k - q}}+\mu)[f^{\prime}(\xi_{\mathbf{k - q}\uparrow})+f^{\prime}(\xi_{\mathbf{k - q}\downarrow})]}{(e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})(e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m } ) } \nonumber \\ & & { } + \frac{f(e_{1\uparrow})+f(e_{1\downarrow})}{(e_{1}-e_{2})(e_{1}-e_{3})}\,\big[\frac{\frac{4}{m}(\xi_{\mathbf{k - q}}+\mu)(e_{1}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m})}{(e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})^{2}}-6\frac{e_{1}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{1}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}\big ] \nonumber \\ & & { } + \frac{f(e_{2\uparrow})+f(e_{2\downarrow})}{(e_{2}-e_{1})(e_{2}-e_{3})}\,\big[\frac{\frac{4}{m}(\xi_{\mathbf{k - q}}+\mu)(e_{2}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m})}{(e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})^{2}}-6\frac{e_{2}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{2}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}\big ] \nonumber \\ & & { } + \frac{f(e_{2\uparrow})+f(e_{2\downarrow})}{(e_{2}-e_{1})(e_{2}-e_{3})}\,\big[\frac{\frac{4}{m}(\xi_{\mathbf{k - q}}+\mu)(e_{3}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m})}{(e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m})^{2}}-6\frac{e_{3}+\mathcal{e}_{kq}+\frac{\mathbf{k\cdot q}}{m}}{e_{3}+\mathcal{e}_{kq}-\frac{\mathbf{k\cdot q}}{m}}\big ] \big\ } \nonumber \\ & & { } -\frac{1}{12mz\delta_{1}^{2}}\sum_{\mathbf{k}}\big\ { [ f(\xi_{\mathbf{k - q}\uparrow})+f(\xi_{\mathbf{k - q}\downarrow } ) ] \big[3+\frac{4\xi^{(1)}_{\mathbf{kq}}(\mathbf{k - q})^2}{m\delta_{1}^{2}}\big ] \nonumber \\ & & { } -[f(e^{(1)}_{1,\uparrow})+f(e^{(1)}_{1,\downarrow})]\big[\frac{3\xi^{(1)}_{\mathbf{kq}}}{e^{(1)}_{kq } } + \frac{2[(\mathbf{k - q})^2+q^2]({e^{(1)}_{kq}}^{2}+{\xi^{(1)}_{\mathbf{kq}}}^{2})}{me^{(1)}_{kq}\delta_{1}^{2}}\big]+\frac{2[f(e^{(1)}_{1,\uparrow})-f(e^{(1)}_{1,\downarrow})]}{m\delta_{1}^2}[(\mathbf{k - q})^2-q^2]\xi^{(1)}_{\mathbf{kq } } \nonumber \\ & & { } + \frac{2}{m}[f^{\prime}(\xi_{\mathbf{k - q}\uparrow})+f^{\prime}(\xi_{\mathbf{k - q}\downarrow})](\mathbf{k - q})^2 - 3\big(1-\frac{\xi^{(1)}_{\mathbf{kq}}}{e^{(1)}_{kq}}\big ) + \frac{2e^{(1)}_{kq}}{m\delta_{1}^{2}}\big(1-\frac{\xi^{(1)}_{\mathbf{kq}}}{e^{(1)}_{kq}}\big)^{2}(\mathbf{k - q})^2\big\}.\end{aligned}\ ] ] here we define @xmath272 and @xmath273 , where @xmath274 .

in this paper , we describe a new methodology for studying persistence of topological features across a family of spaces or point - cloud data sets . this theory of _ zigzag persistence _ generalises the successful and widely used theory of persistence and persistent homology @xcite . moreover , zigzag persistence can handle several important situations that are not currently addressed by standard persistence . the zigzag persistence framework is activated whenever one constructs a _ zigzag diagram _ of topological spaces or vector spaces : a sequence of spaces @xmath0 where each adjacent pair is connected by a map @xmath1 or @xmath2 . the novelty of our approach is that the direction of each linking map is arbitrary , in contrast to the usual theory of persistence where all maps point in the same direction . this paper has three principal objectives : * to describe several scenarios in applied topology where it is natural to consider zigzag diagrams ( section [ sec : intro ] ) . * to develop a mathematical theory of persistence for zigzag diagrams ( sections [ sec : zigzags ] and [ sec : zigzag2filt ] ) . * to develop algorithms for computing zigzag persistence ( section [ sec : algorithms ] ) . there is one subsidiary objective : * to introduce the _ diamond principle _ , a calculational tool analogous in power and effect to the mayer vietoris theorem in classical algebraic topology ( section [ sec : further ] ) . this is a theoretical paper rather than an experimental paper , and we devote most of our effort to covering the mathematical foundations adequately . the technical basis for zigzag persistence comes from the theory of graph representations , also known as quiver theory . we are deeply indebted to the practitioners of that theory ; what is new here is the emphasis on algorithmics and on applications to topology ( particularly sections [ sec : intro ] , [ sec : algorithms ] and [ sec : further ] ) . one of the principal challenges when attempting to apply algebraic topology to statistical data is the fact that traditional invariants such as the betti numbers or the fundamental group are extremely non - robust when it comes to discontinuous changes in the space under consideration . persistent homology @xcite is the single most powerful existing tool for addressing this problem . a typical workflow runs as follows @xcite . the input is a _ point cloud _ , that is , a finite subset of some euclidean space or more generally a finite metric space . after an initial filtering step ( to remove undesirable points or to focus on high - density regions of the data , say ) , a set of vertices is selected from the data , and a simplicial complex @xmath3 is built on that vertex set , according to some prearranged rule . in practice , the simplicial complex depends on a coarseness parameter @xmath4 , and what we have is a nested family @xmath5}$ ] , which typically ranges from a discrete set of vertices at @xmath6 to a complete simplex at @xmath7 . persistent homology takes the entire nested family @xmath8 and produces a _ barcode _ or _ persistence diagram _ as output . a barcode is a collection of half - open subintervals @xmath9 , which describes the homology of the family as it varies over @xmath4 . an interval @xmath10 represents a homological feature which is born at time @xmath11 and dies at time @xmath12 . this construction has several excellent properties : * there is no need to select a particular value of @xmath4 . * features can be evaluated by interval length . long intervals are expected to indicate essential features of the data , whereas short intervals are likely to be artefacts of noise . * there exists a fast algorithm to compute the barcode @xcite . * the barcode is a complete invariant of the homology of the family of complexes @xcite . * the barcode is provably stable with respect to changes in the input @xcite . in contrast , any individual homology group @xmath13 is highly unstable . the major limitation of persistence is that it depends crucially on the family @xmath8 being nested , in the sense that @xmath14 whenever @xmath15 . this applies to the current theoretical understanding as well as the algorithms . zigzag persistence addresses this limitation . if we discretise the variable @xmath4 to a finite set of values , the family of simplicial complexes can be thought of as a diagram of spaces @xmath16 where the arrows represent the inclusion maps . if we apply the @xmath17-dimensional homology functor @xmath18 with coefficients in a field @xmath19 , this becomes a diagram of vector spaces @xmath20 and linear maps , where @xmath21 . such a diagram is called a _ persistence module_. what makes persistence work is that there is a simple algebraic classification of persistence modules up to isomorphism ; each possible barcode corresponds to an isomorphism type . our goal is to achieve a similar classification for diagrams in which the arrows may point in either direction this is zigzag persistence , in a nutshell . we consider some problems which arise quite naturally in the computational topology of data . [ densityprofile ] some of the most interesting properties of a point cloud are contained in the estimates of the probability density from which the data are sampled . deep structure is sometimes revealed after thresholding according to a density estimate ( see @xcite for an example drawn from visual image analysis ) . however , the construction of a density estimation function @xmath22 invariably depends on choosing a smoothing parameter : for instance @xmath23 might be defined to be the number of data points within distance @xmath24 of @xmath25 ; here @xmath24 is the smoothing parameter . it happens that different choices of smoothing parameter may well reveal different structures in the data ; a particularly striking example of this occurs in @xcite . statisticians have invented useful criteria for determining what the ` appropriate ' value of such a parameter might be for a particular data set ; but another point of view would be to analyse all values of the parameter simultaneously , and to study how the topology changes as the parameter varies . the problem with doing this is that there is no natural relationship between , say , the @xmath26 densest points as measured using two different parameter values . this means that one can not build an increasing family of spaces using the change in parameters , and so one can not use persistence to analyze the evolution of the topology . on the other hand , there are natural zigzag sequences which can be used to study this problem . select a sequence of parameter values @xmath27 and a percentage @xmath28 , and let @xmath29 denote the densest @xmath30 of the point cloud when measured according to parameter value @xmath24 . we can then consider the union sequence @xmath31 or the intersection sequence @xmath32 as we see in section [ subsec : strongdiamond ] , there is essentially no difference between the zigzag persistent homology of the union and intersection sequences of a sequence of spaces . here that assertion needs to be filtered through the process of representing the data subsets @xmath29 as simplicial complexes . [ sampling ] suppose we are given a very large point cloud @xmath33 . if it is too large to process directly , we may take a sequence of small samples @xmath34 and estimate their topology individually , perhaps obtaining a persistence barcode for each one . how does this reflect the topology of the original sample @xmath33 ? on one hand , if most of the barcodes have similar appearance , then one might suppose that @xmath33 itself will have the same barcode . on the other hand , one needs to be able to distinguish between a single feature detected repeatedly , and multiple features detected randomly but one at a time . if we detect @xmath35 features in @xmath36 on average , are we detecting @xmath35 features of @xmath33 with detection probability 1 , or @xmath37 features with detection probability @xmath38 ? once again , there is a need to correlate features across different instances of the construction . the union sequence comes to the rescue : @xmath39 in this case , the intersection sequence is not useful at the level of samples , because two sparse samples are unlikely to intersect very much . the approach in this example is analogous to bootstrapping in statistics , where measurements on a large data set are estimated by making repeated measurements on a set of samples . [ landmarking ] in computational topology , there exist several techniques for modelling a point cloud data set @xmath33 by a simplicial complex @xmath3 : the cech complex , the vietoris rips complex , the alpha complex @xcite , the witness complex @xcite , and so on . the witness complex @xmath40 , in particular , depends on the choice of a small subset of ` landmark ' points @xmath41 which will serve as the vertex set of @xmath3 . roughly speaking , a simplex @xmath42 with vertices in @xmath43 is included in @xmath44 if there is some @xmath45 which witnesses it , by being close to all the vertices . how does the witness complex @xmath40 depend on the choice of landmark set ? there is no direct way to compare @xmath40 with @xmath46 for two different choices of landmark sets @xmath47 . however , it turns out that one can define a witness _ bicomplex _ @xmath48 which maps onto each witness complex . the cells are cartesian products @xmath49 , where @xmath50 have vertices in @xmath47 respectively . a cell @xmath49 is included provided that there exists @xmath45 which simultaneously witnesses @xmath42 for @xmath44 and @xmath51 for @xmath52 . given a sequence @xmath53 of landmark subsets , one can then construct the biwitness zigzag : @xmath54 long intervals in the zigzag barcode will then indicate features that persist across the corresponding range of choices of landmark set . the fundamental requirement is then for a way of assessing , in a zigzag diagram of vector spaces , the degree to which consistent families of elements exist . the point of this paper is that there is such methodology . we will interpret the isomorphism classes of zig - zag diagrams as a special case of the classification problem for quiver representations ( see @xcite for background on this theory ) . there turns out to be a theorem of gabriel @xcite which classifies arbitrary diagrams based on dynkin diagrams , and which shows in particular that the set of isomorphism classes of zigzag diagrams of a given length is parametrised by barcodes just as persistence modules are . long intervals in the classification define large families of consistent elements , hence indicate the presence of features stable across samples , landmark sets , or parameter values for a density estimator . in section [ sec : zigzags ] we describe the theory of decompositions of zigzag modules . these decompositions produce zigzag persistence barcodes analogous to the barcodes of persistent homology . the foundational theorem of gabriel is stated without proof . in section [ sec : zigzag2filt ] we develop the machinery of right - filtrations , which turn out to be the right tool for accessing the decomposition structure of a zigzag module . this is an important section for the reader who wishes to make serious use of zigzag persistence . in section [ sec : algorithms ] , we present a general - purpose algorithmic framework for calculating zigzag persistence , and we show how this operates in a practical class of examples . the algorithm is based on a proof of gabriel s theorem for zigzag modules , included for completeness . section [ sec : further ] is devoted to a localisation principle which gives another approach to zigzag barcode calculations . we apply this to prove the diamond principle . we use this in turn to compare the zigzag barcodes for two natural zigzag diagrams obtained from a sequence of simplicial complexes . we work over a field @xmath19 which remains fixed throughout this paper . there is no significance to the choice of @xmath19 . all vector spaces are finite - dimensional . let @xmath55 denote a sequence of vector spaces and linear maps , of length @xmath35 : @xmath56 each @xmath57 represents either a forward map @xmath58 or a backward map @xmath59 . the object @xmath55 is called a * zigzag diagram * of vector spaces , or simply a * zigzag module * , over @xmath19 . the sequence of symbols @xmath60 or @xmath61 is the * type * of @xmath55 . for instance , a diagram of type @xmath62 looks like this : @xmath63 the length of a type @xmath51 is the length of any diagram of type @xmath51 . for example , we say that @xmath64 has length 4 . we will usually be considering zigzag modules of a fixed type @xmath51 of length @xmath35 . such diagrams are called * @xmath51-modules * , and the class of @xmath51-modules is denoted @xmath65 . * persistence modules * ( see @xcite ) are zigzag modules where all the maps have the forward orientation ; in other words , where @xmath66 . as explained in @xcite , persistence modules can be viewed as graded modules over the polynomial ring @xmath67 $ ] . this observation simplifies the analysis of persistence modules quite considerably . more generally , one can consider * graph representations * of arbitrary oriented graphs . zigzag modules constitute the special case where the graph is @xmath68 ( a path with @xmath35 vertices and @xmath69 edges ) and the orientation is specified by the type @xmath51 . in 1972 , gabriel showed that the dynkin coxeter graphs @xmath68 , @xmath70 , @xmath71 , @xmath72 , @xmath73 ( arbitrarily oriented ) have an especially well - behaved representation theory @xcite . the theory of * quivers * was launched from this starting block ; see @xcite for a beautiful and transparent introduction . zigzag persistence is enabled by the good behaviour of @xmath68 graph representations . @xmath65 has the structure of an abelian category . given two @xmath51-modules @xmath74 , a morphism @xmath75 is defined to be a collection of linear maps @xmath76 which satisfy the commutation relations @xmath77 or @xmath78 for each @xmath79 . ( here the forward and backward maps for @xmath80 are written @xmath81 respectively . ) morphisms can be composed in the obvious way , and have kernels , images , and cokernels : for instance @xmath82 is the @xmath51-module with spaces @xmath83 and maps @xmath84 and @xmath85 defined by restriction . the set of morphisms @xmath86 is naturally a vector space over @xmath19 , and the endomorphism ring @xmath87 is a non - commutative @xmath19-algebra . we can view @xmath88 as the subalgebra of @xmath89 defined by the commutation relations . we wish to understand zigzag modules by decomposing them into simpler parts . accordingly , a * submodule * @xmath80 of a @xmath51-module @xmath55 is defined by subspaces @xmath90 such that @xmath91 or @xmath92 for all @xmath79 . these conditions guarantee that @xmath80 is itself a @xmath51-module , with maps given by the restrictions @xmath93 or @xmath94 . we write @xmath95 . a submodule @xmath80 is called a * summand * of @xmath55 if there exists a submodule @xmath96 which is complementary to @xmath80 , in the sense that @xmath97 for all @xmath79 . in that case , we say that @xmath55 is the * direct sum * of @xmath98 and write @xmath99 . as a rule , most submodules are not summands . @xmath100 has the submodule @xmath101 . however , @xmath80 is not a summand because the only possible complement is @xmath102 , and that is not a submodule of @xmath55 . the direct sum can also be defined as an ` external ' operation : given @xmath51-modules @xmath74 their direct sum @xmath103 is defined to be the @xmath51-module with spaces @xmath104 and maps @xmath105 or @xmath106 . ( here the forward and backward maps for @xmath80 are written @xmath81 respectively . ) a @xmath51-module @xmath55 is * decomposable * if it can be written as a direct sum of nonzero submodules , and * indecomposable * otherwise . any @xmath51-module @xmath55 has a * remak decomposition * ; in other words we can write @xmath107 , where the summands @xmath108 are indecomposable . the existence of such a decomposition is proved by induction on the total dimension @xmath109 : if @xmath55 is decomposable , say @xmath99 , then we may assume inductively that @xmath98 have remak decompositions , and therefore so does @xmath55 . ( base case : if @xmath55 is indecomposable , then it has a remak decomposition with one term . ) remak decompositions themselves are not unique . however , the krull schmidt principle from commutative algebra tells us that the summands in a remak decomposition are unique up to reordering : ( krull remak schmidt . ) [ prop : krull ] suppose a @xmath51-module @xmath55 has remak decompositions @xmath110 then @xmath111 and there is some permutation @xmath42 of @xmath112 such that @xmath113 for all @xmath114 . the proof of theorem 7.5 of lang @xcite , which is stated for modules in the ordinary sense , can be applied verbatim to our present context ; all the required algebraic operations can be carried out within @xmath88 . since our @xmath51-modules have finite total dimension , the ascending and descending chain conditions ( acc and dcc ) are automatic . for further context , we refer the reader to an elegant article by atiyah @xcite ; the krull schmidt principle applies in any exact abelian category to objects which satisfy acc and dcc , or a weaker ` bi - chain condition ' defined by atiyah . our category , @xmath65 , is included by this formulation . thus we can use the multiset @xmath115 as an isomorphism invariant of @xmath55 . for this to be useful , we need to identify the set of indecomposable @xmath51-modules . we now describe a natural collection of indecomposables . for each subinterval @xmath116 $ ] of the integer sequence @xmath117 there is an associated @xmath51-module . let @xmath51 be a type of length @xmath35 and consider integers @xmath118 . the * interval * @xmath51-module with birth time @xmath119 and death time @xmath120 is written @xmath121 and defined with spaces @xmath122 and with identity maps between adjacent copies of @xmath19 , and zero maps otherwise . when @xmath51 is implicit , we will usually suppress it and simply write @xmath123 . if @xmath62 then @xmath124 is the zigzag module @xmath125 [ prop : intind ] interval @xmath51-modules are indecomposable . suppose @xmath126 and consider two adjacent terms @xmath19 connected by an identity map . since @xmath74 are submodules , the dimensions of @xmath55 and @xmath80 can not decrease in the direction of the map ; nor , since they are complements , can they increase . thus @xmath127 and @xmath128 are constant over @xmath129 , and in particular one of @xmath130 must be zero . here is the foundation stone for the theory of zigzag persistence . [ thm : interval ] the indecomposable @xmath51-modules are precisely the intervals @xmath123 , where @xmath131 . equivalently , every @xmath51-module can be written as a direct sum of intervals . this is the simplest special case of gabriel s theorem , for the graphs @xmath68 . the original reference ( in german ) is @xcite . see @xcite for an accessible overview . thus , any @xmath51-module can be described completely up to isomorphism as an unordered list of intervals @xmath116 $ ] , which correspond to its indecomposable summands . this is in exact accordance with the special case of ordinary persistence , where the result is comparatively easy to prove : it is simply the classification of finitely - generated graded modules over the polynomial ring @xmath67 $ ] ( see @xcite ) . the philosophical point is that the decomposition theory of graph representations is somewhat independent of the orientation of the graph edges ( see kac @xcite ) . even in our case this is surprising , because there is no obvious congruence between persistence modules and zigzag modules of an arbitrary type @xmath51 . however , if we accept this principle , then the generalisation from ordinary persistence to zigzag persistence is not surprising : interval decomposition for persistence modules implies interval decomposition for zigzag modules . we will devote much of this paper to constructing a stand - alone proof of theorem [ thm : interval ] . this provides technical support towards our two main goals : to provide algorithms for computing the interval summands of a given @xmath51-module ; and to make rigorous statements about the output of those algorithms . we now define zigzag persistence and develop some of its elementary properties . let @xmath55 be a zigzag module ( of arbitrary type ) . the * zigzag persistence * of @xmath55 is defined to be the multiset @xmath132 \subseteq \{1 , \dots , n\ } \mid j = 1 , \dots , n \right\}\ ] ] of integer intervals derived from a decomposition @xmath133 . the krull schmidt principle asserts that this definition is independent of the decomposition . graphically , @xmath134 can be represented as a set of lines measured against a single axis with labels @xmath117 ( the * barcode * ) , or as a multiset of points in @xmath135 lying on or above the diagonal in the positive quadrant ( the * persistence diagram * ) . see figure [ fig : barpd ] for an example presented in each style . , [ 1,3 ] , [ 3,3 ] , [ 3,4 ] , [ 3,4 ] \}$ ] of a zigzag module of length 4.,title="fig : " ] , [ 1,3 ] , [ 3,3 ] , [ 3,4 ] , [ 3,4 ] \}$ ] of a zigzag module of length 4.,title="fig : " ] in the special case of persistence modules , this agrees with the standard treatment ( see @xcite ) except in the following particular : the closed integer intervals @xmath136 \subseteq \{1 , \dots , n\}$ ] are replaced by half - open real intervals @xmath137 in the standard treatment . this is particularly natural when the indexing parameter is continuous : an interval @xmath138 indicates a feature born at time @xmath119 that survives right up to , but vanishes at , time @xmath120 . our convention is motivated by the desire to maintain symmetry between the forward and backward directions . we advise the reader to take particular care in handling the different conventions . the transition from a zigzag module to its interval decomposition presents certain hazards which are not present in the case of persistence modules . we now draw attention to these hazards . let @xmath55 be a zigzag module and let @xmath139 $ ] denote the restriction of @xmath55 to the index set @xmath140 . a * feature * of @xmath55 over the time interval @xmath141 $ ] is a summand of @xmath139 $ ] isomorphic to @xmath142 . with persistence modules , there are several equivalent ways to recognise the existence of a feature . here is a sample result . [ prop : intuitions ] let @xmath55 be a persistence module of length @xmath35 , and let @xmath143 . the following are equivalent : 1 . the composite map @xmath144 is nonzero . 2 . there exist nonzero elements @xmath145 for @xmath140 , such that @xmath146 for @xmath147 . there exists a submodule of @xmath139 $ ] isomorphic to @xmath142 . 4 . there exists a summand of @xmath139 $ ] isomorphic to @xmath142 , i.e. a feature over @xmath141 $ ] . it is easy to verify that ( 1 ) , ( 2 ) , ( 3 ) are equivalent . for ( 1 ) @xmath148 ( 2 ) , begin by choosing @xmath149 that maps to a nonzero element in @xmath150 , and let @xmath151 be the image of @xmath152 in @xmath153 . for ( 2 ) @xmath148 ( 3 ) , define @xmath154 by @xmath155 . for ( 3 ) @xmath148 ( 1 ) , note that the restriction @xmath156 is nonzero . clearly ( 4 ) @xmath148 ( 3 ) . we now show that ( 1 ) @xmath148 ( 4 ) . consider an interval decomposition @xmath139 = { { { \mathbb{i}}}}(b_1 , d_1 ) \oplus \dots \oplus { { { \mathbb{i}}}}(b_n , d_n)$ ] . on each summand , the map @xmath157 is zero unless @xmath158 and @xmath159 . thus at least one of the summands is isomorphic to @xmath142 . the intuitions supported by proposition [ prop : intuitions ] break down in the general case . let @xmath55 be a zigzag module of arbitrary type . statement ( 1 ) has no clear interpretation at this stage ( something can be said in terms of the right - filtration functor of section [ sec : zigzag2filt ] ) . consider the following statements : 1 . there exist nonzero elements @xmath145 for @xmath140 , such that @xmath146 or @xmath160 ( whichever is applicable ) for @xmath147 . 2 . there exists a submodule of @xmath139 $ ] isomorphic to @xmath142 . 3 . there exists a summand of @xmath139 $ ] isomorphic to @xmath142 , i.e. a feature over @xmath141 $ ] . it is easy to verify that @xmath161 and that ( 4 ) implies @xmath162 . however , the next two examples demonstrate that @xmath162 do not in general imply ( 4 ) . [ ex : caution1 ] let @xmath163 and consider the @xmath51-module @xmath55 defined as follows : @xmath164 the interval decomposition is @xmath165 , where the summands are @xmath166 and @xmath167 respectively . if this example appeared in a statistical topology setting , the feature corresponding to the generator of the @xmath19 at @xmath168 would be regarded as unrelated to the feature corresponding to the generator of the @xmath19 at @xmath169 . on the other hand , @xmath55 does have a submodule ( in fact , many submodules ) isomorphic to @xmath170 . indeed , let @xmath171 denote the diagonal subspace of @xmath172 . then @xmath173 is a submodule @xmath95 isomorphic to @xmath170 . the quotient @xmath51-module @xmath174 is isomorphic to @xmath175 but @xmath80 has no complementary @xmath51-module in @xmath55 . indeed , that would contradict the krull schmidt theorem . more concretely , any complement of @xmath80 must be isomorphic to @xmath176 , but that would require a 1-dimensional subspace of @xmath177 . [ ex : caution2 ] we can extend the previous example to arbitrary length . consider the type @xmath178 , of length @xmath179 . let @xmath55 be the @xmath51-module @xmath180 where @xmath181 , and @xmath182 . then @xmath55 is isomorphic to a sum of short intervals @xmath183 but it has a submodule @xmath184 isomorphic to the long interval @xmath185 . in zigzag persistence it is necessary to respect the distinction between submodules and summands . features are defined in terms of summands ; never submodules . we have defined features in terms of a chosen subinterval @xmath141 $ ] . features behave as expected when zooming to a larger or smaller window of observation . the following proposition illustrates what we mean . [ prop : restriction1 ] let @xmath55 be a zigzag module of length @xmath35 and let @xmath143 . the following statements are equivalent . 1 . there exists a summand of @xmath139 $ ] isomorphic to @xmath142 , i.e. a feature over @xmath141 $ ] . 2 . there exists a summand of @xmath55 isomorphic to @xmath186 , for some @xmath187 \supseteq [ p , q]$ ] . indeed , there is a bijection between intervals @xmath141 $ ] in @xmath188)$ ] and intervals @xmath187 \supseteq [ p , q]$ ] in @xmath134 . consider an interval decomposition @xmath189 . by restriction , this induces an interval decomposition of @xmath139 $ ] into intervals @xmath190 $ ] . this induces the claimed bijection , because @xmath136 $ ] restricts to @xmath141 $ ] if and only if @xmath136 \supseteq [ p , q]$ ] . operating invisibly in this proof is the krull schmidt principle , which allows us to select the interval decompositions most convenient to us when calculating @xmath134 and @xmath188)$ ] . sometimes it is useful to reduce the resolution of @xmath191 . let @xmath192 be any subset . we define the * restriction * of @xmath134 to @xmath193 to be the multiset @xmath194 for instance , proposition [ prop : restriction1 ] amounts to the observation that @xmath188 ) = { \operatorname{pers}}({{{\mathbb{v}}}})|_{[p , q]}$ ] . our strategy is to understand ( and construct ) decompositions of a @xmath51-module @xmath55 by an iterative process , moving from left to right and retaining the necessary information at each stage . the bulk of this information is encoded as a filtration on the rightmost vector space @xmath195 . [ def : rfilt ] the * right - filtration * @xmath196 of a @xmath51-module @xmath55 of length @xmath35 takes the form @xmath197 where the @xmath198 are subspaces of @xmath195 satisfying the inclusion relations @xmath199 @xmath196 is defined recursively as follows . base case : * if @xmath55 has length 1 , then @xmath200 . recursive step . suppose we have already defined @xmath201 : * if @xmath202 is @xmath203 , then @xmath204 . * if @xmath202 is @xmath205 , then @xmath206 . to verify that @xmath207 in the two inductive cases is a filtration of the specified form , note that @xmath208 implies that @xmath209 in the first case , and @xmath210 in the second case . moreover @xmath211 , and @xmath212 . here are the right - filtrations for the two length-2 types : @xmath213 here are the right - filtrations for the four length-3 types : @xmath214 see figure [ fig : rfilt3 ] for a schematic representation . , @xmath215 , @xmath216 , @xmath217 of length 3.,title="fig : " ] , @xmath215 , @xmath216 , @xmath217 of length 3.,title="fig : " ] , @xmath215 , @xmath216 , @xmath217 of length 3.,title="fig : " ] , @xmath215 , @xmath216 , @xmath217 of length 3.,title="fig : " ] in the examples above , it is not difficult to see that @xmath196 comprises all the subspaces of @xmath195 that are naturally definable in terms of the maps @xmath218 . each of the @xmath35 subquotients @xmath219 carries information dating back to some earliest @xmath220 in the sequence of vector spaces . [ ex : birthtimef ] the module @xmath221 has right - filtration @xmath222 . the first subquotient @xmath223 corresponds to vectors born at time 1 which survive to time 2 . the second subquotient @xmath224 corresponds to vectors which appear only at time 2 . [ ex : birthtimeg ] the module @xmath225 has right - filtration @xmath226 . the first subquotient @xmath227 corresponds to vectors at time 2 which are destroyed when mapping back to time 1 . the second subquotient @xmath228 is isomorphic to @xmath229 and records those vectors which survive from time 2 back to time 1 . [ def : birthtime ] the * birth - time index * @xmath230 is a vector of integers @xmath231 which indicate the birth - times associated with the subquotients @xmath232 of the right - filtration of a @xmath51-module . this is defined recursively as follows . base case : * if @xmath51 is empty ( i.e. @xmath55 has length 1 ) then @xmath233 . recursive step . suppose we have already defined @xmath230 : * if @xmath234 is @xmath235 then @xmath236 . * if @xmath234 is @xmath237 then @xmath238 . at length 2 we have @xmath239 whereas @xmath240 . this is consonant with the discussion in examples @xmath241 and @xmath242 . here are the birth - time indices for the types of length 3 . @xmath243 in summary , the information in a @xmath51-module @xmath55 which survives to time @xmath35 is encoded as a filtration @xmath244 on @xmath195 . the ` age ' of the information at each level of the filtration ( i.e. at each subquotient ) is recorded in the birth - time index @xmath245 . for a simplified but precise version of this last claim , we now calculate the right - filtrations of interval @xmath51-modules . in the filtration specified in the following lemma , @xmath246 is the only non - zero subquotient , corresponding to the birth time @xmath231 . [ lem : intervalformula ] let @xmath51 be a type of length @xmath35 , with @xmath230 . for each @xmath247 , we have an isomorphism @xmath248 where @xmath249 is the filtration on @xmath19 defined by @xmath250 we refer to the @xmath251 also as _ intervals _ ( but now in the category of filtered vector spaces ) . this is a straightforward calculation by induction on @xmath51 . for the base case , @xmath51 is empty and @xmath233 . then @xmath252 as claimed . now suppose the result is known for @xmath51 , with @xmath253 . suppose @xmath254 or @xmath237 . in both cases , write @xmath255 . case @xmath60 : suppose that @xmath256 ; then @xmath257 and therefore @xmath258 writing @xmath259 , it follows that @xmath260 for @xmath261 , we have @xmath262 , and indeed @xmath263 case @xmath61 : suppose that @xmath264 ; then @xmath265 and therefore @xmath266 writing @xmath267 , it follows that @xmath268 for @xmath269 , we have @xmath270 and then @xmath271 as required . thus the right - filtration ( with the help of the birth - time index ) distinguishes the different intervals @xmath272 . it gives no information about intervals @xmath123 when @xmath273 , since in those cases @xmath274 . consider @xmath275 , so @xmath276 and in general @xmath277 in particular , @xmath278 which is in accordance with lemma [ lem : intervalformula ] . we now consider filtered vector spaces in their own right , independently of the connection to zigzag - modules , and develop the theory of remak decompositions . we will see later that this is the right tool for understanding remak decompositions of zigzag modules . a filtered vector space of depth @xmath35 is a sequence @xmath279 of vector spaces , where @xmath280 . the class of such objects is denoted @xmath281 . the right - filtration @xmath196 of any zigzag module @xmath55 of length @xmath35 belongs to this class , as do the intervals @xmath282 defined in lemma [ lem : intervalformula ] . indeed , if @xmath283 satisfies @xmath284 , then @xmath285 is isomorphic to some @xmath282 . @xmath281 can be given the structure of a category in a natural way , but it is not quite an abelian category since morphisms do not generally have cokernels . a filtered vector space @xmath286 is a * subspace * of @xmath285 if @xmath287 for all @xmath79 . it is appropriate to consider a stronger notion of subspace when dealing with direct - sum decompositions : @xmath288 is an * induced subspace * of @xmath285 if there exists a vector subspace @xmath289 such that @xmath290 for all @xmath79 . in that event , we write @xmath291 . note that @xmath292 . we say that @xmath285 is the * direct sum * of two subspaces , and write @xmath293 , if @xmath294 for all @xmath79 . we claim that @xmath295 must be induced subspaces . note that @xmath296 . for each @xmath79 , then , @xmath297 is a subspace of @xmath198 which contains @xmath298 and meets @xmath299 only at 0 . it follows that @xmath300 for all @xmath79 . thus @xmath301 , and symmetrically @xmath302 . the general form of a direct - sum decomposition is therefore @xmath303 . what are the requirements on @xmath304 to make this a valid decomposition ? the direct sum condition implies that @xmath305 as a vector space . moreover , given a vector space decomposition @xmath305 , the further condition @xmath306 is necessary and sufficient to guarantee @xmath303 . if @xmath293 , the two subspaces @xmath295 are said to be complementary summands . the following fact radically simplifies the decomposition theory of filtered vector spaces . [ prop : complement ] every induced subspace of a filtered vector space has a complementary summand . we are given @xmath291 , and seek to construct @xmath307 such that @xmath293 . we proceed inductively . since @xmath308 we take @xmath309 . now suppose we have chosen @xmath310 so that @xmath311 . in particular , @xmath312 . then @xmath313 thus @xmath310 and @xmath314 are independent subspaces of @xmath315 , and so @xmath310 can be extended to a complement @xmath316 of @xmath314 in @xmath315 . this completes the induction . the indecomposables in @xmath281 are precisely the intervals @xmath282 . thus , every filtered vector space can be decomposed as a finite direct sum of intervals . by proposition [ prop : complement ] , @xmath285 has nontrivial summands if and only if @xmath317 has nontrivial vector subspaces ; this happens exactly when @xmath318 . the dimension of @xmath319 is defined to be the vector of integers @xmath320 where @xmath321 are the dimensions of the successive subquotients of the filtration . [ prop : filtinterval ] let @xmath285 be a filtered vector space of depth @xmath35 , with @xmath322 . for any decomposition of @xmath285 into intervals , the multiplicity of @xmath282 is @xmath323 . thus : @xmath324 let @xmath325 be the multiplicity of @xmath282 . then , for all @xmath17 , @xmath326 by considering the contribution of each summand , whereas @xmath327 by considering the contribution of each subquotient @xmath328 . this is possible only if @xmath329 for all @xmath79 . this concludes our tour of the decomposition theory for filtered vector spaces . now we must leverage this to achieve a decomposition theory for @xmath51-modules . in one direction , the relationship is straightforward : [ prop : directsum ] the right - filtration operation respects direct sums , in the sense that @xmath330 for @xmath51-modules @xmath331 . this is proved by induction on @xmath51 , following the recursive structure of definition [ def : rfilt ] and using the standard facts @xmath332 and @xmath333 from linear algebra . ( for simplicity we are suppressing various indices here . ) however , what we need is a converse to proposition [ prop : directsum ] : if the filtered vector space @xmath334 can be split as a direct sum @xmath335 , we would like to infer a corresponding splitting @xmath336 of @xmath51-modules . in the following two sections we establish such a principle for a particular class : the ` streamlined ' @xmath51-modules . we introduce a special class of @xmath51-module for which the right - filtration functor preserves all structural information . a @xmath51-module @xmath55 is * ( right-)streamlined * if each @xmath337 is injective and each @xmath338 is surjective . similarly , we may say that a @xmath51-module @xmath55 is left - streamlined if each @xmath337 is surjective and each @xmath338 is injective . we will not need to consider left - streamlined modules until section [ sec : further ] . by default , ` streamlined ' will be taken to mean ` right - streamlined ' . [ ex : streaminterval ] intervals @xmath272 are streamlined ( but not @xmath123 for @xmath273 ) . conversely , a streamlined @xmath51-module @xmath55 with @xmath339 is necessarily isomorphic to some @xmath272 . indeed , @xmath127 is a non - decreasing sequence and therefore comprises some @xmath340 zeros ( where @xmath341 ) followed by @xmath342 ones . the maps between the one - dimensional terms are injective or surjective , and therefore isomorphisms . [ prop : splitstream ] a direct sum @xmath336 of @xmath51-modules is streamlined if and only if each summand is streamlined . each @xmath343 in @xmath55 decomposes as @xmath344 and is injective if and only if each @xmath345 is injective . each @xmath346 in @xmath55 decomposes as @xmath347 and is surjective if and only if each @xmath348 is surjective . the proof of the following lemma appears at the end of this section . [ lem : decomp ] let @xmath55 be a streamlined @xmath51-module and let @xmath334 . for any decomposition @xmath349 , there exists a unique decomposition @xmath107 such that @xmath350 for all @xmath114 . [ thm : streamlinedremak ] let @xmath55 be a streamlined @xmath51-module of length @xmath35 , and write @xmath351 and @xmath230 . then there is an isomorphism of @xmath51-modules @xmath352 let @xmath334 . by proposition [ prop : filtinterval ] , there is a decomposition @xmath353 , where the @xmath354 are a collection of @xmath355 intervals with @xmath282 occuring with multiplicity @xmath323 . lemma [ lem : decomp ] produces a decomposition @xmath356 , with @xmath357 for all @xmath114 . each @xmath358 is streamlined ( proposition [ prop : splitstream ] ) with maximum dimension @xmath359 , and is therefore isomorphic to some @xmath272 ( example [ ex : streaminterval ] ) . by lemma [ lem : intervalformula ] , we must have @xmath360 whenever @xmath361 . it follows that the @xmath358 are a collection of @xmath355 intervals with @xmath362 occuring with multiplicity @xmath323 . we complete this chapter with a proof of the decomposition lemma . we may assume that @xmath363 , since the general case follows by iteration . accordingly , suppose that @xmath334 can be written in the form @xmath293 ; we must show that there is a corresponding decomposition @xmath364 . we will argue by induction on @xmath365 . the first step is to determine the splitting @xmath366 . in fact , the stipulation that @xmath367 and @xmath368 forces @xmath369 and @xmath370 . if @xmath371 , then we are done . otherwise , let @xmath372 denote the truncation of @xmath55 to the indices @xmath373 and let @xmath374 . we will shortly establish that @xmath293 induces a unique compatible decomposition @xmath375 . the inductive hypothesis will then provide @xmath376 , which combines with @xmath366 to produce the desired decomposition @xmath99 . that will complete the proof . write @xmath279 . there are two cases . case @xmath377 , injective . we can identify @xmath378 with the subspace @xmath379 of @xmath195 . thereupon we have @xmath380 the unique splitting of @xmath378 compatible with @xmath366 is @xmath381 we must now verify that the induced subspaces @xmath382 and @xmath383 give a valid decomposition @xmath375 of filtered vector spaces . this follows because @xmath384 and similarly @xmath385 , for all @xmath386 ; so @xmath387 as required . case @xmath388 , surjective . here we identify @xmath378 as the quotient @xmath389 . under this identification , @xmath390 in splitting @xmath391 we are compelled to take @xmath392 which induce @xmath393 for the purported splitting @xmath375 . to confirm that this is a genuine decomposition , we note from linear algebra that the twin facts @xmath394 imply that @xmath395 as required . there is a high - level proof of lemma [ lem : decomp ] which in some sense is the natural explanation for the result . we outline this proof now . the first observation is that the transformation @xmath396 is a functor from @xmath65 to @xmath281 : a morphism @xmath75 induces a morphism @xmath397 . indeed , @xmath398 is defined to be @xmath399 ; one must check that this respects the filtrations on @xmath195 and @xmath400 . being a functor , @xmath401 defines a ring homomorphism @xmath402 . the second key fact is that this homomorphism is an isomorphism if @xmath55 is streamlined ( in general it is surjective ) . it is well known that direct - sum decompositions of a module can be extracted from the structure of its endomorphism ring : direct summands correspond to idempotent elements of the ring . it follows that @xmath55 and @xmath196 have the same decomposition structure . here we describe the algorithm for determining the indecomposable factors of a @xmath51-module . we give three versions of the ` algorithm ' . the first version , in section [ subsec : maintheorem ] , is not an algorithm but a proof that every @xmath51-module decomposes as a sum of interval modules ( theorem [ thm : interval ] ) . moreover , the structure of the proof makes it clear how to compute the interval decomposition ( theorem [ thm : main ] ) . the algorithms in the subsequent sections build on this . in section [ subsec : algabst ] we describe an abstract form of the decomposition algorithm , using the language of vector spaces and linear maps . no consideration is given to how the spaces and maps are described and manipulated in practice . in section [ subsec : algconc ] we suppose that the maps @xmath403 are presented concretely as matrices @xmath404 with respect to a choice of bases for the vector spaces @xmath153 . we describe an algorithm which takes these matrices as input and returns the interval decomposition . our present goal is to give a somewhat constructive proof of theorem [ thm : interval ] , which asserts that any @xmath51-module @xmath55 is isomorphic to a direct sum of intervals @xmath123 . we prove a stronger , more precise result , which explicitly determines the multiplicity of each interval within @xmath55 . some notation will help with the theorem statement . if @xmath405 then let @xmath406 = ( v_1 { \stackrel{p_{1}}{{\longleftrightarrow } } } \dots { \stackrel{p_{k-1}}{{\longleftrightarrow } } } v_k)\ ] ] denote the truncation of @xmath55 to length @xmath17 , and let @xmath407 $ ] denote its type ( which is a truncation of @xmath51 ) . [ thm : main ] let @xmath55 be a @xmath51-module . for @xmath408 , define @xmath409).\ ] ] writing @xmath410)$ ] , define @xmath411 ( whichever is applicable ) when @xmath412 , and @xmath413 then @xmath414 [ add : main ] in the situation of theorem [ thm : main ] , write @xmath415 for @xmath416 , and conventionally define @xmath417 for all @xmath79 . then @xmath418 for @xmath419 . the decomposition strategy begins with the following lemma . the idea is to proceed from left to right along the complex , removing streamlined summands at each step . having done this , the remak decompositions of those summands can be determined by counting dimensions , as prescribed in theorem [ thm : streamlinedremak ] . [ lem : cascade ] let @xmath420 be an irreducible @xmath51-module of length @xmath35 . then there exists a direct - sum decomposition @xmath421 where each @xmath422 is supported over the indices @xmath423 and is right - streamlined over that range . the following picture illustrates the decomposition . @xmath424 each row ( i.e. summand ) is right - streamlined , and therefore amenable to analysis via the right - filtration functor . we proceed by induction on the length of @xmath55 . the inductive statement is that @xmath406 = { { { \mathbb{v}}}}^1 \oplus \dots \oplus { { { \mathbb{v}}}}^{k-1 } \oplus { { { \mathbb{w}}}}^k\ ] ] where the @xmath425 are as in the theorem statement , and @xmath426 is itself right - streamlined . for the base case @xmath427 , there is nothing to prove : take @xmath428 $ ] . now suppose the inductive statement is established for @xmath17 , and consider @xmath429 $ ] . this can be written @xmath430 & = & ( { { { \mathbb{v}}}}^1 \oplus \dots \oplus { { { \mathbb{v}}}}^{k-1 } \oplus { { { \mathbb{w}}}}^k ) { \stackrel{p_{k}}{{\longleftrightarrow } } } v_{k+1 } \\ & = & { { { \mathbb{v}}}}^1 \oplus \dots \oplus { { { \mathbb{v}}}}^{k-1 } \oplus ( { { { \mathbb{w}}}}^k { \stackrel{p_{k}}{{\longleftrightarrow } } } v_{k+1})\end{aligned}\ ] ] where the rebracketing is permissible because all of the @xmath425 terms terminate before time @xmath17 , and therefore do not interact with @xmath431 . the goal now is to rewrite @xmath432 as @xmath433 , where @xmath422 terminates at time @xmath17 and both @xmath422 and @xmath434 are right - streamlined . the rightmost term of @xmath426 is @xmath435 , so @xmath436 is a filtration on @xmath435 . case @xmath60 : @xmath437 . in other words let @xmath439 . proposition [ prop : complement ] implies that @xmath288 has a complement in @xmath436 ; say @xmath440 . this corresponds ( lemma [ lem : decomp ] ) to a direct sum decomposition @xmath441 , where both summands are streamlined ( proposition [ prop : splitstream ] ) . this defines @xmath422 , and we set @xmath442 . to check that this works , note that @xmath443 is zero on @xmath444 and is injective on the complementary subspace @xmath445 . thus @xmath422 is a summand of @xmath429 $ ] terminating at time @xmath17 , and @xmath434 is streamlined . case @xmath61 : @xmath446 . in other words let @xmath448 . proposition [ prop : complement ] implies that @xmath288 has a complement in @xmath436 ; say @xmath440 . this corresponds ( lemma [ lem : decomp ] ) to a direct sum decomposition @xmath449 , where both summands are streamlined ( proposition [ prop : splitstream ] ) . this defines @xmath422 , and we set @xmath450 . to check that this works , note that @xmath451 is surjective onto @xmath452 and misses the complementary subspace @xmath453 . thus @xmath422 is a summand of @xmath429 $ ] terminating at time @xmath17 , and @xmath434 is streamlined . this establishes the inductive step , so eventually @xmath454 = { { { \mathbb{v}}}}^1 \oplus \dots \oplus { { { \mathbb{v}}}}^{n-1 } \oplus { { { \mathbb{w}}}}^n\ ] ] and we set @xmath455 to finish the proof . write @xmath456 according to lemma [ lem : cascade ] . we now calculate the decomposition of each @xmath422 into intervals @xmath457 . note that @xmath406 = { { { \mathbb{v}}}}^k \oplus { { { \mathbb{v}}}}^{k+1}[k ] \oplus \dots \oplus { { { \mathbb{v}}}}^n[k].\ ] ] we can write @xmath458 , so then @xmath459 ) = { \mathrm{r}}({{{\mathbb{v}}}}^k ) \oplus { \mathrm{r}}({{{\mathbb{w}}}}^k[k])\ ] ] ( using proposition [ prop : directsum ] ) . this is a filtration on @xmath460 . suppose @xmath461 . we note that @xmath426 is streamlined up to time @xmath462 , whereas @xmath422 is zero at time @xmath462 . the next map in the sequence is @xmath463 in the first case , it follows that @xmath464 and therefore @xmath465 . in the second case , @xmath466 is a complement to @xmath467 in @xmath435 , so @xmath468 . thus @xmath469 ( whichever middle term is applicable ) . when @xmath470 , moreover , we have @xmath471 thus , at last , @xmath472 using theorem [ thm : streamlinedremak ] to decompose the @xmath422 . write @xmath473))$ ] . since @xmath474)$ ] we can take dimensions and obtain the formula @xmath475 note also that @xmath476 ) = { \mathrm{r}}({{{\mathbb{w}}}}^k[k+1])$ ] . moreover , @xmath426 is streamlined up to time @xmath462 . it follows that @xmath477 ) ) = \left\ { \begin{array}{ll } ( w^k_1 , \dots , w^k_k , ? ) & \qquad \mbox{case~$f$ } \\ ( ? , w^k_1 , \dots , w^k_k ) & \qquad \mbox{case~$g$ } \end{array } \right.\ ] ] and therefore @xmath478 which is the desired formula . we now transcribe theorem [ thm : main ] as an abstract algorithm for determining the interval structure of a @xmath51-module @xmath55 of length @xmath35 . this algorithm will serve as a skeleton for the more concrete algorithms developed later . [ alg : main ] we proceed through @xmath479 , computing the filtration @xmath410)$ ] , the birth - time index @xmath480)$ ] , and the dimensions @xmath481 iteratively . begin * initialisation * ( @xmath427 ) : * @xmath482 . * @xmath483 ) = ( 1)$ ] . * iterative step * ( @xmath484 ) : * calculate @xmath485 from @xmath486 using definition [ def : rfilt ] : @xmath487 * calculate @xmath488))$ ] from @xmath480 ) = ( b^k_1 , b^k_2 , \dots , b^k_k)$ ] using definition [ def : birthtime ] : @xmath489 * calculate @xmath490 using the formula in theorem [ thm : main ] : @xmath491 alternatively , use the formula in addendum [ add : main ] : @xmath492 here @xmath493 . * terminal step * ( @xmath470 ) : * calculate @xmath494 . * print results * : * for @xmath419 , the interval @xmath495 occurs with multiplicity @xmath481 . end note that steps ( 35 ) have an ` @xmath60 ' verson and a ` @xmath61 ' version , depending on the direction of the map @xmath496 . this abstract algorithm does not specify how the filtered vector spaces @xmath497)$ ] are stored , nor how the maps @xmath443 or @xmath451 ( which are used in steps ( 3 ) and ( 5 ) ) are represented . in any concrete setting , it is necessary to specify data structures . a good choice will facilitate the calculations in steps ( 3 ) and ( 5 ) . in the next section , we work out the details in a simple scenario . in this section we describe an algorithm to solve the following concrete problem . let @xmath51 be a type of length @xmath35 . we specify a @xmath51-module @xmath55 as follows . set @xmath498 for integers @xmath499 . for each @xmath79 , the map @xmath500 is defined by an @xmath501-by-@xmath502 matrix @xmath503 or else the map @xmath504 is defined by an @xmath505-by-@xmath501 matrix @xmath506 . we are to determine @xmath134 , given @xmath51 and the matrices @xmath503 or @xmath507 . we follow algorithm [ alg : main ] . the substantial task is to calculate the sequence of right - filtrations @xmath410)$ ] , for step ( 3 ) . everything else is book - keeping : the birth - time indices @xmath508 are calculated according to step ( 4 ) ; and the filtration dimensions @xmath509 ( and hence the @xmath510 ) will be easy to read off from the stored description of the filtrations . the algorithm operates on two levels . on the _ conceptual _ level , we proceed by modifying the bases of the spaces @xmath153 by elementary basis transformations . initially each basis @xmath511 is the standard basis of @xmath512 . we perform modifications on @xmath513 in sequence . on the _ pragmatic _ level , what we actually do is apply elementary row and column operations to the matrices @xmath503 or @xmath507 . we make no attempt to track the bases themselves ; instead we implement the effect of those changes on the matrices . suppose we apply elementary basis transformations to @xmath514 on the conceptual level . on the pragmatic level , we must perform @xmath515 and simultaneously perform @xmath516 to enact those transformations . thus , at every stage we must make parallel changes to two matrices simultaneously . usually we are working to put @xmath517 or @xmath518 in a particular form , and while doing so the changes have to be mirrored in @xmath519 or @xmath520 ( paying no attention yet to the structure of that matrix ) . we now make this precise . the * elementary transformation * @xmath521 is defined as follows . on the conceptual level , this is a modification of @xmath522 involving basis vectors @xmath523 and @xmath524 : @xmath525 on the pragmatic level , if @xmath43 is a matrix representing a linear map @xmath526 for some @xmath527 ( this will be @xmath528 or @xmath503 in our situation ) , then we modify the columns of @xmath43 accordingly : @xmath529 else , if @xmath43 represents a linear map of the form @xmath530 ( this will be @xmath531 or @xmath507 in our situation ) then we must apply the dual transformation to the rows of @xmath43 : @xmath532 in spirit , we right - multiply by the matrix @xmath533 $ ] to modify columns , or else left - multiply by the inverse matrix @xmath534 $ ] to modify rows . besides the elementary transformations @xmath521 , it is sometimes appropriate to permute the basis elements . the operation @xmath535 of interchanging @xmath523 with @xmath524 is realised pragmatically by interchanging @xmath536 with @xmath537 , or @xmath538 with @xmath539 , as appropriate . the filtration @xmath410)$ ] on @xmath435 is to be represented as follows . we require the basis @xmath540 to be compatible with the filtration , in a sense that will become clear . assuming such a basis , the filtration @xmath541 is represented as a non - decreasing function @xmath542 so that @xmath543 for @xmath544 . in other words : the first few basis elements ( those @xmath523 with @xmath545 ) form a basis for @xmath546 ; the next few basis elements extend this to a basis for @xmath547 , and so on . the dimension @xmath548 can be read off as the cardinality of @xmath549 . step ( 3 ) boils down to the following task . suppose that @xmath550 and @xmath551 together represent the filtration @xmath552 ; then modify @xmath514 and determine @xmath553 to represent @xmath485 . we now explain how to do this . case @xmath554 : the matrix @xmath555 represents a linear map @xmath556 . we assume that @xmath550 is compatible with the filtration @xmath552 , and that @xmath551 identifies the filtration . this gives a block structure @xmath557\ ] ] where @xmath558 gathers together the columns @xmath559 with @xmath560 . using row operations only , put @xmath555 into ( unreduced ) row echelon form . this means : * each of the top @xmath24 rows contains a 1 ( the _ pivot _ ) as its leftmost nonzero entry . * each pivot lies strictly to the left of the pivots of the rows below it . * the lowest @xmath561 rows are entirely zero . these row operations correspond to elementary operations @xmath562 , and the effect of these operations is felt on the next matrix @xmath519 or @xmath520 , which must be modified accordingly . we now define @xmath553 as follows : @xmath563 see figure [ fig : gaussm ] . from @xmath552.,title="fig : " ] from @xmath552.,title="fig : " ] it is evident in the figure that @xmath564 maps onto @xmath565 for all @xmath79 . case @xmath566 : the matrix @xmath567 represents a linear map @xmath568 . we assume that @xmath550 is compatible with the filtration @xmath552 , and that @xmath569 identifies the filtration . this time we have a vertical block structure @xmath570\ ] ] where @xmath571 gathers together the rows @xmath559 with @xmath572 . using column operations only , put @xmath567 into the column echelon form defined as follows ( this echelon form begins on the bottom left ) : * each of the leftmost @xmath24 columns contains a 1 ( the pivot ) as its _ lowest _ nonzero entry . * each pivot lies strictly lower than the pivots of the columns to the right of it . * the rightmost @xmath561 rows are entirely zero . these column operations correspond to elementary operations @xmath562 , and the effect of these operations is felt on the next matrix @xmath519 or @xmath520 , which must be modified accordingly . we now define @xmath553 as follows : @xmath573 see figure [ fig : gaussn ] . from @xmath552.,title="fig : " ] from @xmath552.,title="fig : " ] it is evident in the figure that @xmath574 is the largest subspace which maps into @xmath575 , for all @xmath79 . this concludes our treatment of the concrete form of the zigzag algorithm . let @xmath55 be a zigzag module of length @xmath35 and let @xmath408 . we consider the problem of determining the set of intervals in @xmath134 which contain @xmath17 , without necessarily computing @xmath134 itself . we shall see that all the necessary information is contained in a pair of filtrations on the vector space @xmath435 . let @xmath55 be a zigzag module of length @xmath35 . the * left - filtration * of @xmath55 is a filtration on @xmath168 of depth @xmath35 , defined as @xmath576 where @xmath577 is the reversal of @xmath55 ; so @xmath578 , with maps @xmath579 or @xmath580 . for any @xmath17 we therefore have two natural filtrations on @xmath435 : @xmath581 ) , \\ { { { \mathcal{l}}}}_k & = & ( l_0 , l_1 , \dots , l_{n+1-k } ) & = & { \mathrm{l}}({{{\mathbb{v}}}}[k , n ] ) ; \end{array}\ ] ] the right - filtration over the index set @xmath582 and the left - filtration over the index set @xmath583 . we also have birth - time and death - time indices @xmath584 ) \\ { \mathrm{d}}_k & = & ( d_1 , \dots , d_{n+1-k } ) & = & n+1 - { \mathrm{b}}(\bar\tau[k , n ] ) \end{array}\ ] ] which indicate the birth and death times associated with the respective subquotients of @xmath552 and @xmath585 . these depend on the type @xmath51 of @xmath55 . [ ex : bifilt ] consider the zigzag module @xmath586 at @xmath587 , for instance , we have @xmath588 and @xmath589 we can now state the main theorem of this section . [ thm : kinterval ] let @xmath55 be a zigzag module of length @xmath35 and let @xmath408 . let @xmath590 denote the right- and left - filtrations at @xmath17 , and let @xmath591 denote the birth - time and death - time indices at @xmath17 . then , for all @xmath592 in the range @xmath593 , @xmath594 , the multiplicity of @xmath595 $ ] in @xmath134 is equal to @xmath596 equivalently , @xmath597 , the dimension of the @xmath598-th bifiltration subquotient . this theorem answers the original question , because every interval containing @xmath17 can be written as @xmath595 $ ] for some choice of @xmath592 . we now work towards a proof of theorem [ thm : kinterval ] . [ prop : kstream ] it is sufficient to prove theorem [ thm : kinterval ] in the special case where @xmath55 is right - streamlined over @xmath582 and left - streamlined over @xmath583 . it is clear from lemma [ lem : cascade ] that we can write @xmath599 where @xmath600 is supported in @xmath601 and @xmath80 is right - streamlined over @xmath582 . indeed , take @xmath602 and @xmath603 . moreover , it is sufficient to prove theorem [ thm : kinterval ] for @xmath80 , because the filtrations @xmath590 remain unchanged from @xmath55 , and the discarded term @xmath600 decomposes into intervals which do not contain @xmath17 . thus , we may assume that @xmath55 is right - streamlined over @xmath604 . repeating this argument from the other side , we may further assume that @xmath55 is left - streamlined over @xmath583 . assume that @xmath55 satisfies the condition in proposition [ prop : kstream ] . it follows that every interval in @xmath134 contains @xmath17 : any other interval in the decomposition would cause a failure of the streamline condition . we can therefore write the interval decomposition of @xmath55 as @xmath605 where @xmath606 indexes the summands , and @xmath607 and @xmath608 identify the interval type of each summand in terms of the birth - time and death - time indices . it is apparent from this formulation that @xmath609 and it remains to compute this in terms of the dimensions @xmath610 . the interval decomposition restricts at index @xmath17 to a direct sum decomposition of @xmath435 into 1-dimensional subspaces @xmath611 , generated by elements @xmath612 , say . then @xmath613 ) = \bigoplus_{a \in a } { { { \mathcal{r}}}}_k \cap u_a \cong \bigoplus_{a \in a } { { { \mathcal{j}}}}(p(a ) , k)\ ] ] where the final isomorphism comes from lemma [ lem : intervalformula ] . now , the filtration subspace @xmath198 is spanned by the terms isomorphic to @xmath614 with @xmath615 . in other words , for @xmath544 we have @xmath616 a similar argument proceeding from the other direction gives the analogous formula @xmath617 for @xmath618 . since the @xmath612 are independent , these formulas give bases for @xmath619 . we now claim that @xmath620 for all @xmath592 . the inclusion @xmath621 is obvious , because each of the spanning vectors @xmath612 belongs to both @xmath198 and @xmath622 . in the other direction , if @xmath623 then write @xmath624 . since @xmath625 , all the coefficients @xmath626 with @xmath627 must be zero . since @xmath628 , all the coefficients @xmath626 with @xmath629 must be zero . thus @xmath630 . this establishes the reverse inclusion @xmath631 and hence the equality . then @xmath632 for all @xmath592 . the formula in the theorem follows easily from this . the salient fact behind this result is that it is possible to find a direct sum decomposition of @xmath435 which simultaneously decomposes the filtered spaces @xmath590 into intervals within their respective categories @xmath633 , @xmath634 . here we achieved this by appealing to the interval decomposition of @xmath55 , but this can also be proved directly for an arbitrary pair of filtrations on a single vector space . the analogous statement for a triple of filtrations is false . for example @xmath635 ( where @xmath171 ) can not be simultaneously decomposed into intervals . consider the following diagram : @xmath636{w_k } \\ \node{v_1 } \arrow{e , t,<>}{p_1 } \node{\cdots } \arrow{e , t,<>}{p_{k-2 } } \node{v_{k-1 } } \arrow{ne , t}{f_{k-1 } } \node[2]{v_{k+1 } } \arrow{nw , t}{g_{k } } \arrow{e , t,<>}{p_{k+1 } } \node{\cdots } \arrow{e , t,<>}{p_{n-1 } } \node{v_n } \\ \node[4]{u_k } \arrow{nw , t}{g_{k-1 } } \arrow{ne , t}{f_{k } } \end{diagram}\ ] ] let @xmath202 and @xmath637 denote the two zigzag modules contained in the diagram : @xmath638 we wish to compare @xmath639 with @xmath640 , particularly with respect to intervals that meet @xmath641 . this requires a favourable condition on the four maps in the middle diamond . we say that the diagram @xmath642 is * exact * if @xmath643 in the following sequence @xmath644 where @xmath645 and @xmath646 . [ thm : diamond ] given @xmath202 and @xmath637 as above , suppose that the middle diamond is exact . then there is a partial bijection of the multisets @xmath639 and @xmath640 , with intervals matched according to the following rules : * intervals of type @xmath647 $ ] are unmatched . * type @xmath648 $ ] is matched with type @xmath649 $ ] and vice versa , for @xmath650 . * type @xmath651 $ ] is matched with type @xmath652 $ ] and vice versa , for @xmath653 . * type @xmath116 $ ] is matched with type @xmath116 $ ] , in all other cases . it follows that the restrictions @xmath654 , @xmath655 to the set @xmath656 are equal . figures [ fig : diamond_barcode ] and [ fig : diamond_pd ] illustrate theorem [ thm : diamond ] in terms of barcodes and persistence diagrams , respectively . and @xmath640 : ( top row ) the five cases where matching occurs ; ( bottom row ) unmatched intervals @xmath647$].,title="fig : " ] and @xmath640 : ( top row ) the five cases where matching occurs ; ( bottom row ) unmatched intervals @xmath647$].,title="fig : " ] and @xmath640 : ( top row ) the five cases where matching occurs ; ( bottom row ) unmatched intervals @xmath647$].,title="fig : " ] and @xmath640 : ( top row ) the five cases where matching occurs ; ( bottom row ) unmatched intervals @xmath647$].,title="fig : " ] and @xmath640 : ( top row ) the five cases where matching occurs ; ( bottom row ) unmatched intervals @xmath647$].,title="fig : " ] and @xmath640 : ( top row ) the five cases where matching occurs ; ( bottom row ) unmatched intervals @xmath647$].,title="fig : " ] and @xmath640 : ( top row ) the five cases where matching occurs ; ( bottom row ) unmatched intervals @xmath647$].,title="fig : " ] to @xmath640 , for @xmath657 , @xmath658 : points in the persistence plane move according to the arrows ; the multiplicity of the point marked @xmath659 changes unpredictably . ] the @xmath660 summands in @xmath639 span the cokernel of @xmath661 , whereas the @xmath660 summands in @xmath640 span the kernel of @xmath662 . the hypothesis of theorem [ thm : diamond ] does not bring about any relation between these spaces ( which is why the @xmath647 $ ] intervals are unmatched ) . in section [ subsec : strongdiamond ] , however , we consider a situation in which the @xmath647 $ ] intervals can be tracked . we use the localization technique of section [ subsec : localize ] to prove theorem [ thm : diamond ] . we begin with birth- and death - time indices . [ prop : diamond ] let @xmath663 denote the zigzag types of @xmath664 respectively . if we write @xmath665 ) = { \mathrm{b}}(\tau^-[1,k-1])\ ] ] for the birth - time index up to time @xmath666 , then @xmath667 ) & = & ( k+1 , b_1 , \dots , b_{k-1 } , k ) . \\ { \mathrm{b}}(\tau^-[1,k+1 ] ) & = & ( k , b_1 , \dots , b_{k-1 } , k+1),\end{aligned}\ ] ] similarly , if we write @xmath668 ) = { \mathrm{d}}(\tau^-[k+1,n])\ ] ] for the death - time index from time @xmath462 , then @xmath669 ) & = & ( k-1 , d_1 , \dots , d_{n - k } , k ) . \\ { \mathrm{d}}(\tau^-[k-1,n ] ) & = & ( k , d_1 , \dots , d_{n - k } , k-1),\end{aligned}\ ] ] this is immediate from the recursive definition of birth - time index . if we write @xmath670 = \tau^-[1,k-1]$ ] then @xmath671 = \tau_0 fg$ ] and @xmath672 = \tau_0 gf$ ] . the death - time index is treated similarly . here is the crux of the matter : [ lem : diamond ] in the situation of theorem [ thm : diamond ] , the following filtrations are equal : @xmath673 ) & = & { \mathrm{r}}({{{\mathbb{v}}}}^-[1,k+1 ] ) \\ { \mathrm{l}}({{{\mathbb{v}}}}^+[k-1,n ] ) & = & { \mathrm{l}}({{{\mathbb{v}}}}^-[k-1,n])\end{aligned}\ ] ] write @xmath674 ) = { \mathrm{r}}({{{\mathbb{v}}}}^-[1,k-1])$ ] . by the recursive formula ( definition [ def : rfilt ] ) , @xmath675 ) = ( 0 , g_k^{-1 } f_{k-1}(r_0 ) , \dots , g_k^{-1 } f_{k-1}(r_{k-1 } ) , v_{k+1})\ ] ] and @xmath676 ) = ( 0 , f_k g_{k-1}^{-1}(r_0 ) , \dots , f_k g_{k-1}^{-1}(r_{k-1 } ) , v_{k+1}).\ ] ] thus we can prove the first statement of the lemma by showing that @xmath677 for any subspace @xmath678 . we use first - order logic . let @xmath679 . we have the following chain of equivalent statements . @xmath680 on the other hand : @xmath681 since @xmath643 by hypothesis , it follows that @xmath682 . this proves the first equality . the second equality follows symmetrically . we adopt the notation of section [ subsec : localize ] , and consider the right- and left - filtrations at @xmath683 , for both @xmath202 and @xmath637 . since @xmath684 = { { { \mathbb{v}}}}^-[k+1,n]$ ] we have @xmath685 and by lemma [ lem : diamond ] we have @xmath686 finally , @xmath687 agrees with @xmath688 except that @xmath689 are interchanged , according to proposition [ prop : diamond ] . thus , when we use theorem [ thm : kinterval ] to calculate the multiplicity of @xmath116 $ ] for @xmath690 , there is perfect agreement between @xmath202 and @xmath637 except that we must interchange @xmath689 when they occur as birth - times . a symmetrical argument can be made , localizing at @xmath691 . when we compute the multiplicity of @xmath116 $ ] for @xmath692 , there is perfect agreement between @xmath202 and @xmath637 except that we must interchange @xmath693 when they occur as death - times . we have covered all cases of the theorem except for intervals which meet neither @xmath666 nor @xmath462 . intervals contained in @xmath694 $ ] are automatically the same for @xmath202 and @xmath637 because they can be computed by restricting to @xmath695 $ ] and @xmath696 $ ] , which are equal . similarly , intervals contained in @xmath697 $ ] are the same for @xmath202 and @xmath637 , by restricting to @xmath684 = { { { \mathbb{v}}}}^-[k+1,n]$ ] . finally , consider intervals @xmath647 $ ] . nothing can be said about those . the diamond principle can usefully be applied to the following diagram of topological spaces and continuous maps . the four maps in the central diamond are inclusion maps , and the remaining maps @xmath698 are arbitrary . @xmath699{a \cup b } \\ \node{x_1 } \arrow{e , t , < > } { } \node{\cdots } \arrow{e , t , < > } { } \node{x_{k-2 } } \arrow{e , t , < > } { } \node{a } \arrow{ne , t } { } \node[2]{b } \arrow{nw , t } { } \arrow{e , t , < > } { } \node{x_{k+2 } } \arrow{e , t , < > } { } \node{\cdots } \arrow{e , t , < > } { } \node{x_n } \\ \node[5]{a \cap b } \arrow{nw , t } { } \arrow{ne , t } { } \end{diagram}\ ] ] let @xmath700 denote the upper and lower zigzag diagrams contained in this picture ; so @xmath701 passes through @xmath702 and @xmath703 , passes through @xmath704 . [ thm : strongdiamond ] given @xmath701 and @xmath703 as above , there is a ( complete ) bijection between the multisets @xmath705 and @xmath706 . intervals are matched according to the following rules : * @xmath647 \in { \operatorname{pers}}(h_{\ell+1}({{{\mathbb{x}}}}^+))$ ] is matched with @xmath647 \in { \operatorname{pers}}(h_{\ell}({{{\mathbb{x}}}}^-))$ ] . in the remaining cases , the matching preserves homological dimension : * type @xmath648 $ ] is matched with type @xmath649 $ ] and vice versa , for @xmath650 . * type @xmath651 $ ] is matched with type @xmath652 $ ] and vice versa , for @xmath653 . * type @xmath116 $ ] is matched with type @xmath116 $ ] , in all other cases . for any @xmath707 , apply the homology functor @xmath708 to the diagram . the central diamond @xmath709 is exact by virtue of the mayer vietoris theorem , according to which @xmath710 is an exact sequence . the diamond principle therefore applies to @xmath711 and @xmath712 , and we have a partial bijection which accounts for all intervals except those of type @xmath647 $ ] . now consider the connecting homomorphism in the same mayer vietoris sequence : @xmath713 by exactness , @xmath714 induces an isomorphism between the cokernel of @xmath661 and the kernel of @xmath662 . but the @xmath647 $ ] summands of @xmath715 precisely span @xmath716 , whereas the @xmath647 $ ] summands of @xmath717 span @xmath718 . this establishes the claimed bijection between the @xmath647 $ ] intervals . [ ex:1param ] let @xmath719 be a sequence of simplicial complexes defined on a common vertex set . suppose these have arisen in some context where each transition @xmath36 to @xmath720 is regarded as being a ` small ' change . there are two natural zigzag sequences linking the @xmath36 . the union zigzag , @xmath721 : @xmath722{x_1 \cup x_2 } \node[2]{\dots } \node{\dots } \node[2]{x_{n-1 } \cup x_n } \\ \node{x_1 } \arrow{ne } \node[2]{x_2 } \arrow{nw } \arrow{ne } \node[3]{x_{n-1 } } \arrow{nw } \arrow{ne } \node[2]{x_{n } } \arrow{nw } \end{diagram}\ ] ] the intersection zigzag , @xmath723 : @xmath724{x_2 } \node[3]{x_{n-1 } } \node[2]{x_{n } } \\ \node[2]{x_1 \cap x_2 } \arrow{nw } \arrow{ne } \node[2]{\dots } \arrow{nw } \node{\dots } \arrow{ne } \node[2]{x_{n-1 } \cap x_n } \arrow{nw } \arrow{ne } \end{diagram}\ ] ] we can think of these as being indexed by the half - integers @xmath725 . we can apply the strong diamond principle @xmath69 times to derive the following relationship between the zigzag persistence of the two sequences @xmath726 and @xmath727 . restricting to the integer indices , there is a coarse equality : @xmath728 more finely , there is a partial bijection between @xmath727 and @xmath726 . intervals @xmath729 $ ] shift homological dimension by @xmath730 ( from the intersection sequence to the union sequence ) . otherwise @xmath116 \leftrightarrow [ b',d']$ ] where @xmath731 is an unordered pair of the form @xmath732 and @xmath733 is an unordered pair of the form @xmath734 ; dimension is preserved . figure [ fig : n2u ] illustrates the complete correspondence as a transformation of the persistence diagram , for @xmath735 . to @xmath736 , for @xmath735 : points in the persistence plane move according to the arrows ; points marked @xmath737 stay fixed and increase homological dimension by 1 . ] we have presented the foundations of a theory of zigzag persistence which , we believe , considerably extends and enriches the well known and highly successful theory of persistent homology . zigzag persistence originates in the work of gabriel and others in the theory of quiver representations . one of our goals has been to bridge the gap between the quiver literature ( which is read largely by algebraists ) and the current language of applied and computational topology . to this end , we have presented an algorithmic form of gabriel s structure theorem for @xmath68 quivers , and have indicated the first steps towards integrating these ideas into tools for applied topology . there are several ways in which this work is incomplete . the most significant omission is an algorithm for computing zigzag persistence in a homological setting ( as distinct from the somewhat sanitised vector space algorithm described in section [ subsec : algconc ] ) . we address this gap in a forthcoming paper with dmitriy morozov @xcite , where we present an algorithm for computing the zigzag persistence intervals of a 1-parameter family of simplicial complexes on a fixed vertex set . we have made no effort in this paper to flesh out the applications suggested in section [ sec : intro ] . there is often a substantial gap between the concrete world of point - cloud data sets and the ideal world of simplicial complexes and topological spaces . we intend to develop some of these applications in future work . meanwhile , we have given priority to establishing the theoretical language and tools . the diamond principle is particularly powerful . in the manuscript with morozov @xcite , we show that the diamond principle can be used to establish isomorphisms between several different classes of persistence invariants of a space with a real - valued ( e.g. morse ) function . in particular , we use zigzag persistence to resolve an open conjecture concerning extended persistence @xcite . this supports our prejudice that zigzag persistence provides the appropriate level of generality and power for understanding the heuristic concept of persistence in its many manifestations . the authors wish to thank greg kuperberg , konstantin mischaikow and dmitriy morozov for helpful conversations and m. khovanov for helpful correspondence . the authors gratefully acknowledge support from darpa , in the form of grants hr0011 - 05 - 1 - 0007 and hr0011 - 07 - 1 - 0002 . the second author wishes to thank pomona college and stanford university for , respectively , granting and hosting his sabbatical during late 2008 .

the theory of inflation provides an explanation for many observed properties of our visible universe , in particular its homogeneity and isotropy ( see e.g. @xcite and @xcite for reviews ) . it also gives an explanation for the origin of small density perturbations which eventually gave rise to the structures in our universe . these aspects are well understood . however , from the particle physics point of view , a suitable candidate for the inflaton field has yet to be found @xcite . among the many candidates , hybrid inflation @xcite is a particularly well motivated one , especially in supersymmetric models @xcite@xcite . there has been a lot of recent work investigating the possibility of realising inflation within string theory ( see e.g. @xcite@xcite for reviews ) . the most natural framework for brane inflation is the brane - antibrane system where an attractive force leads to a potential for the interbrane distance @xcite@xcite . inflation ends with an open string tachyonic instability similar to hybrid inflation . this scenario has been extensively studied and leads to interesting phenomena such as @xmath4-string formation at the end of inflation . one particular issue in these models comes from the need for moduli stabilisation during inflation . if this is not achieved , the potential for the moduli is of runaway type destroying the existence of slow roll inflation . this problem can be circumvented using the kklt scenario whereby the kahler moduli are stabilised once the complex moduli have been stabilised and non - perturbative gaugino condensation has occurred on d7 branes @xcite . the coupling to inflation has been studied in kklmmt where a fine - tuning of the inflaton superpotential has been advocated @xcite . one can also use a shift symmetry to alleviate the @xmath2 problem @xcite@xcite . another way of realising hybrid inflation in string models can be obtained using the d3/d7 system , see e.g. @xcite@xcite . in this case the end of inflation happens at a lower scale than the planck scale , when the charged open string fields between the branes condense . the d3/d7 system can be modelled using an @xmath5 supersymmetric @xmath4-term inflation model @xcite . the interbrane distance plays the role of the inflaton and the charged open strings between the branes are the waterfall fields . when the supergravity corrections are neglected , i.e. in global supersymmetry , and assuming that the kahler moduli have been stabilised , the inflaton direction is flat . slow roll inflation is driven by the one loop logarithmic corrections as the one loop quadratic divergences are inflaton - independent . in this paper , we are investigating @xmath0-term hybrid inflation with the inflationary superpotential @xmath6 where @xmath7 is the @xmath8 gauge coupling , in the presence of additional moduli fields . it would be very interesting to derive the origin of @xmath9 from fundamental string theory @xcite . we treat the supergravity case going beyond the global supersymmetry analysis . we show that the treatment of fayet - iliopoulos @xmath4-terms ( i.e. when @xmath10 ) needs extra care in supergravity . to include such a term would require additional fields and the study of their stabilisation . for simplicity we therefore do not include inflationary @xmath4-terms in our models . inflation is then driven by the presence of the non - vanishing @xmath9 . we show that supergravity corrections , and supergravity - induced interactions with a moduli sector induce a tree level slope and mass for the inflaton @xcite . moreover there are additional contributions from the quadratic divergence part of the one - loop corrections , since they are inflaton - dependent in supergravity , as opposed to inflaton - independent in global supersymmetry . in general these effects spoil the flatness of the inflationary potential . we show that most of the tree - level problems can be removed by including shift symmetries in the inflaton sector , and taking the moduli sector to be no - scale . it then appears that a viable slow - roll inflation model can be found with limited fine - tuning of the moduli sector . however we also find that the coupling between the inflaton and the moduli has the effect of inducing a small variation of the moduli during inflation . despite the small size of this variation , it gives a significant contribution to the inflaton slope and violates slow - roll inflation . the paper is arranged as follows . in section [ sec : v ] , we construct the full potential for our combined model of inflation and moduli stabilisation . we calculate tree and one loop level parts of the potential , and describe symmetries which help maintain the flatness of the inflaton potential . in section [ sec : mod ] we determine the role of moduli fields during inflation , and show their effect is significant . we find the constraints corresponding to the cobe normalisation and the wmap3 spectral index constraint in section [ sec : con ] . our arguments and analysis apply to a general moduli sector , although as a specific example , we also give explicit results for the kklt model , which has just one moduli field @xmath11 . in section [ sec : kklt ] we study particular choices of racetrack superpotentials for the kklt scenario , and find that the running of the moduli field during inflation typically leads to a large value of @xmath2 . a small value of @xmath2 could only be achieved if the moduli superpotential is heavily fine - tuned . finally in section [ sec : nogo ] we consider models where a minkowski vacuum after inflation is obtained without the need for a lifting term . a model with this property is also interesting as it may give a small gravitino mass , as is required to have an low energy sparticle spectrum . we derive a no - go theorem which states that , for no - scale models with no lifting term , no such stable supersymmetry breaking minkowski vacua can be found . we conclude in section [ sec : conc ] . as already discussed , many string theory inflation scenarios can be modelled at low energy with a supergravity description of hybrid inflation . the system uses three fields , the inflaton @xmath12 measuring the inter - brane distance and two charged fields @xmath13 ( which in the d3/d7 system would represent the open strings between the two types of branes ) . the fields interact according to the superpotential ( [ inflasec ] ) with @xmath10 , which has been studied extensively in the literature . in @xcite it was shown how the fayet - iliopoulos @xmath4-term arose from fluxes on the d7 brane , allowing @xmath4-term inflation in this system . here we will focus almost exclusively on hybrid inflation which is driven only by @xmath0-terms ( so @xmath14 ) , although exactly how such terms are embedded in string theory remains to be found . including inflationary @xmath4-terms in a supergravity theory creates extra complications , as we will discuss in the following subsection . using ( [ inflasec ] ) and taking to @xmath15 we obtain @xmath16 this potential has an @xmath17 origin which was derived in @xcite . during inflation , when the waterfall fields @xmath18 , the potential is @xmath19 which gives a positive contribution to the inflation energy . extending this model to a supergravity theory , and working in units with @xmath20 , we replace the above @xmath0-term by @xmath21 where @xmath22 , and @xmath23 is the kahler potential for @xmath12 and @xmath13 . the simplest possibility would be @xmath24 . it was originally thought that extending the @xmath4-term in the above model to sugra would be straightforward . however it was later discovered that in order to have a fayet - iliopoulos term @xmath25 appearing in a supergravity theory , the superpotential must have charge @xmath26 @xcite . thus the extension of ( [ susyinf ] ) to sugra described in the earlier paper @xcite requires reconsideration . to preserve the properties of @xmath12 and its role as the inflaton , it should be uncharged @xcite . this is only possible when @xmath10 and the charges of @xmath13 are shifted to @xmath27 , where @xmath28 . now all the contributions to @xmath29 from any other sectors of the theory would also have to be charged . this would be for instance the case with the mssm superpotential . since they are , _ a priori _ , unrelated to inflation , this seems unnatural . hence we have an incompatibility between a non - vanishing @xmath9 and a fayet - iliopoulos term @xmath25 . in the usual inflationary scenario , inflation is driven by the fayet - iliopoulos term . one possibility is to introduce a fayet - iliopoulos term with a gauge invariant superpotential . this can be achieved in supergravity with a @xmath4-term by making the @xmath30 pseudo - anomalous . we extend the model to include one modulus @xmath31 and matter fields @xmath32 , all of which are charged under the pseudo - anomalous @xmath30 gauge group . the relevant part of the kahler potential for these fields is @xmath33 where the gauge field , @xmath34 has been included ( the usual kahler potential appearing in the scalar potential expression is obtained by putting @xmath35 formally ) . the potential arising from the @xmath4-term is then @xmath36 where we have taken the gauge coupling function to be @xmath37 in order to cancel the @xmath30 anomaly . we see that a fayet - iliopoulos term can give rise to inflation when @xmath10 , assuming that @xmath31 and @xmath32 can be stabilised by a suitable superpotential . during inflation , the waterfall fields @xmath13 vanish . this implies that the @xmath30 @xmath4-term and the @xmath9 dependent part of the @xmath0-term potential give rise to the inflation energy @xmath38 when @xmath10 and the extra fields are stable , this resembles the usual inflation energy in hybrid inflation . if on the other hand we just use the @xmath0-term to give inflation , there is no need to introduce the additional fields @xmath32 and @xmath31 . the fayet - iliopoulos @xmath4-term must then be zero , due to the gauge invariance of @xmath29 , so @xmath39 . we will take @xmath40 and assume this is the case for the rest of the paper . the string theory origin for @xmath9 has yet to be found @xcite . we are interested in the hybrid inflation scenario based on the superpotential above ( [ inflasec ] ) . we are also taking into account the effects of a separate sector , so the total superpotential is @xmath41 the @xmath42 are moduli and matter fields , which are usually assumed to be stable during inflation . with one modulus in @xmath0-theory no stabilisation was obtained @xcite . here we consider the most general case in supergravity with more than one moduli field and generic superpotential @xmath43 . the moduli dependent part of the superpotential @xmath43 is then responsible for stabilising the moduli fields . the total kahler potential has the form @xmath44 with @xmath45 we have imposed two shift symmetries in the @xmath46 sector to alleviate the @xmath2 problem of supergravity inflation . this is an extension to the usual @xmath12 shift symmetry @xcite@xcite . in string theory , this corresponds to the translational invariance of the brane system . for the full superpotential ( [ wfull ] ) , the @xmath0-term part of the potential can be expanded as @xmath47 \ , , \end{aligned}\ ] ] where @xmath48 denotes the real part of @xmath49 , and we have defined @xmath50 there is also a contribution to @xmath51 from the @xmath4-term @xmath52}\left(|\phi^+|^2 - |\phi^-|^2\right)^2 \ , .\ ] ] in this paper we will be particularly interested in kahler potentials with the no - scale property , such as @xmath53 in the kklt scenario . in this case the above expressions reduce to @xmath54 at the end of inflation , the charged fields @xmath13 condense and the inflaton @xmath12 is zero . the @xmath30 @xmath4-term part of the potential is then zero , and @xmath0-term part reduces to @xmath55 . this part of @xmath56 , which comes from the moduli dependent part of the superpotential @xmath43 , is then responsible for stabilising the moduli fields . for @xmath57 its minimum is an @xmath58 vacuum where supersymmetry is preserved , i.e. @xmath59 for all the fields . since we want a minkowski vacuum at the end of inflation , it is necessary to add an additional lifting term , @xmath60 to the potential . in the kklt scenario a non - supersymmetric lifting term is used @xmath61 where @xmath62 depending on the origin of the term , i.e. @xmath63 for anti d3-branes and @xmath64 for fluxes on @xmath65 branes . an alternative possibility is that the minimum is raised by an uplifting @xmath4-term @xcite . this allows one to lift the @xmath58 vacuum while avoiding the need for a non - supersymmetric potential . in this case the set of fields @xmath42 comprises one modulus @xmath11 and matter fields @xmath66 , all of which are charged under a second , pseudo - anomalous @xmath67 gauge group . the kahler potential in the moduli sector can be written as @xmath68 unlike the kklt case it is not no - scale , although it is straightforward to write down a no - scale version . the potential arising from the @xmath4-term for the additional @xmath67 is then @xmath69 if @xmath70 and all @xmath71 are positive , this potential will be nonzero and provide a suitable lifting term . as explained in @xcite , an @xmath58 vacuum in supergravity can not normally be lifted by @xmath4-terms . one can circumvent this argument by using non - analytic superpotentials with positive definite @xmath4-terms , as described in @xcite . the total moduli stabilisation potential is then @xmath72 where @xmath73 is given by , for example , ( [ nslift ] ) or ( [ dlift ] ) . after inflation , the potential possesses a minkowski vacuum thanks to the presence of the lifting term . for the rest of the paper we will only concentrate on the case where there are no additional matter fields @xmath66 . this implies that the gauge invariance of the superpotential can only be maintained if @xmath11 is gauge invariant , and we must use the non - supersymmetric lifting term ( [ nslift ] ) . this is the minimal setting to study hybrid inflation coupled to moduli . the structure of this potential is rather rich due to the possible interplay between the moduli and other fields . during hybrid inflation , the fields @xmath13 vanish , and the inflaton field is real , so @xmath74 . the full ( tree - level ) potential then reads @xmath75 where we have identified @xmath76 and @xmath77 in addition to the terms above , there will be contributions from loop corrections , which we will discuss in the next subsection . note that the inflaton is much smaller than the planck mass , i.e.@xmath78 is such that @xmath79 . during inflation , the minkowski vacuum is lifted by @xmath80 . we see that in general we will get large , tree - level contributions to the slope and mass of the inflaton @xcite ( from @xmath81 and @xmath82 respectively ) . in other words the model has @xmath1 and @xmath2 problems . this new @xmath2 problem comes from the inclusion of the moduli sector , and is distinct from the usual @xmath2 problem which comes from embedding inflation in supergravity . in the case of no - scale models @xmath83 , and so this @xmath2 problem can be removed by a symmetry of the kahler potential @xmath84 . to obtain a small inflaton slope , we will need to fine tune @xmath81 . the form of the potential discussed so far was based on the hybrid inflationary epoch , in which the fields @xmath85 vanish . inflation ends when the mass matrix acquires a tachyonic direction . this usually occurs when one of the two charged fields @xmath13 ( or a combination thereof ) has a negative mass , i.e. when @xmath86 . without moduli fields , this instability occurs at @xmath87 . however , the inclusion of the moduli sector modifies this result . for simplicity we take @xmath83 in what follows ( as is the case for a no - scale @xmath84 ) . working out the mass matrix for the @xmath13 sector , we find @xmath88 where we have assumed @xmath29 to be real . if @xmath89 , we recover the standard result @xmath87 . however , due to the presence of the moduli fields , the value of @xmath90 is altered . note that if we had used a more general @xmath23 , which did not have the shift symmetries , then @xmath90 would have received far more corrections . taking a canonical kahler potential , and neglecting the contributions of order @xmath81 in the regime @xmath91 , the lowest mass is given by @xmath92\ , .\ ] ] we see that if @xmath93 , inflation never ends as the mass is always positive . we naturally expect that @xmath29 will be much larger than the inflation scale @xmath9 in order to stabilise moduli during inflation . we see that both shift symmetries must be included in @xmath94 for inflation to be viable . in the above discussion we have considered how the moduli sector will alter inflation . of course we must not forget that the inflationary sector can also interfere with the moduli stabilisation , since the inflation potential @xmath80 acts as a perturbation of the moduli potential @xmath95 . in the kklt scenario the minimum of @xmath96 is separated from @xmath97 ( which also has @xmath98 ) by a potential barrier . denoting the height of the barrier by @xmath99 , we see that the value of the above potential ( [ inflapotential ] ) must be less than @xmath99 during inflation , or the modulus field @xmath11 will roll off to infinity , and inflation will not finish in the correct vacuum . the slope terms in the above potential must be small ( in order to have slow - roll inflation ) , so we need @xmath100 . this guarantees that the minimum of the moduli potential is only shifted during inflation and not completely destroyed . typically @xmath101 , where @xmath102 is the energy of the ads minimum of @xmath51 ( if @xmath73 were absent ) . this provides a rough estimate of the barrier height . in this case we need @xmath103 , which is reasonable if the inflation scale is far below the moduli stabilisation scale . the one - loop corrections to the effective potential for a general theory , with cut - off @xmath104 , are @xcite @xmath105 the supertrace is @xmath106 . expressions for the boson and fermion mass matrices are given in the appendix . as long as the masses are below the cut - off scale , the log corrections are smaller than the quadratic divergences . this is guaranteed provided the potential @xmath43 is lower than the cut - off , i.e. lower than the planck scale in practice . this is also a phenomenological requirement in order to obtain a hierarchy between the planck mass and the gravitino mass . we therefore concentrate on the quadratic divergences in what follows . we see there are corrections to all parts of the inflationary potential ( [ inflapotential ] ) . in particular we find @xmath107\right ) e^k \ , .\ ] ] in general , the corrections to the other parts of the potential are complicated , and not particularly illuminating . for simplicity we will explicitly consider the case for one modulus field only , i.e.@xmath108 . we will use @xmath109 . in the minkowski background @xmath110 during inflation we have instead @xmath111 we see that the loop corrections are @xmath12 dependent . note that we have included not only the contributions from @xmath13 , but also those from the moduli sector . for the kklt scenario , we find the tree - level potential to be ( after taking @xmath11 to be real ) @xmath112 where @xmath113 . adding on the loop corrections ( [ strmink ] ) gives the full potential for the minkowski background @xmath114 - \frac{w w'}{2t^2}\left[1+\frac{\lambda^2}{8 \pi^2}\right ] -\frac{w^2}{t^3 } \frac{\lambda^2}{32 \pi^2 } \\ \nonumber & & { } + \frac{e}{(2t)^n}\left[1+\frac{\lambda^2 n(n+1)}{48 \pi^2}\right ] \ , . \label{kkltvs}\end{aligned}\ ] ] adding the loop corrections will change the value of @xmath11 at the minimum . furthermore the minimum will no longer be minkowski . the parameter @xmath115 needs to be re - tuned to fix this . during inflation , i.e. @xmath12 non - vanishing , we get @xmath116 to which we need to add the loop corrections ( [ strinf ] ) . in this paper , we take the cut off at the planck scale . hence , as long as @xmath117 and @xmath29 do not vanish during inflation , the radiative corrections induce a slope for the inflationary potential , which is dependent on @xmath29 , and which can in principle be a danger to the flatness of the inflationary potential . this effect is only present in supergravity , as in global supersymmetry the quadratic divergences are @xmath12 independent . for one modulus field @xmath11 , the full potential during inflation now reads @xmath118 for @xmath11 real , the full moduli stabilisation potential is given above ( [ kkltvs ] ) . we denote the value of @xmath11 at the minimum of @xmath119 by @xmath120 . similarly @xmath121 - w \frac{\lambda^2}{4 \pi^2 } \label{kkltv1}\ ] ] @xmath122 as it is a no - scale model , the only contribution to @xmath123 comes from loop corrections . in this section we will discuss the form of the effective inflationary potential , taking into account the moduli stabilisation mechanism . it is usually assumed that moduli fields are stabilised before inflation begins . the energy scale of the corresponding potential is much higher than that of inflation , and so the effects of the inflaton will not significantly alter the values of the moduli fields . it therefore seems reasonable to assume the moduli are fixed during inflation . as we will show , this is not the case . let us now analyse the moduli stabilisation mechanism when the inflation sector is present . we denote the values of the moduli fields at the minimum of @xmath124 by @xmath125 , and define @xmath126 to be the deviation of the moduli fields from this value . expanding the inflationary potential around this minimum we find @xmath127 where @xmath128 are independent combinations of @xmath129 and @xmath130 which satisfy @xmath131 @xmath132 . minimising this with respect to @xmath128 , we find @xmath133 so we see that the moduli fields are now stabilised along a valley depending on @xmath12 . if @xmath134 ( which is perfectly reasonable ) , then @xmath135 , and the variation of the moduli fields is tiny , as is usually assumed . the correction to the inflationary potential is tiny too . however the slope of @xmath51 is also tiny , and so the contribution of the moduli fields can be comparable to the other effects we are considering . hence it is not reasonable to ignore their variation during inflation after all . now if we substitute the above expression ( [ yeq ] ) back into the potential ( [ vy ] ) , we obtain the effective inflationary potential @xmath136 with slope @xmath137 and effective mass @xmath138 \ , .\ ] ] we remind the reader that loop corrections are included in these expressions . hence , we see for a wide range of models ( including the kklt scenario ) that @xmath139 , and so the inflaton is tachyonic . this tachyonic instability is not necessarily a problem if @xmath12 is small , or the mass term is subdominant . since we need the field to roll towards zero during inflation we also require that @xmath140 is positive . it is tempting to use a moduli sector whose superpotential satisfies @xmath141 at the minimum , since @xmath142 in this case , and so we automatically avoid any large contributions to the slow - roll parameter @xmath1 . this type of model has other appealing features , such as a small gravitino mass @xcite . if we consider just one moduli field @xmath11 , and take a no - scale @xmath84 , then we see from ( [ kkltv2 ] ) that loop corrections give a small negative inflaton mass . the effects of the moduli variation will make it even more negative . even if this mass is small , it will still dominate the behaviour of @xmath12 , and cause it to roll away from zero . hence we see that a moduli potential with @xmath143 at its minimum ( i.e. unbroken supersymmetry ) can not be successfully combined with @xmath0-term hybrid inflation . we have established that during inflation the effective potential has the leading order form @xmath144 with @xmath140 positive and @xmath145 negative . the potential has the form of an inverted parabola . since we want inflation to end by the tachyonic instability of the @xmath146-field , @xmath12 has to roll towards zero . this means that during inflation @xmath147 . thus , an additional constraint on the model parameters is that @xmath148 , as given in ( [ phiend ] ) . to get an idea about the allowed parameter ranges for the theory , making only modest assumptions about the moduli sector , we provide some general constraints on potentials of the form ( [ workingmodel ] ) , leaving @xmath149 , @xmath140 and @xmath145 as free parameters , but with @xmath145 negative . note that @xmath149 , @xmath140 and @xmath145 depend on the underlying theory and are _ not _ independent . as shown in the previous section , they depend on the details of the moduli sector as well as the parameters @xmath7 and @xmath9 . an inflationary model has to satisfy certain constraints in order to agree with observations . firstly , in order to realise a period of slow - roll inflation , the two slow - roll parameters @xmath150 and @xmath151 have to be small ( we have approximated @xmath152 during inflation ) . also , in order to get the right amplitude for the density perturbations from inflation , the cobe normalisation must hold @xcite @xmath153 the current wmap+2dfgrs constraint on the spectral index is @xcite @xmath154 ( a weaker bound has been found in @xcite ) . in the case that @xmath155 , the above two constraints are fulfilled when @xmath156 and @xmath157 the constant @xmath140 can be eliminated from the last expression , using ( [ norm ] ) . since @xmath145 is negative and @xmath158 , this restricts the parameters to be in the region ( in planck units ) @xmath159 implying an upper bound for the inflationary scale of order @xmath160 gev . the allowed region in the @xmath161 plane is shown in figure 1 . [ ] [ ] [ 2]@xmath149 [ ] [ ] [ 2]@xmath145 the parameters @xmath80 , @xmath140 and @xmath145 depend on the details of the moduli superpotential as well as on @xmath7 and @xmath9 . in order to generate slow - roll inflation , we require that @xmath162 and @xmath163 note that if we had not included the contribution from the variation of the moduli fields , we would instead have @xmath164 . in general , the above contributions to @xmath2 from the moduli sector are likely to be large . we see this type of model may have another @xmath2 problem , which is not removed by the symmetries we used in section [ sec : v ] to avoid the other two @xmath2 problems . using ( [ v0con ] ) , which follows from the cobe and spectral index constraints ( [ cobe ] ) and ( [ nsconstr ] ) , we find that @xmath165 this is consistent with @xmath166 , as we have assumed ( and is required , for example , for @xmath96 to retain its minimum during inflation ) . finally , for the maximal allowed value of @xmath167 gev , the relative contribution of gravitational waves @xmath168 is 13 percent . however , it can be easily shown that the field variation @xmath169 during inflation , related to the e - fold number by @xmath170 is larger than @xmath171 ( with @xmath172 ) , as long as @xmath173 gev . since field variations below the planck scale are preferred theoretically , its very likely that the predicted amount of gravitational waves is much lower . here we study the stability of the modulus during inflation . in particular , the previous approximation of assuming that the moduli potential is quadratic when perturbed by the inflaton is only valid if the inflaton does not remove the potential barrier between the inflation minimum and infinity . we have to ensure that the term linear in @xmath12 in the potential is not too big , otherwise this term will completely destabilise the modulus @xmath11 . the condition for this to happen is that for at least 60 e - folds of inflation the term linear in @xmath12 in ( [ vinf ] ) is smaller than the barrier height , @xmath99 . typically @xmath174 which implies @xmath175 using ( [ nfold ] ) to eliminate @xmath169 from this equation , we get @xmath176 where we have used @xmath177 ( ignoring loop corrections ) , and have dropped numbers of order one . using the cobe normalisation , we get @xmath178 where @xmath179 . this is the condition guaranteeing that the modulus is not destabilised during inflation . it states that the gravitino mass is never small . in terms of the soft breaking terms and the sparticle spectrum obtained with a large gravitino mass , this implies that the scalar sparticles have very large masses . this is the type of scalar spectrum advocated in split susy models @xcite . of course if one is to observe scalars at the lhc , the scalars can not be very massive and the gravitino mass must be reduced . this requires a superpotential which gives a small gravitino mass , but which still produces a stabilisation potential with a high barrier . the model in subsection [ ssec : eg2 ] has these properties . it is even possible to have @xmath180 and still stabilise the moduli in a minkowski vacuum , as described in @xcite . in this case there is no lifting term @xmath73 , and the vacuum is supersymmetric . we could then obtain a gravitino mass from additional supersymmetry breaking effects . as we will show in section [ sec : nogo ] , if the vacuum is be minkowski , the susy - breaking must involve a lifting term . otherwise the vacuum will be unstable . as has already been noted , the parameters @xmath149 , @xmath140 and @xmath145 are not independent , but depend on details of the moduli sector as well as on @xmath7 and @xmath9 . as a simple example we will discuss the kklt scenario @xcite , which has just one modulus field @xmath11 and uses a non - supersymmetric lifting term ( [ nslift ] ) . during inflation the full potential is then given by ( [ kkltvs ] ) and ( [ vinf])([kkltv2 ] ) . the stabilisation potential s minimum is at @xmath181 , so @xmath182 . this allows us to eliminate @xmath115 , and one other parameter . following @xcite , we will use a racetrack superpotential @xmath183 where @xmath184 , @xmath185 , @xmath186 and @xmath187 are real constants . the above superpotential comes from gaugino condensation . we take @xmath188 and @xmath189 to guarantee the validity of the supergravity approximation and the weak gauge coupling limit . let us first focus on the kklt model obtained with @xmath190 and @xmath191 . to leading order this gives @xmath192 and @xmath193\ , , \qquad \tilde v_2 = -\frac{\lambda^2}{4 \pi^2 } \ , , \nonumber \\ & & ( e^k\tilde v_1)'\approx -\frac{a a^2}{4t_0 ^ 2 } e^{-at_0 } \left[1+\frac{\lambda^2}{8\pi^2}\right ] \ , , \qquad \lambda \approx \frac{a^2 a^4}{3t_0}e^{-2at_0 } \left[1+\frac{\lambda^2}{8\pi^2}\right ] \ , , \label{eg1}\end{aligned}\ ] ] where @xmath194 . during inflation we also require the slow roll parameter @xmath195 ^ 2 < 1 $ ] . we see above that @xmath196 , and so we need @xmath197 . we observe that @xmath123 is small , suggesting @xmath2 is small too . however there are also corrections to @xmath2 from the modulus field variation during inflation ( see section [ sec : mod ] ) , which can not be ignored . these imply @xmath198')^2}{e^k \lambda}\right ) \ , . \label{neweta}\ ] ] we see that for the above choice of @xmath0-term model , the slow - roll parameter is found to be negative , independent of the parameters , and large : @xmath199 . this value is not compatible with the slow - roll condition and , hence , not compatible with observational constraints . in general there could also be corrections to @xmath2 from variations of the imaginary part of @xmath11 . however for the examples in this section we find that the potential is always minimised at @xmath200 , even during inflation . another constraint on the model comes from requiring that the minimum of @xmath201 does not disappear during inflation . this would be the case if @xmath202 . for the above example we find that the maximum of the potential is at @xmath203 , with @xmath204/[n+(5n-6)\lambda^2/(24\pi^2)]$ ] . the barrier height is then @xmath205 \ , .\ ] ] we see that @xmath206 , as we assumed in the previous section . we need @xmath207 , implying @xmath208 . however this contradicts @xmath209 , which was required to have small @xmath1 . hence the moduli superpotential in this subsection fails on both slow - roll parameters . as a second case , consider the race - track models with @xmath210 , and take the limit @xmath211 . in this case we find @xmath212 and @xmath213 \ , , \qquad \tilde v_2 \approx -\frac{\lambda^2}{4 \pi^2 } \ , , \nonumber \\ & & ( e^k \tilde v_1 ) ' \approx -a(a - b ) e^{-a t_0 } \frac{a}{4t_0 ^ 2 } \left[1+\frac{\lambda^2}{8\pi^2}\right ] \ , , \nonumber \\ & & \lambda \approx a^2 ( a - b)^2 e^{-2at_0 } \frac{a^2}{3t_0 } \left[1+\frac{\lambda^2}{8\pi^2}\right ] \ , . \label{eg2}\end{aligned}\ ] ] small @xmath1 requires @xmath214 which , as in the previous subsection , implies @xmath215 . this suggests that it will again be impossible to have small @xmath1 and a stable modulus @xmath11 . using an asymptotic analysis of @xmath201 , we find that this time the maximum of the potential occurs at @xmath216 , and the height of the barrier is @xmath217 \ , .\ ] ] in contrast to the previous subsection we find @xmath99 is not of order @xmath218 . in this case it is possible to have @xmath219 and @xmath220 . unfortunately when we calculate @xmath2 , we run into the same problem as before . if the modulus field @xmath11 were fixed at @xmath120 , @xmath2 would be small . however this does not occur , and ( [ neweta ] ) applies instead . we find again @xmath199 , ruling out this case too . we will now use approximate analytic arguments to determine if any choices of @xmath29 could give viable slow - roll inflation . using the expressions for @xmath201 ( [ kkltvs ] ) , and @xmath221 ( [ kkltv1 ] ) , we obtain @xmath222 v_1 + \frac{(n-4)}{24 t^3 } v_1 ^ 2 + \mathcal{o}\left(\frac{\lambda^2}{8 \pi^2}\right ) = 0 \label{approx}\end{aligned}\ ] ] at @xmath181 . now the requirement that @xmath223 implies that @xmath224 , and since @xmath166 we can take @xmath225 to find approximate solutions to ( [ approx ] ) . noting that @xmath226 , the effect of the loop corrections will also be small , allowing us to neglect them . for small @xmath81 ( [ approx ] ) is only satisfied if either @xmath29 or @xmath227 is very small . for the former we obtain @xmath228 using this , the expression for @xmath229 simplifies to @xmath230 , to leading order in @xmath81 . from ( [ kkltv1 ] ) we obtain @xmath231 to leading order in @xmath81 this is @xmath232 . substituting all this into the expression for @xmath2 ( [ neweta ] ) , which includes the effect of varying @xmath11 , we find @xmath233')^2}{e^k \lambda } \approx -6 \ , .\ ] ] so we retrieve an @xmath2 problem along this branch of solutions . the models discussed in the previous two subsections fall into this category . the second approximate solution to ( [ approx ] ) has @xmath234 this again allows us to simplify @xmath235 , and we find @xmath236 . expanding in @xmath81 , ( [ dv1app ] ) implies @xmath237 . combining all this , ( [ neweta ] ) implies @xmath238 noting that @xmath239 in order to get successful inflation , we need both @xmath240 and @xmath241 at the minimum @xmath181 . if this fine - tuning is not satisfied then slow roll inflation does not occur . this fine - tuning of the moduli superpotential is in addition to the fine - tuning required for the stabilisation potential @xmath96 to have a minkowski minimum . furthermore , @xmath29 must also be such that the maximum of @xmath96 is greater than the inflationary scale ( so that minimum of the moduli potential does not disappear ) . this all adds up to a huge amount of fine - tuning , and the two - term racetrack potential used above ( [ racetwo ] ) does not seem to have enough parameters to satisfy all the conditions simultaneously . this suggests that a heavily fine - tuned , three - term @xmath29 would be needed to give viable inflation . it is hard to see how such a fine - tuned model could arise naturally . in this section we consider minkowski vacua with a non - vanishing gravitino mass . this is the traditional framework used in particle physics . at the end of inflation , the vacuum is a non - supersymmetric configuration with a vanishing cosmological constant . one would also like to impose @xmath242 in order to have a hierarchy between the planck scale and the sparticle masses . supersymmetry breaking could be achieved with @xmath0-terms , although as we will show this leads to an unstable vacuum . it is therefore necessary to include a lifting term in the theory , such as those discussed at the end of subsection [ ssec : vv ] . consider a theory with only one modulus field , no lifting term , and the following no - scale kahler potential @xmath243 and @xmath108 . it is convenient to transform the field according to @xmath244 in terms of the new variable @xmath49 , one obtains @xmath245 and @xmath246 the kahler transformation ( [ kahlertrafo ] ) preserves @xmath247 , which is the only relevant combination in supergravity . using the new variable @xmath49 one obtains @xmath248 where primes denote here @xmath249 . imposing that the potential has an extremum with a vanishing cosmological constant leads to the following relations @xmath250 and @xmath251 at the extremum when @xmath57 and @xmath252 , i.e. when supersymmetry is broken at the extremum . the extremum equations are also satisfied for @xmath253 . the mass at the extremum involves @xmath254 when @xmath57 and @xmath252 , the extremum equations give @xmath255 at the extremum @xmath256 . as @xmath257 is generically non - zero , this implies that the extremum is a saddle point where the direction along @xmath258 ( for @xmath257 chosen real positive ) is tachyonic . this is a generic property of no - scale models and implies that the modulus can not be stabilised with a vanishing potential and broken supersymmetry . on the contrary when supersymmetry is preserved , we find @xmath259 as long as @xmath227 does not vanish , guaranteeing the stability of the model @xcite . this argument can be generalised to an arbitrary number of fields in the no - scale case @xmath260 where the @xmath66 s are , for example , matter fields . indeed putting @xmath261 and collectively @xmath262 , the kahler superpotential becomes @xmath263 and the superpotential @xmath264 is turned into @xmath265 with these redefinitions , the scalar potential reads @xmath266 the vanishing of the cosmological constant at the extremum implies that @xmath267 the minimum equation reads @xmath268 when supersymmetry is broken . in the supersymmetric case , the minimum equations are @xmath269 . the stability of the extremum depends on the mass matrix which reads @xmath270 \ , .\end{aligned}\ ] ] using the extremum equation we find that @xmath271 in the supersymmetry breaking case . this mass matrix has nice algebraic properties . when @xmath272 has zero eigenvalues , the mass matrix @xmath273 vanishes in these directions . in the directions orthogonal to the zero eigenstates of @xmath272 , all the eigenvalues of @xmath273 are positive but one which vanishes along the eigenvector @xmath274 where @xmath275 . hence the mass matrix @xmath273 vanishes along @xmath276 and the zero eigenstates of @xmath272 . in the supersymmetric case , we find @xmath277 \ , .\ ] ] defining a vector space basis with @xmath278 where @xmath279 , and orthogonal vectors @xmath280 , we can use the sum rule @xmath281 to find that @xmath282 \ , .\end{aligned}\ ] ] which is a positive definite hermitian matrix as @xmath283 for any @xmath284 provided the matrix @xmath272 does not have zero eigenstates corroborating the results in @xcite . hence we have found that the mass matrix @xmath285 has zero eigenstates when supersymmetry is broken . now for a generic supersymmetry breaking model , the holomorphic mass matrix is such that @xmath286 . expanding along the direction @xmath287 where @xmath288 is the minimum leads to a potential @xmath289 notice that this leads to a tachyonic direction as before , implying that the supersymmetry breaking extremum with zero cosmological constant is not stable . hence we see that a supersymmetry breaking minkowski vacuum from a no - scale theory which does not use a lifting term can not be stable . since such a setup does not even stabilise the moduli fields , it is certainly not suitable to use in a hybrid inflation model . in this paper we considered @xmath0-term hybrid inflation in supergravity , when moduli fields are present . in the literature it is usually assumed that moduli fields are stabilised before the period of slow - roll inflation begins and that they therefore have no impact on the dynamics of the inflaton field . we have shown that this is not necessarily the case . firstly , the tree - level coupling of moduli and inflaton fields in supergravity induces a slope and , in general , a mass term in the potential for the inflaton field . on top of this , loop corrections from both the inflation and moduli sectors of theory contribute to the slope and curvature of the potential . a further problem comes from the evolution of the moduli fields during inflation . since they are stabilised by a steep potential , they are roughly constant during inflation . however , despite the small size of their variation , we have shown that they still give a significant contribution to the effective inflaton potential . this generally leads to a large and tachyonic mass for the inflaton . we find that this class of model has up to three different @xmath2 problems . firstly there is the usual one coming from embedding hybrid inflation in supergravity ( which is avoided with the help of shift symmetries ) . secondly there are tree - level contributions to the inflaton mass from the coupling of the inflaton to the moduli sector ( although these are absent for no - scale models ) . finally there is the negative contribution to the effective inflaton mass from the rolling of the moduli fields during inflation . it is not obvious that this third @xmath2 problem can be removed with natural additional symmetries . the model also has an @xmath1 problem due to the tree - level contributions to the inflaton slope from the moduli sector . we found that in the popular model for moduli stabilisation , the kklt scenario , the induced mass term for the inflaton field is generally too big to give a period of slow - roll inflation . it seems that only with a heavily fine - tuned superpotential for the moduli sector may viable slow - roll inflation take place . to conclude , we have shown that the coupling of moduli fields to hybrid inflation can have an important impact on the effective inflationary potential . neglecting this coupling is not consistent and , as we have seen , generally leads to incorrect conclusions about the inflationary dynamics . these considerations are also relevant to other inflationary scenarios . in each case it should be checked that the moduli fields , even if thought to be stabilised , do not spoil the flatness of the inflationary potential , and therefore the dynamics of the inflaton field . we would like to thank k. dasgupta and r. kallosh for comments on a previous version of the manuscript and n. chatillon and u. ellwanger for discussions . acd thanks cea saclay for hospitality while some of this work was in progress . acd and cvdb were supported in part by pparc . scd was supported by the swiss science foundation and the netherlands organisation for scientific research ( nwo ) . phb acknowledges support from rtn european programme mrn - ct-2004 - 503369 . we have latexmath:[$m^2_{\rm ( boson ) } = 2 \mathcal{k}^{i\bar \jmath } \partial_i \partial_{\bar \jmath } v + 6f_r^{-1 } \sum_{i=\pm } assuming the only gauge group in the theory is @xmath30 . if there were additional gauge groups , the additional gauge sector would also give contributions to the mass matrices . for the fermions @xmath291 where @xmath292 with @xmath293 , and @xmath294 where @xmath295 , and @xmath296 otherwise . define @xmath297 \ , .\ ] ] the diagonal entries of the boson mass matrix for the minkowski background are @xmath298 and non - zero elements of fermion mass matrix are @xmath299 hence we obtain @xmath300 for the inflationary background , the diagonal entries of the boson mass matrix are @xmath301e^k \ , , \ \nonumber \\ m^2_{\pm \bar \pm } & = & [ 2(g^2x^4 + g^2[2x^4 + 2x^2 + 1]\phi^2 -\sqrt{2}g \phi \re(w)[1 + 2x^2 ] + |w|^2 ) \nonumber \\ & & \hspace{1 in } { } + v_s - 2\sqrt{2}gx^2 \phi \re ( v_1)+2 g^2x^4\phi^2 v_2]e^k \ , , \ m^2_{i \bar j } & = & 2g^2x^4 e^k ( k_{i\bar j}+k_i k_{\bar j } ) + ( e^k v_s + v_\mathrm{lift})_{,i\bar j } \nonumber \\ & & \hspace{1 in } { } - 2\sqrt{2}gx^2 \phi ( e^k \re v_1)_{,i\bar j } + 2 g^2 x^4 \phi^2 ( e^k v_2)_{,i\bar j } \ , , \end{aligned}\ ] ] while fermion mass matrix has @xmath302 - w ) \ , , \nonumber \\ m_{\phi j } & = & -e^{k/2}\sqrt{2 } g x^2 k_j \ , , \qquad m_{\lambda\lambda } = e^{k/2 } \frac{\partial_i f k^{i\bar j}}{2f_r } ( d_{\bar j } \bar w-\sqrt{2}g x^2 \phi k_{\bar j } ) \ , , \nonumber \\ m_{ij } & = & m^{(0)}_{ij } -\sqrt{2 } g x^2\phi m^{(1)}_{ij } \ , , \qquad m_{3/2 } = e^{k/2 } ( w-\sqrt{2}g x^2 \phi ) \ , .\end{aligned}\ ] ] this implies @xmath303g^2x^4e^k \nonumber \\ & & { } - 4\sqrt{2}g x^2\phi \re \biggl[(w+3 v_1 ) e^k + k^{i\bar j } ( e^k v_1)_{,i\bar j } \nonumber \\ & & \hspace{0.3 in } { } - m^{(0)}_{ij}k^{i\bar a } \bar m^{(1)}_{\bar a\bar b}k^{j\bar b } -\frac{e^k}{4 f_r^2}(k^{i\bar j } k_{\bar j } \partial_i f ) ( k^{b\bar a}d_b w \partial_{\bar a } \bar f ) \biggr ] \nonumber \\ & & { } + 4g^2x^4\phi^2 \biggl[(1+v_2 ) e^k + k^{i\bar j}(e^k v_2)_{,i\bar j } \nonumber \\ & & \hspace{1 in } { } - |m^{(1)}_{ik}k^{k\bar j}|^2 -\frac{e^k}{4f_r^2}|k^{i\bar j } k_{\bar j } \partial_i f|^2\biggr ] \ , .\end{aligned}\ ] ] 99 a. d. linde , particle physics and inflationary cosmology , contemp . concepts phys . * 5 * ( 2005 ) 1 [ hep - th/0503203 ] .

the gas and dust that is ejected by dying stars and from which new stars will form , is constantly being processed chemically as well as physically ( _ e.g. _ , * ? ? ? * ) . a particularly rich and diverse route is followed by carbon . in the cool surroundings of carbon stars evolved stars that have dredged up large amounts of carbon to the stellar surface more than 60 individual molecular species have been identified , including benzene , polyynes and cyanopolyynes @xcite . in addition , the spectra reveal diverse dusty materials such as hydrogenated amorphous carbon ( hac or a - c : h ) and silicon carbide ( sic ) . as the mass - loss process strips the outer layers of the star , a hot ( @xmath0 k ) central white dwarf becomes exposed , whose strong uv irradiation further processes the ejecta and makes them visible as a planetary nebula ( pn ) . a key component of the carbon dust inventory in pne are polycyclic aromatic hydrocarbons ( pahs ) and pah - like species , a class of large and hardy carbonaceous species whose formation mechanisms are unclear @xcite . although not a single pah member has been identified in space , their characteristic spectral features at infrared wavelengths are observed throughout the universe , from which it is inferred that they reside ubiquitously in space and could lock up as much as 15% of the cosmic carbon ( _ e.g. _ , * ? ? ? two other large aromatic species have been recently identified in space : c@xmath1 and c@xmath2 . these are the best - known members of the family of fullerenes , a class of molecules made of hexagonal and pentagonal aromatic carbon rings , fused in the shape of a hollow sphere or ellipsoid . the most abundant of these molecules , the buckminsterfullerene c@xmath3 , has the structure of an old - fashioned black and white soccer ball . the near - spherical carbon configuration of these two molecules , which have a closed surface , a closed - shell electron distribution and an almost unstrained network , results in a very high stability against dissociation and prolonged exposure to high temperature . fullerenes , in particular c@xmath3 , have peculiar and appealing photochemical , electrochemical and physical properties , which can be exploited in various fields , from nanotechnology to medicine , to space science . c@xmath3 and c@xmath4 were discovered during laser ablation experiments on graphite targets , aiming to study long carbon chains in interstellar clouds @xcite . because of their remarkable stability , fullerenes appeared particularly suited to survive the harsh conditions of the interstellar medium . their unique properties indicate that they may play an important role in organic astrochemistry and astrobiology . cosmic fullerenes remained elusive until the recent discovery of c@xmath3 and c@xmath4 in the planetary nebula tc 1 by @xcite . fullerenes have since been confirmed in many more evolved star environments @xcite , as well as in young stellar objects @xcite , reflection nebulae ( rne ) @xcite and photodissociation regions @xcite . fullerenes can be efficiently synthesized in the laboratory by the vaporization of carbon rods in an electric arc @xcite and from hydrocarbon combustion under optimised conditions @xcite . however , the formation routes of fullerenes in space are still unknown . we review the currently known formation mechanisms of fullerenes in the laboratory ( [ earth_sec ] ) , showing why these methods would not work in space ( [ not_working_sec ] ) . we propose an alternative pathway , consistent with astrophysical conditions , based on the photo - processing of a family of carbonaceous species which we name `` arophatics '' ( [ our_mechanism_sec ] ) . we discuss the observations of fullerenes in pne ( [ observations_sec ] ) and finally we summarize our conclusions ( [ conclusions_sec ] ) . the original discovery of c@xmath3 and c@xmath4 dates back to 1985 with the experiment of @xcite . in this experiment , carbon species were vaporized from a graphite target into a he flow with tunable pressure / density . the vaporization was done by a pulsed laser , the resulting carbon clusters were expanded in a supersonic molecular beam , photoionized and their masses measured by a time - of - flight mass spectrometer . vaporization at low he pressure ( less than 10 torr ) leads to a broad distribution of clusters , with 38 to 120 carbon atoms ( always even numbers ) , with c@xmath3 and c@xmath4 present but not the dominant species . at a pressure of 760 torr the c@xmath3 and c@xmath4 peaks clearly dominated . when the expansion of the he+carbon cluster mixture was delayed , the resulting mass distribution was completely dominated by c@xmath3 - 50% of total large cluster abundance , and c@xmath5 , with 5 % of total large cluster abundance . these distributions have been interpreted as due to the increasing number of he - cluster and cluster - cluster collisions , resulting in the `` survival of the fittest '' , i.e. the more stable clusters . the famous icosahedral soccer ball structure was proposed for the first time for c@xmath6 , and later confirmed by spectroscopic studies @xcite . however , the elementary reaction mechanisms occurring during the `` thermalization '' process remained unknown . @xcite experimentally studied the gas - phase formation of carbonaceous compounds . their first experiment was essentially a replication of the kroto experiment , i.e. laser vaporization of a graphite target in a quenching atmosphere of he or he / h@xmath7 with pressures between 3.3 and 26.7 mbar . the vibrational temperature of the laser - induced plasma was estimated to be between 4000 and 6000 k for the laser power densities used . in a second set of experiments , the laser - induced pyrolysis of c@xmath7h@xmath7 and c@xmath8h@xmath8 using a pulsed laser with high power densities was studied . the resulting condensation temperatures were above 3500 k , comparable with the vaporization experiment . in a third set of experiments , laser - induced pyrolysis of the same hydrocarbon precursors was carried out , but this time at a much lower temperature ( max 1700 k ) . the analysis of the condensation products showed a striking effect of the temperature . during the high temperature experiments , only fullerene - like soot and fullerenes were produced . during the low temperature experiments , only soot and pahs were formed ( 100 % pahs at @xmath9 1000 k ) . the results clearly tell that the temperature determines which kind of condensates will be formed . moreover , the two pathways seem to be mutually exclusive , at least under these experimental conditions : fullerenes and pahs can not be formed together . it is important to note that high pressure has been used during the experiments to concentrate the condensation within a small volume , i.e. to mantain a high density of the condensing species . the laboratory experiments described in the previous sections do not tell us how fullerene is formed from graphite and hydrocarbons , but only under which conditions this happens . the elementary reaction mechanisms involved in the formation of fullerenes can be investigated performing quantum chemical molecular dynamics simulations @xcite . in one set of simulations a hot carbon vapor was reproduced by putting 40 randomly oriented c@xmath7 molecules in a tiny 20 cubic box @xcite . the system was let to evolve , and extra c@xmath7 molecules were added at fixed times in order to simulate an open environment . during the first stage , the system was kept at a constant temperature of 2000 k and giant carbon cages self - assembled . during the second stage of the simulation the temperature was raised to 3000 k , and the shrinking of these hot giant fullerenes down to c@xmath10 was observed @xcite . the formation of the giant cages starts with nucleation of polycyclic structures ( hexagons and pentagons ) from entangled polyyne chains . this is followed by growth of the structure and finally cage closure similar to the self - capping of open - ended single - walled nanotubes . because of the rapid gain of energy due to cage closure , the giant fullerenic cages are produced in a vibrationally highly excited state . the excess energy has to be dissipated , either by unimolecular dissociation or collision with other carbon clusters or carrier gas atoms . the simulations show that the newly produced giant fullerenes inevitably shrink to smaller sizes . all the road maps to fullerene formation proposed by previous models were associated with intermediate structures that are in thermodynamic equilibrium , while a hot carbon vapor is indeed a system far from thermodynamic equilibrium . in the model developed by irle , morokuma and collaborators , for the first time the dynamic self - assembly of fullerene molecules occurs as an irreversible process emerging naturally under the nonequilibrium conditions typical of a hot carbon vapor . the experimental results of @xcite have presented evidence of the spontaneous tendency of electron - irradiated graphite to include pentagons in its hexagonal network , hence form curved structures . this has been further confirmed by @xcite , who have shown that fullerenes can form directly from graphene in a similar fashion . in the experiment of chuvilin and co - workers , a sheet of graphene was exposed to a beam of energetic 80 kev electrons . the energy transferred to the graphene results in the loss of carbon atoms from the edges of the sheet . if the size of the graphene flake is not too big ( less than a few hundreds of carbon atoms ) , the loss of carbon atoms at the edge will lead to the formation of pentagons , which triggers the curving of graphene into a bowl - shaped structure . further carbon removal from the edges using the electron beam will reduce the size of the curved structure until it is sufficiently small to zip up its open edges and isomerize into a closed fullerene structure . @xcite claim having identified the fundamental processes leading to the formation of fullerenes in a recent experimental work , based on laser vaporization of pure c@xmath3 in the presence of carbon vapor . according to this study , fullerenes would self - assemble bottom - up through a closed network growth ( cng ) mechanism based on the ingestion of atomic carbon and c@xmath7 clusters . it should be noted that the work from @xcite provides experimental evidence of the growth of larger fullerenes from pre - existing c@xmath3 only . because the experiments show that the cng of larger fullerenes does not result in the production of c@xmath6 , it is deduced that c@xmath3 formation must be a result of cng from smaller fullerenes . however , the initial formation mechanism of such smaller fullerenes is still under debate . although all the chemical ingredients and the required temperatures for graphite vaporization and hydrocarbon combustion can be found in astrophysical environments , the densities are far too low to proceed to fullerene formation on reasonable timescales . in space , the most favourable conditions for fullerene formation via vaporization / combustion are found in a post - shock gas . in such environments the fullerene building blocks , i.e. c@xmath11 groups and polycyclic species coming from the fragmentation of pahs @xcite , ought to exist in the required vibrationally excited states . however , the post - shock carbon densities are low and represent a serious obstacle to fullerene formation . we have derived the scaling rule ( eq . [ scaling_eq ] ) relating the time @xmath12 , required for fullerene condensation , with the initial density @xmath13 of the carbon gas . the details of the calculation are reported in the appendix . this scaling rule has then been used to estimate the time necessary for the coagulation of fullerenes from a carbon gas with interstellar densities ( @xmath14 10@xmath15 carbon @xmath16 ) . the scaling rule is : @xmath17 where @xmath18 and @xmath19 are the condensation timescales corresponding to densities @xmath20 and @xmath21 respectively . knowing the density and timescale in simulations , @xmath20 and @xmath18 , and the density of the interstellar gas , @xmath21 , we can then calculate the fullerene condensation time in space , @xmath19 . we adopt @xmath20 = 10@xmath22 carbon @xmath16 and @xmath18 = 50 ps @xcite . the hydrogen density of the post - shock gas is @xmath9 1 @xmath16 , but carbon is 10@xmath15 less abundant , so @xmath21 = 10@xmath15 carbon @xmath16 . from eq . [ scaling_eq ] we obtain @xmath19 = 1.6 @xmath23 10@xmath24 yr , which is longer than the age of the universe . we also estimate the carbon density @xmath21 required to form fullerenes in a shocked gas before the temperature drops below 1000 k. at this temperature the condensation of fullerenes can not occur . for a typical interstellar shock with a velocity of 125 km s@xmath25 the cooling time ( after which the temperature of the post - shock gas drops below 1000 k ) is about 2@xmath2310@xmath26 yr . under these conditions , the required carbon atom density for the condensation of fullerenes is @xmath21 = 3@xmath2310@xmath27 carbon @xmath16 , and in space such densities do not exist outside of stellar atmospheres . @xcite propose a scheme for the direct formation of c@xmath6 from pahs in the reflection nebula ngc 7023 based on graphene dissociation experiments from @xcite . in ngc 7023 fullerenes and pahs appear to coexist in the same location close to the central star , while only pahs are detected further away from the star . according to @xcite , a similar phenomenon to the one described by @xcite occurs in ngc 7023 , but induced by photo processing instead of electron bombardment . the absorption of uv photons from the central star first completely dehydrogenates the pahs @xcite , producing graphene flakes . further photon absorption induces the loss of carbons from the edges of the graphene sheet . this , according to @xcite , gives rise to pentagonal defects which will induce the curvature of the sheet and the subsequent closure of the cage to form a c@xmath3 molecule . there are some issues related to the applicability of the results from such experiments to interstellar conditions . first of all , it is not possible in space to tune the number of carbons removed from the graphene flake in order to get the right size required for cage closure . the number of ejected carbons depends on the rather slow photodissociation process , which is governed by the dissociation parameter @xmath28 . this parameter describes the dissociation rate of a highly excited pah molecule using an arrhenius law @xcite , and to a first approximation can be seen as the energy that needs to be concentrated over a bond in order to break it and release a fragment . the dissociation probability ( and hence the dissociation rate ) decreases for increasing @xmath28 because more energy is required in the bond that has to be broken . the value of @xmath28 has been derived from experimental data only for very small pahs ( up to 24 c - atoms ) with a very open carbon skeleton . the extrapolation of such results to larger ( @xmath9 50 c - atoms ) and compact astrophysically relevant pahs dissociating under interstellar conditions is very uncertain , leading to values of @xmath28 ranging from 3.65 to 5.6 ev @xcite . this corresponds to values for the dissociation probability going from 0.5 to 3@xmath2310@xmath29 respectively . to form fullerenes during the lifetime of ngc 7023 , the mechanism proposed by @xcite needs a specific tuning of the parameters involved . bacause the size of pahs can not be controlled as during experiments , the number of carbons in the precursor molecule is severely constrained to about 70 atoms , very close to the 60 carbons required to form c@xmath3 . more importantly , the most favourable conditions for dissociation are chosen , i.e. , the lowest value for the dissociation parameter , @xmath28 = 3.65 ev . however , this choice does not appear physically motivated . indeed , if another legitimate value is adopted for the dissociation parameter , e.g. , @xmath28 = 4.6 ev ( the value falling in the middle of the range determined for @xmath28 see * ? ? ? * ) , the decrease of the dissociation rate is such that the fragmentation process is far too slow and the conclusion suggested in @xcite no longer holds : pahs with 70 carbon atoms require longer than the lifetime of ngc 7023 to fragment , therefore they can not be the precursors of the c@xmath3 observed at the present epoch . thus , none of these mechanisms are feasible in the astrophysical settings where fullerenes have been detected , and hence , alternative routes are needed . trace amounts of fullerenes have been identified among the dissociation products of uv - irradiated hacs @xcite . together with fullerene emission , the three pne tc 1 , smp lmc 56 and smp smc 16 show broad emission plateaus between 6 9 @xmath30 m and between 10 13 @xmath30 m , compatible with emission from h - rich amorphous carbon nanoparticles @xcite . this suggests that a viable formation route in space could exist that starts from hac processing ( _ e.g. _ , * ? ? ? * ; * ? ? ? in this study we focus on the formation of fullerenes in the circumstellar environments of proto - planetary nebulae ( ppne ) and pne . hacs are an abundant carbonaceous dust component in both these types of objects @xcite . in addition , one of the main characteristics of pne is their strong radiation field , which will process the hacs . in their experiments , @xcite and @xcite showed that hac / a - c : h decomposition leads to a low density material ( @xmath31 g @xmath16 ) . in particular , @xcite and @xcite showed that the decomposition end - product is an aerogel - like material consisting of weakly - connected aromatic `` proto - graphitic '' or pah - like clusters in a `` friable network '' . the dehydrogenation of small carbonaceous particles , containing less than a few hundred c atoms , will likely lead to structures that resemble the particles seen by @xcite in their experiments and/or the locally aromatic polycyclic hydrocarbons ( laphs ) proposed by @xcite . the particles in the @xcite experiments are found to be small alkanes , unsaturated carbon - chain radicals and small dehydrogenated pah - like species . after the release of these lighter molecules @xcite observed a strong mass peak near 500amu ( @xmath32 c atom , proto - graphitic , aromatic clusters ) having ir spectra similar to that of hac / a - c(:h ) with absorption and emission features at 3.3 , 3.4 and @xmath33 m . the results of these experiments and recent modelling @xcite therefore imply an intimate mix of aromatic clusters and aliphatic groups in small carbonaceous species . we note that the laph structures studied by @xcite have compositions in the range c@xmath34h@xmath35 to c@xmath36h@xmath37 and contain aromatic domains with only a few rings per domain , which are linked by bridging aliphatic ring systems , and should have ir spectral properties similar to hac / a - c(:h ) materials ( _ e.g. _ , * ? ? ? * ; * ? ? ? * ; * ? ? ? the conditions required for hac dehydrogenation are common in space , we can thus expect the formation of laph - like species where the aromatic components are bridged by olefinic carbon - containing rings and chains . the key component in these structures is the presence of the aliphatic bridging groups linking the aromatic clusters , in what we here call `` arophatic '' structures . such structures , analogous to an aliphatic periphery on pah molecules ( _ e.g. _ , * ? ? ? * ) , could provide sites for molecular hydrogen formation in pdrs @xcite . size - dependent @xmath38 band considerations indicate that the intrinsic aromatic clusters are indeed most likely to consist of ` isolated ' two- and three - ring systems in such small hydrocarbon particles @xcite . the non - aromatic carbon atoms are in the form of `` more flexible '' short olefinic and/or aliphatic bridging structures such as ch = ch and ch@xmath7ch@xmath7. this generic type of structure was also proposed by @xcite , who suggested that aliphatic cyclo - hydrocarbons are the precursors to aromatic cluster formation in hac materials . this kind of configuration , with aliphatic chains attached to aromatic sites , is also found in aromatic - aliphatic linked hydrocarbon ( aalh ) structures , which appear to have an important role in the nucleation of nanoparticles in soot formation in flames @xcite . it should be noted that aalhs are small and relatively simple molecules , such as 1,2-di(naphthalen- 2-yl)ethane ( dne ) , while our proposed `` arophatic '' clusters are bigger and more complex structures . the aliphatic bridging structures must be quasi - orthogonal to the aromatic species that they bridge and will be unable to transform into aromatic structures without compromising ( _ i.e. _ , destroying ) the entire structure . these active sites could , likely , interchange between @xmath39 olefinic and radical @xmath40 states with dangling bonds , in a kind of isomerisation `` flipping '' process , which could play a key role in cage closure and fullerene formation in the solid state . the work of @xcite indicates that small clusters within a - c(:h ) materials are chain - like , while the larger clusters are three - dimensional , cage - like , @xmath39 structures and as @xcite show these cage - like structures have a natural tendency to close - up and form fullerene - like and fullerene structures . the `` arophatic '' clusters are structures evolving from larger , amorphous hydrocarbon particles and , from a topological point of view , they will inevitably end up with a tube - like , cup - like or cage - like appearance ( _ e.g. _ , see the two right hand species in fig . 1 of * ? ? ? such structures are ` curled - up ' or ` folded - over ' , defected graphite ( dg ) or graphene sheets @xcite , and we can therefore represent them by their ` flattened - out ' equivalents as per fig . 12 in @xcite but with aromatic domains reduced in size to only a few rings per aromatic cluster ( fig . [ fig_arophatics_0 ] ) . [ b ] the aromatisation of the aliphatic bridges to form pentagons can provide a viable route to fullerene formation in the solid phase via these `` arophatic '' species . the key process is the dehydrogenation of the aliphatic and olefinic `` bridges '' which naturally leads to aromatic pentagon formation , as shown in fig . [ fig_schemes ] for small a - c : h particle fragments . it appears that there is a link between small carbonaceous particle dehydrogenation in intense uv radiation fields and the formation of fullerenes , which could explain the observation of fullerenes in circumstellar regions associated with small hydrocarbon grains ( _ e.g. _ , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the mechanism that we propose for fullerene formation in space is the following . the starting point is large , nm - sized , h - rich a - c : h particles containing more than a hundred carbon atoms . uv photolysis by photons with energy above 10 ev dehydrogenates these grains @xcite and leads to further changes in the molecular structure a process that we call * stir * ( * s*tructural * t*ransformations in * i*ntense * r*adiation fields ) . stir processing requires constant heating ( by photon absorption ) to keep the particle in a vibrationally excited state that promotes structural transformations leading to the closing of cage - like structures within the a - c : h nanoparticle . amorphous carbon nanoparticles can be maintained at thermal - equilibrium temperatures of the order of 100 - 150 k by photon absorption ( for h - rich a - c : h particles with band gaps larger than 1 ev ) even at thousands of astronomical units from the central stars of pne , due to their low emissivity at the long wavelengths where they radiate thermally @xcite . however , at these same distances , almost completely dehydrogenated nanoparticles have temperatures of only tens of kelvin , as expected . in order to form fullerenes , the arophatic clusters resulting from stir processing must be large , containing more than a hundred carbon atoms , as exemplified in figs . [ fig_arophatics_2 ] and [ fig_movie2_shot ] . a key step in the route to fullerenes is the formation of 5-membered rings . this can occur via dehydrogenation during the stir processing of the parent a - c : h or of the emerging arophatic cluster , in a similar fashion as shown in fig . [ fig_schemes ] for small a - c : h particles . dehydrogenation is a viable way to introduce 5-membered rings , as indicated by @xcite . this work shows that structures with a high degree of curvature result from dehydrogenation reactions occurring in aromatic hydrocarbon compounds . these reactions involve the loss of hydrogen with rearrangement of the structure and ring closure which leads to five - membered rings , and hence curvature . uv irradiation of hac nanoparticles not only causes dehydrogenation ( * ? ? ? * and references therein ) , but also forces the structure to curl up because of the introduction of pentagonal rings . in fact , the uv - induced dehydrogenation _ is _ the cause of the formation of 5-membered rings . thus , in a single step , dehydrogenation fulfills two essential conditions for the formation of fullerene molecules : 1 ) removing hydrogen atoms and 2 ) triggering pentagonal ring formation in the evolving structure . h@xmath41 ) , precursor of cosmic fullerenes . carbon is represented in dark grey , hydrogen in light gray . the color version of this figure ( carbon in red , hydrogen in gray ) is available in the electronic edition of the _ astrophysical journal_. ] h@xmath41 , precursor of cosmic fullerenes and schematically represented in fig . [ fig_arophatics_2 ] . carbon is in dark grey , hydrogen in light gray . the movie , with carbon in red and hydrogen in gray , is available in the electronic edition of the _ astrophysical journal_. ] the fullerene precursor is a big and complex arophatic cluster , with the aromatic domains being non - co - planar . the 3-d structure favors the formation of 5-membered aromatic rings , and then curvature , even in such a big structure , via photo - induced dehydrogenation . the particle emerging from the dehydrogenation process is a giant fullerenic cage similar to the ones formed in simulations @xcite . the initial , stir - driven ( catastrophic ) photolysis and the following cage closure induced by the formation of pentagonal rings will result in a highly excited structure . the cage will release the excess energy via ejection of c@xmath42 molecules , shrinking down to the islands of stability represented by c@xmath3 and c@xmath4 @xcite . the down - sizing occurs through unimolecular decomposition of the vibrationally excited cage , resulting in the ejection of carbon molecules . to model the unimolecular dissociation we adopt an arrhenius law @xcite , which describes the evolution of the dissociation rate as a function of the temperature of the dissociating species : @xmath43 the quantity @xmath44 is the rate for carbon ejection measured in carbon atom lost per second ( c / s ) , @xmath45 the vibrational temperature of the fullerenic cage and @xmath46 the boltzmann constant . the unimolecular decomposition is governed by two parameters : the pre - exponential factor @xmath47 and the dissociation energy @xmath28 . to estimate the value of these two parameters , we use two sets of simulations which follow the shrinking of giant fullerenic cages at two different temperatures , 2000 and 3000 k , and provide the corresponding dissociation rates . the average dissociation rate @xmath44 is 1.5@xmath2310@xmath27 c / s and 3@xmath2310@xmath27 c / s for @xmath45 = 2000 k and @xmath45 = 3000 k respectively @xcite . we obtain the following values for @xmath47 and @xmath28 : @xmath48 the time @xmath49 required for a big fullerenic cage to shrink down to c@xmath1 is given by : @xmath50 where @xmath51 is the number of carbons that have to be ejected from the cage and @xmath44 is the ejection rate from eq . [ eq : arrhenius_generic ] . we consider a fullerenic cage containing 144 carbon atoms , i.e. with size comparable to the size of the particles considered in simulations @xcite , therefore , @xmath51 = 84 . the shrinking timescales for various temperatures of the initial cage are reported in table [ tab : shrink_time ] . .[tab : shrink_time ] timescale , @xmath49 , for shrinking of the fullerenic cage c@xmath52 to c@xmath1 . [ cols="^ , < , < " , ] this process requires relatively short time scales : to shrink a large ( 144 carbon atoms ) fullerene cage to c@xmath1 requires only about 0.8 s at 200 k , and increases to 27 years at 100 k. to the best of our knowledge , our extrapolation is the first attempt to characterize the shrinking of a giant fullerenic cage under interstellar conditions ( e.g. , very low temperatures with respect to the usual laboratory conditions ) . the value of 0.36 ev that we have calculated for @xmath28 should be considered as a lower limit , having been determined under the most favourable conditions for fullerene formation on earth , as indicated by simulations ( nevertheless the optimized conditions for industrial production of fullerene are quite similar ) . the issue that we are facing here is another example of the well known problem related to the extrapolation to extraterrestrial conditions of the results from statistical unimolecular dissociation theories @xcite . the proper evaluation of @xmath28 and of the uncertainty associated with it requires the development of a specific model which will be the subject of a following paper . our present scope is to provide a first indication of the timescales associated with the shrinking process in space . in the case of pah dissociation under extraterrestrial conditions , the detailed evaluation of the dissociation parameter @xmath28 resulted in a factor of 10 uncertainty on the calculated fragmentation timescales @xcite . because of the similarities between large pahs and fullerenes @xcite , we can consider a @xmath53 uncertainty for the fullerene shrinking timescale as a very conservative assumption . we obtain then a maximum shrinking time of 2670 years at @xmath45=100 k ( the least favourable condition ) . this is fully consistent with the age of young pne where fullerenes have been observed . c@xmath3 and c@xmath4 are the only , and therefore smallest , c@xmath54 fullerenes with @xmath55 70 satisfying the isolated pentagon rule ( ipr ) . this implies ( almost ) spherical , thermodynamically favored , closed shell structures particularly suited to survive harsh conditions . the shrinking stops at c@xmath1 because of the very high arrhenius dissociation energy @xmath28 for c@xmath11 elimination to c@xmath56 , @xmath57 10 ev @xcite . the estimated temperatures of a few hundreds of kelvin for fullerenes detected in planetary nebulae @xcite are far from the approximately 1000 k required to overcome this decomposition barrier @xcite . our model proposes the formation of a single c@xmath3 ( c@xmath5 ) molecule from a single large arophatic cluster . this implies that the parent a - c : h particles must be nm - sized , and therefore contain a hundred or more carbon atoms , because of the dissociative loss of carbon ( and hydrogen ) atoms that will occur during structural transformation . note that the large arophatic clusters can be further fragmented by the same radiation field that caused their formation from the parent a - c : h grain . most of the formed arophatics will fragment into structures other than fullerenes , resulting in a fullerene formation process that is rather inefficient , consistent with the low fullerene abundance observed in space so far . the peculiarities of the pne where fullerenes have been detected are their intense uv radiation fields , sufficient to sustain photo - chemical processing of hacs , and the ablation of carbonaceous material from the dense circumstellar matter formed during their agb phase . thus , in these objects abundant , nm - sized , h - rich , a - c : h dust is being liberated and undergoing catastrophic uv - photolysis , which leads to stir processing . in tc 1 , c@xmath3 and c@xmath4 have been detected far for the central star ( 8000 au * ? ? ? * ) , but even at this distance the uv irradiation is sufficient to keep the particles in a vibrationally excited state ( @xmath58 100 - 130 k ) and to trigger and sustain the stir processing of the a - c : h dust and the shrinking of the resulting giant fullerenic cages down to c@xmath3 and c@xmath4 . it appears that the young pne where fullerenes have been observed are the optimised factories for their formation because their stellar emission peaks in the required 10 ev photon energy regime . in their pre - pne stages these objects do not have sufficient uv photons to dehydrogenate and heat the dust , and in their later stages the radiation field is so much harder that dissociative losses will destroy the h - rich carbonaceous particles before they can re - structure to form fullerenes . we studied the formation of fullerenes c@xmath3 and c@xmath4 in planetary nebulae , for which a satisfactory formation path is still missing . in space , the low density of the gas precludes the formation of fullerene materials following known vaporization or combustion synthesis routes even on astronomical timescales . the direct formation of c@xmath3 through dissociation - induced curvature of dehydrogenated pahs requires a very specific tuning of the dissociation parameters , and thus appears unlikely . the scheme for fullerene formation in space that we propose is based on a top - down approach . the top of the tree is represented by large a - c : h / hac particles , i.e. materials whose presence in space has been firmly established . uv photolysis promotes structural transformations in the parent a - c : h particle with the formation of what we call `` arophatic '' clusters : 3-d , hollow structures characterized by the presence of aromatic clusters linked by aliphatic bridging groups . uv - induced dehydrogenation of the emerging arophatic cluster ( or the parent a - c : h ) introduces pentagonal rings , and hence curvature , in the evolving structure . the result is a large , vibrationally - excited fullerenic cage that shrinks down to the stable molecules c@xmath3 and c@xmath4 through emission of c@xmath7 groups . the very high energy barrier for c@xmath7 ejection from c@xmath3 prevents further dissociation . the young pne where fullerenes have been observed appear as the optimised factories for the formation of these molecules . in their pre - pne stages the formation of fullerene - like structure is not favoured because these objects lack sufficient uv photons to dehydrogenate and heat the dust . in their later stages the radiation field is so much harder that it will destroy h - rich carbonaceous particles before they can re - structure to form fullerenes . we would like to thank g. lavaux for the realization of the movie and fig . [ fig_arophatics_2 ] . e.r.m , j.c . and e.p . acknowledge the support from the national science and engineering research council of canada ( nserc ) . wishes to acknowledge the support from a marie curie intra - european fellowship within the 7th european community framework program under project number 272820 . this research has made use of nasa s astrophysics data system . 52 natexlab#1#1 , j. , cami , j. , peeters , e. , et al . 2012 , , 757 , 41 , o. , & tielens , a. g. g. m. 2012 , proceedings of the national academy of science , 109 , 401 , j. , bernard - salas , j. , peeters , e. , & malek , s. e. 2010 , science , 329 , 1180 , j. , bernard - salas , j. , peeters , e. , & malek , s. e. 2011 , in iau symposium , vol . 280 , iau symposium , 216227 , j. , heras , a. m. , tielens , a. g. g. m. , et al . 2001 , , 546 , l123 , i. , le teuff , y. h. , williams , p. m. , & tielens , a. g. g. m. 2000 , , 357 , 572 , s .- h . , & violi , a. 2011 , proc . inst . , 33 , 693 chuvilin , a. , kaiser , u. , bichoutskaia , e. , besley , n. a. , & khlobystov , a. n. 2010 , nature chemistry , 2 , 450 , g. c. , de marco , o. , whitney , b. a. , et al . 2011 , , 142 , 54 dunk , p. w. , kaiser , n. k. , hendrickson , c. l. , et al . 2012 , nat . commun . , 3 , 855 , a. , van loon , j. t. , woodward , c. e. , et al . 2012 , , 421 , l92 , a. c. , & robertson , j. 2004 , phil . lond . a , 362 , 2477 , k. a. k. , mutschke , h. , & jger , c. 2011 , , 528 , a56 , d. a. , kameswara rao , n. , & lambert , d. l. 2011 , , 729 , 126 , d. a. , manchado , a. , garca - lario , p. , et al . 2010 , , 724 , l39 , d. a. , iglesias - groth , s. , acosta - pulido , j. a. , et al . 2011 , , 737 , l30 , c. , cami , j. , bouwman , j. , peeters , e. , & min , m. 2011 , , 536 , a54 , m. , gaessler , w. , hayano , y. , et al . 2003 , , 589 , 419 , v. i. , & duley , w. w. 2000 , , 543 , l85 , v. i. , tereszchuk , k. , duley , w. w. , & bernath , p. 2001 , , 558 , l129 , j. b. , mckinnon , j. t. , makarovsky , y. , lafleur , a. l. , & johnson , m. e. 1991 , , 352 , 139 irle , s. , zheng , g. , wang , z. , & morokuma , k. 2006 , the journal of physical chemistry b , 110 , 14531 , pmid : 16869552 , c. , huisken , f. , mutschke , h. , jansa , i. l. , & henning , t. 2009 , , 696 , 706 , a. p. 2012 , , 540 , a1 + . 2012 , , 540 , a2 + . 2012 , , 542 , a98 + , w. , fostiropoulos , k. , & huffman , d. r. 1990 , chem . lett . , 170 , 167 , w. , lamb , l. d. , fostiropoulos , k. , & huffman , d. r. 1990 , , 347 , 354 , h. w. , heath , j. r. , obrien , s. c. , curl , r. f. , & smalley , r. e. 1985 , , 318 , 162 , s. 2009 , international journal of astrobiology , 8 , 161 , s. d. , goodwin , d. g. , anderson , m. s. , & anderson , j. r. 1995 , , 51 , 9973 , e. r. , jones , a. p. , & tielens , a. g. g. m. 2010 , , 510 , a36 + . 2010 , , 510 , a37 + . 2004 , masters of the flame : industrial production of fullerenes becomes a reality ( http://www.nano-c.com/technologies.as ) , j. r. , cernicharo , j. , goicoechea , j. r. , gulin , m. , & asensio ramos , a. 2007 , , 661 , 250 , e. , tielens , a. g. g. m. , allamandola , l. j. , & wolfire , m. g. 2012 , , 747 , 44 , s. , stranger , r. , & duley , w. w. 2003 , , 594 , 869 , e. , & hornekr , l. 2008 , , 679 , 531 , k. r. g. , smith , k. t. , & sarre , p. j. 2012 , , 421 , 3277 , r. h. , simpson , j. p. , odell , c. r. , et al . 2011 , , 410 , 1320 , a. , & duley , w. w. 1996 , , 472 , l123 + , a. , duley , w. w. , & pinho , g. p. 1997 , , 489 , l193 + , k. , werner , m. w. , ingalls , j. g. , et al . 2010 , , 722 , l54 , f. w. 1984 , journal of applied physics , 55 , 764 taylor , r. , hare , j. p. , abdul - sada , a. k. , & kroto , h. w. 1990 , j. chem . , 1423 , a. g. g. m. 2005 , the physics and chemistry of the interstellar medium ( cambridge univ . press , cambridge ) , s. , andersen , j. u. , gottrup , c. , hvelplund , p. , & pedersen , u. v. 2001 , phys . , 87 , 073401 , d. 1992 , , 359 , 707 , a. 2004 , combustion and flame , 139 , 279 , y. , & kwok , s. 2011 , , 730 , 126 zheng , g. , irle , s. , & morokuma , k. 2005 , the journal of chemical physics , 122 , 014708 , g. , wang , z. , irle , s. , & morokuma , k. 2007 , j. nanosci . nanotechnol . , 7 , 1662 we report here the details of the calculation of the scaling rule ( eq . [ scaling_eq ] ) , which relates the time @xmath12 required for fullerene condensation with the initial density @xmath13 of the carbon gas . consider an ensemble of @xmath59 particles that can occupy @xmath60 cells . the probability of having one particle in one cell is @xmath61 . the probability @xmath62 of having two particles in the same cell at the same time is described by the poisson statistics : @xmath63 we have that @xmath64 , where @xmath65 is the total volume and @xmath66 is the volume of a single cell . we can then rewrite the single occupancy probability as @xmath67 . the probability @xmath68 of having interaction between two particles at time @xmath69 can be written as : @xmath70 where @xmath71 , @xmath72 being the time and @xmath73 the characteristic time of a single interaction ( depending on the nature of the interaction itself ) . the term @xmath74 represents the probability of interaction between two particles at a generic time @xmath72 . the probability of having interaction between two particles up to the time @xmath69 is then given by the following equation : @xmath75 if @xmath74 is small , we can rewrite eq . [ tilde2_eq ] as : @xmath76 the probability @xmath74 can be written in the following way : @xmath77 the term @xmath78 includes the characteristics intrinsic to the interaction ( which we do not need to specify ) while @xmath79 is the probability from eq . [ poisson_eq ] . the approximation resulting in the right - hand term of the equation is valid for @xmath80 . combining eqs . [ tilde3_eq ] and [ lambda_eq ] , with @xmath81 , we obtain @xmath82 from eq . [ tilde4_eq ] we can derive the timescale for condensation : @xmath83 because both @xmath73 and @xmath84 are independent from the initial particle density @xmath13 , we can write the following scaling rule @xmath85 where @xmath18 and @xmath19 are the condensation timescales corresponding to densities @xmath20 and @xmath21 respectively .

quantum mechanics ( qm ) , particularly featured by @xmath0 and @xmath1 , has been always facing the fundamental critics since its establishment @xcite . the exact physical meaning of `` quantum '' is unclear , which leads to the well - known `` weirdness '' . the development of quantum field theory ( qft ) , which is often viewed as an improved form of qm , does not resolve the seminal problems in qm , such as the direct physical meaning of wave function and phase , the role of quantization and measurement , etc . the problem of the foundation of qm has been widely concerned again due to the progress of quantum information and quantum computing ( qiqc ) @xcite . the concepts of nonlocality and entanglement have proven their importance , yet , there exist confusions between them @xcite . for the research of foundation of qm , we can briefly clarify two sub - fields , one is the interpretation of qm @xcite , the other is the post - qm @xcite . the problems and confusions on basic concepts even philosophy , such as the classical @xcite , hidden variable @xcite , collapse @xcite , etc are mostly addressed in the interpretations . post - qm also devotes to the complete mathematical form of qm , such as the information - theoretic approach @xcite , the khler structure approach @xcite , etc . the standard qm bases on several assumptions , and the physical explanations of the copenhagen ( orthodox ) interpretation are not satisfying . uncertainty and complementarity are emphasized ; however , for uncertainty , there are different explanations , either based on measurement , knowledge , propensity , or reality @xcite . for the statistical meaning of the wave function and measurement , the disagreement is more notable . born viewed statistics as inherent , while einstein , by epr - argument @xcite , viewed statistics as a result of incompleteness of qm . the notable `` jump '' and `` collapse '' do not have clear physical pictures . the most primary problem is that the copenhagen interpretation did not give an exact and direct meaning to the wave function , e.g. , why @xmath1 is complex ? indeed , there are direct efforts to explain the physical meaning of wave function , which led to the hidden variable method and bohm mechanics @xcite . every particle is endowed with definite coordinate and trajectory under the restraint of guiding wave . also , there exists the physical collapse theory aimed to explain collapse @xcite . however , it modifies the schrdinger equation and the linearity , which is shown to be the analogy of the approaches based on decoherence @xcite . coherence and decoherence , which were not demonstrated in the standard qm , have gained lots of attentions these years . for instance , the many - world interpretation @xcite views the universe as a coherent entity , and there exist the inner observer and relative state . the consistent ( decohered ) history approaches @xcite take the time evolution and measurement into account , and demonstrate the structure of quantum logic . in zurek s exsistential interpretation @xcite , the uncertainty principle is re - explained from the view of information , and the born s rule is deduced via the symmetry of envariance . recently , kinds of information - theoretic theories @xcite are developed , where qm is viewed as a kind of information processing theory . with all these progresses , however , the exact essence of quantum has not been drawn . at the same time , the continuous argument on foundation of qm also arose the reflection of the primary philosophical conception , which can not be avoided in the quantum physics . the problem that why we can not easily understand qm just brings the problem that why we think the classical mechanics ( cm ) is normal . in this paper , our standing point is to view qm as a special kind of `` description '' of motion , from which , there is no good or bad of qm and cm @xcite . the theories , with equations and models , deal with the same nature only revealing different aspects and properties . the state vector in hilbert space can capture the properties of coherence , e.g. , interference , which is exotic for cm . on the contrary , trajectory is the basic idea in cm and the macroscopic world , yet , not in qm , due to the uncertainty . we note the trajectory in the path - integral method , which borrows ideas from cm , only has mathematical meaning . in practice ( reality ) , to employ which description depends on its efficiency . in this work , we start from one new concept , `` scope '' , to check the foundation of qm . we focus on the basic concepts and ideas in qm , mostly relating to superposition . we do not intend to construct a broader theory with the standard qm as the special case , the methods based on scope mainly belong to the interpretation of qm . scope describes the `` structure of motion '' , i.e. , the logic and systematic potential action region of the movement of a certain object ( system ) . that is , from scope we can get the ability and the whole structure of the motion of the matter . the merits of the concept of scope are multi - folds . firstly , the direct one , the weirdness of qm can be resolved based on scope , the wave function and superposition can get their physical meaning . secondly , the abstract qm can gain solid natural foundation and physical essence , i.e. , qm is a totally new kind of description different with cm ( including statistical mechanics ) and also relativity . thirdly , scope can bring new ideas . for instance , the wave function and superposition principle can also be used in the mesoscopic and macroscopic scales , that is , the concept of scope is universal , can not be restricted by the scale . also , qm may not rely on the hilbert space , since scope itself can form a kind of space , also there can be other kinds of space , e.g. , tangnet @xcite , relating to quantum information and entanglement . forth , another point , the concept of scope may have new indication of the basic ideas of nature and philosophy , e.g. , it means every object has finite ability of motion . this work is divided into four parts . in sec . [ sec : open ] , we start from the concept of openness , which is central for qm , to clarify several basic physical ideas . in sec . [ sec : scope ] , we introduce the new concept of `` scope '' , from which we discuss the physical meaning of superposition and entanglement , and to address the foundation of qm . in sec . [ sec : entanglement ] , we study several kinds of quantum states , and we analyze the degree of entanglement and information . several topics are discussed in the appendix . one is the differences between scope and other existed methods , where we state that the concept of scope has never been proposed before . another one is the model for the reduced entanglement , by comparison with the two - body problem in cm . last , we study the physical meaning of entanglement measures at present , like negativity , relative entropy of entanglement . in sec . [ sec : conclusion ] , we conclude and discuss briefly the physical roles of scope in the foundation of qm . in this section , before the study on the concept of scope and entanglement , we show that the concept of openness is fundamental in qm . openness means that the quantum object can not be separated from the outside world , and the basic subject in qm is the open system , just the opposite of cm @xcite . the physical reason for openness is the existence of quantum coherence . we show that all the basic equations in qm demonstrate the openness and coherence . the quantum coherence leads to many crucial facts , such as the double - slit interference , coherent and squeezed state of light , also the heisenberg s uncertainty principle , etc . relating to measurement , the uncertainty principle states that any measurement disturbs the state of the quantum system , since the coherence is disturbed . on the contrary , the nature light from sun is not coherent as it is a kind of mixture . in cm , there is no coherence , which is the main distinction between quantum and classical dynamics @xcite . usually , in qm the method of density matrix @xcite is employed to capture coherence , which is represented by the non - diagonal elements . a diagonal density matrix is said to be classical as the coherence vanishes . in qft , there exists coherence since the quantum field itself is the coherent entity , such as vacuum , electromagnetical field etc , although the density matrix is not often used . `` coherent '' means that the field as a whole has fixed amplitude and phase . the micro - particles , as the excited state of field , have coherence since interactions between them are due to the field . the facts of particle creation , classicality , and decoherence lead to the degeneracy of coherence , yet , particle can go back to its ground state via the annihilation operator , thus , re - excite the coherence @xcite . in addition , we note that the standard qm does not focus on the properties of field , e.g. , the creation and vanish of particles , also the origin of ensemble , which we will discuss below . in qm , there was the argument that whether quantum theory describes the dynamics of the single system or the ensemble @xcite . according to the standard qm , the existence of ensemble is _ a priori _ fact . for instance , in the double - slit interference experiment , the ensemble of electrons is used , the electrons are viewed as identical , and the state of electrons are the superposition of two modes corresponding to the two slits . here , qm can not explain the origin of the ensemble ; in contrast , qft can give the creation - annihilation picture of the ensemble of electrons . the quantum states of each electron before entering the slit are not necessarily orthogonal , and there exists field among the electrons , that is , the existence of ensemble does not mean there is no coherence . yet , the methods of qm are believed as universal @xcite , thus , it should provide the physical picture for ensemble . recently , there are efforts to indicate the quantum origin of ensemble via entanglement @xcite . it is shown that the system within the coherent universe behaves as the canonical ensemble , as if the universe were in the mixed state , in which each pure state has the equal probability . physically , there can be kinds of coherence and decoherence processes in qm , which relates to the distinction between the fine - grained and coarse - grained description of states of the system @xcite . we will further analyze this in the study of density matrix below . in addition , we note that the concept of ensemble is similar to `` mixture '' , which is also widely employed in qm . however , the `` mixed state '' is not an exact expression , e.g. , it does not indicate the state is of one single object or many objects . we view mixture as the classical concept . the coherence indicates the openness of quantum dynamics . here we show that the basic equations of qm all demonstrate the openness . the schrdinger equation @xmath2 , where @xmath3 is viewed as the pure state vector of the system , from which the eigenstate and the eigen - energy can be deduced . the single quantum system can be , e.g. , one electron , field , or ensemble of electrons , as long as there exists coherence . here , `` single '' is different with that in cm , where the system is labeled by mass . the wave function has the property of holistic , and we can say that the system is `` open inside '' . now , for schrdinger equation , it is easy to get the following equations @xmath4 . let the density matrix ( or statistical matrix ) @xmath5 , and @xmath6,\ ] ] which is the liouville equation . in reality , it is often impossible to get the wave function , i.e. , to collect all the information of the system this can be described by the well - known @xmath7 model , where the universe @xmath8 contains one system @xmath9 embodied by one environment @xmath10 . the wave function distributes across @xmath9 and @xmath10 . according to the method of entanglement and schmidt decomposition @xcite : @xmath11 then the density matrix @xmath12 where coherence is the non - diagonal elements . the state of the system is deduced by tracing out the environment @xmath13 with @xmath14 , which is often viewed as the origin of the classical measurement results . further , interestingly , the wave function can also be written as @xmath15 where the set @xmath16 is also wave function , thus , not necessarily orthogonal , and the parameters @xmath17 are complex , with @xmath18 . the state in eq . ( [ eq : wfes ] ) is the `` wave function of ensemble state '' ( wfes ) , the detailed properties are discussed in appendix [ sec : wfesd ] . this decomposition is coarse - grained , i.e. , each @xmath19 can further be decomposed as the superposition of orthogonal states . in the ensemble , each party has the property of openness that there can be ( but not necessarily ) coherence between any two of them . last , another well - known form is the wigner function @xcite in the quantum phase - space , which is often written as @xmath20 where the integral is across the whole space . there exists a whole class of distribution functional , with the wigner function as the simplest and straightforward one @xcite . here , in the phase - space , the system is represented by the coordinate and its movement is momentum @xmath21 . contrast with classical method , in wigner function , there exists another coordinate @xmath22 , which is integrated out . the physical meaning is that the effects of the outside world ( environment ) should be deleted ( tracing out ) , so that we can confirm the existence of the system . from this point , we can say that the phase - space approach in qm is also natural since it captures the essence of openness , the negative value of @xmath23 is one expression . to sum up , in this section we show that the existence of coherence is central for quantum behaviour , which results in the openness demonstrated by all the basic equations of qm . we note that here we do not study the detailed properties and classifications of quantum coherence . in the next section , we will further show that quantum coherence and the property of openness can be based on the new concept of `` scope '' , and from which , we will analyze the direct physical meaning of the wave function and also entanglement . in this section , we study the foundation of qm from the concept of `` scope '' . briefly , scope , or quantum scope , describes the `` structure of motion '' , which means the action region of the motion , and the correlations among states and observable , rather than the casual dynamics , neither the structure of matter , nor matter . by comparison , according to cm , the object , labeled by mass @xmath24 , associates with a certain force , satisfying @xmath25 . here , in qm every motion has a certain scope . the wave function is indeed the scope of the quantum dynamics . further , the concept of scope is not restricted within qm , it is the universal method , and it has the connections with other methods , such as relativity . below , we present the theorems and properties of scope . before that , we first note two points . firstly , the concept of scope is in consistent with qm , since it is directly based on qm . secondly , scope can give physical meaning to wave function and also superposition , then complete the basic principle of qm . firstly , we present the theorems relating to scope itself , we state that the principles of superposition and entanglement are universal . every motion has one scope @xmath26 . here `` motion '' includes the condition when the observed velocity of one object is zero . strictly , there is no object at rest . scope @xmath26 belongs to the motion of a certain object , not the object itself . the concept of scope describes the ability of motion in a systematical way . scope @xmath26 is composed with state @xmath27 and its weight @xmath28 , where @xmath29 , @xmath30 , @xmath31 decides the space of @xmath32 , labeled as @xmath33 . the concept of `` state '' originates from qm and statistical physics . @xmath27 can be a vector in the hilbert space , or the point in the tangnet space @xcite . within one scope @xmath26 , there can be many physical states @xmath27 . one motion can only be in state within its own scope . the principle of superposition has two parts : * _ the principle of superposition of scope @xmath26 _ : + the scope of motion can be expressed as the superposition of all the states within as @xmath34 there is a set of povm `` active operator '' @xmath35 which associates with @xmath26 and satisfies + @xmath36 + which means @xmath37 can act state @xmath38 out from scope @xmath26 . here , we employ dirac s `` bra - ket '' symbol @xcite . @xmath37 satisfies @xmath39 which means @xmath40 is the eigenvalue of the `` activer '' @xmath37 . this is the alternative of the `` probabilistic interpretation '' of the coefficients @xmath40 in qm . also @xmath41 which means all the states are acted out , that is , the whole scope @xmath26 is realized @xmath42 , thus @xmath43 . this is the origin of the `` normalization regulation '' . from the probabilistic view , it means the sum of all the probabilities equals to 1 . and , @xmath44 also , there exists the `` anti - active operator '' 1-@xmath45 . we do not study the active operator in detail here . * _ the principle of superposition of state @xmath1 _ : + the state of motion can be the coherent superposition state as @xmath46 where normalization means the norm of a state vector is @xmath47 , which is the result , yet not the same with the property of scope . note that the superposition for mixed , pure , also classical states are all included . there exists entanglement between two or multi- scopes . the entangle process @xmath48 for two systems @xmath49 and @xmath50 can be written as @xmath51 we note that the analysis can be generalized to multi - systems directly , here we focus on bi - party system for clarity . @xmath52 represents the entangle process . there are two parts of the principle of entanglement : * _ the principle of entanglement of scope @xmath26 _ : + the entangled scope of motion can be @xmath53 where the sets of relative states @xcite @xmath54 and @xmath55 correlate with each other one - to - one , which can be labeled as the `` 1 - 1 branch '' rule . the details which @xmath38 is the relative state to @xmath56 is determined by nature , e.g. , the interaction . + there also exist the set of active operator @xmath57 , which satisfies @xmath58 where the branch latexmath:[$|\psi_i\rangle \otimes |\phi_i\rangle \equiv * _ the principle of entanglement of state @xmath1 _ : + the entangled state can be a part of the entangled scope @xmath60 three kinds of states have to be classified : ( 1 ) if @xmath61 , the state is truly entangled ; if @xmath62 , the state is separable , the same as the usual form in eq . ( [ eq : separable ] ) ; this means the separable state is also the result of the entangle process of scope thus containing quantum information as we will show below ; ( 3 ) if @xmath63 , the state reduces to a simple product state . in reality , the branches within the entangled scope are not all necessarily realized , thus , the entangled state is the fragment of the scope as long as satisfying the normalization rule . in addition , there may be interactions during the entangle process , however , entanglement describes the information transition or coherent correlations between different systems , instead of the energy transition . we will define the degree of entanglement @xmath64 in sec . [ sec : degree ] . next , relating to qm , particularly the dynamics of the scope and state , there are two theorems . there exist representations , corresponding to different complete sets formed by the commutative observable , labeled as `` the scope under representation '' . in qm , the physical reason for representation is that there exist different complete sets , which is demonstrated by bohr s complementarity principle @xcite . we note that , according to the consistent history approach @xcite , there exist several frameworks , which also satisfy the complementarity principle . this theorem indicates again that the concept of scope is the systematical , instead of casual , description of the potentiality and structure of the motion . there exist pictures , corresponding to different frames and time choice . there are three basic kinds of pictures as follows : * _ the frame is set on the motion itself _ : schrdinger picture . + the scope @xmath65 ( can also be labeled as @xmath9 ) satisfies the schrdinger equation @xmath66 where @xmath67 is the hamiltonian . the solution of this equation is the `` structure of the motion '' . here , `` @xmath68 '' means the change of space rather than the evolution of time @xcite . in fact , `` @xmath68 '' means the `` gradually evolving of the scope '' . the dynamical variables do not change with time . we mention that in quantum cosmology , eq . ( [ eq : sch ] ) becomes @xmath69 , where the hamiltonian includes the gravitational field plus all matter sources in the universe , the external timing parameter @xmath68 vanishes naturally @xcite . * _ the frame is set external _ : heisenberg picture . + the dynamical variable @xmath70 satisfies @xmath71,\ ] ] which means the change of dynamical variable in different states within the fixed scope . * _ the frame is set in - between _ : dirac ( interaction ) picture @xcite . + @xmath72 , @xmath73 , satisfying @xmath74,\end{aligned}\ ] ] where @xmath75 . the existence of pictures and the problem of the external parameter @xmath68 reveal that scope mostly describes the logic structure of the states , from which the commutation relations of observable do not change with time . last , we show another kind of property of scope . there exist relations between the concept of scope and relativity , which describes the coherent properties of the space - time metric of the object . here , we only present the mostly crucial conjecture as below . the scope @xmath26 can be influenced by mass @xmath76 . this theorem sets the connection between the motion of a certain object and the object itself . it means that objects with different masses can not have the same scope . every scope is unique . also , the shape of scope is not regularly euclidian . in general , it is topologically non - euclidian . please see the geometrical properties @xmath77 of scope below . in this subsection , we study some properties and consequences deduced from the theorems above of scope . scope @xmath26 has a certain geometrical shape . the number of state @xmath31 can not be zero . the states within the scope can form certain shape . for instance , if there is one single state , @xmath31=1 , the corresponding shape of scope is just a point ( in geometry ) . if @xmath31=2 , the shape is one segment line , straight or carved , with a certain length . if @xmath31=3 , the shape is one triangle face , with a certain area . if @xmath31=4 , the shape is one tetrahedron . if @xmath78 , the shape is one ball . we should state that the shape is in the three - dimensional space , euclidian or non - euclidian , different from the shape in the high - dimensional space , such as the hilbert space . the shape not only has the `` shell '' with boundary and face , also it has inner structure . particularly , when @xmath78 , if we neglect its structure , the ball can be viewed as one point . this is the alternative of the `` classical limit '' of the orthodox qm by taking @xmath79 . we note that the geometric properties of quantum states are under research , see ref . @xcite for instance . scope @xmath26 has a certain magnitude . the magnitude is defined as the length ( or the area , the volume ) of its shape . to calculate , we have to specialize the `` distance '' between different states . for instance , we can use the energy @xmath10 to define the distance @xmath80 , such as @xmath81 the magnitudes of @xmath26 of different movements should be different , since they have physical meanings . the bigger the magnitude , the better the ability of the motion , thus , the more `` freedom '' in the scope @xmath26 . the magnitude is not necessarily equal to one , which is different with the normalization condition . one particular scope @xmath26 has its own space - time matric . this is the direct result of theorem @xmath82 . when the effect of mass is not obvious , we can set the space - time matric of different scopes the same with each other , which is another kind of classical limit , or the low - energy limit . the scope @xmath26 has the properties of `` reality '' ( actuality ) and `` propensity '' . when @xmath37 acts , the certain state @xmath38 is realized , from potential to real , and the scope @xmath26 `` gradually evolves to its whole form '' . there is the transition between reality and propensity . actually , the two aspects of scope were studied a long time ago by aristotle in the ancient greece @xcite . the method was mostly ignored and was claimed as `` metaphysical '' useless for physics . yet , the idea of propensity has gained quite a lot of attentions in recent years @xcite . the concept of @xmath26 manifests that qm is the systematical description of the structure of movement , which is totally new compared with the classical dynamical method , from which the `` local realism '' @xcite originates . thus , we view the critics based on local realism as the misunderstanding of the essence of qm . here , we do not discuss the problems of local realism and nonlocality in detail . also , there is no `` collapse '' @xcite . collapse means that when there is measurement , the state collapses from superposed state to the eigenstate . or , some others view the collapse happens in our knowledge @xcite . in fact , with the process of decoherence @xcite , when the initial state is superposed , it decoheres instead of collapse when there is measurement or environment . measurement is a process of entanglement . this is a simple result of the fact that the apparatus also has one scope with the states relative to the observed system . the measurement process is the disturbance of system to apparatus ( or environment ) @xcite . the measurement problem in qm @xcite , as well as the observe effect in relativity , can be well explained with the method of entanglement , and the change of our knowledge is another story @xcite . there are mainly two kinds of states : time - quasistatic ( tq ) state and ensemble - isotactic ( ei ) state . at present , there is no unified classification of quantum states . briefly , there are many kinds due to different criterions , such as separable / entangled , pure / mixed , local / non - local states , etc . the non - unification just indicates that the underlying physical reason for the classification is not clear . we give a new classification as follows : * onefold state * * eigenstate * * superposed state * * * time - quasistatic ( tq ) superposed state * * * ensemble - isotactic ( ei ) superposed state * multifold state * * factorizable state : product , separable , etc * * entangled state * * * time - quasistatic ( tq ) entangled state * * * ensemble - isotactic ( ei ) entangled state firstly , state is differentiated according to whether it is a multifold ( many - body ) state . for a single object , its state is onefold state . there can be superposed state according to the principle of superposition . for multifold system which contains several inner parts , the state is multifold state , and there can be entangled state . the separable state can be viewed as a kind of factorizable state , which we will study in the next section . the distinction of time - quasistatic ( tq ) state and ensemble - isotactic ( ei ) state bases on the theorem of ergodicity of statistical physics and the principle of identity in qm @xcite . for example , the double - slit state of the electron ( or other particles ) is the ei superposed state , which characterizes the electron ensemble . in this state , for one particular electron , it can only pass through one slit at one time , thus , the state of a single electron is not superposed . another example , as we know , we can use laser to control a certain two - level atom , and drive it to the so - called rabi oscillation @xcite . this is the tq state , the population transfers from one state to another periodically . the famous bell basis are the ei entangled state , such as the light source from type - ii spdc @xcite in the teleportation experiment , describing the ensemble of photons , the state of one photon is not entangled or superposed . an excellent example of tq entangled state is the electrons forming the chemical bonding in the molecules . the difference between ( ei ) state and ( tq ) state is the `` inner '' dynamics of the state itself , thus , this kind of definition is obviously not mathematical . however , in our study below , we do not care about this difference without loss of generality . , thus , qu@xmath83it ) . the correlation between scopes or states is depicted as straight line ( string ) . ] qm often deals with the properties of many - body ( instead of single or infinite ) system . to characterize the quantum features , at present , there exist three widely studied methods : nonlocality @xcite , nonclassicality @xcite , and entanglement @xcite . traditionally , nonclassicality was introduced in the quantum phase - space in the context of quantum optics . the state of light , e.g. , squeezed light , contains nonclassicality compared with the coherent light , which sets the boarder between quantum and classical states . we note that the continuous variable ( cv ) system in phase - space can be translated into the fock space , from which the entanglement can be defined . interestingly , after the seminal work of epr @xcite and bell @xcite , the nonlocality of entangled state was demonstrated with nonclassical light @xcite , where the three methods are all involved . in this work , we do not intend to analyze the relations of them ; instead , we focus on entanglement relating to superposition and the concept of scope . part of our study on entanglement has been presented in ref . we will study the physical role of entanglement in various quantum states , and the degree of entanglement , also we compare with different entanglement measure in appendix [ sec : measure ] . entanglement is widely studied in qiqc , and it is believed to be the key for information processing and computing . however , entanglement is not indispensable . in quantum cryptography , the bb@xmath84 does not rely on entanglement @xcite ; instead , particularly the b@xmath85 protocol shows that it is the nonorthogonality ( also no - cloning ) that ensures the security of the key @xcite . the recently studied continuous variable ( cv)-quantum key distribution ( qkd ) @xcite and decoy - state qkd @xcite , also do not rely on entanglement , which indeed are the improvement of the bb@xmath84 protocol . this indicates that the quantum nature of information acts even when there is no entanglement , i.e. , the quantum information does not necessarily depends on entanglement . in quantum computing , the well - known qdc@xmath47 model is shown not directly rely on entanglement @xcite ; instead , the quantum discord plays the central role @xcite . in another line of research , there are attempts to build qkd on the foundation of nonlocality and nonlocal box , where entanglement does not play the central role @xcite . the examples above indicate that entanglement is not such quantity viewed from the information - theoretic framework ( itf ) , which is the theoretic paradigm or the world view correlating with the research of qiqc to re - consider quantum theory and its foundation from the information point , where information is believed as physical and fundamental @xcite . on the contrary , especially from the concept of scope , entanglement is the direct generalization of superposition , not necessarily related to information ( entropy ) , physically . the original cv - entangled state of epr @xcite and ( discrete variable ) dv - entangled state of spin of bohm @xcite do not relate to information . the analysis of schrdinger just reveals that information can be effected by entanglement @xcite . also , entanglement does not only play roles in qiqc . for instance , in quantum chemistry , entanglement plays roles in the formation of molecules and chemical reactions . further , the so - called `` quantum information '' is quite misleading , as the zeilinger s principle @xcite states that there is no quantum information , there is only information processing with quantum sources . that is to say , information can be effected by the quantum features , including nonorthogonality ( superposition ) , entanglement , also nonlocality and nonclassicality . in this section , we will systematically study the quantum correlations of the bi - party system , the results for the multi - party system will be presented in other places . also , without loss of generality , we assume here that the branches within the entangled scope are all realized , i.e. mathematically , the entangled state is the same with the entangled scope . relating to the analysis in sec . [ sec : ensemble ] , the entangled state can be realized both for single quantum system and the ensemble of quantum system . the quantum states we analyze include : * product state ; entangled qudit ; decohered qudit ; * ensemble - product state ; separable state ; ensemble - entangled qudit ; ensemble - decohered qudit . for clarity of the physical picture , we develop a kind of diagram to represent the scope . as shown in fig . [ fig.product](a ) , the circle stands for the scope , and the points within stand for the orthogonal eigenstates ( with number @xmath83 ) . the correlation between scopes or states is depicted as straight line ( string ) , the dashed line stands for decohered string . the states we consider are shown in fig . [ [ fig.product]-[fig.ensemble ] ] . _ product state . _ as well - known , product state contains no correlation . here , we can describe three kinds of product states . suppose the two sub - systems are @xmath86 and @xmath87 , with orthogonal eigenstate @xmath88 and @xmath89 , respectively . the state @xmath90 is product state . the more complex product form is @xmath91 where @xmath92 , @xmath93 . we note that the states for @xmath86 and @xmath87 are not ensemble . this product state is shown in fig . [ fig.product](b ) . the third kind is the ensemble - product state , shown in fig . [ fig.product](c ) . the wave functions of the systems @xmath86 and @xmath87 can be written as @xmath94 which is the coarse - grained decomposition , since there exist the sub- wave functions @xmath95 , @xmath96 , and generally @xmath97 . with @xmath98 , the density matrix can be easily written as @xmath99 the ensemble - product state is @xmath100 the state @xmath101 contains quantum information , although there is no entanglement . there is coherence for the @xmath102 and @xmath103 , that is , there can be decoherence process . the result of the decoherence is that the local state of @xmath86 and @xmath87 become @xmath104 , and @xmath105 . yet , the global state @xmath106 remains as product . _ entangled qudit . _ the general form of the entangled qudit is written as @xmath107 and the density matrix is expressed as @xmath108 with @xmath109 . the state @xmath110 , shown in fig . [ fig.entangled](a ) , can be viewed as a part of the product state @xmath111 in case ( b ) above , by selecting out the branches according to the permutation symmetry . the coherence of the state @xmath110 is represented by the non - diagonal elements . when decoherence occurs by some disturbance , the state @xmath110 becomes diagonal , which is said to be classical . the decohered ( classical ) qudit is @xmath112 let @xmath113 , @xmath114 , and introduce the parameter @xmath115 , then @xmath116 which is separable . that is , for the single decohered qudit , it is classical as well as separable . this state is shown in fig . [ fig.entangled](b ) . _ separable state . _ the well - known separable state @xcite in fig . [ fig.ensemble](a ) is defined as @xmath117 with the normalization relation @xmath118 . we show that the separable state @xmath119 is the analogy of entangled state @xmath110 , except that the former one employs the coarse - grained decomposition , while the latter one employs the fine - grained decomposition , i.e. , the eigenstates are orthogonal . the decomposition is the same with those in equations ( [ eq : cg1 ] ) and ( [ eq : cg2 ] ) . then equation ( [ eq : separable ] ) can be written as @xmath120 ) . the ensemble ( sets ) @xmath121 and @xmath122 are not classical , while the ensemble @xmath123 and @xmath124 are classical . this reveals that there exists quantum information in the separable state . then , when the separable state goes to the classical state ? consider the case when only one eigenstate each is acted out , that is @xmath125 , and also let latexmath:[$\langle \psi_{\xi}^{a(b ) } @xmath119 in equation ( [ eq : separable2 ] ) becomes the same with the classical state @xmath127 in equation ( [ eq : decoher ] ) . we can label this as the `` classical reduction procedure '' . let us give a simple example of the @xmath128 system . it is reasonable to set @xmath129 , and @xmath130 , then the separable state is @xmath131 which is also classical . generally , when @xmath132 , the separable form of the density matrix does not rely on the local coherence within @xmath102 and @xmath103 , which just indicates that there can be quantum information in the separable state . _ ensemble - entangled qudit . _ the ensemble of entangled state is often studied in qiqc , such as the ensemble of bell s state . for single entangled qudit , it can be written in the schmidt basis as @xmath133 with @xmath134 . the density matrix of this ensemble can be expressed as @xmath135(b ) . it is easy to see that when applies the classical reduction procedure , @xmath125 , and @xmath136 , the state @xmath137 reduces to classical and also separable . when decoherence occurs , each entangled qudit becomes classically correlated , and the ensemble - decohered qudit is @xmath138 which is classical but not separable , shown in fig . [ fig.ensemble](c ) . further , to make the distinctions of the quantum states clear , we can classify four kinds of correlations : entanglement , decohered classicality , nonorthogonality , and coarse - grained classicality . the quantum characters in the states we studied are different . in the single entangled qudit , the entanglement is the global shared coherence , there is also the classical correlation . for the decohered qudit , the shared coherence disappears , thus there is only classicality . the ensemble - entangled qudit , which relies on the coarse - grained decomposition , contains another kind of shared coherence due to nonorthogonal local states , namely , @xmath139 , we name it as nonorthogonality . correspondingly , there exists another kind of classicality , via the parameter @xmath140 , instead of the correlations of the eigenstates between @xmath86 and @xmath87 . we name this kind of correlation as coarse - grained classicality , and the former one as decohered classicality . from this analysis , we can briefly present the main characters of quantum state in table i. [ tab:1 ] [ cols="^,^,^,^,^,^,^",options="header " , ] in this section , we study superposition and entanglement quantitatively . before our analysis , we note that the superposition and entanglement are both due to the existence of coherence , local and global ( shared ) , respectively . entanglement is not the quantum information ; instead , the classical information ( correlation ) can be effected by the entanglement ( shared coherence ) , thus , the so - called `` quantum information '' . we can say that entanglement and quantum information ( entropy ) are the two related kinds of characters of the entangled state . below , we present the definitions of the degree of superposition , entanglement , and nonorthogonality , which all relate to the quantum coherence . + degree of superposition @xmath141 . + for a superposed state vector @xmath142 , where @xmath143 [ def:1 ] notice that , by introducing the @xmath144-norm as @xmath145@xmath146 , then the @xmath147-norm is @xmath148@xmath149 , thus , @xmath150 . this indicates that the @xmath147-norm has physical meaning , i.e. , describing the degree of superposition . let us show some examples . when @xmath151 , @xmath152 . @xmath153 , @xmath154 , @xmath155 , and max(@xmath141)@xmath156 when @xmath157 . @xmath158 , @xmath159 , @xmath160 , and max(@xmath141)@xmath161 when @xmath162 . then generally , for @xmath163 , max(@xmath141)@xmath164 etc . that is , for @xmath31 , the maximal degree of superposition @xmath141 is @xmath165 , and the minimum approaches to _ zero_. if there are more eigenstates in a superposed state , then the degree of superposition it can get is bigger . it means , physically , that although a state vector in the hilbert space has to be normalized , the degree of superposition is not a relative quantity , it characterizes the amount of coherence . degree of entanglement @xmath166 . + for a general entangled state with @xmath24 subsystems @xmath167 with each @xmath31-dimensional : @xmath168 , the degree of entanglement @xmath169 [ def:2 ] the same with the degree of superposition , we can get @xmath170 . our definition @xmath171 is similar with negativity ( see appendix [ sec : measure ] ) , thus , we do not present the proof of the entanglement monotone @xcite , which states that the entanglement can not increase under the local quantum operation and classical communication ( locc ) . according to our understanding based on scope and coherence , the locc constraint is the property of quantum information ( entropy ) instead of entanglement ( globally shared coherence ) . yet , it is easy to check that entanglement obviously satisfies the locc constraint since the quantum coherence can not be created via classical way . the definition @xmath171 defines entanglement in a positive way . as a result , the locc constraint is not sufficient to define entanglement , and not all non - separable states are entangled ( sec . [ sec : states ] ) . next we study some basic applications . we assume the coefficients are all positive , without loss of generality . for @xmath128 system , if @xmath172 label @xmath173 , @xmath174 . there are two kinds of entangled state according to the nature of the branches ( strings ) . if @xmath175 of @xmath86 correlates with @xmath175 of @xmath87 , it is named as `` direct '' ( @xmath176 ) ; if @xmath175 correlates with @xmath177 , it is named as `` cross '' ( @xmath178 ) : @xmath179 and the degree of entanglement @xmath180 generally , @xmath181 . after some algebra , we can get @xmath182 introduce the reduced degree of entanglement @xmath183 then we can get the fundamental relation @xmath184 when @xmath185 , @xmath186 , @xmath187 , @xmath188 , @xmath189 , and @xmath190 . this is the maximal condition for the qubit . in addition , for the reduced degree of entanglement of the qubit system , we will further study the physical properties by comparing to the classical two - body problem in appendix [ sec : reduced ] . for @xmath191 system , if @xmath192 there are four kinds of entangled state , which belongs to the greenberger - horne - zeilinger ( ghz ) state @xcite : @xmath193 and it is easy to write the four degree of entanglement ( we do not present here ) , after some algebra , we get @xmath194 also , introduce the reduced degree of entanglement @xmath195 then @xmath196 for the maximal condition , when all the coefficients equal to @xmath197 , @xmath198 , @xmath199 , and @xmath200 . generally , for @xmath201 system , the maximal reduced entanglement is @xmath202 . this means when we add the numbers of systems to be entangled , the reduced entanglement becomes smaller and smaller , that is , the entanglement of the whole system becomes more fragile , when one subsystem is disturbed , the total entanglement vanishes . for @xmath203 system , if @xmath204 and the degree of superposition @xmath205 there are six kinds of entangled state according to the symmetry between the branches within the entangled state . here we only give two examples , the other four are easy to be drawn : @xmath206 the entanglements of these six states have the form @xmath207 . after some algebra , we can get @xmath208 for the maximal condition , when @xmath209 , @xmath210 , @xmath211 , @xmath212 . for this bi - qutrit entangled state , there is no definite form of reduced entanglement . for @xmath213 system , there exist totally twenty - four different states , each has four branches . the entanglement also can be written as @xmath214 . we get the following equations @xmath215 for the maximal condition , when @xmath216 , @xmath217 , @xmath218 , @xmath219 . generally , for @xmath220 system , there are @xmath221 different entangled states , set @xmath207 , we can get the general equations as @xmath222 for the maximal condition , when @xmath223 , @xmath224 , @xmath225 , @xmath226 . that is , for the two - party systems , when the number of states entangled together increases , the max(@xmath166 ) also increases , like the superposition @xmath141 . this is reasonable since when more states are entangled together , the amount of entanglement should increase , and the state becomes more robust . for the higher dimensional systems , the math is a little complex , and the physical picture is not so clear . we do not analyze them here . next , we turn to the quantization of the four kinds of correlations we defined in the last subsection . for the entangled qudit , the entanglement is the shared coherence , so the degree of entanglement @xmath166 can be easily calculated , according to the definition [ def:2 ] above . also , as we have discussed , we can employ different forms of entropy to characterize the quantum information . at present , the well - known quantities include von neumann entropy , relative entropy , quantum discord , squashed entanglement , entanglement of formation , cost , and distillation , etc @xcite . we will study some of them in appendix [ sec : measure ] . for the decohered qudit , there is pure classical correlation without shared quantum coherence . also , there exist the complete projective measurements , under which the classical density matrix remains @xcite . the information is characterized by the von neumann entropy . for the four kinds of ensemble state , see table i , there exists nonorthogonality . this kind of coherence is not globally shared by the two parties @xmath86 and @xmath87 , yet , it is shared by the local states @xmath227 of @xmath86 , and the local states @xmath228 of @xmath87 . the degree of nonorthogonality can be defined as @xmath229 in the sum , each element is smaller than one , yet , the sum can be arbitrarily big even infinity . this is reasonable since the quantity of coherence increases with the amount of the ensemble . particularly , we note that the separable state contains the quantum nonorthogonality thus quantum information , although there is no entanglement . the coarse - grained classicality results from the classical parameter @xmath140 , so the classical information can be easily calculated via the von neumann entropy @xmath230 . last , we make some comments on the mixed state , which , in deed , is a rather broad and rough method . the mixed state with purity smaller than unity is a `` fragment '' of a global pure state with an unknown part lost , thus , it is not easy to decide the exact structure of the mixed state . at present , there exist lots of ways to detect the entanglement , and different methods can be unified with the method of entanglement witness in a way @xcite . yet , even we can find the entanglement , we can not decide the structure of the mixed state . that is to say , to find how the mixed state is formed is more valuable and practical . the six kinds of states studied above can serve as some standard or assistant to detect the quantum correlations . when a mixed state @xmath231 can be written as similar to any standard state , labeled as @xmath232 , that is , if @xmath233 , then the quantum correlation ( entanglement , nonorthogonality , etc ) of @xmath231 can be defined as that of @xmath232 times @xmath234 . @xmath235 can be viewed as the disturbance or perturbation to @xmath232 . this method is similar with the relative entropy method @xcite and the lewenstein - sanpera ( ls ) decomposition @xcite . to sum up , in this section , we quantify the degree of superposition and entanglement , and also nonorthogonality . physically , entanglement is directly based on superposition , and nonorthogonality is also the result of superposition . the four kinds of correlations can be easily calculated . the weirdness of qm is a result of the nature of quantum coherence , the abstractness of quantum description , and the confusions on quantum logic and philosophy . the concept of scope , which is exotic for cm , relativity , statistical mechanics , also classical wave mechanics , just represents the quantum origin . it describes the possible ( potential ) action region and the systematic and coherent structure of movement of a certain object . superposition and entanglement are the natural results of scope with definite physical meaning . also , we discussed the properties of four kinds of quantum correlations and the six kinds of quantum states , and we show that entanglement is related yet different with entropy . further , the concept of scope brings new methods to the standard qm . one is that the scope itself has certain geometric features , just similar to the matter itself . this may indicate that the structure of the motion of the matter is more crucial than the structure of the matter itself . the other is that , rather than the information - theoretic framework , there exists a new kind of space , i.e. , tangnet , on which the quantum structure with entanglement exists @xcite . as a result , the concept of scope lays the new starting point to the foundation of qm . in the future , further efforts on , e.g. , the geometrical properties of tangnet , the comparison of scope with other interpretations , also the relation between nonlocality and entanglement , are needed . in addition , the physical roles of entanglement in qiqc are also widely concerned , we hope our approach can bring new ideas on some tough problems , such as the bound entanglement @xcite , the analogy of entanglement to energy or entropy , etc . in hilbert space , one state vector can be decomposed via orthogonal basis also non - orthogonal basis . physically , the two decompositions can be endowed with different physical contents : for single system , there only exists orthogonal physical basis ; while for ensemble , there can be non - orthogonal basis . for example , the spin state in one quantum dot can be up or down , or to the left or right in the diagonal basis ; for a collection of spins in quantum dots , the superposed states of spins form the non - orthogonal basis . mathematically , based on the measurement theory , especially the positive operator - valued measure ( povm ) @xcite , we can see the wfes has the corresponding properties as the wave function of pure state . for the pure state @xmath236 , it satisfies : ( 1 ) normalization : @xmath237 ; ( 2 ) completeness relation : @xmath238 ; ( 3 ) projective measurement : there exist a set of projective operators @xmath239 satisfying @xmath240 , the expectation value @xmath241 ; note that when the non - projective povm operator acts on the pure state , it can not distinguish its basis ; ( 4 ) probabilistic interpretation : @xmath242 means the probability to find state @xmath243 with projector @xmath244 ; ( 5 ) observable : the expectation value of observable @xmath245 is @xmath246 , with @xmath247 ; ( 6 ) evolution : @xmath248 also the basis @xmath243 satisfy schrdinger equation ; ( 7 ) decoherence : the non - diagonal elements in @xmath249 disappear , leading to state @xmath250 , each part @xmath251 forms the classical trajectory ; ( 8) quantum reference frame @xcite : to determine the preferred basis @xmath252 , the reference frame and certain interaction are needed to form the entangled state , i.e. , the coherence within @xmath248 delocalizes to form entanglement ; thus , by ignoring the reference frame , the state @xmath248 decohered . for the wfes , the state in eq . ( [ eq : wfes ] ) , the properties are similar while more complicated . we firstly note that if @xmath253 , it reduces to the pure state case ; if the states @xmath16 are orthogonal with each other , it goes to the totally classical ( decohered ) state . instinctively , there exists one pure density matrix , labeled as @xmath254 , which is @xmath255 to form the density matrix @xmath231 , the non - diagonal elements disappear , with coefficients @xmath256 , which is actually a kind of decoherence . the state @xmath257 indicates that before the formation of the ensemble , there exists one decoherence process . physically , the state @xmath3 stands for the `` field '' of the ensemble , and the ensemble emerges from the field after the decoherence process , i.e. , the creation of particles @xcite . note that here we do not study this issue with quantum field theory . we name this kind of coherence as _ sub - coherence _ , or coarse - grained coherence , and the decoherence as _ sub - decoherence_. to arrive at the classical state , the standard decoherence is then needed . if we only account coherence by ignoring sub - coherence , we can view @xmath3 as the wfes , from which , the density matrix @xmath231 for the ensemble emerges . further , the sub - decoherence can be introduced by one quantum reference frame @xmath258 , which with @xmath257 forms @xmath259 which is exactly the separable state @xcite . note that for pure state , the system and its reference frame form the entangled state . correspondingly , the basic properties of wfes are : ( 1 ) normalization : @xmath260 ; ( 2 ) completeness relation : @xmath261 ; ( 3 ) measurement : there exist a set of povm operators @xmath262 satisfying @xmath263 , the expectation value @xmath264 ; note that the measurement does not act on the fine - grained orthogonal basis directly ; ( 4 ) probabilistic interpretation : @xmath265 means the probability to find state @xmath266 with operator @xmath267 ; ( 5 ) observable : the expectation value of observable @xmath245 is @xmath268 , with @xmath269 ; ( 6 ) evolution : states @xmath3 and @xmath266 both satisfy schrdinger equation , with which the liouville equation for @xmath231 is equivalent ; ( 7 ) sub - decoherence : the non - diagonal elements in @xmath270 disappear , leading to quantum ensemble @xmath231 ; decoherence : the party @xmath266 each can further decohere , leading to classical state and classical trajectory , respectively ; ( 8) quantum reference frame : to determine the preferred basis @xmath271 , the reference frame @xmath232 and certain interaction are needed to form the separable state , i.e. , the sub - coherence within @xmath3 delocalizes to form separability ; thus , by ignoring the reference frame , the state sub - decoheres ; further , if there exists the entanglement - type reference frame , the state eventually decoheres to classical trajectory . last , to illustrate the difference between coherence and sub - coherence , we give one simple example : suppose there are only two parties , @xmath272 and @xmath273 , in the ensemble , and each party with only two orthogonal basis @xmath274 and @xmath275 , and @xmath276 , @xmath277 . note that even if party @xmath273 is not in the basis @xmath278 , there can be rotation which can rotate it to this basis , with effects absorbed in the coefficients . thus , the density matrix of wfes is written as @xmath279 and its conjugate , when they disappear , the state @xmath257 sub - decoheres to the quantum ensemble @xmath231 . the coherence is represented as the coefficients @xmath280 , @xmath280 and their conjugate , when they disappear , the state decoheres to the classical state , labeled as @xmath281 , with diagonal elements @xmath282 , @xmath283 , and @xmath284 . _ scope and qft . _ in qft , the wave function is further viewed as the operator , which is the well - known `` second quantization '' . the total number of particles is not conserved as there always exist the fluctuations of field and creation / annihilatin of particles . particle itself is viewed as the excited state of field , and field is viewed as a kind of matter , whose motion exists within itself . thus , the second quantization of wave function , i.e. , scope , means to view the structure of the motion as the equivalence of the structure of matter , also to view the motion of the matter as the same with the matter itself , which is reasonable for field . this is also consistent with the mass - energy relation @xmath285 from relativity . however , for particles , atoms etc in qm , the object itself is different with its motion , thus , it is not proper to view scope as a kind of matter ; instead , scope should be viewed as a kind of description of the logical structure of the motion of the object . in history , another physical picture similar with qft is mainly developed by de broglie , namely , the `` pilot wave '' approach @xcite , in which the particle with definite local form sits at a certain place in the real pilot wave , and the particle can be viewed as the high - energy concentrate singularity of the wave . the pilot - wave describes the collective motion of two objects : particle and wave , which indeed goes to the method of qft . this relates to the `` wave - particle duality '' that the particle can perform both classical particle property and wave property , but not both of them at the same time @xcite . also , the concept of `` matter wave '' was developed . however , according to the concept of scope , the relation @xmath286 should be explained as that the object has the particle property ( @xmath24 ) and wave property ( @xmath234 ) within its own scope , consistent with the wave - particle duality . as a result , the matter wave can not be viewed as a kind of matter , and indeed , the so - called matter wave does not exist . we note that the other theory , bohmian mechanics @xcite , in which the particle has the trajectory restrained by the wave function , is consistent with the methods of qtf and scope . _ consistent history ( ch ) _ this is a widely studied form of qm , yet , there exist different approaches . this theory has the similar character with our methods based on scope , particularly the active operator . however , in ch , there is no reason for the physical meaning of the decomposition of the identity . next , we present our understanding for the ch . generally , ch takes the evolution of state in time into account . the main quantity , the decoherence functional @xcite is defined as @xmath287 where @xmath231 is the density matrix . the class operator is @xmath288 , with @xmath289 , @xmath290 means that @xmath291 acts on the scope instead of state . @xmath291 is the projective decomposition of the identity , @xmath292 when @xmath293 , @xmath294 , which is the probability of the history , and the normalization @xmath295 . when @xmath296 , if @xmath297 , then we say the history is decohered . in figure [ fig.ch ] , we show one consistent history . we put state and operator on the equal footing , then form the state - operator space . @xmath298 is the eigenstate set as the ordinate , the abscissa is the @xmath291 , where @xmath299 associates with time @xmath300 . for simplicity , we set @xmath301 , i.e. , there are six states within the scope . from the figure , we can easily get the activers of the history : @xmath302 . in addition , there can be other kinds of history , e.g. , several activers can act together as @xmath303 causing the superposed state . eigenstates within the scope , and the number of the projection ( activer ) @xmath299 represents the evolve of time . when the states become continuum , the lattice space becomes continuous space . the black polylines stands for the consistent history.,scaledwidth=33.0%,scaledwidth=25.0% ] _ kraus operator for open system _ @xcite . we study via the widely concerned model of the pure universe @xcite . let universe @xmath8 be composed with system @xmath9 and environment @xmath10 , labeled as @xmath7 . the initial state is @xmath304 and @xmath305 , correspondingly , the initial state for the universe is thus @xmath306 . the evolution of the wave function is @xmath307 , with the evolution operator @xmath308 . the reduced density matrix for the system is @xmath309 where @xmath310 , @xmath311 , and we let @xmath312 . the normalization rule is @xmath313 the kraus operator representation demonstrates the openness and the evolution in time of quantum system . however , although there is evolution in time , the logical relations between operators remains . for instance , the relation @xmath314 does not depends on time . below , we show that this method is consistent with the ch method . the physical role of the evolution operator @xmath308 can be substituted by the class operator @xmath315 , the state of the system at time @xmath316 is @xmath317 , where @xmath304 is the initial state . we know @xmath318 and @xmath319 thus @xmath320 , which is the same as the normalization of the kraus operator , physically . _ interacting faculties_. the concept of `` potentiality '' , originated from aristotle @xcite , is quite widely studied in qm . the actuality ( reality ) and potentiality were systematically studied recently , where the authors stated that potentiality could be a mode of existence , and the concept of `` faculty '' was introduced to replace `` entity '' @xcite . the faculty describes a level which does not pertain to things but rather to potential actions . possessing a faculty offers the ability to do something . according to our understanding , yet , the concept of faculty is the analogy of potentiality . this approach relates to the property @xmath321 in section [ sec : properties ] , yet it does not aim to provide the direct physical meaning of wave function and the superposition principle . consider the mixtures of bell states latexmath:[$|\psi\rangle=(|01\rangle \pm @xmath3 is the `` cross '' state , and @xmath323 is the `` direct '' state . so the mixture ( ensemble ) can be viewed as the mixture of two kinds of systems . the classical analogy is , e.g. , the mixture of two kinds of gases . there exists the center of mass for each of the gases , so the problem can be simplified as the two - body problem in classical mechanics , as shown in fig . [ fig.red](a ) . the two objects @xmath324 and @xmath325 exist in the configuration space , the coordinates are @xmath326 and @xmath327 , respectively . the kinetic energy is @xmath328 where @xmath329 is the total mass , @xmath330 is the reduced mass , @xmath331 is the coordinate of the mass center of the system , and @xmath332 is the relative displacement between the two objects . now , consider velocity @xmath333 . as we know , there exists the maximum speed as the light speed @xmath334 . let @xmath335 , thus , the set of speed @xmath336 forms the probability space with @xmath337 . the kinetic energy is written as @xmath338 consider the mixtures of bell states . the density matrix can be written as @xmath339 the entanglement of the mixture is @xmath340 where by introducing the probability amplitude @xmath341 , @xmath342 , @xmath343 , and @xmath344 is the reduced entanglement . the situation is depicted in fig . [ fig.red](b ) . by comparison , the entanglement is the analogy of mass , the probability is the analogy of @xmath345 . for high - dimensional and many - body problems , there is no obvious form of reduced mass , also the reduced entanglement . yet , we can apply the rough bi - party partition to deduce the apparent reduced mass or entanglement . however , we have to say , there are significant differences . for instance , the quantum probabilities @xmath346 and @xmath347 each exist in the probability space , and also they together exist in another probability space . but , the classical speed ( square ) @xmath348 and @xmath349 do not exist in the probability space with each other . this relates to the problem of special relativity , this simple mode may have implication for the relation between qm and relativity . according to our study , entanglement and quantum information ( entropy ) are different . entanglement is the shared coherence by the parties of the entangled state . also , we find nonorthogonality is the shared coherence within the local states of each party of the whole state . quantum information ( @xmath350 ) is the improved form of classical information ( correlation ) ( @xmath351 ) due to nonorthogonality ( @xmath352 ) or entanglement ( @xmath166 ) . thus , @xmath350 can be the function of @xmath352 or entanglement @xmath166 . here , we compare our definition with the usual entanglement measure . _ entanglement of formation . _ in qiqc , the mixtures of bell states , i.e. , ensemble of qubits , are often prepared , the entanglement of formation , also cost and distillation are studied @xcite . the entanglement of formation , which measures quantum information instead of entanglement , is defined as @xmath353 and the density matrix is written as @xmath354 as follows . ( b ) . `` computation basis '' @xmath361 , @xmath362 , @xmath363 , @xmath364 . the qubit is written as @xmath365 , and concurrence is @xmath366 it is easy to see that the basis in case ( a ) can be viewed as the special case of ( b ) . schmidt basis @xmath367 . the qubit is written as @xmath368 , with @xmath369 , the density matrix is @xmath370 and concurrence is @xmath371 the schmidt basis can be rotated to the `` computation basis '' in case ( b ) by the local unitary transformation @xmath372 @xcite . the above three kinds of basis give the same results of concurrence . the entanglement of formation is defined as @xmath373 , where the binary entropy function @xmath374 . we can form the entropy - concurrence matrix ( @xmath375 ) as follows @xmath376 then , it is obvious that the entropy @xmath377 describes the property of the diagonal elements ( population ) , and the concurrence @xmath378 describes the non - diagonal elements ( coherence ) . _ negativity and robustness . _ for pure state , the definition of negativity @xcite is the same with our definition of the degree of entanglement . negativity is usually viewed as the absolute value of the sum of the negative eigenvalues of @xmath379 @xmath380 where @xmath381 is the negative eigenvalues . in deed , the negative eigenvalues are the results of the shared coherence of the entangled state . for pure qubit , the negativity is the same with concurrence , that is to say , the definition of negativity is more general than concurrence for bi - party pure state . the robustness @xcite is the twice of the negativity for pure state . physically , it describes the robustness of the entanglement to the minimal amount of external classical noise . the role of noise is to disturb and destroy the coherence . physically , the robustness can describe the shared coherence , thus give the right quantity of entanglement . however , it is not easy to calculate since it is the indirect way to measure entanglement . in addition , we discuss a little of the physical meaning of concurrence and negativity for the mixed state condition . it is well - known that the two are often different for mixed state , though the same for the pure state case . the physical reason is quite clear : they are the results of two different entanglement witness ( operator ) @xcite . physically , in the mixed state there can be nonorthogonality , which can influence concurrence and negativity without being detected . that is to say , it is not proper to generalize concurrence and negativity to the mixed state , since the result is not necessarily entanglement . where the set @xmath383 can be taken as the set of separable states , states with positive partial transpose , nondistillable states , etc . the quantum discord @xcite and squashed entanglement @xcite , which we have studied in ref . @xcite , can be viewed as the relative entropy . physically , @xmath384 does not quantify entanglement directly , yet it relates to entanglement . the relation between @xmath384 and entanglement also nonorthogonality can not be deduced easily especially for mixed state , since the detailed structure of the mixed state and the distribution of coherence are not clear . as a result , the relative entropy is only an effective way to measure quantum information instead of entanglement . 99 d. bohm , _ quantum theory _ ( prentice - hall , inc . , englewood cliffs , new jersey , 1951 ) . m. jammer , _ the philosophy of quantum mechanics _ ( wiley , new york , 1974 ) . j. s. bell , _ speakable and unspeakable in quantum mechanics _ ( cambridge university press , cambridge , england , 1987 ) . m. a. nielsen and i. l. chuang , _ quantum computation and quantum information _ ( cambridge university press , cambridge , england , 2000 ) . w. h. zurek , rev . * 75 * , 715 ( 2003 ) . m. schlosshauer , rev . * 76 * , 1267 ( 2004 ) . c. g. timpson , arxiv : quant - ph/0611187 . a. einstein , b. podolsky , and n. rosen , phys . rev . * 47 * , 777 ( 1935 ) . e. schrdinger , phys . rev . * 28 * , 1049 ( 1926 ) . d. bohm , phys . rev . * 85 * , 166 ( 1952 ) . j. von neumann , _ mathematical foundations of quantum mechanics _ ( princeton university press , 1955 ) . h. everett , rev . * 29 * , 454 ( 1957 ) ; b. s. dewitt and n. graham , _ the many worlds interpretation of quantum mechanics _ ( princeton university press , princeton nj , 1973 ) ; d. deutsch , int * 24 * , 1 ( 1985 ) . m. gell - mann and j. b. hartle , in _ complexity , entropy and the physics of information _ , santa fe institute studies in the science of complexity vol . viii edited by w. zurek ( addison - wesley , redwood city , ca , 1990 ) ; phys d * 47 * , 3345 ( 1993 ) ; r. b. griffiths , j. stat . phys . * 36 * , 219 ( 1984 ) ; r. omns , rev . 64 * , 339 ( 1992 ) ; h. f. dowker and j. j. halliwell , phys . d * 46 * , 1580 ( 1992 ) ; j. j. halliwell , phys . rev . d * 58 * , 105015 ( 1998 ) ; phys . * 83 * , 2481 ( 1999 ) . r. clifton , j. bub , and h. halvorson , found . phys . * 33 * , 1561 ( 2003 ) ; g. brassard , nature phys . * 1 * , 2 ( 2005 ) ; j. oppenheim and b. reznik , arxiv : quant - ph/0312149 ; g. m. dariano , arxiv : quant - ph/1012.0535 ; jae - weon lee , arxiv : hep - th/1011.1657 ; p. goyal , arxiv : quant - ph/0702124 ; arxiv : quant - ph/0702149 . as bohr said : `` there is no quantum world . there is only an abstract quantum physical description . it is wrong to think that the task of physics is to find out how nature is . physics concerns what we can say about nature . '' from n. bohr , _ atomic theory and the description of nature _ ( cambridge univ . press , new york , 1934 ) . e. calzetta and f. d. mazzitelli , phys . d * 42 * , 4066 ( 1990 ) ; f. lombardo and f. d. mazzitelli , phys . d * 53 * , 2001 ( 1996 ) . s. habib , y. kluger , e. mottola , and j. p. paz , phys . rev . lett . * 76 * , 4660 ( 1996 ) ; g. mahajan and t. padmanabhan , gen . rel . grav . * 40 * , 661 ( 2008 ) ; a. giraud and j. serreau , phys . lett . * 104 * , 230405 ( 2010 ) . r. j. glauber , phys . rev . * 131 * , 2766 ( 1963 ) ; a. miranowicz _ et al . _ , a * 82 * , 013824 ( 2010 ) , and references therein . g. vidal , j. mod . opt . * 47 * , 355 ( 2000 ) ; m. b. plenio and s. virmani , quantum inf . comput . * 7 * , 1 ( 2007 ) ; r. horodecki , p. horodecki , m. horodecki , and k. horodecki , rev . * 81 * , 865 ( 2009 ) . a. aspect , j. dalibard , and g. roger , phys . lett . * 49 * , 1804 ( 1982 ) . c. h. bennett and g. brassard , _ proceedings of the ieee international conference on computers , systems and signal processing _ ( ieee computer society , new york , 1984 ) , pp . 175 - 179 . c. h. bennett , phys . 68 * , 3121 ( 1992 ) . f. grosshans _ et al . _ , nature * 421 * , 238 ( 2003 ) . lo , x. ma , k. chen , phys . lett . * 94 * , 230504 ( 2005 ) . j. a. wheeler and k. ford , _ geons , black holes @xmath385 quantum foam : a life in physics _ ( new york : w.w . norton @xmath385 co , 1998 ) . j. a. wheeler , in w. zurek , editor , _ complexity , entropy and the physics of information _ , pages 3 - 28 . ( addison - wesley , redwood city , ca 1990 ) . r. landauer , proc . workshop on physics and computation physcomp 92 ( ieee comp . press , los alamitos , 1993 ) pp . 1 - 4 . d. m. greenberger , m. a. horne , and a. zeilinger , _ going beyond bell s theorem in bell s theorem , and conceptions of the universe _ ( kluwer academic , dordrecht , 1989 ) ; phys . today * 8 * , 22 ( 1993 ) ; d. m. greenberger , m. a. horne , a. shimony , and a. zeilinger , am . j. phys . * 58 * , 1131 ( 1990 ) . a. peres , phys . lett . * 77 * , 1413 ( 1996 ) . g. vidal and r. f. werner , phys . a * 65 * , 032314 ( 2002 ) . c. h. bennett , h. j. bernstein , s. popescu , b. schumacher , phys . a * 53 * , 2046 ( 1996 ) . s. hill and w. k. wootters , phys . lett . * 78 * , 5022 ( 1997 ) ; w. k. wootters , phys . rev . * 80 * , 2245 ( 1998 ) . c. h. bennett , d. p. divincenzo , j. a. smolin , and w. k. wootters , phys . a * 54 * , 3824 ( 1996 ) . v. vedral and m. b. plenio , phys . rev . a * 57 * , 1619 ( 1998 ) g. vidal and r. tarrach , phys . a * 59 * , 141 ( 1999 ) . m. steiner , phys . a * 67 * , 054305 ( 2003 ) . h. ollivier and w. h. zurek , phys . lett . * 88 * , 017901 ( 2001 ) ; w. h. zurek , phys . a * 67 * , 012320 ( 2003 ) ; l. henderson and v. vedral , j. phys . a * 34 * , 6899 ( 2001 ) . s. luo , phys . a * 77 * , 022301 ( 2008 ) . m. christandl and a. winter , j. math . phys.(n.y . ) * 45 * , 829 ( 2004 ) . f. g. s. l. brando , phys . a * 72 * , 022310 ( 2005 ) . o. gittsovich , o. ghne , p. hyllus , and j. eisert , phys . a * 78 * , 052319 ( 2008 ) . m. lewenstein and a. sanpera , phys . lett . * 80 * , 2261 ( 1998 ) .

modified gravity ( for a review , see e.g. @xcite ) constitutes an interesting dynamical alternative to @xmath0cdm cosmology in that it is also able to describe with success the current acceleration in the expansion of our universe , the present dark energy epoch . the specific class of modified @xmath1 gravities ( for a review , see e.g. @xcite ) has undergone many studies which suggest that this family of gravitational alternatives for dark energy is able to pass the constringent solar system tests . the investigation of cosmic acceleration as well as the study of the cosmological properties of @xmath1 models has been done in refs . the possibility of a natural unification of early - time inflation with late - time acceleration becomes a realistic and quite natural possibility in such models , as is demonstrated e.g. in ref . @xcite . recently , the importance of modified gravity models of this kind has been reassessed with the appearance of the so - called ` viable ' @xmath1 models @xcite . those are theories which satisfy both the cosmological as well as the local gravity constraints , which had caused in the past a number of serious problems to some of the first - generation theories , that had to be considered now as only approximate descriptions from more realistic theories . the final aim of all those phenomenological models is to describe a segment as large as possible of the entire history of our universe , as well as to recover all local predictions of einstein s gravity that have been already verified experimentally , with very good accuracy , at the solar system scale . it is remarkable that , as was demonstrated in refs . @xcite , some of these realistic models lead to a natural unification of the early - time inflation epoch with the late - time acceleration stage . let us recall that , in general ( see e.g. @xcite , for a review ) , the total action for the modified @xmath1 gravitational models can be written as [ xxx7 ] s = d^4 x + s_(m ) . here @xmath2 is a suitable function of the scalar curvature @xmath3 , which defines the modified gravitational part of the model . the general equation of motion in @xmath4 gravity with matter is given by [ xxx22 ] g _ f(r ) - r _ f(r ) - g _ f(r ) + _ _ f(r ) = - t_(m ) , where @xmath5 is the matter energy - momentum tensor and @xmath6 is the derivative of @xmath1 with respect to its argument @xmath3 . for a generic @xmath1 model is not easy to find exact static solutions . however , if one impose some restrictions , one can proceed along the following lines . first , we may require the existence of solutions with _ constant _ scalar curvature @xmath7 , and we arrive at f(r_0)r_=g_. [ geneqrc ] taking the trace , we have the condition 2f(r_0)=r_0f(r_0 ) [ b ] and this means that the solutions are einstein s spaces , namely they have to satisfy the equation r_=g_= g _ , [ e ] @xmath8 being a solution of eq . ( [ b ] ) . this gives rise to an effective cosmological constant , namely _ eff==. [ l ] the purpose of our work will be to study the appearance of multiply de sitter space solutions in several realistic models of modified gravity . the occurrence of multiply de sitter solutions plays a fundamental role in modified gravity because it permits to describe the inflation stage as well as current @xmath0cdm cosmology in terms of this theory alone without any need for either fine - tuning of a cosmological constant nor of introducing extra scalar fields . in summary , this is a minimal and at the same time very powerful approach which circumvents some of the hardest problems of present day physics . one may argue that these theories are equivalent to introducing extra scalar fields , but this equivalence has been proven to hold at the classical level only , not at the quantum one . in addition , the cosmological interpretation of modified gravity solutions is different from that of scalar field cosmology . the paper is organized as follows . in the next section we discuss a viable modified gravity model @xcite in an attempt to study de sitter solutions there . it is shown that , for some values of the parameters , it is possible to find several de sitter spaces ( as a rule , numerically , for small values of the curvature exponent ) ; one of them may serve for the inflationary stage , while the other one can be used for the description of the dark energy stage . the evolution of the effective equation of state parameter is investigated numerically . 3 is devoted to the analysis of the same question in a slightly generalized model which is known to describe the unification of the early - time inflation with the late - time acceleration epochs @xcite . the same numerical investigation is carried out , and a multiply de sitter universe solution is constructed . in section 4 , the corresponding problem is investigated for a viable model of tangential modified gravity , proposed in ref . @xcite . having in mind the possibility to use a sds universe for the description of cosmic acceleration , those solutions for modified gravity are investigated in sect . comparison of the free energies of the de sitter and the sds solutions , for the viable model of the second section , is then made . finally , some outlook is given , together with the conclusions , at the end of the paper . in this section we will study de sitter solutions in realistic modified gravity models using a dynamical system approach . we shall start from the following form for the initial action @xmath9 , \label{1.1}\ ] ] where @xmath10 is some function of the geometrical invariants . in order to make use of the method of dynamical systems , we work in the metric corresponding to a spatially - flat frw universe , namely @xmath11 from here , the frw equation can be written in the following way ( for more details see ref . @xcite ) @xmath12 where @xmath13 the `` dot '' over the symbol means derivative with respect to time @xmath14 , while @xmath15 and @xmath16 represent the derivative of @xmath10 with respect to @xmath17 and @xmath18 respectively . in what follows we will mainly concentrate on the simplest case @xmath19 . let us consider the following choice for the function @xmath20 @xcite , which represents a very interesting subclass of viable modified gravitational models @xmath21 where the constant @xmath22 has dimension of mass , while @xmath23 , @xmath24 , @xmath25 are some positive dimensionless constants . note that here @xmath26 $ ] , @xmath27 , @xmath28 . after making variation , it is chosen @xmath29 . introducing a nonzero constant @xmath25 is here equivalent to introducing a shift in the effective cosmological constant , and for this reason we will assume @xmath30 in our further computations . finally , we can write eq . ( [ 1.3 ] ) for the function in ( [ 1.5 ] ) as @xmath31 , \end{array } \label{1.6}\ ] ] where @xmath32 . this equation can be rewritten as a dynamical system , namely @xmath33 it is easy to see that the critical points of this system are the de sitter points ( @xmath34 ) . to investigate the nature of these points we need to determine them explicitly , a non - trivial problem in the general case . it is however easy to obtain from ( [ 1.6 ] ) the equation satisfied by the critical points @xmath35 : @xmath36 the same result can be obtained directly starting from equation ( [ b ] ) of the previous section . it is convenient to introduce the notation @xmath37 , for further investigation of this equation . so , finally , we have @xmath38 first of all we find one ( trivial ) root of this equation @xmath39 ( it corresponds to @xmath40 ) , what allows us to reduce the order of the equation . ( note also that , if @xmath41 , eq . ( [ 1.9 ] ) takes a more complicate form and then it does not have the trivial root @xmath39 . ) but nevertheless the equation still is of @xmath42-order and in the interesting case ( @xmath43 ) ; this is too high and the roots can not be found algebraically . we need a specific discussion of this problem . it is clear that if there are too many ( say 10 or 20 ) de sitter points in the theory , it looks like a classical analogue ( at a reduced scale , of course ) @xcite of the string landscape vacuum structure in which case it will be far from trivial to obtain the standard cosmology . nevertheless , something can be done even in this case , by comparing the energies of the corresponding de sitter solutions , which should in fact differ , as is discussed in sect . v. using descartes rule of signs , we find that eq . ( [ 1.9 ] ) can have 2 or 0 positive roots . more detailed information can be obtained by using sturm s theorem . unfortunately it is not possible to apply it in the general case ( for arbitrary @xmath44 ) , so we need to investigate separately each of the different cases , @xmath45 , and see which one is more interesting from a cosmological viewpoint . we have therefore undertaken here a systematic analysis of all possible cases in this region of values of @xmath44 . computations are rather involved ( specially for larger @xmath44 ) , but the final results are not so difficult to describe analytically . we have found , for any of the values of @xmath44 considered , in the range above , that the number of roots depends on the parameter @xmath46 only , and that there is an @xmath47 such that , for @xmath48 , eq . ( [ 1.9 ] ) has no positive roots , while for @xmath49 , eq . ( [ 1.9 ] ) has two positive roots . we here now enumerate our results systematically : @xmath50 ; @xmath51 ; @xmath52 ; @xmath53 ; @xmath54 ; @xmath55 ; @xmath56 ; @xmath57 . note also that , near the critical point @xmath47 , both solution are very close to each other when they exist ( a real value of @xmath49 ) , and they are complex conjugate to each other for @xmath48 . this means that , in a situation where these two roots differ substantially , it must necessarily be @xmath58 . the ds - points described above can be used for the construction of inflationary or late - time acceleration behavior ( depending on the value of the scale factor @xmath59 ) . to this aim , the corresponding ds point must be unstable in the inflationary stage but can be either unstable or stable for the late - time acceleration one . however , this is in fact not a strictly necessary condition , since even for stable ds inflation , the exit from it can be achieved by a coupling with matter , through the effect of small non - local term or by some other mechanism . unfortunately , an all - round investigation of stability of the ds point turns out to be very difficult in the general case . for the recent analysis of critical points in more general modified gravity theories depending on all geometrical invariant see ref . @xcite . here we can carry out the analysis on the stability of de sitter points for the model we are dealing with in this section , only for some chosen numerical set of parameters . such computations show that one of the ds points is very likely to be always unstable ( with a smaller value of @xmath60 ) and that another one is very likely to be always stable ( with larger value of @xmath60 ) . this means that there is actually an easy way for the inflationary stage construction ( if the initial conditions lie sufficiently close to the unstable ds point , see fig . [ fig.1 ] ) . however , only within the model under consideration it might be a problem with late - time acceleration , originated from the stable point . the only possibility for late time acceleration is to choose very special parameters , in which the stable point is situated very far away from the unstable one . of course , the problem disappears when one describes only late - time acceleration within such a model , or when one takes into account other terms like the local ones . note however that specific values of the parameters in which both points are unstable can be chosen too , which give rise to a sound theory which provides a unified description of the inflationary and late - time acceleration epochs . ) near the equilibrium points . stars denote the beginning of the trajectories . dashed and dot dashed lines lead to bottom.,scaledwidth=50.0% ] near an unstable point for the dashed and dot - dashed lines of fig . [ fig.1 ] . the time origin corresponds to the stars from the previous graph.,scaledwidth=50.0% ] let us now consider specific numerical results . first , note that a full numerical investigation starting from the inflation ( or from our ds point ) and extending to the late - time acceleration epoch and going through a frw - like stage is not feasible because of the presence of numerical instabilities that prevent such possibility . for this reason , we will investigate the regions near the equilibrium points only , and especially near the unstable one . in fig . [ fig.1 ] the trajectories in the phase plane ( @xmath61 ) are presented for the set of parameters : @xmath62 , @xmath63 , @xmath64 , @xmath65 . such small values of @xmath23 and @xmath24 are motivated in @xcite , but the picture is typical in any case . the position of the ds point is specified . we may see one stable point ( with bigger value of @xmath60 ) and another , unstable one . the stars in the graph correspond to the beginning of the trajectories . we see clearly that trajectories can pass sufficiently close to the ds point , and that they escape from it , lead by the two attractors . one of them ( the upper one ) leads to a singularity , while another one ( heading towards the bottom ) , leads to a solution going somewhere in the frw - like region ( this is difficult to analyze exactly , due to numerical instabilities ) . in fig . [ fig.2 ] the evolution of @xmath66 is depicted for two trajectories : the upper one corresponds to the dashed line of fig . [ fig.1 ] , the bottom one to the dot - dashed line of fig . [ fig.1 ] ( both of them lead to the bottom in fig . [ fig.1 ] ) . we see that there actually are trajectories which start from phantom - like solutions and tend to a normal evolution ( when this is possible in the model ) . of course , these are qualitative considerations only , owing to the fact that no matter is taken into account . most probably , inclusion of matter will change the situation . however , it is just remarkable that both the early - time inflation epoch as well as the late - time accelerating one could be obtained in a unified and natural way in such a model , owing to the presence of several ds points . a slightly different choice for the function @xmath2 is motivated by the realistic and viable model which was proposed for the unification of early - time inflation and late - time acceleration in ref . @xcite , namely @xmath67}. \label{2.1}\ ] ] here we imply that @xmath68 , @xmath69 and @xmath8 is positive , what seems reasonable . now , our frw - like equation ( [ 1.3 ] ) reads @xmath70 , \end{array } \label{2.2}\ ] ] where @xmath32 . rewriting this equation under the form of a dynamical system ( [ 1.7 ] ) we find , by introducing the new variable @xmath71 , that the equilibrium points of this system ( which are ds points indeed ) are obtained from the following equation @xmath72x^m+2\kappa^2r_0f_0mx^{m-1}+\\ \\ ( f_0+f_1r_0^m)^2x+r_0(f_0+f_1r_0^m ) ( f_0+f_1r_0^m-4\kappa^2r_0^{m-1})=0 . \end{array } \label{2.3}\ ] ] it is clear that , in the general case ( even for fixed @xmath73 ) , solving this equation will not be an easy thing . we can make use of the following hint . let us require that @xmath74 be a solution of eq . ( [ 2.3 ] ) . this is possible only if there is some relation among the constants of our theory . as we can easily see from ( [ 2.3 ] ) , it must be that @xmath75 . moreover , since @xmath76 , we need that @xmath77 . substituting this condition into ( [ 2.3 ] ) we find an essential simplification of the former equation , namely @xmath78 where the root @xmath74 is already excluded . we now discuss the physical meaning of finding a solution @xmath74 . as a rule this will imply that @xmath8 is the value of the scalar curvature at present time , but in principle @xmath8 is just a parameter of the theory , which could be given any likely value . we can , for instance , fix @xmath8 to have the ds point value which is situated exactly at present time , @xmath79 , or at @xmath80 years . and there is also the possibility to explain late time acceleration if this point is stable . now let us consider other possible roots of eq . ( [ 2.4 ] ) . using sturm s theorem as in the previous section , we find several positive roots ( which correspond to ds points in the past ) and a number of negative roots ( needless to say , only values which are @xmath81 have a physical meaning ) , corresponding to ds points in the future . unfortunately , computations are harder than in the previous case and our calculations have been performed for three specific values of the parameter @xmath82 : 3 , 5 and 7 , only . in any case , the results are very similar for the three situations , what gives us hope that for larger values of @xmath82 result will be also similar . we may predict the upper limit of the ds points in the future only , which is 2 , but in any case this point is not much interesting . concerning the ds points in the past , the picture turns out to be much like the one in the previous case : the number of ds points depends only on a dimensionless parameter for all investigated values of @xmath82 . that is , @xmath83 . and there are two values , @xmath84 and @xmath85 , which are different for each @xmath82 , so for @xmath86 and @xmath87 there are two ds points and for @xmath88 there is no ds point . we have not been able to obtain exact analytical expressions for @xmath84 and @xmath85 and thus give here numerical results only . for @xmath89 @xmath90 , @xmath91 ( as positive roots of @xmath92 ) . for @xmath93 @xmath94 , @xmath95 ( as positive roots of @xmath96 ) . for @xmath97 @xmath98 , @xmath99 ( as positive roots of @xmath100 ) . as we can see , the interval where there are no ds points shrinks when the value of the parameter @xmath82 is increased . if we assume that for larger @xmath82 the number of ds points depends on the parameter @xmath101 only as clearly happens in the investigated cases we do find @xmath84 and @xmath85 for any value of @xmath82 . such investigations show that @xmath84 and @xmath85 slowly change when @xmath82 increases and that the interval without ds points is present even for very big value of the parameter @xmath82 . for example , for @xmath102 we have @xmath103 and @xmath104 . a simple numerical investigation of stability for the existing ds points shows that one of them ( the one with the smaller value of @xmath60 ) is always unstable , but the other one ( with the larger value of @xmath60 ) can either be stable or unstable . from our numerical results , the appearance of unstable ds points pairs relative to the interval @xmath86 seems most probable . note also that the condition @xmath76 means that it actually must be @xmath105 and that in this case it is most likely the one stable and one unstable point show up . now let us consider the evolution equation ( [ 2.2 ] ) from a different point of view . as we already noted , @xmath8 is a parameter of the theory which corresponds to the value of the scalar curvature at some epoch . this means that during the normal evolution of our universe , from large to little ( or zero ) curvature , it becomes @xmath7 at some moment . but , as we can see from eq . ( [ 2.2 ] ) , this means that that at this point the coefficient of the higher derivative term ( @xmath106 ) is equal to zero . this is a well - known mathematical problem , which needs special investigation . as we know from mathematics there are two possibilities : the solution of the perturbed equation ( with a higher derivative term ) may tend to the solution of the degenerate equation ( without higher derivatives ) , and then the coefficient of the higher derivative term may tend to zero or not tend to the solution of the degenerate equation . a special investigation of this problem shows that solutions of eq . ( [ 2.2 ] ) tend to solutions of the degenerate equation , which is @xmath107 when @xmath108 but not so when @xmath109 . this means that the point @xmath8 is reachable during the evolution from large @xmath3 to zero , but it is not the final point of the evolution , because there is an instability in the future . on the other hand , we have @xmath7 at this point , and therefore @xmath110 . substituting this into ( [ 2.5 ] ) we find a relation among the parameters of the theory which were introduced before , @xmath75 . this means that , generically , if one wants the point @xmath8 to be reachable during evolution , a relation of this sort among the parameters of the theory must be fulfilled . that is , we have two independent parameters only ( or even just one if we consider that @xmath8 is strictly related to our epoch ) . note also that it is impossible to use standard numerical methods near this point for solving eq . ( [ 2.2 ] ) because all available methods need solving the equation with respect to the highest derivative , and it turns out that , near this point , the numerical solution is unstable . ) near equilibrium points . stars denote beginning of trajectories . dashed and dot dashed lines lead to bottom.,scaledwidth=50.0% ] near unstable point for dashed and dot - dashed lines of fig.[fig.1 ] . zero in time corresponds to the stars from the previous graph.,scaledwidth=50.0% ] now let us consider numerical results near the stable points . since the value of @xmath8 is very small in our epoch ( @xmath111 @xmath112 ) , it is impossible to use real values in computations . thus , we have produced a qualitative analysis using the following set of parameters : @xmath113 , @xmath114 , @xmath115 , @xmath97 . as we can see , @xmath116 , and there must be two ds points corresponding to positive @xmath117 . the evolution lines near these points are represented in fig . [ fig.3 ] , where we can trace one stable and one unstable point , stars denoting the beginning of trajectories , as in the previous case . actually , there are three ds points for these chosen parameters . one of them , @xmath118 , which corresponds to @xmath74 , is degenerate and situated out of the graph . the time evolution of @xmath119 is represented in fig . [ fig.4 ] by the dashed and dot - dashed lines . the picture is very similar to the one in the previous section , and the same comments also apply here . thus , we have here shown that the unification of early - time inflation with late - time acceleration is in principle possible , and even very likely , due to the appearance of several ds points in the evolution of the universe . for a more realistic study , the presence of ( minimally and non - minimally coupled ) matter should be taken into account with care . as a third example , now we discuss a model proposed in @xcite . such model is defined by means of the function [ tan1 ] f(r)=r - f(r)=r - a , where @xmath120 , @xmath121 and @xmath8 are arbitrary parameters . one immediately sees that @xmath122 , as required and , moreover , that [ tan3 ] _ rf(r)=2_eff a. if @xmath123 in the present universe then @xmath124 plays the role of the effective cosmological constant . we also observe that the derivative [ tan4 ] f(r)=1- has a minimum when @xmath7 , which reads [ tan5 ] f(r_0)=1- . in order to avoid antigravity one needs to require [ tan6 ] 0<f(r)<f(r_0)= 1- . the model given by eq . ( [ tan1 ] ) is able to describe late acceleration , since , in general , de sitter critical points exist . they are the solutions of ( [ b ] ) . then , we set k(r)=2f(r)-rf(r ) , [ rkr ] and study ( numerically ) the zeros of this transcendental function ( see fig . [ p : tanh1 ] ) . we see that for suitable choices of the parameters there are one or two de sitter critical points . for @xmath125 and different choices of the parameter @xmath121 : @xmath126 , @xmath127 ( picture on the left ) ; @xmath128 , @xmath129 ( picture on the right).,title="fig:",scaledwidth=40.0% ] for @xmath125 and different choices of the parameter @xmath121 : @xmath126 , @xmath127 ( picture on the left ) ; @xmath128 , @xmath129 ( picture on the right).,title="fig:",scaledwidth=40.0% ] now , we impose @xmath8 to be a de sitter critical point for the model in eq . ( [ tan1 ] ) . this means that ( [ rkr ] ) has to be satisfied for @xmath7 and such a condition fixes one parameter , say @xmath120 . then , we get a= , and introducing , for convenience , the dimensionless variables x=,b_0=r_0b , we obtain f(r_0)=1- . _ eff= , k(r)=- . in order to have @xmath130 and , at the same time , to avoid antigravity , the parameter @xmath131 has to be negative and in the range @xmath132 . by varying @xmath131 in that range , the value of @xmath124 can acquire any desired value . moreover the model has always two de sitter critical points with constant curvatures @xmath133 and @xmath134 . one can see that @xmath135 , while @xmath136 . in fig . ( [ p : tanh2 ] ) we have plotted @xmath137 for some values of @xmath131 . , with @xmath8 as a fixed de sitter solution , for different choices of the parameter @xmath131 : @xmath138 , @xmath139 ( picture on the left ) ; @xmath140 , @xmath141 ( picture on the right).,title="fig:",scaledwidth=40.0% ] , with @xmath8 as a fixed de sitter solution , for different choices of the parameter @xmath131 : @xmath138 , @xmath139 ( picture on the left ) ; @xmath140 , @xmath141 ( picture on the right).,title="fig:",scaledwidth=40.0% ] thus , our study shows the existence of multiply - de sitter universes in more complicated , tangential models . one can prove that other viable @xmath2 gravities introduced in ref . @xcite naturally lead to de sitter solutions too , as demonstrated in ref . in this section we will discuss spherically symmetric exact solutions for the modified gravities above . we mainly concentrate on multiply black hole solutions and on their related entropies and free energies . we have seen in the introduction that , requiring constant curvature solutions , one may consider the simplified form of the equations of motion given in ( [ e ] ) . as a result , @xmath1 modified gravity models admit the following class of general static neutral black hole solutions in four dimensions and with a non vanishing cosmological constant . in order to describe them , recall the metric ds^2=-v(r)dt^2++r^2d_k^2,v(r)>0 , [ ds ] where @xmath142 and the possible horizon manifolds are @xmath143 , the two dimensional sphere , @xmath144 , the two dimensional torus , and @xmath145 , the two dimensional compact riemann surface . the scalar curvature for the _ ansatz _ ( [ ds ] ) reads r=- [ r^2v+4rv+2v-2k ] and this means that every constant curvature solution with @xmath7 has to satisfy the equation r^2v+4rv+2v-2k =- r^2 r_0 . the general solution of this differential equation depends on two integration constants , @xmath121 and @xmath146 , and reads v(r)=+k -- r^2 . the ricci tensor and the scalar curvature have to satisfy the equations r_=g_,r = r_0 , [ e1 ] where @xmath8 is implicitly given by ( [ b ] ) , that is r_0= , now , it is easy to verify that eqs . ( [ e1 ] ) are satisfied only if @xmath147 , while @xmath146 is an arbitrary parameter , which is usually assumed to be non - negative and related to the mass @xmath148 of the black hole by @xmath149 . the special case @xmath150 is also admissible . then , we have v(r)=k -- r^2,c0,v(r)>0 . [ vr ] in principle there are physical solutions of the latter equation for @xmath151 and @xmath152 , which give rise , respectively , to a torus topology and a hyperbolic topology for the horizon manifold ( the so called topological black holes @xcite ) . here we are mainly interested in the usual spherical symmetric horizons and so we only consider in detail the case @xmath153 . as it is well known , in the special case @xmath154 , @xmath155 in ( [ vr ] ) is always positive when @xmath156 , and this corresponds to the anti de sitter ( ads ) solution . on the contrary , when @xmath157 , @xmath155 is positive , for @xmath158 , and this corresponds to the de sitter solution . if @xmath153 and @xmath159 , one has black hole solutions but only if c^2r_0 - 032mg1 . [ bkc ] we see that ( [ bkc ] ) is always satisfied if @xmath156 . this corresponds to the schwarzschild - anti - de sitter ( sads ) black hole . in this case @xmath160 , @xmath161 being the positive root of @xmath162 ( horizon radius ) . if @xmath157 , there are solutions only if @xmath163 . in this case @xmath164 , @xmath161 and @xmath165 being the positive roots of @xmath162 ( horizon and cosmological radius respectively ) . the extremal case @xmath166 is also admissible ( nariai solution ) , but the thermodynamics of such a black hole have to be discussed separately @xcite . at this point , we provide a brief discussion regarding the thermodynamical properties of the above black hole solutions . if one make use of the noether charge method for evaluating the entropy associated with the black hole solutions with constant curvature @xcite in modified @xmath1 gravity models , one has @xcite s= f(r_h ) . [ ve ] where the factor @xmath1 has to be evaluated on the horizon , with area @xmath167 . as a consequence , one obtains a modification of the `` area law '' . several examples have been discussed in @xcite . in the case of a constant curvature solution one has simply @xmath168 , @xmath8 being the solution of eq . ( [ b ] ) , which has been investigated in previous sections . with regard to this , we have found sufficient conditions to have two de sitter solutions . thus , we may investigate their thermodynamical behavior evaluating the associated free energy . the free energy @xmath169 is related to the canonical partition function @xmath170 by = - . on the other hand , a semiclassical approximation gives ze^-i_e , where @xmath171 is the euclidean classical action associated with the de sitter solution . a direct calculation leads to i_e=- and from ( [ b ] ) and ( [ ve ] ) it directly follows that i_e =- s_h=- , [ fe ] @xmath172 and @xmath173 being , respectively , the entropy and the inverse temperature of the black hole . for a generic schwarzschild - de sitter ( sds ) solution , eq . ( [ fe ] ) reads = - . [ fe1 ] the temperature @xmath174 is related to the horizon radius @xmath161 by t_h=== |-| @xmath161 being a positive solution of the algebraic equation r_h^3-r_h+=0 . [ ho ] as a result , we finally have = -|r_h-3mg| . [ fe2 ] in the pure de sitter case @xmath175 , @xmath176 , and so one has [ fds ] for sds one needs to consider separately the two admissible cases @xmath166 and @xmath163 . here we discuss the second one only , that is , the proper sds solution where @xmath177 . in this case , eq . ( [ ho ] ) has one negative root and two distinct positive roots , @xmath161 , the event horizon and @xmath165 , the cosmological horizon , with @xmath178 . the positive roots can be written in the form r_c=,12<_c<1 , r_h= , 0<_h<12 , where _ c=12and _ h=12(3-_c ) . in this way , we get _ c = f_ds , [ fc ] _ h = f_ds , [ fh ] @xmath179 being the expression ( [ fds ] ) . it has to be noted that , in the latter equations , the factors in front of @xmath179 are always smaller than 1 and thus the energies @xmath180 and @xmath181 are always smaller than @xmath179 , independently of the value of the mass . moreover , the factor @xmath182 is a monotone decreasing function of @xmath183 , which is equal to 1 for @xmath184 and to 0 for @xmath166 , while the factor @xmath185 is exactly equal to 0 for @xmath186 and quite small otherwise . its maximum value , a little bit more than @xmath187 , is reached for @xmath188 . this means that the minimum value for the free energy @xmath181 of a sds black hole is obtained for @xmath189 . now , we shall study some models and explicitly compute the corresponding free energy . first of all , as a trivial example , we consider the @xmath190 model described by @xmath191 . this has a stable de sitter solution , with @xmath192 . then , we immediately have _ ds=- . the corresponding sds cosmological and black - hole free energies are plotted in fig . [ p : lacdm ] , as a function of the mass @xmath148 ( in energy units @xmath193 ) . and @xmath181 sds free energies for the @xmath0cdm model.,scaledwidth=40.0% ] as a second example we consider the viable model in ( [ 1.5 ] ) , with @xmath63 , @xmath194 , @xmath195 , @xmath30 , as discussed in sect . ii . with this choice of parameters , eq . ( [ b ] ) has three real solutions for @xmath196 ( minkowski ) , @xmath197 and @xmath198 . the last one ( @xmath134 ) is stable , while the other two are unstable . for the free energies of the two de sitter solutions we get ^(1)_ds=-23 -,^(2)_ds=-23 - . [ minimods ] we see that ^(2)_ds < f^(1)_ds,~3.636 . independently of the parameter @xmath22 and so the stable de sitter solution with curvature @xmath134 is favorite , from the energetic point of view , with respect to the one with smaller curvature @xmath133 . the sds free energies are plotted in fig . [ p : fcfh ] , as functions of the mass @xmath148 ( in energy units @xmath199 , for @xmath200 . however , it has to be stressed that , for the sds corresponding to the de sitter solution with curvature @xmath134 , the mass needs to be restricted to @xmath201 and thus , for a fixed mass @xmath148 , we can compare the corresponding free energies in this range only . for both the cosmological and black - hole horizons there are critical values of the black hole mass @xmath148 , say @xmath202 , for which @xmath203 . then , we can distinguish three different regions : + ( a ) @xmath204 : in this first case @xmath205 ; + ( b ) @xmath206 : in this second case @xmath207 , but @xmath208 ; + ( c ) finally , for @xmath209 , we have @xmath210 . as we have seen from general considerations , the free energies corresponding to the cosmological horizons are monotonous functions of the mass , while the ones corresponding to the black - hole horizons are convex functions , which reach the minimum at @xmath211 . for our two cases , @xmath133 and @xmath134 , we get m_1~~ , ^(1)_h , min~- , m_2~~ , ^(2)_h , min~- . [ minimum ] then we see that , for the model in ( [ 1.5 ] ) we are dealing with , the configuration with the minimum free energy corresponds to sds with mass @xmath212 and curvature @xmath213 . the associated free energy is given in ( [ minimum ] ) . we also observe that this is smaller than the free energy of the pure de sitter configuration as it is clear from ( [ minimods ] ) thus , it is demonstrated here that multiple de sitter solutions can appear also under the form of sds solutions . in other words , the number of multiply solutions becomes significantly bigger . with the appearance of both ds and sds universe solutions one can suggest various scenarios for the universe evolution . for instance , we can conjecture that a ( pre-)inflationary universe is described by some sds spacetime . as time proceeds , this universe decays and enters into the well known radiation / matter dominance phase . in its further evolution , the universe transits to the ds ( or almost ds ) era , by stability and least - energy principle considerations . the future universe may again appear as an sds spacetime . ): cosmological free energies on the left and black - hole free energies on the right.,title="fig:",scaledwidth=40.0% ] ): cosmological free energies on the left and black - hole free energies on the right.,title="fig:",scaledwidth=40.0% ] . in summary , we have investigated in this paper several viable models of modified gravity which satisfy both the constraints of local as well as cosmological tests . by means of a numerical study , it is demonstrated that some versions of highly non - linear models exhibit multiply de sitter universe solutions which often appear in pairs , being one of them stable and the other unstable . the numerical evolution of the effective equation of state parameter is presented too . as a result , these models can be considered as natural candidates for the unification of early - time inflation with late - time acceleration through ds critical points . moreover , based on the de sitter solutions , multiply sds solutions can also be constructed . further , we have investigated the thermodynamic properties of these sds universes : their corresponding entropies and free energies have been calculated and compared . sds universe might also appear at the ( pre-)inflationary stage . owing to the highly non - linear structure of the theories under discussion , the ds universes had to be constructed with numerical tools mainly . moreover , in order to simplify the problem , at this first stage of the investigation we did not to consider matter contributions . it is clear that , in the next step , we must necessarily include matter and reconsider the problem in its presence . this has the potential to lead to a sufficiently realistic quantitative description of the universe expansion history , in which modified gravity would be responsible for both acceleration stages : the inflation epoch and the dark energy one . this quantitative analysis will be presented elsewhere . * acknowledgements . * this paper is an outcome of the collaboration program infn ( italy ) and dgicyt ( spain ) . it has been also supported in part by mec ( spain ) , projects fis2006 - 02842 and pie2007 - 50i023 , by agaur ( generalitat de catalunya ) , contract 2005sgr-00790 and grant 2008be1 - 00180 , and by rfbr , grant 06 - 01 - 00609 ( russia ) . the work of p.t . was partially supported by rfbr , grant 08 - 02 - 00923 , and with the scientific school grant 4899.2008.2 of the russian ministry of science and technology . pt thanks alexey toporensky for some useful discussions . the research of ee is partly based on work done while on leave at the department of physics and astronomy , dartmouth college , 6127 wilder laboratory , hanover , nh 03755 , usa . s. nojiri and s.d . odintsov , arxiv:0807.0685[hep - th ] ; t. sotiriou and v. faraoni , arxiv:0805.1726[gr - qc ] ; s. capozziello and m. francaviglia , gen.relat.grav . * 40 * , 357 ( 2008 ) ; f. lobo , arxiv:0807.1640[gr - qc ] . s. capozziello , int.j.mod.phys . d * 11 * , 483 ( 2002 ) ; m. ruggiero and l. iorio , arxiv : gr - qc/0607093 ; a. cruz - dombriz and a. dobado , arxiv : gr - qc/0607118 ; n. poplawski , arxiv : gr - qc/0610133 ; m. c. b. abdalla et al , class . quant . grav . * 22 * , l35 ( 2005 ) [ arxiv : hep - th/0409177 ] ; a. brookfield , c. van de bruck and l. hall , arxiv : hep - th/0608015 ; h. oyaizu , m. lima and w. hu , arxiv:0807.2462[astro - ph ] ; b. li , k. chan and m. chu , arxiv : astro - ph/0610794 ; x. jin , d. liu and x. li , arxiv : astro - ph/0610854 ; t. sotiriou and s. liberati , arxiv : gr - qc/0604006 ; t. sotiriou , arxiv:0805.1160[gr - qc ] ; i. navarro and k. van acoleyen , arxiv : gr - qc/0611127 ; a. bustelo and d. barraco , arxiv : gr - qc/0611149 ; f. briscese , e. elizalde , s. nojiri and s. d. odintsov , phys . lett . b * 646 * , 105 ( 2007 ) [ arxiv : hep - th/0612220 ] ; b. li and j. barrow , arxiv : gr - qc/0701111 ; v. faraoni , arxiv : gr - qc/0703044 ; arxiv:0706.1223[gr - qc ] ; arxiv:0805.1249[gr - qc ] ; arxiv:0810.2602[gr - qc ] ; o. bertolami , c. boehmer , t. harko and f. lobo , arxiv:0704.1733[gr - qc ] ; o. bertolami and j. paramos , arxiv:0805.1241[gr - qc ] ; s. carloni , a. troisi and p. dunsby , arxiv:0706.0452[gr - qc ] ; arxiv:0707.0106[gr - qc ] ; arxiv:0809.3673[astro - ph ] ; j. santos , j. alcaniz , f. carvalho et al , arxiv:0708.0411[astro - ph ] ; arxiv:0807.2443[astro - ph ] ; h. mohseni sadjadi , arxiv:0709.2435[gr - qc ] ; n. lanahan - tremblay and v. faraoni , arxiv:0709.4414[gr - qc ] ; n. goheer , j. leach and p. dunsby , arxiv:0710.0814[gr - qc ] ; arxiv:0809.5247[gr - qc ] ; o. lecian and g. montani , arxiv:0807.4428[gr - qc ] ; t. harko , arxiv:0810.0742[gr - qc ] ; b. vakili , arxiv:0809.4591[gr - qc ] ; s. srivastava , arxiv:0809.1950[gr - qc ] ; d. saez , arxiv:0809.1311[hep - th ] ; k. bamba et al , arxiv:0803.3384[hep - th ] , phys.rev . d*77 * , 123532 ( 2008 ) ; arxiv:0801.0954[astro - ph ] , jcap * 0804*,024 ( 2008 ) ; arxiv:0807.2575[hep - th ] . p. mahato , annales de la fondation louis de broglie 32 ( 2007 ) 297 - 310;arxiv:0710.3821 [ astro - ph ] . s. nojiri and s. odintsov , gen . grav . * 36 * , 1765 ( 2004 ) [ arxiv : hep - th/0308176 ] ; phys . b * 576 * , 5 ( 2003 ) [ arxiv : hep - th/0307071 ] ; phys . * b599 * , 137 ( 2004 ) [ arxiv : astro - ph/0403622 ] ; arxiv:0801.4843[astro - ph ] ; g. cognola , e. elizalde , s. nojiri , s. d. odintsov and s. zerbini , phys . rev . * d73 * , 084007 ( 2006 ) , [ arxiv : hep - th/0601008 ] ; s. capozziello et al , arxiv : astro - ph/0501426 ; arxiv : hep - th/0512118 ; g. allemandi , a. borowiec , m. francaviglia and s. d. odintsov , arxiv : gr - qc/0504057 ; g. allemandi , m. francaviglia , m. ruggiero and a. tartaglia , arxiv : gr - qc/0506123 ; t. multamaki and i. vilja , arxiv : astro - ph/0612775 ; arxiv:0709.3422[astro - ph ] ; j. a. r. cembranos , phys . rev . d * 73 * , 064029 ( 2006 ) [ arxiv : gr - qc/0507039 ] ; t. koivisto and h. kurki - suonio , arxiv : astro - ph/0509422 ; t. clifton and j. barrow , arxiv : gr - qc/0509059 ; i. brevik , arxiv : gr - qc/0601100 ; t. faulkner , m. tegmark , e. bunn and y. mao , arxiv : astro - ph/0612569 ; s. capozziello et al , arxiv:0804.2163[gr - qc ] ; arxiv:0712.2980[gr - qc ] ; arxiv:0810.2549[gr - qc ] ; s. nojiri , s. d. odintsov and p. tretyakov , arxiv:0704.2520[hep - th ] ; m. movahed , s. baghram and s. rahvar , arxiv:0705.0889[astro - ph ] ; c. boehmer , t. harko and f. lobo , arxiv:0710.0966[gr - qc ] ; j. miritzis , arxiv:0708.1396[gr - qc ] ; d. vollick , arxiv:0710.1859[gr - qc ] ; j. evans , l. hall and p. caillol , arxiv:0711.3695[astro - ph ] ; t. clifton , arxiv:0807.4682[gr - qc ] ; h. stefancic , arxiv:0807.4428[gr - qc ] ; a. dev et al , arxiv:0807.3445[hep - th ] ; t. kobayashi and k. maeda , arxiv:0807.2503[astro - ph ] ; s. nojiri , s. d. odintsov , phys . d * 74 * , ( 2006 ) 086005 [ arxiv : hep - th/0608008 ] ; arxiv : hep - th/0611071 ; s. capozziello , s. nojiri , s. d. odintsov and a. troisi , phys . b * 639 * , 135 ( 2006 ) [ arxiv : astro - ph/0604431 ] . s. tsujikawa , k. uddin and r. tavakol , arxiv:0712.0082 ; s. fay , s. nesseris and l. perivolaropoulos , arxiv : gr - qc/0703006 . g. cognola , m. gastaldi and s. zerbini , arxiv : gr - qc/0701138 ; j.l . cortes and j. indurain , arxiv:0805.3481[astro - ph ] ; s. tsujikawa and t. tatekawa , arxiv:0804.4343[astro - ph ] ; n. deruelle , m. sasaki and y. sendouda , phys.rev . d * 77*,124024 ( 2008 ) ; k. kainulainen and d. sunhede , arxiv:0803.0867[gr - qc ] ; g. cognola and s. zerbini , j.phys . a * 39 * , 6245 ( 2006 ) . s. a. appleby and r. a. battye , arxiv:0705.3199[astro - ph ] ; l. pogosian and a. silvestri , arxiv:0709.0296[astro - ph ] ; s. tsujikawa , arxiv:0709.1391[astro - ph ] ; s. capozziello and s. tsujikawa , arxiv:0712.2268[gr - qc ] . i.brevik , s. nojiri , s.d . odintsov and l. vanzo , hep - th 0401073 , phys.rev . d * 70 * , 043520 ( 2004 ) ; g. cognola , s. nojiri , s.d . odintsov and s. zerbini , hep - th/0501096 , jcap * 0502 * , 010 ( 2005 ) ; f. briscese and e. elizalde , phys . d * 77 * , 044009 ( 2008 ) [ arxiv:0708.0432 [ hep - th ] ] .

physics of graphene perception begins when one compares landau levels in two dimensional schrdinger and dirac theories . such spectacular phenomenon as unconventional quantum hall effect @xcite is caused by the anomaly of the lowest landau level ( lll ) @xcite which for dirac fermions in graphene is field independent and can accommodate only half the usual number of the states from the conduction band and takes the other half from the valence band . the easiest way to accomplish this peculiar feature of the lll is to solve a pair of the dirac equations that describe excitations near two inequivalent @xmath0 points of graphene s brillouin zone . normally this is done in a constant homogeneous magnetic field , although this property of the lll for dirac fermions is topologically protected for inhomogeneous field configurations and in the presence of ripples @xcite . the simplest inhomogeneous field configuration which contains nontrivial aharonov - bohm physics can be created by an infinitesimally thin solenoid . in practice such magnetic field configuration may be obtained when a type - ii superconductor is placed on top of graphene or semicondunducting heterojunction hosting a 2d electron gas ( 2deg ) with quadratic dispersion . while graphene devices still have to be fabricated , devices like this with a superconducting film grown on top of a semiconducting heterojunction ( such as gaas / algaas ) hosting a 2deg have in fact been fabricated twenty years ago @xcite and theoretically well studied ( see e.g. refs . ) . theoretically a problem of the dirac fermions in the field of aharonov - bohm flux was encountered in the context of cosmic strings by gerbert and jackiw @xcite . while for the solutions of the dirac equation with nonzero angular momentum the square integrability requirement specifies which of the two independent solutions should be taken , they noticed this is not the case for the zero angular momentum . for the zero momentum there is an ambiguity as both solutions are square integrable , but divergent as @xmath1 at the origin , where @xmath2 is the space coordinate . the ambiguity of the solution selection is caused by the singular nature of the infinitesimally thin solenoid vector potential at the origin . this problem has initiated a vast theoretical literature which addresses interesting aspects related to the rigorous treatment of the solutions of the dirac equation with the aharonov - bohm potential ( see e.g. refs . for a review ) . in the condensed matter context the dirac fermions in the field of solenoid emerged during the study of the dirac - bogolyubov - de gennes quasiparticles in the vortex state of @xmath3-wave superconductors @xcite ( see also @xcite for a review ) . due to the divergence of the zero mode solutions theory predicts a formation of nonzero local density of states ( ldos ) near the vortex center . however , this theoretical prediction based on the dirac nature of quasiparticles in @xmath3-wave superconductors does not agree with the results of scanning tunneling spectroscopy ( sts ) measurements @xcite in high - temperature superconductors . finally we mention a related problem of the description of topological defects in graphene based on the dirac equation with a pointlike pseudomagnetic vortex which has also been studied intensively , see e.g. refs . . the purpose of the present paper is to study the electronic excitations in graphene in the field of the aharonov - bohm flux and compare them with the corresponding results for 2deg with a quadratic dispersion . we rely on the existing studies of the dirac fermions in the aharonov - bohm potential , but focus on the specific features of graphene such as the presence of two inequivalent @xmath0 points which implies that one should consider the solutions for both inequivalent irreducible representations of the dirac @xmath4 matrices . also to avoid unnecessary formal complications we consider the physical regularization of the problem modeling a finite radius flux tube created by the abrikosov vortex . we utilize the simplest case of magnetic field concentrated in a thin cylindrical shell of small but finite radius @xmath5 when @xmath6 @xcite . in contrast to high - temperature superconductors dirac description of the quasiparticles in graphene is proven valid under the different conditions . in particular , sts measurements of graphene flakes on graphite @xcite exhibit the structural and electronic properties expected of pristine graphene such as the development of a single sequence of pronounced landau level peaks corresponding to massless dirac fermions in a homogeneous magnetic field . we propose to perform sts measurements for graphene penetrated by vortices from a type - ii superconductor , because the dirac theory predicts rather peculiar behavior of ldos not expected for the 2deg with a quadratic dispersion of carriers . the paper is organized as follows . in sec . [ sec : model ] we introduce the model hamiltonians and discuss the regularization of the aharonov - bohm potential used in this work . [ sec : schrodinger ] is devoted to the nonrelativistic case , and the relativistic case is discussed in detail in sec . [ sec : dirac ] . the structure of both sections is the same : we consider the solution of the corresponding schrdinger or dirac equation , construct the green s function ( gf ) with coinciding arguments , obtain the dos and study the behavior of the ldos . in sec . [ sec : concl ] our final results are summarized . we consider nonrelativistic and relativistic hamiltonians . the 2d nonrelativistic ( schrdinger ) hamiltonian has the standard form @xmath7 where @xmath8 , @xmath9 with the vector potential @xmath10 , planck s constant @xmath11 and the velocity of light @xmath12 describes a spinless particle with a mass @xmath13 and charge @xmath14 . the dirac quasiparticle in graphene is described by the hamiltonian @xmath15 where the matrices @xmath16 and @xmath17 are defined in terms of the pauli matrices as @xmath18 here @xmath19 labels two unitary inequivalent representations of @xmath4 gamma matrices , so that one considers a pair of dirac equations corresponding to two inequivalent @xmath20 points of graphene s brillouin zone . in eq . ( [ hamilton - rel ] ) @xmath21 is the fermi velocity and @xmath22 is the dirac mass ( gap ) , which is introduced in the hamiltonian for generality . note that we consider the simplest case when the gap has the same sign for @xmath19 [ see ref . for a discussion of more general cases ] . while tight binding calculations show that the quasiparticle excitations in graphene have a linear dispersion at low energies @xcite and are described by the massless dirac equation with @xmath23 @xcite , recent sts measurements revealed a mass gap near the dirac point in a single layer graphene sample suspended above a graphite substrate @xcite . since this gap and its origin are intensively studied both theoretically and experimentally in the last few years , here we consider a generic case with a finite value of @xmath22 . the vector potential of a vortex at the origin directed in the @xmath24 direction is @xmath25 where @xmath26 is the flux of the vortex expressed via magnetic flux quantum of the electron @xmath27 with @xmath28 , . ] where the value @xmath29 corresponds to the flux created by the abrikosov vortex . the magnetic field is then @xmath30 the essential difference between the schrdinger @xmath31 and dirac @xmath32 equations in this case can be seen if one squares the latter : @xmath33 \right ) \psi = ( e^2 - \delta^2 ) \psi,\ ] ] where the commutator @xmath34 = -\frac{e}{\hbar c } b_z(\mathbf{r})\ ] ] which introduces pseudo - zeeman term which is related to the sublattice rather than the spin degree of freedom . it should be mentioned that in the case of graphene the components of the spinor @xmath35 are associated with a sublattice rather than a spin degree of freedom . since the hamiltonian ( [ hamilton - rel ] ) originates from a nonrelativistic many - body theory , the zeeman interaction term has to be explicitly added to this hamiltonian . this resembles the case of the nonrelativistic hamiltonian ( [ hamilton - nonrel ] ) which becomes pauli one when the interaction between the magnetic moment of the spin and an external magnetic field is added . in the present paper we do not include the spin degree of freedom neither in ( [ hamilton - nonrel ] ) nor in ( [ hamilton - rel ] ) . ( [ dirac - squared ] ) and ( [ commutator ] ) identify the origin of complications @xcite in the problem with a singular vortex ( [ a - singular ] ) when a singularity in @xmath36 occurs at a singular point of the differential equation ( [ dirac - squared ] ) . to avoid these complications one can consider a vortex with a finite radius flux tube @xcite , i.e. with the magnetic field and vector potential written in cylindric coordinates @xmath37 : @xmath38 where @xmath39 is a profile function with a compact support satisfying the normalization @xmath40 and connected to the profile function @xmath41 by the relation @xmath42 . the simplest choice of the field distribution @xmath39 which regularizes the problem with the solutions solely expressed in terms of bessel functions is a magnetic field concentrated on the surface of the cylinder of radius @xmath5 , @xmath43 . then , the corresponding profile function @xmath44 in the limit @xmath45 we recover the aharonov - bohm potential ( [ a - singular ] ) but avoiding formal complications . as shown in ref . @xcite , there is no dependence on the detailed form of @xmath39 in the limit @xmath6 provided that @xmath46 . in the present paper we restrict ourselves by considering the profile function ( [ step - profile ] ) . in this section we recapitulate the results of refs . for nonrelativistic case . they are important not only for comparison with relativistic case , but also because the relativistic result is constructed using the nonrelativistic one . in the limit @xmath47 the admissible solution of the schrdinger equation is always a regular solution which in polar coordinates @xmath48 takes the form @xmath49 where @xmath50 is the bessel function with the wave vector @xmath51 which is related to the quasiparticle energy @xmath52 via @xmath53 . the eigenfunction expansion for the retarded schrdinger gf reads @xmath54 or after substituting the wave function ( [ solution - nonrel ] ) it becomes @xmath55 where @xmath56 . since an analytic continuation of the gf ( [ nonrel - gf ] ) on the imaginary axis in the complex momentum plane , @xmath57 is free of singularities , it is convenient to work with the corresponding gf @xmath58 where @xmath59 in eq . ( [ nonrel - gf - continued ] ) we already set two arguments coinciding , @xmath60 , because in the present work we consider the dos only . as we will see below , the function @xmath61 is also used in the representation of the dos for the dirac fermions . after the calculation of the gf ( [ g - continued ] ) is done , the ldos per spin projection can be found by returning back to the real momentum axis @xmath62 the gf @xmath63 was calculated in ref . using the contour integration technique . a weak point of this calculation was discussed in @xcite , where the same method was applied to obtain the @xmath64-dependent contribution to the gf , @xmath65 with coinciding arguments when the approach of @xcite is valid . referring to the derivation of @xcite , here we simply start from the corresponding expression for @xmath66 obtained in ref . @xmath67 notice that eq . ( [ delta - g - marino ] ) coincides with the corresponding expression from @xcite up to a coefficient . changing the variables to @xmath68 , @xmath69 one can obtain from ( [ delta - g - marino ] ) the final expression @xmath70}{\cosh x } e^{-2 \mathcal{q } r \cosh x\cosh y}\ ] ] which we will use in what follows . it turns out that the investigation of the full dos is simpler than the analysis of the ldos . thus in the next sec . [ sec : dos - nonrel ] we firstly consider the dos and return to the ldos ( [ ldos - def ] ) below in sec . [ sec : ldos - nonrel ] . the full dos per spin projection is obtained from the ldos ( [ ldos - def ] ) by integrating over the space coordinates @xmath71 since we have the integral representation ( [ delta - g ] ) for @xmath72 , it is straightforward to calculate directly the perturbation of dos , @xmath73 induced by the aharonov - bohm potential . here @xmath74 is a free dos of 2deg per unit area and @xmath75 is the 2d volume ( area ) of the system . firstly integrating @xmath72 over space , one obtains @xcite @xmath76 then integrating over @xmath77 and @xmath78 we obtain that @xmath79 returning to the real @xmath80 axis we reproduce the usual aharonov - bohm depletion of the dos with respect to the free dos @xmath81 @xcite : @xmath82 writing eq . ( [ norel - dos - depletion ] ) we have also included a case of the opposite field direction . now we come back to the ldos ( [ ldos - def ] ) . as in the case of the full dos , it is convenient to consider the excess ldos , @xmath83 . then the value @xmath84 can be obtained by calculating the function @xmath85 given by eq . ( [ delta - g ] ) and substituting the result to eq . ( [ nonrel - gf - continued ] ) . the analytic continuation @xmath86 described by eq . ( [ ldos - def ] ) has to be done at the very last step of the calculation . thus our purpose is to derive a simple representation for the function @xmath87 . first , one can rewrite it in the form @xmath88 where we used that @xmath89 . differentiating ( [ delta - g ] ) we get the integrand of the last expression @xmath90\int_0^\infty dy \cosh y e^{-2\mathcal{q } r\cosh x\cosh y}.\ ] ] using the integral representation for the modified bessel function @xmath91 @xcite @xmath92 and the formula ( 2.16.13.2 ) from @xcite @xmath93 we come to the equation @xmath94 integrating the last expression with mathematica , we get @xmath95 - \ln \frac{\mathcal{q } r}{2 } \\ & - \frac{\mathcal{q}^2 r^2}{4 \eta(1-\eta)}\,_3f_4(1,1,3/2;2,2,2-\eta , 1+\eta;\mathcal{q}^2 r^2 ) + \frac{1}{2}\psi(1-\eta)+ \frac{1}{2}\psi(\eta ) , \end{split}\ ] ] where @xmath96 is the generalized hypergeometric function and @xmath97 is the logarithmic derivative of the gamma function @xmath98 . after the analytic continuation @xmath99 is made only two terms in the square brackets ( with the hypergeometric function itself remaining real ) and logarithmic term contribute in @xmath100 , so that @xmath101 -1\}\ ] ] with @xmath102 note that the last term @xmath103 in eq . ( [ ldos - function ] ) arises from the logarithmic term of ( [ hypergeometric - integral ] ) . in the important limiting cases the last expression is greatly simplified . for example , in the limit @xmath104 one obtains @xmath105 also in the physically important case @xmath29 the ldos is expressed in terms of the sine integral , @xmath106 as follows @xmath107.\ ] ] using the asymptotic of the integral sine @xcite @xmath108 for @xmath109 from ( [ ldos - si ] ) we recover the previous expression ( [ ldos - large ] ) valid for @xmath29 and @xmath104 . in the opposite limit @xmath110 for @xmath111 , we see that @xmath112 $ ] . in fig . [ fig:1 ] we show the dependence ( [ ldos - si ] ) of the induced ldos @xmath113 on the distance from the center of the vortex @xmath2 . we observe that in the case of nonrelativistic 2deg the presence of the vortex induced the depletion of the ldos for small @xmath114 . the function @xmath115 crosses zero near @xmath116 and for @xmath117 the function it reaches the maximal value @xmath118 . in fig . we model a situation when the stm tip is positioned at some distance from the center of vortex and a bias voltage is swept to explore the energy dependence of the ldos . to take into account the presence of a finite carrier density in 2deg , we introduce a finite fermi energy @xmath119 , so that the ldos at zero energy , @xmath120 , corresponds its value at the fermi level , i.e. @xmath121 . to choose the appropriate units we set the distance scale @xmath122 to be the order of the lattice constant . then the energy scale , @xmath123 is the order of the bandwidth . the dimensionless variable @xmath124 can now be rewritten as follows @xmath125 . the dependence of @xmath126 is shown in fig . [ fig:2 ] for three values of @xmath127 : for @xmath128 solid ( blue ) curve , for @xmath129 long - dashed ( red ) curve and for @xmath130 short - dashed ( black ) curve . the chemical potential is taken @xmath131 . we observe that only for the smallest ratio @xmath132 the values of the ldos are significantly depleted below the free ldos @xmath133 . as we saw , the depletion of the ldos occurs for @xmath134 . since the presence of the fermi surface makes the value of @xmath80 large , so that the region of small @xmath135 is accessible only for @xmath136 . indeed , we observe that only for the smallest ratio @xmath132 the values of the ldos are depleted to a half of the value of free ldos @xmath133 . since the realistic values of the vortex core size are at least of the order of magnitude larger than @xmath122 , this implies that the region of a significant depletion of the ldos is not accessible experimentally . still , due to the slow decay of @xmath137 even for @xmath138 the amplitude of @xmath139 oscillations is of order of @xmath140 , so that this behavior can be probably observed experimentally . in this section we examine the density of states for dirac particles in the potential of the infinitesimally thin solenoid . to avoid formal complications we consider the physical regularization of the problem with the magnetic field concentrated in a thin cylindrical shell of small but finite radius @xmath5 and take the limit @xmath45 at the end of the calculation . all answers are presented in the form convenient for comparison with the case of the schrdinger equation . the dirac equation ( [ dirac - eq ] ) in the field of the regularized vortex ( [ profile ] ) , ( [ step - profile ] ) ( see also ( [ a - regular ] ) ) is solved in appendix [ sec : a ] . in our consideration we follow refs . . the profile ( [ a - regular ] ) implies that for @xmath141 the particle obeys the dirac equation ( [ system : r < r ] ) for a free particle , while for @xmath142 the particle moves in the field of aharonov - bohm vortex ( [ system : r > r ] ) . accordingly , for @xmath141 the squared dirac equation ( [ dirac - squared ] ) is equivalent to free schrdinger equations for the components of the spinor @xmath143 notice that in the appendix [ sec : a ] the definition ( [ psi - def - appendix ] ) for @xmath144 explicitly includes the factor @xmath145 . for @xmath142 the components @xmath35 satisfy the schrdinger equations with aharonov - bohm potential and the commutator ( [ commutator ] ) is singular at @xmath146 : @xmath147=-\frac{\eta}{r}\delta(r - r).\ ] ] the solution of the problem can found by matching the solutions obtained in the domains @xmath148 and @xmath142 [ see e.g. ( [ solution - psi1 ] ) ] . the radial components of the spinor @xmath149 have to be continuous : @xmath150 and the singularity of the commutator ( [ commutator - cylinder ] ) is taken into account by a condition on the derivatives @xmath151 we stress that in contrast to the dirac equation case , for the nonrelativistic case when the solution ( [ solution - nonrel ] ) is obtained using the same regularization procedure , both the wave function and its derivative should be continuous . this as we saw from eqs . ( [ dirac - squared ] ) , ( [ commutator ] ) and ( [ components - cartesian ] ) is related to the pseudo - zeeman term . after the limit @xmath45 is taken we obtain for the case @xmath152 the following solutions : @xmath153 for positive value of the energy @xmath154 and @xmath155 ; @xmath156 for negative value of energy @xmath157 and @xmath155 . here the upper and lower sign of the second spinor s component corresponds to @xmath158 and @xmath159 solutions , respectively . the @xmath160 solution turns out to be special , while the upper component is regular at @xmath161 , the lower component diverges as @xmath162 : @xmath163 for @xmath164 and @xmath165 for @xmath166 . one can check this property by using the matching conditions for @xmath167 component and then finding @xmath168 from eq . ( [ system : r > r - for2 ] ) . on the other hand , one can discover this singularity in the limit @xmath6 by directly analyzing the matching conditions for @xmath168 and then finding nonsingular @xmath167 from eq . ( [ system : r > r - for1 ] ) . comparing eqs . ( [ m=0.e>0 ] ) and ( [ m=0.e<0 ] ) we observe that a singular at @xmath161 zero - mode solution @xmath169 is hole - like , because a singular electron - like solution vanishes due to the @xmath170 factor . concluding the discussion of the solutions for the case @xmath152 we stress that for the opposite field direction it is the upper component of the spinor which is singular in the solution equivalent to eqs . ( [ m=0.e>0 ] ) and ( [ m=0.e<0 ] ) . moreover , the nonvanishing zero mode is now electron - like . as it was firstly noticed in ref . , this behavior under the change in the field direction breaks the symmetry under @xmath171 ( see also ref . for a discussion ) . the set of solutions for the case @xmath172 is the following : @xmath173 for @xmath164 and @xmath174 for @xmath166 . the prescription for the upper and lower sign of the second spinor s component is the same as for @xmath152 . the @xmath160 solutions in the case @xmath172 are the following : @xmath175 for @xmath164 and @xmath176 for @xmath166 . we observe that in this case it is the upper component of the spinor which is singular at @xmath161 and nonvanishing is the electron - like zero mode . now we are at the position to construct the gf for the dirac fermions using the presented above solutions . however , before going to this , we should stress that as shown in@xcite the field configuration with @xmath177 confined to the surface of a cylinder of radius @xmath5 is not essential for the main result . in simple words , an more general and not singular at the origin potential can be considered as a set of concentric shells . or putting more formally , a profile function @xmath39 should satisfy the condition given below eq . ( [ step - profile ] ) which excludes delta - funtion at the origin . the eigenfunction expansion for the retarded dirac gf s now includes both positive and negative energy solutions @xmath178 where depending on the sign of @xmath179 the sum is taken over either @xmath152 or @xmath172 solutions . accordingly , for the diagonal matric elements of @xmath180 , for instance , in the @xmath152 case we obtain @xmath181 and @xmath182 , \end{split}\ ] ] where @xmath183 . then the gf with coinciding arguments , @xmath60 acquires the following form @xmath184 \end{split}\ ] ] for @xmath152 , and @xmath185 , \\ g^{\mathrm{d}}_{\eta22}(\mathbf{r},\mathbf{r},e+i0;\zeta=-1 ) & = \frac{e-\delta}{2\pi(\hbar v_f)^2}g_\eta(\mathbf{r},q ) , \end{split}\ ] ] for @xmath172 . here the function @xmath186 is related to the function @xmath61 defined in eq . ( [ g - continued ] ) by analytic continuation @xmath187 . note that while the function @xmath61 is identical for both nonrelativistic and relativistic cases , the real momentum @xmath80 function @xmath186 has a different analytical properties in these cases reflecting the fact that in contrast to the nonrelativistic case the relativistic spectrum contains positive and negative energy branches . similarly , the function @xmath188.\ ] ] is the analytic continuation to the imaginary axis of the function @xmath189 . the @xmath190 contribution to @xmath191 originates from the zero mode solutions of the dirac equation . using the integral [ ( 2.12.32.12 ) from@xcite ] @xmath192 where @xmath193 is the modified bessel function and @xmath194 is the macdonald function , we obtain the following simple result for @xmath190 @xmath195 due to the matrix structure of the gf and the presence of the valley degree of freedom , @xmath196 , in contrast to the nonrelativistic expression ( [ dos - nonrel ] ) , the full dos in the relativistic case involves not only the integration over area , but also summing over the diagonal components of the gf and @xmath20 valleys . for a better understanding of the final result for the full dos , it is instructive to consider separate expressions for the dos corresponding separate @xmath20 points @xmath197.\ ] ] recall that for @xmath198 the free dos of the dirac quasiparticles per spin and one valley is equal to @xmath199 . using eqs . ( [ dirac - g - zeta=1 ] ) and ( [ dirac - g - zeta=-1 ] ) one can express the dos ( [ dirac - dos - zeta ] ) via the functions @xmath200 and @xmath201 defined by eqs . ( [ g - continued ] ) and ( [ f - continued ] ) , respectively , as follows @xmath202 . \end{split}\ ] ] the integral of used in the nonrelativistic case function @xmath200 , or to be precise of the function @xmath87 , over the space coordinates is calculated in eq . ( [ g - integral ] ) . the corresponding space integration of the function @xmath201 given by eq . ( [ f - macdonald ] ) produces the result @xmath203 notice that since @xmath204 for @xmath198 , there is no need to introduce a function @xmath205 . having these results for the spatial integration we can calculate the imaginary part . since both eqs . ( [ g - integral ] ) and ( [ f - integral ] ) depend on @xmath206 , the imaginary part is obtained by using a simple prescription @xmath207/(\hbar v_f)^2 $ ] . this gives the final expression for the perturbed dos , @xmath208 by the aharonov - bohm vortex @xmath209 + ( [ dirac - dos - final ] ) which is @xmath210 originates from @xmath211 part of eq . ( [ dirac - dos - g&f ] ) and resembles the nonrelativistic result ( [ norel - dos - depletion ] ) . for @xmath23 it turns out to be twice larger than ( [ norel - dos - depletion ] ) simply because in eq . ( [ dirac - dos - final ] ) we summed over the diagonal components of the gf @xmath180 which are related to the two sublattices of graphene . the last term of ( [ dirac - dos - final ] ) originates from @xmath190 part of eq . ( [ dirac - dos - g&f ] ) and thus is related to the zero mode solution of the dirac equation . the mentioned above fact that the singular component of the zero mode solution is hole - like for @xmath152 [ see eq . ( [ m=0.e<0 ] ) ] and it is electron - like for @xmath172 [ see eq . ( [ m=0.e>0-negative - zeta ] ) ] finds its reflection in the asymmetric form of the last term of eq . ( [ dirac - dos - final ] ) which also corresponds to the holes ( electrons ) for @xmath212 ( @xmath172 ) . in accordance with the behavior of the zero mode solutions described in sec . [ sec : dirac - solutions ] , when the direction of the field is reversed the expressions @xmath213 and @xmath214 are interchanged . the full excess dos @xmath215\ ] ] is obviously symmetric in energy . in contrast to the nonrelativistic case , in the dirac case the aharonov - bohm vortex induces the excess of the states which is related to the presence of the last term of eq . ( [ dirac - dos - final ] ) and caused by the zero modes . we note that in ref . the corresponding term of eq . ( [ dirac - dos - final ] ) has a wrong sign . the positiveness of @xmath216 can also be understood by the following simple argument . for @xmath217 the free dos @xmath218 . therefore , since the dos has to be positive , the value @xmath216 should also be positive . now we investigate the ldos for the dirac case . while it was useful to consider each valley separately , especially because field theoretical studies of the problem often involve only one unitary inequivalent representation of @xmath219 gamma matrices [ see e.g.@xcite ] , the ldos measurement picks up both valleys together . on the other hand , ldos distinguishes sublattices . thus we consider separately the ldos for @xmath220 and @xmath177 sublattices which are defined as follows @xmath221 , \\ n_\eta^{\mathrm{d } ( b)}(\mathbf{r},e)=-\frac{1}{\pi } \mbox{im } \left [ g_{\eta 22}(\mathbf{r},\mathbf{r},e+i0;\zeta=1 ) + g_{\eta22}(\mathbf{r},\mathbf{r},e+i0;\zeta=-1 ) \right ] . \end{split}\ ] ] again we consider the perturbation of the ldos induced by the aharonov - bohm potential , @xmath222 . using eqs . ( [ dirac - g - zeta=1 ] ) and ( [ dirac - g - zeta=-1 ] ) we rewrite the ldos s @xmath223 in terms of the functions @xmath224 and @xmath201 as follows @xmath225.\ ] ] here the upper ( lower ) sign corresponds to @xmath220 ( @xmath177 ) sublattice . to obtain the final expression for ldos we should make the analytic continuation @xmath226 . for the function @xmath227 given by eq . ( [ hypergeometric - integral ] ) one obtains @xmath228 - 1\right\},\ ] ] where the function @xmath229 is defined in eq . ( [ f - eta ] ) . one can easily see that because in the nonrelativistic case the analytic continuation to the real momentum @xmath99 is different from the analytic continuation @xmath226 in the relativistic case , the function ( [ g - real - rel ] ) differs from the function ( [ ldos - function ] ) . using the relationships @xcite @xmath230 between the macdonald function of the imaginary argument , the hankel function of the second kind @xmath231 and the bessel function of the first @xmath232 and second @xmath233 kinds , we obtain that @xmath234,\ ] ] where we used the property @xmath235 . accordingly , the analytic continuation of the function @xmath236 from eq . ( [ f - macdonald ] ) takes the form @xmath237 - 2\cos\pi\eta j_\eta(qr)y_\eta(qr)\}.\ ] ] substituting ( [ g - real - rel ] ) and ( [ f - real - rel ] ) in eq . ( [ ldos - ab - f&g ] ) we arrive at the final main result @xmath238 - 1\\ + & \frac{\sin^2(\pi\eta)}{2 } [ y^2_\eta(qr)- j^2_\eta(qr)]- \frac{\sin ( 2 \pi\eta)}{2 } j_\eta(qr)y_\eta(qr ) \biggr\ } , \qquad q = \frac{\sqrt{e^2 - \delta^2}}{\hbar v_f } , \end{split}\ ] ] where @xmath239 is free dos of the dirac quasiparticles per spin and one sublattice ( or valley ) for @xmath23 . the first part of eq . ( [ ldos - ab - final ] ) which includes @xmath240 and @xmath241 is identical to the nonrelativistic expression ( [ ldos - function ] ) , while the second part of eq . ( [ ldos - ab - final ] ) with bessel functions originates from the zero mode contribution . in the limit @xmath104 the last expression acquires a simple form @xmath242 comparing eqs . ( [ ldos - ab - large - qr ] ) and ( [ ldos - large ] ) we conclude that in the dirac case the impact of the vortex is more localized than in the nonrelativistic case . as we saw in sec . [ sec : ldos - nonrel ] in the physically important case @xmath29 the expression ( [ ldos - function ] ) is significantly simplified to the result ( [ ldos - si ] ) . the same remains true for eq . ( [ ldos - ab - final ] ) , because half - integer bessel functions are expressed in terms of the elementary functions and we obtain that @xmath243.\ ] ] from eq . ( [ ldos - ab - si - cos ] ) we immediately observe the main difference between the relativistic and nonrelativistic cases . the presence of zero modes causes a positive divergence of the ldos , @xmath244 for @xmath114 near the center of the vortex . integrally this results in the excess of the states in the full dos ( [ dirac - dos - full ] ) . using the asymptotic expansion of @xmath245 given below eq . ( [ ldos - si ] ) we recover the previous expression ( [ ldos - ab - large - qr ] ) valid for @xmath29 and @xmath104 . in fig . [ fig:3 ] we show the dependence ( [ ldos - ab - si - cos ] ) of the induced ldos @xmath246 on the distance from the center of the vortex @xmath2 for @xmath23 . we observe the features expected from the analytic expressions such as the excess of the ldos for small @xmath114 and faster than in 2deg decay of @xmath247 for @xmath104 . in fig . [ fig:4 ] we consider a situation similar to fig . [ fig:2 ] . we fix the distance at @xmath248 and plot the energy dependence of the relativistic ldos ( [ ldos - ab - si - cos ] ) . we consider the most interesting case of undoped graphene with zero carrier density . in contrast to the 2deg , in graphene is easily tuned to this regime . again , we introduce the distance scale @xmath122 of the order of the lattice constant . then for the dirac case the energy scale is @xmath249 , and accordingly the dimensionless variable is @xmath250 . to make sublattices inequivalent we introduce a finite gap @xmath251 which introduces the asymmetry between the the ldos on @xmath220 and @xmath177 sublattices . the ldos @xmath252 shown as a solid ( blue ) curve has a sharp peak near @xmath253 and practically no peak at @xmath253 , while the ldos @xmath254 shown as dashed ( red ) curve has a peak near @xmath255 and no peak at @xmath256 . when @xmath257 is increasing the ldos very quickly reduces to its value for the dirac system in the absence of the vortex . we note that this behavior of the ldos is seen for the large value of @xmath138 , i.e. far from the center of the vortex , indicates that it should be possible to observe this excess of the ldos in experiments on graphene . finishing our discussion of the dirac case , we mention a recent work @xcite , where the effect of the vacuum polarization in the field of an infinitesimally thin solenoid at the distances much larger than the radius of solenoid was studied . constructing the gf the authors neglected the delta - function ( [ discontinuity ] ) motivated by the fact that they are interested in the regime @xmath258 . one of their interesting conclusions is that for @xmath29 the induced current is zero . the main motivation of this work was to address a natural question whether one can distinguish graphene from 2deg measuring the ldos near the abrikosov vortex penetrating them . just by comparing figs . [ fig:2 ] and [ fig:4 ] we find a positive answer . indeed , by putting the stm tip close to the vortex , for example , at the distance , @xmath259 , where @xmath122 is the lattice constant , we observe that in 2deg the value of the ldos will be close to its background value @xmath260 in the free system , while in undoped graphene one should observe a strong ldos enhancement near the fermi level . or to be more precise , if there is no gap @xmath22 in the quasiparticle spectrum , the peak of the ldos should be observed at the fermi level , or if @xmath261 the peaks should be observed at the energies @xmath262 and the sign of the peak energy should depend on the sublattice . this peaked behavior of the ldos with @xmath244 for @xmath114 reflects the specific feature of the dirac fermions such as the presence of the divergent as @xmath1 at the origin zero mode solution of the dirac equation . thus the observation of this feature in sts measurements would contribute to the expanding list of the experimental manifestations of the dirac fermions in graphene . our second conclusion is that while in the nonrelativistic case the presence of the aharonov - bohm vortex leads to a depletion of the full dos , in the dirac case the full dos is enhanced . this result can likely be checked by analyzing the sts maps . we remind at the end that in this paper we did not consider the effect of disorder in the presence of magnetic field which will definitely affect the behavior of the ldos in magnetic field @xcite and should be taken into account in the analysis of the sts data . we thank e.v . gorbar and v.p . gusynin for many stimulating discussions . s.g.s . thanks c.g . beneventano and e.m . santangelo for an interesting discussion and bringing ref . to his attention . this work was supported by the scopes grant no . iz73z0`_`128026 of swiss nsf , by the grant simtech no . 246937 of the european fp7 program and by the program of fundamental research of the physics and astronomy division of the national academy of sciences of ukraine . was also supported by the ukrainian state foundation for fundamental research under grant no . a positive energy solution of the time - dependent dirac equation has a form @xmath263 with a two - component spinor @xmath35 satisfying the time - independent dirac equation ( [ dirac - eq ] ) : @xmath264\psi({\mathbf r})=0.\ ] ] where @xmath19 distinguishes two unitary inequivalent representations of @xmath265 gamma matrices . it is convenient to denote the components of two - component spinor @xmath35 @xmath266 with the factor @xmath145 explicitly included in the definition of the lower component . then we can rewrite dirac equation ( [ dirac - eq - append ] ) in the components as follows : @xmath267 since we consider a cylindrically symmetric configuration of the field with a vector potential @xmath268 , the system ( [ components - cartesian ] ) has to be rewritten in the polar coordinates @xmath269 : @xmath270 as discussed in sec . [ sec : model ] , to analyze the problem with a singular at @xmath161 aharonov - bohm potential ( [ a - singular ] ) one has to do a self - adjoint extension of the dirac operator , see e.g. refs . . to avoid this complication we consider a regularized field configuration ( [ profile ] ) suggested in refs . with the profile function ( [ step - profile ] ) , so that @xmath271 from now on , we consider the specific case @xmath212 and seek for a solution of eq . ( [ components - polar ] ) in the following form @xmath272 for @xmath273 we obtain a system of the radial equations for a free dirac particle [ 2nd - order : r < r ] @xmath277\psi_1(r)=0 , \label{psi1:r < r } \\ & \frac{d^2}{dr^2}\psi_2(r)+\frac{1}{r}\frac{d}{dr}\psi_2(r)- \left[\frac{m^2}{r^2}-\frac{e^2-\delta^2}{(\hbar v_f)^2}\right]\psi_2(r)=0 , \label{psi2:r < r}\end{aligned}\ ] ] [ 2nd - order : r > r ] @xmath278\psi_1(r)=0 , \label{psi1:r > r } \\ \frac{d^2}{dr^2}\psi_2(r)+\frac{1}{r}\frac{d}{dr}\psi_2(r)- \left[\frac{(m+\eta)^2}{r^2}-\frac{e^2-\delta^2}{(\hbar v_f)^2}\right]\psi_2(r)=0 \label{psi2:r > r}\end{aligned}\ ] ] for @xmath142 . the solutions of eqs . ( [ 2nd - order : r < r ] ) and ( [ 2nd - order : r > r ] ) are expressed in terms of the usual bessel functions . for example , for the solutions of the equations ( [ psi1:r < r ] ) and ( [ psi1:r > r ] ) for the component @xmath167 are given by where @xmath280 , @xmath281 , and @xmath282 are constants and the @xmath283 s are the bessel functions . the solution ( [ solution - psi1:r < r ] ) is standard due to normalizability and the absence of delta function at @xmath161 . the coefficients @xmath280 , @xmath281 , and @xmath282 to be found from the matching conditions ( [ continuity ] ) and ( [ discontinuity ] ) . to find the second component , @xmath168 one can substitute the result for @xmath167 in eq . ( [ system : r > r - for2 ] ) . or , equivalently , one can start from eqs . ( [ psi2:r < r ] ) and ( [ psi2:r > r ] ) for the component @xmath168 which also have a solution in the form ( [ solution - psi1 ] ) , find the corresponding constants from the matching conditions and then use eq . ( [ system : r > r - for1 ] ) to obtain @xmath167 . finally , the overall factor before the solution is determined by the normalization condition @xmath284

multilayers built from ferromagnetic layers with various spacer layers include a wide variety of magnetic film systems that have been intensively studied during last years . due to remarkable phenomena as giant magnetoresistance , exchange - spring behaviour and/or exchange bias , and surface enhanced magnetic anisotropy , such nanostructures have already found a number of applications and are considered as promising candidates for nonvolatile magnetic recording media.@xcite on the other hand , nanoscale superlattices and similar structures provide convenient model systems to study different aspects of surface magnetism and magnetic ordering in confining geometry . in particular , much attention has been given to multilayers composed of antiferromagnetically coupled ferromagnetic nanolayers . @xcite such layered synthetic antiferromagnets can be separated into two classes according to the symmetries ruling their magnetic properties : superlattices with relatively strong uniaxial anisotropies include low - symmetry multilayers with effective uniaxial magnetic anisotropy in the layer planes , e.g. epitaxial systems deposited on ( 110 ) , ( 211 ) faces of cubic substrates.@xcite also multilayer systems with perpendicular anisotropy belong to this class.@xcite magnetization processes in these nanostructures are strongly influenced by the uniaxial anisotropy which is responsible for specific phenomena such as `` surface spin - flop '' @xcite or field - induced metamagnetic jumps.@xcite on the other hand , the uniaxial anisotropy may be absent in layered systems with higher symmetry . these represent another large and intensively investigated class of synthetic antiferromagnetic nanostructures . superlattices with planar magnetization grown on ( 001 ) faces of cubic substrates , e.g. multilayers from materials combinations as co / cu @xcite , fe / si @xcite co / cr , @xcite or fe / cr , @xcite belong to this class . in the case of weak four - fold anisotropy , their magnetic properties are mostly determined by the interlayer exchange interactions which may include an important _ biquadratic _ contribution . @xcite evidence of strong biquadratic exchange interaction has been given in a number of experimental papers for fe / cr two - layers and multilayers @xcite , for fe / si(001 ) multilayers @xcite ) . strong effective four - fold anisotropies have been found in systems such as co / cr(001 ) or fe / cr(001 ) . @xcite the competition between magnetic anisotropy , applied fields and exchange energies may cause complicated magnetic effects and processes . in fact , a great number of novel magnetic configurations and remarkable reorientational effects in external fields have been found in such superlattices . @xcite in particular , recent experimental results using modern depth - resolving techniques reveal spatially inhomogeneous magnetic structures , e.g. in fe / cr(001 ) superlattices,@xcite and specific reorientation effects imposed by four - fold planar ( tetragonal ) anisotropy.@xcite the understanding and interpretation of the complex magnetization processes found in such systems @xcite requires a theoretical underpinning . theoretical activity in this field is largely based on analytical and numerical calculations mostly within phenomenological approaches . @xcite these studies have demonstrated the general validity of the phenomenological models to describe the magnetization processes in antiferromagnetic nanostructures . @xcite for the system under discussion the phenomenological theory has been developed to describe effects of 90 degree couplings and a concomitant complex evolution of domain structures in fe / cr / fe layers.@xcite four - fold anisotropy effects have been theoretically investigated in ref . for sandwich structures with @xmath2=2 ferromagnetic layers coupled through a spacer . multilayer systems provide also experimental models to study effects of the confining surfaces on antiferromagnetic structures.@xcite in this context , specific inhomogeneous magnetic states described for theoretical models in ref . , have recently been observed in fe / cr superlattices.@xcite the existing theoretical results , however , are restricted to special cases and are not sufficient for an exhaustive description of the magnetic states and field - induced reorientional effects observed in recent experiment . @xcite in this paper we provide a theoretical analysis of magnetic states and magnetization processes in planar antiferromagnetic superlattices with and without four - fold anisotropy in magnetic fields applied within the multilayer plane . in such magnetic superlattices the exchange interlayer coupling is an oscillatory function of the spacer thickness.@xcite depending on the spacer thickness an alternating sequence of ferromagnetic and antiferromagnetic interlayer couplings is realized , and , by adjusting the spacer thicknesses , very different strengths of antiferromagnetic coupling can be realized . on the other hand , the four - fold anisotropy includes interface - induced contributions which implies a strong dependency of the effective average anisotropy of each ferromagnetic layer on the layer thickness.@xcite thus , the effective magnetic interactions can vary in extremely broad ranges for such multilayers in dependence on the chosen materials combinations and thicknesses , see e.g. ref . , where for a co / cu(001 ) two - layer system with a wedged spacer layer the ratio between four - fold anisotropy energy and the exchange coupling is changed by orders of magnitude . in contrast to bulk planar antiferromagnets , where an essentially fixed hierarchy for the strengths of the magnetic interactions holds,@xcite in these artificial antiferromagnetic systems the ratios between different magnetic energies , respectively the phenomenological parameters in the magnetic free - energy , may assume practically arbitrary values . moreover , as the interlayer exchange is weak compared to direct exchange interactions , the fields to induce spin - reorientation phenomena are similarly weak and experimentally accessible . the rich phase diagrams for these systems precludes an analysis in all details . the phase space in terms of the phenomenological parameters includes a large variety of different magnetic states with a corresponding multitude of spontaneous and field - induced phase - transformations . in a first step to such an analysis , all laterally homogeneous states in such multilayers must be found . they are the building blocks for a domain theory.@xcite for the case of laterally homogeneous states of each ferromagnetic sublayer one has a system that behaves like an antiferromagnetically coupled chain of stoner - wohlfarth particles . this simplified one - dimensional model for the behaviour across the multilayer stack also yields the limiting case for the magnetization processes with maximum hystereses . again , a direct analysis of all magnetic states even for these one - dimensional models yields an intricate succession of phase diagrams and magnetization curves.@xcite in this paper , we avoid the cumbersome task of listing and classifying _ all _ solutions and transitions . instead , we provide a broad physically intuitive picture of the physical effects due to the different exchange or anisotropic forces , and those imposed by the confining geometry of the system . to this end , we study limiting cases of the model . this includes the case of strictly zero anisotropy with and without biquadratic exchange , and the case of infinite anisotropy with fixed four - fold orientation of the magnetizations in each layer . for the antiferromagnetic two - layer systems ( @xmath2=2 ) , i.e. the sandwich structures ( experimentally realized as ferromagnetic / spacer / ferromagnetic trilayers ) , we provide a detailed investigation of the magnetic phase - diagram for arbitrary orientation of fields in the layer planes . based on this , we can understand the _ basic _ magnetic configurations in the multilayers , and we can give a map of the _ topologically different types _ of magnetic phase diagrams . we use standard methods to analyse magnetic phases and transitions within the phenomenological approach and the theory of phase transitions.@xcite the one - dimensional chain models are considered as composite order - parameters with many components ( @xmath2 components in a multilayer stack composed of @xmath2 ferromagnetic / spacer bilayers ) and a characteristic structure of couplings between the components defined by interlayer exchange and the surfaces . from this point of view , the very rich phase - diagrams and correspondingly complex sequences of magnetic configurations can be understood . for the general cases of the model , the equations for equilibrium and phase stability can be solved only by numerical methods . with the methods and results expounded below , one can extend the analysis to specific experimental cases in all detail . the solutions for the one - dimensional chain models include various field - induced canted and inhomogeneous states with a net magnetization . based on the phase diagrams of these models , the evolution of laterally inhomogeneous ( domain ) states and magnetization processes can be discussed.@xcite in this connection , the coexistence regions of different phases in the vicinity of discontinuous ( first - order ) magnetic phase transitions are important . in external fields , thermodynamically stable domain - configuration from these competing phases can be established in extended multilayers this is crucial for an understanding of the hysteretic magnetization processes under coercivity mechanisms . the paper is organized as follows . after introducing the model and mathematical tools ( sec . ii ) we consider the effects of the bilinear and biquadratic exchange interactions in the next section ( sec . iii ) . in sec . iv we investigate in detail four - fold anisotropy effects in antiferromagnetically coupled two - layers and then discuss the generalization of these finding to the case of multilayers . in sec . iv we discuss domain states and magnetization processes by using qualitative arguments . in the concluding part we discuss possible extensions of the theory , and we suggest some useful experiments to enhance our understanding of magnetization processes in antiferromagnetic superlattices . let us consider a stack of @xmath2 ferromagnetic plates infinite in @xmath3- and @xmath4-directions and with finite thickness along @xmath5-axis . the magnetization of the layers is @xmath6 , and there are indirect interlayer - exchange couplings through spaces between them . the phenomenological energy of the system can be written in the following form @xmath7 -\mathbf{h}\cdot \sum_{i=1}^{n}\zeta_i \mathbf{m}_i - \frac{1}{2}\sum_{i=1}^{n } k_i\left ( \mathbf{m}_x^4 + \mathbf{m}_y^4 \right ) \label{energy1}\end{aligned}\ ] ] where @xmath8 ( @xmath9 ) are unity vectors along the @xmath10-th layer magnetization . @xmath11 designate deviations of the magnetization in the @xmath10-th layer from the averaged value @xmath12 . we assume that @xmath13=0 , i.e. the layer magnetizations are restricted to the layer plane . @xmath14 and @xmath15 are constants of bilinear and biquadratic exchange interactions , respectively ; @xmath16 are constants of the in - plane four - fold anisotropy . the functional ( [ energy1 ] ) generalizes similar models considered earlier in a number of studies on exchange @xcite and anisotropy @xcite effects in planar antiferromagnetic systems . within this approach the ferromagnetic layers are considered as homogeneously magnetized blocks with constant values of the magnetic interactions . this assumption deserves some comment . it is well - established that in magnetic nanostructures surface / interface exchange and relativistic interactions strongly modify electronic and magnetic properties within all volume of the magnetic constituents @xcite . this means that the values of the exchange or anisotropy parameters , and the magnetizations include large interface / surface - induced components which may strongly vary across the thickness of the individual layers.@xcite however , the hypothesis of magnetic homogeneity in the ferromagnetic nanolayers in the models of type ( [ energy1 ] ) has a solid physical basis and is justified by successful applications of these models to describe magnetization processes in layered ferro- and antiferromagnetic nanostructures.@xcite this relies on the fact that in ferromagnetic nanolayers the intrinsic ( direct ) exchange coupling are usually very strong and overcome surface / interface induced interactions . thus , they play the dominating role to determine the magnetic order _ within _ the layers , which reacts also very stiffly on all external and induced magnetic forces . furthermore , in these planar systems the stray field forces confine the magnetization of the layers into their plane . as a result , in most of these systems the ferromagnetic layers preserve essentially homogeneous in - plane magnetized states even under the influence of strongly inhomogeneous induced interactions.@xcite in this connection it is important to stress here that the materials parameters in the phenomenological model ( [ energy1 ] ) should be treated as averages over a multilayer _ period_. they comprise ( in integral form ) all intrinsic and induced energy contributions of the magnetic states in the ferromagnetic layers . in particular , for the systems under consideration they may also include a contribution from magnetism of the spacer layers . thus , in contrast to their bulk counterparts which are considered as constants of the magnetic material , in nanoscale multilayer systems the phenomenological parameters strongly depend on many physical and geometrical factors and may considerably vary from sample to sample . parity of @xmath2 also plays an important role . the model ( [ energy1 ] ) with _ even _ number of ferromagnetic layers represents systems with fully compensated magnetization . such superlattices can be treated as analogues to bulk collinear antiferromagnets . superlattices with odd numbers of layers or with unequal thicknesses of layers own a noncompensated magnetization which strongly determines their global magnetic properties . in their response to an applied field , these structures are similar to bulk ferrimagnets . they could be studied by similar methods as used below , but have to be considered as separate class of systems . concentrating on the properties of systems with fully compensated total magnetization , we consider only superlattices with even @xmath2 and equal magnetization in all layers ( @xmath17 ) . in this contribution dedicated to general properties of these antiferromagnetic superlattices , we avoid secondary effects which are related to their micro - structure and interface properties , such as strains , chemical intermixing etc . with that in view we will study the model ( [ energy1 ] ) mainly for the case of _ identical _ ferromagnetic layers assuming that the exchange and anisotropy constants are equal for all the layers , @xmath18 , @xmath19 , and @xmath20 . some of the analytical results can be generalized to the isotropic ( i. e. with @xmath21 ) system with mirror symmetry about the center of the layer stack , i.e. @xmath22 and @xmath23 . in our problem the magnetization of the _ i_th layer is confined in planar orientations and can be described by the angle @xmath24 between the vector @xmath25 and @xmath3-axis . thus , calculations of the magnetic states for the model ( [ energy1 ] ) can be reduced to the optimization of the function @xmath26 . following the theory of bulk antiferromagnetism it is convenient to introduce here the vectors of the _ total magnetization _ @xmath27 and the _ staggered _ ( or _ antiferromagnetism _ ) vector @xmath28 @xmath29 the energy ( [ energy1 ] ) is invariant under transformation @xmath30 , and , therefore , all magnetic states in this model are degenerate with respect to the sign of the staggered vectors . in the following only solutions with a definite sign of the staggered vector will be discussed . however , one should keep in mind that the full set of solutions includes also those with opposite sign of @xmath28 . the magnetic states with opposite sign of @xmath28 behave identically in an applied magnetic field . thus , for the magnetization processes there is no need to distinguish these solutions . for the particular case of a two - layer ( @xmath0 ) the energy ( [ energy1 ] ) can be transformed to the following form @xmath31 where @xmath32 , @xmath33 and @xmath34 is the angle between the @xmath35 and the @xmath3-axis . this case of a two - layer is of special interest . the energy ( [ energy2 ] ) functionally coincides with that of a bulk two - sublattice antiferromagnet . this offers the opportunity for useful physical relations and analogies with bulk antiferromagnetism . on the other hand , the two - layers represent the simplest model of antiferromagnetically coupled layered nanostructures . the equations minimizing the function @xmath26 have strongly nonlinear character and no analytical solution are available generally . the body of our results have been obtained by numerical methods . we solve the coupled equations for equilibria @xmath36 by an efficient conjugate gradient minimization.@xcite for systems with large anisotropy the configurations with the magnetizations oriented in the four - fold easy - axis directions are close to mostly metastable minima of the energy . these configurations have been systematically searched for global absolute stability and checked for stability . numerically , this is feasible for @xmath2 not too large with present day computers . stability limits and phase - transitions are determined from the evolution of the smallest eigenvalue @xmath37 of the stability matrix @xmath38 under changing parameters in ( [ energy1 ] ) , i.e. ratios @xmath39 , @xmath40 , or the applied magnetic field . basic spin - configurations in isotropic antiferromagnetically coupled two - layers ( @xmath2=2 ) . at zero field ( a ) the magnetization vectors @xmath25 are antiparallel in adjacent layers but the structure has no fixed orientation in the plane ( infinite degeneracy ) . an in - plane magnetic field lifts the degeneracy and stabilizes states with * l * perpendicular to the field and with * m * along the field - this is a spin - flop ( sf ) phase ( b ) . in a sufficiently strong magnetic field the sf phases transform into the saturated ( _ flip _ ) phase ( c),width=321 ] we start the investigations of the static configurations minimizing energy ( [ energy1 ] ) from the isotropic case ( i.e. with @xmath16 = 0 ) . depending on relations between bilinear and biquadratic constants different collinear and noncollinear configurations are stable.@xcite ) the equations @xmath41 , @xmath42 , @xmath43 determine the region in the parameter space where a collinear _ antiferromagnetic _ ( af ) phase is the zero - field ground - state consisting of blocks of adjacent layers with antiparallel orientation of the magnetization ( fig . we restrict our analysis to this practically important case . the af phase is infinitely degenerate with respect to orientations of the staggered vector * l * in planar directions . the magnetic states for the isotropic two - layer ( @xmath0 ) in the applied field can be obtained by minimization of the energy ( [ energy2 ] ) with @xmath44 . the solutions ( @xmath45 , @xmath46 ) describe the states with the staggered vector perpendicular to the applied field ( so called _ spin - flop _ ( sf ) phase ) . in an increasing magnetic field the magnetization vectors rotate into the field direction ( fig . 1 ( b ) ) , and finally a continuous transition into the ferromagnetic ( _ flip _ ) phase with @xmath47 . this ( _ spin - flip _ transition ) occurs at the exchange - field @xmath48 ( fig . 1 ( c ) ) , which is lower than the corresponding spin - flip field @xmath49 for a bulk system . in superlattices with @xmath1 the magnetic field applied in the plane also lifts the degeneracy and stabilizes the sf phase ( the state with @xmath50 and @xmath51 ) ( figs . however , the magnetic configurations in these sf states and their evolution in the applied field are markedly different from those in bulk antiferromagnets or in the two - layer systems . it turns out that magnetic structures of the multilayers with @xmath2 divisible by four ( @xmath52 called here _ even - even _ ) differ from _ even - odd _ systems with @xmath53 ( @xmath54 ) . in low fields ( @xmath55 ) the solutions for the sf phase are given by a set of linear equations @xmath56 where @xmath57 for even - even and @xmath58 even - odd systems . it is clear that for the isotropic model the direction of the magnetic field in the film plane plays no role . following the definition of the angles @xmath24 we assume here that the magnetic field is applied along the @xmath3-axis . the solutions ( [ sol1 ] ) describe small deviations of the magnetization vectors , from the directions perpendicular to the easy axis ( see the magnetization profiles in figs . 2 , 4 and the configuration in fig . 3 ( b ) ) . in all internal layers ( @xmath59 ) neighbouring pairs retain antiparallel orientations ( e.g. , ( @xmath60 ) and ( @xmath61 ) in fig . 3b ) . for even - even systems the magnetization vectors of the central layers ( @xmath62 , @xmath63 ) preserve perpendicular orientation even in finite ( but weak ) fields ( see layers 6 and 7 in fig . 3 ) . for even - odd systems the magnetization vectors of all layers have finite deviations from perpendicular orientation . towards top and bottom layer @xmath64 or @xmath2 in the stack , these deviations increase . due to antiparallel orientation the pairs have zero net magnetization . the total magnetization arises only due to the top and bottom layers @xmath64 and @xmath2 . note that in these configurations a number of layers rotates against the applied field ( in fig . 2 - 4 the layers with @xmath10 = 2 , 4 and their symmetric counterparts with @xmath10 = 11 , 9 ) . this occurs because , in weak fields , the exchange interactions favouring antiparallel magnetizations in adjacent layers play the dominant role . an increasing magnetic field counteracts and slows down this reverse rotation , when the non - linear evolution for the magnetization structure is reached , and finally the sense of rotation is changed at characteristic fields @xmath65 where @xmath66 . with increasing @xmath2 the number of layers increases which display this reverse rotation , the deviations from orientations perpendicular to the applied low field become larger near the surface layers , and the fields @xmath65 are reached at lower fields ( see inset in fig . 2 for @xmath67 ) . a further set of characteristic fields @xmath68 defines the points where the projection of @xmath69 on the field directions changes the sign ( the field @xmath70 and @xmath71 are indicated in fig . 2 . ) in increasing field these characteristic fields initially are reached for the central layers and at higher fields for those closer to the boundaries . for fields @xmath72 the magnetization directions of all layers have positive projections onto the field direction . in the model with equal exchange constant there is another special field @xmath73 ( independent of @xmath2 ) , where all inner layers have the same projection onto the field direction ( @xmath74 , @xmath75 ) . the parameters of this `` knot '' point are determined from a set of equations @xmath76 the functions @xmath77 , @xmath78 , @xmath79 are plotted in fig . [ knotpoint ] . in particular for superlattices with zero biquadratic exchange @xmath80 one has @xmath81 and @xmath82 . near saturation , the sf phase has only positive projections of the magnetization on the direction of the magnetic field which decreases towards the center similarly to spin configurations obtained numerically in ref . . evolution of spin - states in multilayers with a field @xmath83 applied in the plane of the layers and for zero anisotropy @xmath84 and biquadratic exchange @xmath80 , hence @xmath85 . main figure : rotation angles of the layers in the plane @xmath24 @xmath86 for superlattice with @xmath87 . at point labeled `` r '' the sense of rotation changes for the magnetization direction @xmath24 ( here , @xmath88 ) ; at point `` @xmath89 '' the magnetization of this layer is perpendicular to the applied field again . point `` @xmath90 '' is the `` knot - point '' ( eq . ( [ knot ] ) ) . inset : @xmath67 only @xmath24 for even @xmath88 and @xmath91 shown . , width=321 ] spin - states in multilayers with @xmath87 for zero anisotropy @xmath84 and field @xmath83 applied in the plane : ( a ) @xmath92 ( b ) @xmath93 ( c ) @xmath94 ( d ) @xmath95 ( e ) @xmath96 , and ( f ) @xmath97 , i.e. below the transition to the flip - phase ( cf . [ evolutionkzero ] ) . all states have mirror - symmetry about the center of the multilayer stack ( marked by the dashed line ) . , width=321 ] the spatial inhomogeneity of the sf phases in the multilayers and their remarkable evolution in the applied field are due to the particular _ finite - size _ effects in this type of magnetic nanostructures . in such layered nanostructures all internal layers interact with two adjacent layers but the top and bottom layer have only one neighbouring layer . thus , their exchange coupling is weakened compared to internal layers within the multilayer stack . this strong disbalance of the exchange forces at the boundaries affects the magnetic ordering within the entire multilayer and causes the reorientational processes in the sf phases . this mechanism is also responsible for the occurrence of inhomogeneous states in other noncollinear structures in the multilayers ( see iv.c and v.a ) . note , this important role of the cut exchange - couplings is specific to antiferromagnetic confined systems with noncompensated moments at surfaces . in contrast , the cutting of exchange couplings does not affect _ ferromagnetically _ coupled nanostructures where relative orientation of the magnetization in the layers essentially depend only on magnetic anisotropy , applied fields , and the demagnetizing stray - fields . evolution of spin - flop profiles for the multilayer with finite biquadratic exchange is plotted in fig . 4 . for the systems with collinear ground - states the biquadratic exchange does not induce reorientional transitions but it has rather strong influence on the distribution of the magnetization in the multilayers ( see insets in fig . 4 ) and the value of the characteristic fields ( fig . 5 ) . the effects of a biquadratic exchange @xmath98 , if present in multilayers with collinear antiferromagnetic ground - state , can be easily understood . for magnetic configurations close to the antiparallel orientation of neighbouring layers , it softens the linear exchange forces ; for nearly parallel orientation the system becomes stiffer instead . the evolution of the magnetic states in the external field is distorted accordingly by the presence of the biquadratic exchange @xmath98 ( fig . [ evolutionkappa ] ) . in low fields close to the antiferromagnetic ground - state , the system reacts more strongly on the external field . thus , the fields @xmath99 , where the rotation of the ithe layer reverses , and the field @xmath68 , where perpendicular orientation with respect to field is regained , respectively , are reached at relatively lower fields ( labels `` r '' and `` @xmath89 '' in fig . [ evolutionkappa ] ) . conversely , the ferromagnetic state is reached only close to the enhanced exchange fields @xmath100 reflecting the stiffening of the effective exchange couplings in nearly ferromagnetic configurations . thus , the magnetization curves acquire a nonlinear character with increasing @xmath98 ( inset ( a ) in fig . [ evolutionkappa ] ) . experimental observation of the fields @xmath99 and the special field @xmath101 ( fig . [ knotpoint ] ) could be used to determine the relative strength of @xmath98 . for large values of @xmath102 the reversal of the rotation of the magnetizations of individual layers takes place at very low fields ( see , inset ( b ) in fig . [ evolutionkappa ] ) . evolution of magnetic states in antiferromagnetic multilayer @xmath87 with zero anisotropy @xmath84 and biquadratic exchange @xmath103 , i.e. with enhanced @xmath49 ( compare with fig . [ evolutionkzero ] for @xmath104 ) . @xmath83 is applied in the layer plane . insets ( a ) : magnetization curves in the range of values of @xmath105 with antiferromagnetic ground - state in zero field . ( b ) : evolution of states ( only shown for layers with even numbers ) in small fields for @xmath106 . , width=321 ] dependence of `` knot '' point @xmath107 ( eq . ( [ knot ] ) on strength of biquadratic exchange @xmath105 . inset shows the two angles @xmath108 , @xmath109 which characterise this special configuration independent on @xmath2 . , width=321 ] spin - flip fields for antiferromagnetic multilayers ( zero anisotropy @xmath84 ) in dependence on number of layers @xmath2 . , width=321 ] near the spin - flip transition from the sf - phase to the ferromagnetic states , the deviations of @xmath25 from the field directions are small ( @xmath110 ) , and the energy of the system ( [ energy1 ] ) can be expressed by the quadratic form @xmath111 , where the matrix @xmath112 has a tridiagonal band structure with nonzero elements only in the main diagonal @xmath113 and with the side diagonal elements @xmath114 . the spin - flip field @xmath115 is determined as the largest root of the equation @xmath116 . in particular for the model with equal parameters these solutions are plotted in fig . the spin - flip field gradually increases from @xmath117 for the two - layer to the `` bulk '' value @xmath118 in the limit of infinite @xmath2 ( fig . this dependence reflects the increase of the average number of exchange bonds from the value 1 for the two - layer and approaching 2 as @xmath2 tends to infinity . effects of the four - fold magnetic anisotropy on the states in the antiferromagnetic superlattices are revealed in detail by applying fields in arbitrary directions . we approach the general case by a detailed investigation on the case of the two - layer . to set the stage , we discuss the highly symmetric phase - diagrams with fields in direction of easy and hard - axes in the plane . this completes earlier work by dieny et al . @xcite on the @xmath2=2 systems . then , we present phase diagrams for arbitrary field directions in the layer plane . as we have seen in the last section , the primary effect of biquadratic exchange is a distortion of phase diagrams in various regions . thus , here we consider only models with @xmath119 to avoid such quantitative effects which will not affect the general topological features of the phase diagrams . optimization of the function @xmath120 eq . ( [ energy2 ] ) yields solutions for the magnetic states and their stability limits . in zero field the tetragonal anisotropy lifts the rotational degeneracy of the af phase and stabilizes two different states with perpendicular orientations of @xmath28 ( fig . 7 ( a ) ) . magnetic configurations in planar antiferromagnetically coupled two - layers with four - fold anisotropy . easy directions are indicated by thin continuous lines , hard directions by dashed lines . in zero field two antiferromagnetic configurations ( i ) ( a ) and ( ii ) ( b ) with the staggered vectors along the easy directions correspond to the ground - state . a magnetic field along the hard directions distorts them into the canted phases ( c ) , ( d ) which transform into the spin - flop state ( e ) . corresponding configurations of net magnetization and staggered vectors are shown for the two canted phases ( c ) ( d ) , and for the spin - flop phase ( e ) . , width=287 ] magnetic phase diagram of the two - layer system in magnetic fields along _ hard _ directions . the tricritical point ( @xmath121 , @xmath122 ) separates the continuous ( dotted line ) and discontinuous ( thick line ) transitions between the canted and sf phases . a further dotted line gives the critical field of the second - order transition between the sf and flip phases @xmath123 ( [ tran2 ] ) . thin lines give the lower stability limit @xmath124 of the sf phases and the upper limit @xmath125 of the canted phases . the grey area is the region of the phase equilibrium between the two canted phases ( see fig . [ tetragonconfigs ] ( c),(c ) and ( d),(d ) ) . magnetization curves for low and high anisotropy , as indicated in the main panel , are plotted in ( a ) and ( b ) . , width=287 ] in a field applied along one of the hard axes , the af configurations of the ground - state ( fig . 7(a),(b ) ) are distorted into low symmetry configurations ( so called _ canted _ phases - fig . 7(c),(d ) ) . both these magnetic configurations preserve mirror symmetry with respect to the field direction and have the same energy . the total magnetizations * m * of these canted states deviate from the field directions to different sides at equal angles ( fig . 7(c),(d ) ) . an oblique magnetic field deviating from the hard direction to one or the other side violates the phase balance between these two canted states . it favours the canted phase with the larger projection of * m * onto the field direction . this is the typical situation of a first - order ( or discontinuous ) transition . in an increasing field the vectors * m * in both canted phases rotate into the field direction and reach it at a certain critical field @xmath126 . at this point both phases transform into the common configuration corresponding to the sf phase ( fig . 7(d ) ) , i.e. a second - order ( or continuous ) phase transition from low symmetry canted phases into the high - symmetry sf - phase occurs in this field . standard analysis yields the following expression for the parameters of the critical point @xmath127 where @xmath128 determines the equilibrium sf configuration at the critical field . the two canted phases are competing phases related by a first - order transition . at the field @xmath124 the discontinuity between these phases disappears , i.e. this is a _ critical end - point _ of the first - order transition . after the transition into the sf phase , the system further evolves by rotation of the magnetizations @xmath25 into the direction of the field . this process is terminated at another critical field @xmath129 ( fig . the transition between the canted and sf phases is continuous only below a certain anisotropy strength , @xmath130 , and becomes discontinuous at higher anisotropy . the parameters of the corresponding tricritical point @xmath121 , @xmath122 have been calculated numerically from the equation @xmath131 together with eqs . ( [ tran2 ] ) . numerically calculated transition fields @xmath132 and the upper stability fields of the canted phases @xmath125 are plotted in fig . 8 . in this picture the grey area describes the region of the first - order transition , respectively the region of phase - coexistence between the canted phases . a magnetic field applied in one of the easy directions violates the energy balance between the two af states . the state af(i ) with staggered vector parallel to the field does not change its configuration and exists as a metastable collinear state up to the critical field @xmath133 . the phase af(ii ) with the staggered vector perpendicular to the magnetic field transforms into a sf phase which corresponds to the global energy minimum of the system . in an increasing magnetic field this phase evolves similarly to the sf phase along the hard directions , and continuously transforms into the flip states at the critical fields @xmath134 for low anisotropy . the transition between the sf and flip phases changes the order in another tricritical point ( @xmath135 ) . at a critical field @xmath136 ( [ stab1 ] ) the sf phase becomes unstable with respect to flipping distortions , i.e. modes that redirect the magnetization vectors into the field direction but preserve the symmetry of the sf phase . the parameters of the tricritical point ( @xmath137 ) . and the stability limits @xmath138 have been calculated in @xcite . after some algebra the first - order transition field @xmath139 also can be derived in analytical form @xmath140\ , , \label{tran1}\end{aligned}\ ] ] at the transitional field we have @xmath141 . magnetic phase diagram of the two - layer system in magnetic fields along _ easy _ directions . the tricritical point ( @xmath142 , @xmath143 ) separates the continuous @xmath134 ( [ stab1 ] ) ( dashed line ) and discontinuous @xmath134 ( [ tran1])(thick line ) transitions between sf and flip phases . the tricritical point ( @xmath144 , @xmath145 ) separates the continuous @xmath134 ( [ stab1 ] ) ( dashed line ) and discontinuous @xmath134 ( [ tran1])(thick line ) transitions between sf and flip phases . thin lines indicate the lability lines of the individual phases ( [ stab1 ] ) . the grey area designates the stability region of the canted phases . for the anisotropy @xmath146 = 0.47596 ( point @xmath147 ) , the critical field of the sf phase switches from an instability against symmetric `` flipping '' at the line @xmath136 to an instability against canting at @xmath148 given by the two branches @xmath149 from eq . ( [ stabsf2 ] ) . ( inset in main figure shows the details of this process ; see text following eq . ( [ stabsf2 ] ) for further explanation . ) small panels ( a ) and ( b ) show magnetization curves for low and high anisotropies as indicated in the main figure . full thick lines are for the evolution of the equilibrium ground - states , in ( a ) from sf phase to a continuous spin - flip transition into saturation . dotted lines are the evolution starting from the metastable af - phase . in ( b ) the evolution of the saturated state in decreasing fields is given by a thin line . the inner small hysteresis loop shows the transitions from and into the metastable canted phase . , width=287 ] the ( k , h)-phase diagram for this cases is plotted in fig . 9 . with increasing strength of the anisotropy , the `` landscape '' described by the energy function @xmath120 acquires additional folds . in particular , metastable canted phases arise in a certain region of the magnetic fields along the easy directions . ( this region is indicated by grey colour in fig . [ kheasy ] ) . for @xmath150 with @xmath151 = 0.28 + 0.08@xmath152 = 0.47596 the sf phase undergoes an instability with respect to a transition into the canted phases . this instability occurs at lability fields which are given by the two branches of the following parametric equation @xmath153 where @xmath154 is given by the the configuration in the sf - state , @xmath155 . both branches of the lability field ( [ stabsf2 ] ) start at the point @xmath156 ( point @xmath147 in fig . [ kheasy ] ) ; @xmath157 meets the lability line @xmath136 of eq . ( [ stab1 ] ) at the `` beak '' ( @xmath158 ) ( point @xmath159 in fig . [ kheasy ] ) . the transition between the metastable sf phase and the metastable canted phases along @xmath160 is continuous between the points @xmath159 and a further tricritical point @xmath161 . for stronger anisotropy the transition between these metastable phases is discontinuous at a first - order transition field @xmath162 . this transition line along with the stability limits of the canted phases have been numerically determined as given in fig . [ kheasy ] . note that such processes between metastable phases may be realized only in systems with high coercivity where the transition into the thermodynamically stable flip phase is suppressed at the field @xmath139 ( [ tran1 ] ) . now we consider the magnetic states in applied fields deviating from the symmetric directions and construct ( @xmath163)-phase diagrams for different values of the parameter @xmath164 . for the ( @xmath163)-diagrams in figs . [ hxhyweak ] to 14 it is assumed that the easy directions coincide with the @xmath3- and @xmath4-axis . ( a ) schematic ( @xmath163)-phase diagram of the two - layer for @xmath165 . the astroid ( continuous line ) gives the stability limit of the canted metastable phases . thick lines in diagonal directions ( hard axis ) are first order transitions between canted phases . the line for the ( continuous ) transition into the saturated spin - flip state is given by the dotted line . note that the real size of the astroid in the ( @xmath163)-diagram is much smaller than shown here . ( b ) evolution of spin - flip field , respectively lower stability limits of the saturated ( ferromagnetic ) phase ( fm ) . phase fm is stable outside the closed curves for the various values of anisotropy @xmath166 . for @xmath167 the fm - phases are ( meta)stable even in reverse fields and the existence regions overlap for states saturated in different easy directions . , title="fig:",width=287 ] ( a ) schematic ( @xmath163)-phase diagram of the two - layer for @xmath165 . the astroid ( continuous line ) gives the stability limit of the canted metastable phases . thick lines in diagonal directions ( hard axis ) are first order transitions between canted phases . the line for the ( continuous ) transition into the saturated spin - flip state is given by the dotted line . note that the real size of the astroid in the ( @xmath163)-diagram is much smaller than shown here . ( b ) evolution of spin - flip field , respectively lower stability limits of the saturated ( ferromagnetic ) phase ( fm ) . phase fm is stable outside the closed curves for the various values of anisotropy @xmath166 . for @xmath167 the fm - phases are ( meta)stable even in reverse fields and the existence regions overlap for states saturated in different easy directions . , title="fig:",width=287 ] this corresponds @xmath168 in the energy eq . ( [ energy2 ] ) . diagrams for @xmath169 can be obtained by rotation of those for @xmath168 through an angle @xmath170 . in the limit of weak anisotropy , @xmath171 , independent minimization with respect to the angle @xmath172 yields @xmath173 and the energy ( [ energy2 ] ) is simplified to the following form @xmath174 \ , . \label{energy3}\end{aligned}\ ] ] the set of equations for the stability limits of solutions , @xmath175 and @xmath176 , yield two closed lines of critical fields in the ( @xmath177)-phase diagram ( fig . [ hxhyweak ] ) . one of them @xmath178 describes the second - order transition into the flip phase with @xmath179 . the other closed curve can be written in a parametric form @xmath180 where @xmath181 , @xmath182 . it describes an astroid with eight - cusps that confines the region of the canted metastable states ( fig . [ hxhyweak](a ) ) . first quadrant of the ( @xmath163)-phase diagram for antiferromagnetic two - layers with ( a ) @xmath183 ( b ) @xmath184 . easy axes of magnetic anisotropy are along the axes of the diagrams . inset in ( a ) magnifies the region of the `` swallow tail '' for the discontinuous transitions in magnetic fields oriented close to the hard axes . thick lines are lines of first - order transitions . the dotted lines @xmath185 @xmath186 are continuous spin - flips from spin - flop phase to the saturated ( spin - flip ) state . thinner lines are the limits of stability , defining the coexistence regions for the corresponding phases . , title="fig:",width=287 ] first quadrant of the ( @xmath163)-phase diagram for antiferromagnetic two - layers with ( a ) @xmath183 ( b ) @xmath184 . easy axes of magnetic anisotropy are along the axes of the diagrams . inset in ( a ) magnifies the region of the `` swallow tail '' for the discontinuous transitions in magnetic fields oriented close to the hard axes . thick lines are lines of first - order transitions . the dotted lines @xmath185 @xmath186 are continuous spin - flips from spin - flop phase to the saturated ( spin - flip ) state . thinner lines are the limits of stability , defining the coexistence regions for the corresponding phases . , title="fig:",width=287 ] the cusps along the easy directions coincide with the stability field of the af phase @xmath133 eq . ( [ stab1 ] ) , and those along the hard directions with @xmath187 eq . ( [ tran2 ] ) . thick black lines within the astroid and along the hard directions indicate the first - order transitions between the canted phases from fig . [ tetragonconfigs ] . this topology of the ( @xmath163)-phase diagram fig . [ hxhyweak](a ) is preserved up to the first tricritical point @xmath188 . for @xmath189 a first - order transition between sf phase and two canted phases arises along the hard directions ( point @xmath190 in fig . [ hxhyo3](a ) ) . for stronger anisotropies , @xmath191 , discontinuous phase transitions exist also for finite deviations of the field direction from the hard axes . the corresponding line of these first - order transitions have critical end - points , where the difference between the competing phases disappears ( analogously to the end - point of a coexistence - line in gas - liquid phase diagrams ) . for @xmath192 and for fields along the easy directions , the spin - flip into the saturated ( induced ) ferromagnetic state occurs discontinuously . thus , another line of first - order transitions develops for @xmath193 also in oblique fields . correspondingly , the transition lines for spin - flips in ( @xmath163)-phase diagrams consist of continuous and discontinuous sections joined by the tricritical points @xmath185 , @xmath186 . the calculated ( @xmath163)-phase diagrams for @xmath194 in fig . [ hxhyo3 ] ( a ) and for @xmath195 in fig . [ hxhyo3 ] ( b ) represent the main topological features of these phase diagrams . point ` @xmath190 ' indicates the field of phase equilibrium between the sf phase and two canted phases and corresponds to the critical field @xmath132 from fig . 8 ; points @xmath196 , @xmath197 are the end - points of the first - order transitions between one of the canted phases and the sf phase . points @xmath198 @xmath199 correspond to the transition field @xmath139 from fig . 9 ; points @xmath185 , @xmath186 are the tricritical points , where the discontinuous transition between the flip phase and the distorted sf phase ends . the lability lines in the vicinity of the transition field @xmath132 have the shape of a `` swallow tail '' . similar phase diagrams arise in uniaxial ferromagnets with strong fourth - order anisotropy.@xcite as the anisotropy constant increases in the region @xmath193 the lines of the first - order transitions extend and near the special value @xmath200=@xmath201 the points @xmath196 , @xmath197 reach the easy directions . for larger @xmath202 canted phases exist as metastable state also for fields in easy directions ( compare fig . however , for increasing fields close to this direction , the discontinuous transition from the spin - flop state into the spin - flip state occurs before the transition into such a canted phase can take place , i.e. the two first - order lines @xmath190 @xmath196 and @xmath185 @xmath198 cross each other close to the easy axis directions in fig . brhl et al . @xcite reported evidence of an intermediate canted state in a co / cu / co / cu(001 ) system with field in an easy direction , following the theoretical prediction of such states by dieny et al . @xcite however , the state was found at lower fields than expected . this may be due to a misorientation of the external field and/or a mosaic of the epitaxial layer system because then the canted state becomes stable already at lower fields , as can be seen from line @xmath190 @xmath196 in phase diagram fig . generally , it is not sufficient to investigate only magnetization behaviour in hard and easy axes directions for a thorough understanding of the magnetization phases in these multilayer systems . to understand the transformation of ( @xmath163)-diagrams with increasing tetragonal anisotropy the limiting case is useful where @xmath25 are strictly oriented along the easy axes . in this case with _ infinite _ anisotropy , i.e. @xmath203 , our model can be considered as a chain of an antiferromagnetic four - state clock - model or planar potts - model@xcite in a transverse external field . recently , such four - state - clock models were employed to analyse the spin configurations of bulk tetragonal metamagnets with large four - fold anisotropies , such as rare - earth nickel - borocarbides @xcite or rare - earth silver - antimonides.@xcite rich experimental ( @xmath163)-phase diagrams have been obtained and analysed in terms of four - state - clock models , e.g. for honi@xmath204b@xmath204c @xcite or dyagsb@xmath204.@xcite the values of the angles @xmath205 0 , @xmath206 , @xmath207 , or @xmath208 in these four - fold states for @xmath209 are symbolically given by @xmath210 , @xmath211 , @xmath212 , and @xmath213 , e.g. the af and sf phases are ( @xmath210@xmath212 ) and ( @xmath213@xmath211 ) for @xmath0 . in addition to the collinear af and flip states there are phases with perpendicular orientation of the magnetization , or `` 90 degree - folded '' phases ( fig . [ hxhyinfinite ] ) . the states created by all combinations of these four angles @xmath24 for the magnetization in the multilayer stack exist as metastable states in arbitrary magnetic fields because the energy wells corresponding to these solutions are separated by infinitely high potential barriers . the regions of the absolute stability of these phases are separated by first - order transitions lines shown in fig . [ hxhyinfinite ] . in the first quadrant the transition between the af states and the canted phase ( @xmath214 ) occurs at the line @xmath215 ; the transitions lines from the canted phase ( @xmath214 ) into the flip states ( @xmath216 ) and ( @xmath217 ) are @xmath218 and @xmath219 , correspondingly ( fig . 12 ) . in the points ( 1,0 ) and ( 0,1 ) four phases coexist . thus , under increasing strength of the anisotropy ( @xmath220 ) the ( @xmath163)-phase diagram evolves from that plotted in fig . [ hxhyweak](a ) for @xmath221 to that in fig . [ hxhyinfinite ] for infinitely large values of @xmath222 . ( @xmath163)-phase diagram of the @xmath0-layer system in the limit of an infinite positive four - fold anisotropy . , width=321 ] the equilibrium magnetic configurations in two - layer systems arise as a results of the competition between the interlayer exchange coupling and tetragonal anisotropy . for multilayers the disbalance of the exchange forces at the boundaries ( sec . ii ) additionally influences the magnetic states . we first describe the structure of the one - dimensional solutions for laterally homogeneous states in finite antiferromagnetic superlattices . these magnetic states are determined by the interplay between cut exchange at the surfaces and the four - fold anisotropy . then we discuss some consequences for multidomain states and magnetization processes , and we discuss the physical nature of other effects which may play a role for the magnetic behaviour of real experimental systems . first , we investigate the ( @xmath163)-diagrams of magnetic states of multilayers with @xmath1 in the limit of infinite four - fold anisotropy . ( @xmath163)-phase diagram of the multilayer in the limit of infinite positive four - fold anisotropy . the phase - region of the l@xmath223x@xmath223-phases coincides with that of the collinear ferrimagnetic f@xmath223 , not shown here . ( for details see text . ) , width=287 ] the zero - field ground - states are the af phases @xmath224 and @xmath225 we may restrict the field to be oriented in directions @xmath34 in the interval @xmath2260,@xmath207/4@xmath227 $ ] . the configurations for other values of @xmath34 follow from symmetry . then , we may distinguish the four ground - state ( zero - field ) domains ( af1)@xmath228= @xmath229 , ( af2)@xmath228= @xmath230 , ( sf1)@xmath228= @xmath231 , and ( sf2)@xmath228= @xmath232 , where @xmath233 signifies the number of repetitions of the pair in a domain . in external fields @xmath234 configurations with net magnetization will be stabilized . these configurations must have pairs of adjacent moments flipped by 90 degree l=@xmath235 or x=@xmath236 , and by 180 degree f=@xmath237 . the 90-degree - folded configuration `` x '' is less favourable than `` l '' for field orientations in the chosen range , except for fields in easy - axis direction @xmath238 . thus , the magnetized configurations with lowest magnetization and smallest expense of exchange energy are those with only one pair of type `` l '' . in a short - hand , we write l@xmath223 for these configurations of type ( af1)@xmath239l@xmath240(sf2)@xmath241 ( with @xmath242 . ) next , we may form configurations with higher magnetization and smallest expense of exchange energy using one `` l '' and one `` x''-pair : l@xmath223x@xmath223= ( af1)@xmath239l@xmath240(sf2)@xmath243x@xmath240(af1)@xmath244 , ( @xmath245 ) . these configurations have the same energy and , therefore , are degenerate with configurations containing only one `` f''-pair : f@xmath223=(af1)@xmath239f@xmath240(af2)@xmath241 . note that these states are highly degenerate because the `` l''- , `` x''-pair , or `` f''-pair may be placed at arbitrary positions in the stack of @xmath2 layers . in higher fields configurations with various combinations of `` l''- , `` x''- , and `` f''-pairs may be stabilized by an external field . however , the structures with lowest energy , i.e. the absolutely stable states , are rather simple because of the following considerations . if , in external fields , more than one `` l''-pair can be formed starting from the l@xmath223 structure , the formation of the maximum number of l - pairs gives the most favourable configuration . in particular , the structure l@xmath246=(@xmath214)@xmath246 is the lowest energy structure for fields pointing in hard axis - directions @xmath247 in the limit of infinite h. the state with higher magnetization than l@xmath246 for fields pointing closer to the easy axis direction , @xmath248 , and with the smallest expense of exchange energy is in our notation l@xmath249f . the saturated state fm@xmath250f@xmath246 is the most favourable state which is reached whenever energy can be gained in external fields by flipping more moments into position @xmath210 in the states l@xmath249f , f@xmath223 , or l@xmath223x@xmath223 . thus , for general @xmath2 we have only the following phases for infinite positive four - fold anisotropy : af or sf as zero - field ground - states ; for fields with orientation close to the hard direction @xmath247 l@xmath223 and l@xmath246 ; otherwise in intermediate fields two degenerate phases f@xmath223 and l@xmath223x@xmath223 ; at high fields an asymmetric l@xmath249f - phase and the fully saturated ferromagnetic phase fm . we did numerical checks to ascertain that no further energetically stable phases do exist in external fields of arbitrary strength and direction , indeed . thus , we searched for the states of lowest energy by sampling all possible configurations for models with @xmath2=4 , @xmath251 , 12 corroborating our arguments . based on this set of magnetic configurations the resulting ( @xmath163)-phase diagram for general @xmath252 can be calculated analytically ( fig . [ hxhyinfinite2 ] ) . as in the simpler case of fig . [ hxhyinfinite ] for the two - layer system all these states are separated by infinitely high potential barriers and remain metastable for arbitrary fields . the first - order transitions between different phases occur along straight lines as shown in fig . [ hxhyinfinite2 ] . for finite strength of the four - fold anisotropy and under the influence of the exchange interactions , the basic structures are derived from the phases in fig . [ hxhyinfinite2 ] . under the influence of the field , they are elastically distorted into spatially _ inhomogeneous _ configurations . we have numerically investigated models for such cases with @xmath253 and various values of anisotropy @xmath222 . [ mvshn8 ] displays the general features for the example @xmath254 . the numerically calculated magnetization curves corresponding to the lowest energy states for sufficiently high anisotropies show the sequence of phases present in the infinite - anisotropy phase diagram fig . [ hxhyinfinite2 ] . however , the degeneracy of these phases is lifted because the distortions possible at finite anisotropy yield different gains of energy for the different configurations of the phases l@xmath223 , f@xmath223 , and l@xmath223x@xmath223 . generally , depending on field - orientation , large domains with nearly spin - flop - like configurations @xmath50 are favoured because these configurations can be more easily distorted by the fields yielding a corresponding gain of energy . in particular , phases derived from the ferrimagnetic collinear f@xmath223-phase are disfavoured compared to the l@xmath223x@xmath223-type phases . the nearly collinear ferrimagnetic configurations seem to exist only as metastable states . for fields closer to the hard - axis direction the phases with either one l - pair or @xmath255 l - pairs occur . under the influence of an applied field further discontinuous transitions occur at intermediate anisotropies , as seen e.g. in fig . [ mvshn8 ] ( a ) for @xmath256 . these transitions are jumps from one energy basin to another which are formed by distorting degenerate configurations mainly of the type l@xmath223 and l@xmath223x@xmath223 . examples of magnetization in an antiferromagnetic multilayer with @xmath254 . curves corresponding to the evolution of lowest energy states and for various values of anisotropy are shown with field oriented in directions of an easy axis ( a ) , in an oblique field ( b ) and in hard - axis direction ( c ) . @xmath34 is angle between magnetic field and easy axis direction . in each case , various magnetic configurations are shown for the discontinuous evolution of the magnetization curves with highest anisotropy . , title="fig:",width=287 ] examples of magnetization in an antiferromagnetic multilayer with @xmath254 . curves corresponding to the evolution of lowest energy states and for various values of anisotropy are shown with field oriented in directions of an easy axis ( a ) , in an oblique field ( b ) and in hard - axis direction ( c ) . @xmath34 is angle between magnetic field and easy axis direction . in each case , various magnetic configurations are shown for the discontinuous evolution of the magnetization curves with highest anisotropy . , title="fig:",width=287 ] examples of magnetization in an antiferromagnetic multilayer with @xmath254 . curves corresponding to the evolution of lowest energy states and for various values of anisotropy are shown with field oriented in directions of an easy axis ( a ) , in an oblique field ( b ) and in hard - axis direction ( c ) . @xmath34 is angle between magnetic field and easy axis direction . in each case , various magnetic configurations are shown for the discontinuous evolution of the magnetization curves with highest anisotropy . , title="fig:",width=287 ] at low anisotropies the phase diagram attains the behaviour discussed for zero anisotropies in section iii . here the magnetic states are only influenced by the cuts of exchange bonds at the surfaces and spatially inhomogeneous spin - flop states slightly distorted by four - fold anisotropy are realized ( see figs . ( @xmath163)-diagrams for this isotropic case consist of the region of the inhomogeneous sf phase ( fig . 2 , 3 ) separated from the saturated flip state by the critical line @xmath115 ( fig . these global features of the phase - diagrams are similar to the case of antiferromagnetic multilayers with uniaxial anisotropy.@xcite up to now we have discussed the effects of the competition between the bilinear exchange and four - fold anisotropy . it is clear that finite _ biquadratic coupling _ may substantially change the critical fields and stability regions of the various phases as it affects the elastic stiffness of the system . we note further , that the simple structure of the phase - diagram ruled by the phases present in the infinite anisotropy limit of fig . [ hxhyinfinite2 ] is valid only for the case of equal exchange constants @xmath257 in the multilayer stack . for arbitrary sets of values for @xmath14 in energy ( [ energy1 ] ) the phase diagrams may become considerably more complex and may contain further phases with different combinations of flipped 90 and 180 degree spin - pairs . even in such cases , the outline of the magnetic phase - diagrams , described here for finite equal - constant superlattices , should generally hold : ( i ) the high anisotropy limit is comparably simple with few phases determined by the competing lowest - energy basins of the anisotropy . non - equal constants may lift some degeneracies that are present in superlattices . ( ii ) in intermediate anisotropy range many elastically distorted phases appear , which are derived from stable and metastable high - anisotropy phases . ( iii ) for vanishing anisotropy the phase - diagrams become simple again as only inhomogeneous spin - flop - like phases and the saturated ferromagnetic phase remain in external fields . so far , we have analyzed single domain magnetic configurations . magnetization processes in real systems , however , are usually accompanied by complex reconstruction of the multidomain patterns as those recently observed in fe / cr multilayers.@xcite there are two main physical mechanism for multidomain states in the systems under discussion . four - fold degeneracy of the ground - state leads to creation of _ antiferromagnetic multidomain structures _ with 90 degree domain walls . unlike the case of magnetic materials with nonzero total magnetization , where multidomain states are caused by demagnetization effects , in antiferromagnets such domains are _ metastable _ and arise during the formation of the ordering state , i.e. they have a kinetic origin.@xcite hence , multidomain patterns observed in antiferromagnetic coupled two - layers and multilayers have irregular morphologies and depend on thermal and magnetic - field histories.@xcite however , another type of multidomain structures arises in the vicinity of field - induced discontinuous transitions.@xcite such thermodynamically stable _ transitional domain structures _ are formed by domains from states corresponding to the coexisting phases at first - order transitions . these domains are analogous to the domains of a demagnetized ferromagnet . in principle , the equilibrium parameters of such multidomain structures and their boundaries can be calculated by standard methods.@xcite in @xcite such calculations have been carried out for bulk easy - plane tetragonal antiferromagnets . _ magnetoelastic interactions _ lead to modification of the inhomogeneous states and decrease the regions of the multidomain states up to their complete suppression.@xcite this and other coercivity mechanism partly block the development of the equilibrium states . as a result , in real systems the evolution of multidomain states @xcite is accompanied by rather strong hystereses.@xcite for experiments , there are two important consequences related to the starting states and history dependence of magnetization processes . the `` texture '' of a real antiferromagnetic state in zero - field depends on the detailed kinetics imposed by e.g. cooling rates or deposition conditions . on the other hand , field cycling by inner loops for an antiferromagnetic multilayer may yield various metastable configurations and multidomain structures owing to the very wide coexistence region of the domains whenever sizable magnetic anisotropies are present in the multilayer stack within a phenomenological approach we develop the theory of reorientation transition in antiferromagnetically coupled superlattices with in - plane magnetization . detailed investigations of the surface effects in the isotropic multilayers ( sec . iii ) and four - fold anisotropy effects in two - layer systems ( sec . iv ) reveal the most important features of the system : ( i ) complex evolution of the inhomogeneous states ( figs . 2 - 4 ) is imposed by the strong disbalance of the exchange coupling , i.e. by the _ cut of the exchange bonds_. ( ii ) remarkable field - induced reorientational processes occur due to enhanced four - fold anisotropy ( figs . 8 to [ mvshn8 ] ) . the model used here , corresponding solutions and phase - diagrams include as special cases earlier theoretical studies on surface @xcite and four - fold anisotropy @xcite effects . the results are in essential accordance with existing experimental observations on inhomogeneous distribution in the spin - flop phase near the saturation field @xcite , and some effects of four - fold anisotropy in co / cu(001 ) wedged two - layers @xcite and fe / cr superlattices.@xcite our approach and results enable a _ qualitative _ analysis of the magnetization processes in the multilayered systems ( sec . v ) . in spite of the rather complex phase - diagrams of these systems , the analysis can be extended towards a quantitative description of real systems belonging to the class of artificial layered antiferromagnets described by eq . ( [ energy1 ] ) . so far most experimental work is carried out only for special conditions , often data are collected only with fields along easy - axes . these results only cover small regions of the ( @xmath258,@xmath259)-phase diagrams ( figs . [ hxhyweak ] to [ hxhyinfinite2 ] ) and do not capture the rich varieties of magnetizations processes in such systems . it is remarkable that many interesting effects , such as 90-degree folded phases , transitions into asymmetric canted phases etc . , are present already in antiferromagnetically coupled two - layers.@xcite as we have seen , phase diagrams for multilayer systems with @xmath260 become very complex . hence , systems with few layers are probably a better starting point for detailed investigations on magnetization processes . such experiments could be used to assess magnetic parameters and quality of such systems . generally , magnetization processes and checks for their dependence on magnetic and thermal pre - history should be made by applying fields in oblique directions and/or under rotating fields . only then , the behaviour of the ( @xmath258,@xmath259)-phase diagrams can be usefully compared with detailed theoretical investigations . further theoretical work may address models with non - equal constants . also effectively ferrimagnetic systems with odd numbers @xmath2 of layers and with different layer thicknesses may be interesting . some experimental data for magnetization processes in such systems exist , e.g. for co / cr two - layers with different ferromagnetic layers thicknesses,@xcite and for odd - numbered multilayers.@xcite interesting reorientational effects should arise also in magnetic fields , which are applied perpendicular or under arbitrary angle to the layer plane . antiferromagnetically coupled superlattices may also undergo transitions into perpendicularly magnetized states for certain thicknesses of the individual ferromagnetic layers , as observed for co / cr(001).@xcite in such cases the stray field must be taken into account already for laterally homogeneous states . concluding we state that the magnetic effects and phenomena discussed in this paper can be used for detailed investigations on such aspects of nanomagnetism as interlayer - exchange interactions , reorientational , and multidomain processes . g. j. strijkers , s. m. zhou , f. y. yang , and c. l. chien , phys . b * 62 * , 13896 ( 2000 ) ; k. y. kim , s. h. jang , k. h. shin , h. j. kim , and t. kang , j. appl . phys . * 89 * , 7612 ( 2001 ) ; a. moser _ et al . _ , j. phys . d * 35 * , r157 ( 2002 ) . m. a. howson _ et al . _ , b * 49 * , 9560 ( 1994 ) . r. w. wang , d. l. mills , e. e. fullerton , j. e. mattson , and s. d. bader , phys . lett . * 72 * , 920 ( 1994 ) ; s. g. e. te velthuis , j. s. jiang , s. d. bader , and g.p . felcher , phys . lett . * 89 * , 127203 ( 2002 ) . p. steadman _ et al . _ , lett . * 89 * , 077201 ( 2002 ) ; s. g. e. te velthuis , j. s. jiang , g. p. felcher , appl . . lett . * 77 * , 2222 ( 2000 ) . o. hellwig , t. l. kirk , j. b. kortright , a. berger , and e. e. fullerton , nature materials * 2 * , 112 ( 2003 ) . k. ounadjela _ _ , phys . b * 45 * , 7768 ( 1992 ) ; s. hamada , k. himi , t. okuno , and k. takanashi , j. magn . magn . * 240 * , 539 ( 2002 ) . h. itoh , h. yanagihara , k. suzuki , and e. kita , j. magn . mater . * 257 * , 184 ( 2003 ) . a. n. bogdanov and u. k. rler , phys . b * 68 * , 012407 ( 2003 ) ; u. k. rler and a. n. bogdanov , cond - mat/0307494 ( unpublished ) . k. brhl , s. di nunzio , f. schreiber , th . zeidler , and h. zabel , j. appl . * 75 * , 6184 ( 1994 ) ; k. brhl , th . zeidler , f. schreiber , a. schreyer , and h. zabel , j. magn . mater . * 130 * , l1 ( 1994 ) . a. van der graaf , m. valkier , j. kohlhepp , and f. j. a. den broeder , j. magn . * 165 * , 157 ( 1997 ) . r. w. e. van de kruijs _ et al . _ , phys . b * 65 * , 064440 ( 2002 ) . t. zeidler , f. schreiber , and h. zabel , j. appl . phys . * 79 * , 4793 ( 1996 ) . j. j. picconatto , m. j. pechan , and e. e. fullerton , j. appl . phys . * 81 * , 5058 ( 1997 ) . j. z. hilt , j. j. picconatto , a. obrien , m. j. pechan , and e. e. fullerton , j. magn . magn . mater . * 198 * -*1999 * , 387 ( 1999 ) . v. v. ustinov , m. a. milyaev , l. n. romashev , t. p. krinitsina , e. a. kravtsov , j. magn . * 226 * -*230 * , 1811 ( 2001 ) . c. chesman _ _ , phys . rev . b * 58 * , 101 ( 1998 ) . v. lauter - pasyuk , h. j. lauter , b. p. toperverg , l. romashev , and v. ustinov , phys . 89 * , 167203 ( 2002 ) ; v. lauter - pasyuk _ et al . _ , j. magn . magn . mater . * 258 * -*259 * , 382 ( 2003 ) . k. temst , physica b * 276 - 278 * , 684 ( 2000 ) . d. l. nagy _ et al . _ , lett . * 88 * , 157202 ( 2002 ) ; l. bottan _ et al . _ , j. magn . mater . * 240 * , 514 ( 2002 ) . h. lauter _ et al . _ , j. magn . * 258 * -*259 * , 338 ( 2003 ) . s. o. demokritov , j. phys . d : appl . * 31 * , 925 ( 1998 ) . b. dieny , j. p. gavigan , and j. p. rebouillat , j. phys . : condens . matter * 2 * , 159 ( 1990 ) ; b. dieny and j. p. gavigan , j. phys . : condens . matter * 2 * , 187 ( 1990 ) . f. c. nrtemann , r. l. stamps , a. s. carrio , and r. e. camley , phys . b * 46 * , 10847 ( 1992 ) ; a. l. dantas and a. s. carrio , phys . rev . b * 59 * , 1223 ( 1999 ) . a. hubert , r. schfer , _ magnetic domains _ ( springer - verlag , berlin , 1998 ) . d. l. mills , j. magn . mater . * 198 * -*199 * , 334 ( 1999 ) . p. grnberg , r. schreiber , y. pang , m. b. brodsky , and h. sowers , phys . lett . * 57 * , 2442 ( 1986 ) ; s. s. p. parkin , n. more , and k. p. roche , phys . rev . lett . * 64 * , 2304 ( 1990 ) . m. d. stiles , j. magn . mater . * 200 * , 322(1999 ) . v. g. baryakhtar , a. n. bogdanov , and d. a. yablonskii , usp . * 156 * , 47 ( 1988 ) [ sov . usp . * 31 * , 810 ( 1988 ) ] . l. d. landau , e. m. lifshitz , _ statistical physics _ part i. course of theoretical physics ; v. 5 ( butterworth - heinemann , oxford 1980 ) . f. y. wu , rev . mod . phys . * 54 * , 235 ( 1982 ) . mller and v. n. narozhnyi , rep . phys . * 64 * , 943 ( 2001 ) ; p. c. canfield and s. l. budko , j. alloys compd . * 262 * , 169 ( 1997 ) ; g. andr _ et al . _ , physica b * 292 * , 176 ( 2000 ) . p. c. canfield _ et al . _ , b * 55 * , 970 ( 1997 ) ; v. a. kalatsky and v. l. pokrovsky , phys . b * 57 * , 5485 ( 1998 ) . k. d. myers , p. c. canfield , v. a. kalatsky , and v. l. pokrovsky , phys . b * 59 * , 1121 ( 1999 ) . a. hubert , _ theorie der domnenwnde in geordneten medien _ ( springer , berlin , 1974 ) .

flow rate measurements of liquid metals are required in various technological processes ranging from the cooling of nuclear reactors to the dosing and casting of molten metals.@xcite electromagnetic flowmeters are essential in the diagnostics and automatic control of such processes . a variety of electromagnetic flowmeters have been developed starting from the late 1940s and described by shercliff.@xcite the standard approach is to determine the flow rate by measuring the potential difference induced between a pair of electrodes by a flow of conducting liquid in the magnetic field.@xcite this approach is now well developed and works reliably for common liquids like water,@xcite but not so for liquid metals . major problem in molten metals , especially at elevated temperatures , is the electrode corrosion and other interfacial effects , which can cause a spurious potential difference between the electrodes . the electrode problem is avoided by contactless eddy - current flowmeters , which determine the flow rate by sensing the flow - induced perturbation in an applied magnetic field.@xcite the main problem with this type of flowmeters is the weak field perturbation which may be caused not only by the flow . we showed recently that the flow - induced phase shift of ac magnetic field is more reliable for flow rate measurements than the amplitude perturbation.@xcite another contactless techniques for flow rate measurements in liquid metals is the so - called magnetic flywheel invented by shercliff,@xcite who prescribes _ a `` plurality '' _ of permanent magnets distributed equidistantly along the circumference of a disk , which is mounted on an axle and placed close to a tube carrying the liquid metal flow . the eddy currents induced by the flow across the magnetic field interact with the magnets by entraining them , which makes the disk rotate with a rate proportional to that of the flow . this type of flowmeter , described also in the textbook by shercliff@xcite and extensively used by bucenieks,@xcite was recently successfully reembodied under the name of the lorentz force velocimetry ( lfv).@xcite recently , we suggested an alternative and much more compact design of such a flowmeter , which conversely to shercliff s flywheel uses just a single magnet mounted on the axle it can freely rotate around and magnetized perpendicularly to it.@xcite we also introduced a basic mathematical model and presented first experimental implementation of this type of flowmeter.@xcite when such a magnet is placed properly at a tube with the liquid metal flow , it starts to revolve similarly to shercliff s flywheel . but in contrast to the latter , which is driven by the electromagnetic force acting on separate magnets , the single magnet is set into rotation only by the torque . this driving torque is due to the eddy currents induced by the flow across the magnetic field . as the magnet starts to rotate , additional eddy currents are induced , which brake the rotation . an equilibrium rotation rate is attained when the braking torque balances the driving one , and this rate depends only on the flow velocity and the flowmeter arrangement , whereas it is independent of the electromagnetic torque itself . thus , the equilibrium rotation rate is affected neither by the magnet strength nor by the electrical conductivity of the liquid metal provided that the friction on the magnet is negligible . this a major advantage of the single - magnet rotary flowmeter over the lfv approach , which relies on direct force measurements.@xcite in this paper , we present an extended theory of the single - magnet rotary flowmeter and compare it with experimental results . two limiting cases of long and short magnets are analyzed using linear - dipole and single - dipole approximations . we obtain simple analytic solutions for the force and torque on slowly moving and rotating magnets due to eddy currents in the layers of infinite horizontal extent and arbitrary depth . this allows us to find the equilibrium rotation rate of the magnet at which the torques due to the translation and rotation balance each other . we also consider an active approach , where the force on the magnet is used to control its rotation rate so that the resulting force vanishes . this rotation rate , similarly to the equilibrium one , is proportional to the layer velocity and independent of its conductivity and the magnet strength . the torque on a magnetic dipole rotating about an axis normal to a thin sheet has been calculated by smythe using an original receding image method.@xcite reitz uses this method to calculate the lift and drag forces on the coils of various geometries moving with constant velocity above a conducting thin plate.@xcite the lift force , which at high speeds approaches the force between the coil and its image located directly below it , varies as the velocity squared in the low - speed limit considered in this paper . the drag force is found to vary inversely and directly with the velocity at low and high speeds , respectively . palmer finds analytical expressions for the eddy current forces on a circular current loop moving with a constant velocity parallel to a thin conducting sheet.@xcite the force on a rectangular coil moving above a conducting slab has been calculated numerically by reitz and davis using the fourier transform method.@xcite the same problem for the magnetic dipole of arbitrary orientation placed next to a thin slowly moving slab is addressed by kirpo _ @xcite the force and torque on a transversely oriented dipole above a slowly moving plane layer of arbitrary thickness has been found analytically in the context of the lfv.@xcite fast computation of forces on moving magnets are of interested also for the eddy current force testing techniques.@xcite this paper is organized as follows . the following section presents two simple mathematical models of the single - magnet rotary flowmeter , which are used to calculate analytically the force and torque on the magnets moving and rotating slowly above a layer of infinite lateral extent . the limits of long and short magnets , which are approximated by linear and point dipoles , are considered in secs . [ sub:2d ] and [ sub:3d ] , respectively . section [ sec : exp ] presents the flowmeter implementation details and test results . the paper is concluded by a summary in sec . [ sec : con ] . schematic view of the single - magnet rotary flowmeter with the @xmath0-axis directed out of the plane of figure . ] consider a horizontally unbounded planar layer of electrical conductivity @xmath1 occupying the lower half - space and moving as a solid body with a constant velocity @xmath2 parallel to a permanent magnet placed at a distance @xmath3 above its surface and rotating with a constant angular velocity @xmath4 around an axis parallel to the surface and perpendicular to @xmath5 velocities are assumed sufficiently low for the magnetic field of induced currents to be negligible compared to the field of the magnet . the origin of cartesian coordinates is set at the center of the magnet with the @xmath6 @xmath0 and @xmath7 axis directed along @xmath8 @xmath9 and downward normally to the surface , respectively , as shown in fig . [ cap : sketch ] . in the following , two limiting cases will be considered in which the magnet will be assumed either much longer or much shorter than @xmath10 we start with a long cylinder magnetized perpendicularly to its axis about which it can freely rotate.@xcite in this case , the magnetic field is approximated by that of a two - dimensional ( linear ) dipole with the vector potential @xmath11 which has only the @xmath0-component @xmath12 where @xmath13 is the vacuum permeability and @xmath14 is the linear dipole moment , which is perpendicular to @xmath15 and directed at the angle @xmath16 from the positive @xmath17 axis ; @xmath18 is the radius vector from the magnet axis , and @xmath19 and @xmath20 are the cylindrical radius and the polar angle in the cylindrical coordinates around the @xmath0-axis . the magnetic field of linear dipole is @xmath21 eddy currents are induced by two effects : the translation of the layer and the temporal variation of the magnetic field due to its rotation . the latter vanishes in the co - rotating frame of reference , where the magnetic field is stationary , while the layer appears to move with the resulting velocity @xmath22 which contains also an apparent rotational motion of the layer @xmath23 opposite to that of the magnet . the density of eddy currents is given by ohm s law for a moving medium @xmath24 where @xmath25 is the electric potential . in this case , no electric potential is induced because the e.m.f . , @xmath26 is both solenoidal and tangential to the surface . if the induced magnetic field is negligible as originally assumed , eddy currents can be represented as the superposition @xmath27 where @xmath28 and @xmath29 are the currents induced by the translation and rotation , respectively . it is important to note that in the approximation under consideration with a fixed magnetic field distribution , eddy currents are determined only by the instantaneous velocities . besides that eddy currents are coplanar to the surface and , thus , mutually independent over the depth of the layer consequently , a layer of finite thickness may be represented as a semi - infinite one with zero velocity at @xmath30 where @xmath31 is the distance of the lower boundary of finite - thickness layer from the magnet . this , in turn , is equivalent to the superposition of two semi - infinite layers with the second layer at @xmath32 moving oppositely to the first one at @xmath33 so that the resulting velocity vanishes at @xmath34 in the following , this approach allows us to construct the solution for a finite - thickness layer by taking the difference of two half - space solutions , which is subsequently denoted by @xmath35_{h_{1}}^{h_{2}},$ ] where @xmath36 stands either for the force or the torque due to the eddy currents in a half - space . moreover , by the same arguments , this approach can easily be extended to @xmath7-dependent velocity distributions @xmath37 , for which general solution can be constructed as a superposition of solutions for thin layers moving with various velocities given by @xmath38 where @xmath39_{h_{1}}^{h_{2}}/\left[z\right]_{h_{1}}^{h_{2}}$ ] for @xmath40 and @xmath41 the linear force density experienced by an infinitely long magnet due to the layer translation , which according to the momentum conservation law is opposite to that acting upon the layer itself , can be written as @xmath42 where the integral is taken over the @xmath43-cross - section of the layer . the @xmath0-component of force is absent due to the @xmath44 reflection symmetry . in the low - speed limit under consideration , when force varies linearly with the velocity , there is also no @xmath7-component of force . this is the case because according to the linearity @xmath45 when @xmath46 while the latter transformation is equivalent to the rotation of the coordinate system by @xmath47 around the @xmath7-axis , which leaves the @xmath7-components invariant . in polar coordinates with the surface defined by @xmath48 we obtain @xmath49 the linear torque density , which because of the aforementioned symmetries has only the @xmath0-component , can be found as @xmath50 where @xmath51 the linear force density due to the rotation , defined by @xmath52 is found as @xmath53 the linear torque density due to the rotation , as that due to the translation above , has only the @xmath0-component @xmath54 which is not defined for a semi - infinite layer because the last integral diverges . nevertheless , expression ( [ eq : m1-gen ] ) can still be evaluated for the layer of finite depth by substituting the infinite limit in the last integral by @xmath55 which results in @xmath56_{h_{2}}^{h_{1}}.\label{eq : m1-h2}\ ] ] the solution above becomes unbounded as @xmath57 and , thus , inapplicable to thick layers . this implies that for a half - space layer the induced magnetic field can not be neglected however slow the rotation of magnet . the induced magnetic field is related to eddy currents by ampere s law @xmath58 which combined with expressions ( [ eq : b2d ] ) and ( [ eq : ohm ] ) leads to @xmath59 as suggested by the external magnetic field ( [ eq : a2d ] ) , we search for the vector potential in the complex form @xmath60.\label{eq : a - cmp}\ ] ] then eq . ( [ eq : a ] ) for the complex amplitude @xmath61 takes the form @xmath62 where the prime stands for the derivative with respect @xmath63 the general solution of eq . ( [ eq : a - rot ] ) can be written as @xmath64 where @xmath65 is an unknown constant to be determined by matching the induced and externally imposed magnetic fields ; @xmath66 is the modified bessel function of the second kind,@xcite and @xmath67 is the skin depth due to the rotation . at distances @xmath68 smaller than the skin depth @xmath69 @xmath70 is expected to approach the imposed field @xmath71 which yields @xmath72 ) and @xmath73 substituting expressions ( [ eq : a - cmp ] ) and ( [ eq : a - hat ] ) into integral ( [ eq : m1-gen ] ) , the torque on the magnet can be represented as @xmath74 where @xmath75r\,\d r$ ] and the asterisk denotes the complex conjugate . using eq . ( [ eq : a - rot ] ) , after some algebra we obtain @xmath76=\left(\frac{\mu_{0}}{2\pi}\right)^{2}\re\left[k_{0}((1+i)r / d)k_{1}((1-i)r / d)(1+i)r / d\right].\ ] ] for the low rotation rates that satisfy @xmath77 , we obtain @xmath78=\left(\frac{\mu_{0}}{2\pi}\right)^{2}(\ln(2d / r)-\gamma),\ ] ] where @xmath79 is euler s constant . finally , taking into account that @xmath80 and @xmath81 we obtain@xmath82 where @xmath83 is the effective skin depth , which owing to the low velocities under consideration is supposed to be large relative to @xmath84 . now we can use the results above to find the magnet rotation rate depending on the velocity of layer . first , if the magnet rotates steadily without a significant friction , the torques due to the translation and rotation are at equilibrium : @xmath85_{h_{2}}^{h_{1}}=0,$ ] which yields @xmath86 note that this equilibrium rotation rate @xmath87 depends neither on the magnet strength nor on the layer conductivity unless @xmath88 in the latter case , the skin effect becomes important and , thus , @xmath31 has to be substituted by @xmath89 in the expression above . the velocity of layer can be determined also in another way by measuring the force on the magnet , as in the lfv , to control the magnet rotation rate so that the resulting force vanishes , @xmath90_{h_{2}}^{h_{1}}=0.$ ] this results in@xmath91 which again depends linearly on the layer velocity , but does not depend on its conductivity or the magnet strength . in the other limiting case of a short magnet , when the distance to the surface is large or at least comparable to the size of magnet , the latter can be considered as a dipole with the scalar magnetic potential @xmath92 where @xmath93 is the dipole moment and @xmath94 is the fundamental solution of laplace s equation , which satisfies @xmath95 with the dirac delta function on the r.h.s . in the following , we use simplified notation @xmath96 where @xmath97 is the unit vector . then the dipole magnetic field is given by @xmath98 where the last term is added to ensure the solenoidality of @xmath99 also at @xmath100 the associated dipole current distribution is @xmath101 which easily leads to the classical expressions for the force and torque used later on . in the following , we will be using also the spherical and cylindrical coordinates associated with the cartesian ones in the usual way . as in the previous section , we change to the co - rotating frame of reference and consider the eddy currents due to translation and rotation separately . for the translation with @xmath8 the charge conservation @xmath102 applied to eq . ( [ eq : ohm ] ) results in laplace s equation for @xmath103@xmath104 at the surface @xmath105 the normal component of electric current vanishes : @xmath106 in order to find the induced electric potential , firstly , it is important to notice that @xmath99 being a free - space magnetic field satisfies the laplace equation itself . consequently , the cartesian components of @xmath99 satisfy this equation , too . secondly , if @xmath107 in bc ( [ bc : phi0 ] ) satisfies laplace s equation , then @xmath108 satisfies not only bc ( [ bc : phi0 ] ) but also eq . ( [ eq : phi0 ] ) because the integration along a straight line , similarly to the differentiation , are interchangeable with the laplacian . by the same argument , we can interchange integration and differentiation in expression ( [ sol : phi0-gen ] ) , which yields @xmath109 where @xmath110 and @xmath111 is the zeroth degree associated legendre function of the second kind;@xcite @xmath112 is the spherical polar angle from the positive @xmath7-axis ( see fig . [ cap : sketch ] ) and @xmath68 is the corresponding cylindrical radius . the `` constant '' of integration @xmath113 which similarly to the first term @xmath114 is supposed to be axisymmetric and also to satisfy the laplace equation , is chosen to regularize @xmath115 at @xmath116 by removing the logarithmic singularity @xmath117 for @xmath118 this results in @xmath119 and@xmath120 where @xmath121 is the spherical radius . for a transversal dipole @xmath122 considered also by thess _ _ at al.__@xcite , solution ( [ sol : phi0-h ] ) reduces to @xmath123 where the last term represents the magnetic potential of a dipole aligned with the @xmath0-axis . for a general dipole orientation , expression ( [ eq : h ] ) substituted into solution ( [ sol : phi0-h ] ) after some algebra yields@xmath124,\label{sol : phi0}\ ] ] where @xmath20 is the azimuthal angle from the positive @xmath17-axis in the @xmath125-plane . the electric potential distribution ( [ sol : phi0 ] ) allows us to calculate the force acting upon the magnet , which , as noted above , is opposite to that acting upon the layer , i.e. , @xmath126 where the integral of the lorentz force density @xmath127 is taken over the layer volume @xmath128 for the longitudinal force component , we obtain@xmath129+\int_{h}^{\infty}(\partial_{xm}^{2}g)^{2}\,\d z\right\ } \,\d s\nonumber \\ & = & \frac{\mu_{0}^{2}\sigma v}{512\pi h^{3}}(3m_{x}^{2}+m_{y}^{2}+4m_{z}^{2}),\label{eq : fx}\end{aligned}\ ] ] where the first integral is taken over the surface @xmath130 at @xmath131 and can be swapped with the second one over the layer depth . for a transversal dipole @xmath132 , the expression above coincides with that of thess _ _ et al.__@xcite as well as with the result of reitz@xcite for a thin sheet in the limit of a slowly moving dipole . as seen , the force is the strongest on a transversal dipole and reduces on longitudinal and spanwise dipoles by factors of @xmath133 and @xmath134 , respectively . thus , the force ( [ eq : fx ] ) , in contrast to force ( [ eq : f0-gen ] ) for a long magnet , varies with the dipole orientation in the @xmath43-plane . on a horizontally inclined dipole @xmath135 , there is also a spanwise force component@xmath136 but there is no vertical force whatever the dipole orientation:@xmath137 this is consistent with the result of reitz@xcite stating that in the low - speed limit the lift force is proportional to the velocity squared , while the approximation under consideration takes into account only the part of force proportional to the velocity . alternatively , the force and torque acting on dipole can be found using the associated current distribution ( [ eq : j ] ) and the induced magnetic field @xmath138 which lead straightforwardly to the classical expressions@xcite @xmath139 where the integrals are taken over the space @xmath140 above the layer . this is a bit longer but algebraically more straightforward approach , which will be pursued in the following . in order to find the induced magnetic field @xmath138 it is important to notice that solution ( [ sol : phi0-gen ] ) satisfies condition ( [ bc : phi0 ] ) not only at @xmath131 but at any @xmath141 thus , the @xmath7-component of the current is absent not only at the surface but throughout the whole layer . then @xmath102 and @xmath142 imply @xmath143 where @xmath144 is the electric stream function , whose isolines coincide with the eddy current lines . substituting this expression into eq . ( [ eq : ohm ] ) and taking the @xmath7-component of the _ curl _ of the resulting equation , we obtain@xmath145 note that in contrast to the electric potential in eq . ( [ eq : phi0 ] ) , no boundary conditions are required for @xmath144 at the surface . this is because , firstly , eq . ( [ eq : psi - gen ] ) contains no derivatives in @xmath7 and , secondly , the absence of the @xmath7-component of current at the surface is explicitly ensured by expression ( [ eq : j - psi ] ) . for translational motion with @xmath8 the r.h.s of eq . ( [ eq : psi - gen ] ) takes the form @xmath146 which by the same arguments as above satisfies laplace s equation and equals to @xmath147 with @xmath148 which satisfies the same equation and , thus , represents the solution of eq . ( [ eq : psi - gen ] ) . for a transversal dipole @xmath149 ) , the general solution ( [ eq : psi0-h ] ) simplifies to @xmath150 where the last term represents the magnetic potential of a longitudinal dipole . thus , in this case , the eddy current lines coincide with the isolines of electric potential ( [ sol : phi0 ] ) rotated by @xmath151 about the @xmath7-axis . for a longitudinal dipole @xmath152 the comparison of solutions ( [ eq : psi0-h ] ) and ( [ sol : phi0-h ] ) shows that the eddy current lines and the electric potential isolines are swapped with the corresponding distributions induced by a spanwise dipole @xmath153 and rotated by @xmath151 about the @xmath7-axis . for an arbitrarily oriented dipole , expression ( [ eq : h ] ) substituted into the general solution ( [ eq : psi0-h ] ) after some algebra yields@xmath154.\label{sol : psi0}\ ] ] figure [ fig : mov - xyz ] shows the isolines of the electric potential ( [ sol : phi0 ] ) and those of @xmath155 which represent the eddy current lines , in the @xmath125-plane @xmath156 for three basic dipole orientations along the @xmath17- , @xmath0- and @xmath7-axis . note that the patterns are self - similar in the @xmath125-plane with the characteristic length scaling directly with @xmath141 namely , these distributions are functions of spatial angles only , while according to ( [ sol : phi0 ] ) and ( [ sol : psi0 ] ) the magnitude of both @xmath103 and @xmath157 falls off as @xmath158 isolines of the electric potential @xmath103 ( dashed ) and of the stream function @xmath157 ( solid ) in the @xmath125-plane , where the latter represent the eddy current lines , induced by translation in the magnetic field of the dipole aligned with the @xmath17- ( a ) , @xmath0- ( b ) , and @xmath7-axis ( c ) . , title="fig:"](-30,0)(a)isolines of the electric potential @xmath103 ( dashed ) and of the stream function @xmath157 ( solid ) in the @xmath125-plane , where the latter represent the eddy current lines , induced by translation in the magnetic field of the dipole aligned with the @xmath17- ( a ) , @xmath0- ( b ) , and @xmath7-axis ( c ) . , title="fig:"](-30,0)(b ) isolines of the electric potential @xmath103 ( dashed ) and of the stream function @xmath157 ( solid ) in the @xmath125-plane , where the latter represent the eddy current lines , induced by translation in the magnetic field of the dipole aligned with the @xmath17- ( a ) , @xmath0- ( b ) , and @xmath7-axis ( c ) . , title="fig:"](-30,0)(c ) in order to satisfy the solenoidality condition @xmath159 the induced magnetic field is sought as @xmath160 where @xmath161 is the vector potential , which is supposed to satisfy the coulomb gauge @xmath162 then ampere s law leads to @xmath163 which , in turn , results in @xmath164 substituting expression ( [ eq : j - psi ] ) into the last integral , after some algebra we obtain @xmath165 where @xmath166 is the same as used by thess _ _ et al.__@xcite the previous expression implies that @xmath167 similarly to its source @xmath168 has no @xmath7-component . then the induced magnetic field can be written as @xmath169 where @xmath170 represents the scalar magnetic potential , which completely defines @xmath171 outside the layer , where @xmath172 further , using the tensor notation with the einstein summation convention , expressions ( [ eq : f ] ) and ( [ eq : m ] ) can be represented in terms of @xmath173 as @xmath174 where @xmath175 is the anti - symmetric tensor and the subscript after the semicolon denotes the differential with respect to the corresponding coordinate with indices @xmath176 @xmath177 and @xmath178 standing for the @xmath6@xmath0 and @xmath179 directions , respectively . taking into account the symmetry of @xmath180 with respect to the interchange of observation and integration points , the derivatives of @xmath173 at the dipole location @xmath181 in expressions ( [ eq : fi ] ) and ( [ eq : mi ] ) are found as @xmath182 where @xmath183 integrals ( [ eq : x - ij ] ) and ( [ eq : x - ijk ] ) can be evaluated analytically using , for example , the computer algebra system mathematica,@xcite which also allows us to carry out all other analytical transformations . in such a way , we firstly verify that eq . ( [ eq : fi ] ) with expression ( [ eq : psi0-h ] ) substituted into integral ( [ eq : x - ijk ] ) indeed reproduces previous results ( [ eq : fx ] ) , ( [ eq : fy ] ) and ( [ eq : fz ] ) . secondly , eq . ( [ eq : mi ] ) with the same expression for @xmath157 substituted in integral ( [ eq : x - ij ] ) results in@xmath184 which for a transversal dipole again coincides with the results of thess _ _ et al.__@xcite for a solid - body rotation with the velocity @xmath185 which appears in the co - rotating frame of reference , the charge conservation @xmath102 applied to eq . ( [ eq : ohm ] ) results in@xmath186 while the vanishing of the normal current component at the surface @xmath131 requires@xmath187 in order to solve this problem , firstly , it is important to notice that since @xmath188 satisfies the laplace equation , @xmath189 is a particular solution to eq . ( [ eq : phi1 ] ) . comparing expressions ( [ eq : phi1-b ] ) and ( [ sol : phi0 ] ) shows that @xmath190 where @xmath103 is given by solution ( [ sol : phi0-h ] ) . searching for the solution as @xmath191 reduces eq . ( [ eq : phi1 ] ) to the laplace equation for @xmath192 while bc ( [ bc : phi1 ] ) takes the form @xmath193 by the usual arguments , we obtain @xmath194 where @xmath195 and @xmath196 is the first order associated legendre function of the second kind.@xcite using the expressions above , after some transformations we obtain@xmath197.\label{sol : phi1}\end{aligned}\ ] ] which is the electric potential in the co - rotating frame of reference . instantaneous isolines of the electric potential @xmath198 ( dashed ) and of the stream function @xmath199 ( solid ) , which represent the eddy current lines , induced by a solid - body rotation in the magnetic field of the dipole aligned with the @xmath17- ( a ) , @xmath0- ( b ) , and @xmath7-axis ( c).,title="fig:"](-30,0)(a)instantaneous isolines of the electric potential @xmath198 ( dashed ) and of the stream function @xmath199 ( solid ) , which represent the eddy current lines , induced by a solid - body rotation in the magnetic field of the dipole aligned with the @xmath17- ( a ) , @xmath0- ( b ) , and @xmath7-axis ( c).,title="fig:"](-30,0)(b ) instantaneous isolines of the electric potential @xmath198 ( dashed ) and of the stream function @xmath199 ( solid ) , which represent the eddy current lines , induced by a solid - body rotation in the magnetic field of the dipole aligned with the @xmath17- ( a ) , @xmath0- ( b ) , and @xmath7-axis ( c).,title="fig:"](-30,0)(c ) as before , condition ( [ bc : phi1 ] ) is satisfied not only at @xmath131 but any @xmath7 and , thus , the @xmath7-component of the current vanishes throughout the layer . consequently , the electric current can again be expressed as ( [ eq : j - psi ] ) with @xmath200 for which eq . ( [ eq : psi - gen ] ) takes the form@xmath201 since the r.h.s . of eq . ( [ eq : psi1 ] ) satisfies the laplace equation , after some algebra we obtain@xmath202 the eddy current lines in the @xmath125-plane , which are the isolines of @xmath200 along with the isolines of the electric potential ( [ sol : phi1 ] ) are shown in fig . [ fig : rot - xyz ] . again , the patterns are self - similar and scale directly with the distance @xmath7 from the dipole , while the magnitude of both @xmath199 and @xmath198 falls off as @xmath203 in contrast to the translation considered above , the eddy currents induced by rotation are time - dependent and vary periodically as the dipole orientation changes from the @xmath17- to the @xmath7-axis , which are shown in figs . [ fig : rot - xyz](a ) and ( c ) , respectively . only the electric potential , but no current , is induced by a solid - body rotation when the dipole is aligned with the axis of rotation ( @xmath0 ) . in his case , when the magnetic field is symmetric about the axis of rotation , the induced e.m.f . is irrotational and , thus , compensated by the electric potential gradient , which equals to @xmath204 where @xmath205 is the cylindrical radius from the axis of rotation , @xmath206 is the azimuthal unity vector , and @xmath207 is the azimuthal component of the vector potential , which can be used to describe a general axially symmetric poloidal magnetic field . for the dipole field , we have@xmath208 where @xmath209 is the associated dipole current ( [ eq : j ] ) and @xmath210 substituting this into expression ( [ eq : v1xb ] ) and equating it to the gradient of the electric potential @xmath211 we obtain @xmath212 which is the free - space electric potential in the co - rotating frame of reference shown in fig . [ fig : rot - xyz](b ) and coinciding with ( [ sol : phi1 ] ) when @xmath213 this potential vanishes in the laboratory frame of reference , where the magnetic field is invariant with respect to the rotation around the symmetry axis . further , using expression ( [ sol : psi1 ] ) for @xmath199 in integrals ( [ eq : x - ij ] ) and ( [ eq : x - ijk ] ) , which can be evaluated together with eqs . ( [ eq : fi ] ) and ( [ eq : mi ] ) in the same way as for the translation in the previous section , we obtain @xmath214 the first point to note is that both the force and torque vanishes when dipole is aligned with the axis of rotation @xmath215 which , as discussed above , is due to the absence of eddy currents in this case . secondly , the longitudinal @xmath216 component of force ( [ eq : f1 ] ) , in contrast to that due to the translation given by expression ( [ eq : fx ] ) , is independent of the dipole orientation in the @xmath43-plane and depends only on the magnitude of the dipole moment in this plane : @xmath217 thirdly , there is also a non - zero transversal @xmath218 force , which in contrast to the longitudinal one is purely oscillatory and varies periodically with the dipole orientation in the @xmath43-plane as @xmath219 where @xmath16 is the poloidal angle of the dipole orientation in the @xmath43-plane from the positive @xmath17-axis , which is the same as that for the linear dipole in expression ( [ eq : a2d ] ) ( see fig . [ cap : sketch ] ) . moreover , when dipole is inclined to the axis of rotation ( @xmath220 also a spanwise @xmath221 force appears , which similarly to the transversal one alternates periodically around zero as @xmath222 with the dipole rotation . as seen from expression ( [ eq : m1 ] ) , inclination also gives rise to alternating torque components around both the @xmath17- and @xmath7-axis . the @xmath0-component of torque , which is negative and , thus , opposing the rotation , has not only a constant but also an alternating part , which varies with the double frequency of rotation as @xmath223 instantaneous rotation rate ( [ eq : omgt - trq ] ) versus time ( solid ) and orientation ( dashed ) for a negligible inertia of the magnet.,scaledwidth=50.0% ] thus , in contrast to the long magnet in @xmath224[sub:2d ] , there is no steady rotation rate , which could balance the @xmath0-component of the constant torque ( @xmath225 ) due to the translation . the variation of the magnet rotation rate in response to oscillatory torque is constrained by the inertia of magnet . for the oscillatory part of the rotation rate @xmath226 to be negligible compared to the mean one @xmath227 we need @xmath228 where @xmath229 is the moment of inertia of magnet . if this condition is satisfied , the oscillatory component may be neglected in the balance of torques ( [ eq : m0 ] ) and ( [ eq : m1 ] ) around the axis of rotation , which is the @xmath0-axis . for a layer of finite thickness , this results in @xmath230 the expression above differs only by a factor of @xmath231 from result ( [ eq : omg - frc ] ) for a long magnet in the case of a vanishing longitudinal force . in the opposite limit of a negligible inertia , the torque balance yields@xmath232 this instantaneous rotation rate @xmath233 which is shown in fig . [ fig : omg - t ] versus both the time @xmath234 and orientation @xmath235 alternates between @xmath236 and @xmath237 of the mean value ( [ eq : omgb - trq ] ) @xmath238 the last relation follows from the integration of eq . ( [ eq : omgt - trq ] ) as @xmath239 which also defines @xmath87 parametrically versus @xmath234 in fig . [ fig : omg - t ] . note that the oscillations of the rotation rate do not affect its mean value @xmath240 which is defined by ( [ eq : omgb - trq ] ) independently of condition ( [ eq : omeg - osc ] ) . alternatively , when the magnet rotation rate is actively controlled so that to balance the longitudinal force ( [ eq : fx ] ) with the @xmath17-component of force ( [ eq : f1 ] ) , the instantaneous rotation rate is @xmath241 where the temporal mean value @xmath242 follows from the solution of differential equation ( [ eq : omgt - frc ] ) , which can be written as @xmath243 single - magnet rotary flowmeter with a _ smco_-type magnet ( 1 ) of @xmath244 and @xmath245 in diameter and height , respectively , installed at the side of a rectangular stainless steel duct ( 2 ) of @xmath246 in cross - section carrying the liquid sodium flow . the rotation rate is measured with an inductive magnetic proximity sensor ( 3 ) and also monitored with a photoelectric barrier sensor ( 4 ) . , scaledwidth=50.0% ] the laboratory model of the single - magnet flowmeter shown in fig . [ exp - pic ] consists of a cylindrical _ smco_-type permanent magnet with diameter @xmath247 and height @xmath248 , magnetized perpendicularly to its axis with @xmath249 surface induction , which holds up to @xmath250 the magnet is mounted on a stainless steel axle held by a housing made of `` macor '' ceramic , which ensures a low mechanical friction and can withstand up to @xmath251 the flowmeter was tested on a sodium loop using about 90 liters of molten _ na _ at @xmath252 with the electrical conductivity of @xmath253 the flow was driven by a linear electromagnetic pump , which provided a maximum velocity of @xmath254 at the flow rate of @xmath255 in a stainless steel duct of @xmath246 cross - section . the frequency of magnet rotation was determined with an inductive magnetic proximity sensor ( sick mm12 - 60aps - zu0 ) , which was fixed at the distance of @xmath256 perpendicularly to the side surface of magnet at its midheight as shown in fig . [ exp - pic ] . the sensor briefly switched off and then on again as the magnetic field along the axis of the sensor changed its direction . the frequency of sensor transition from the off to the on state , which happens twice per revolution , was measured with keithley 2000 multimeter using the reciprocal frequency counting techniques with the measurement gate time set to @xmath257 the frequency was also monitored with tektronix tds 210 oscilloscope . after changing the pump current , the flow was allowed to develop for @xmath258 and then six measurements were taken with the intervals of approximately @xmath259 the standard deviation in the measured rotation frequencies was typically a few tenths of @xmath260 the rotation rates measured at two gap widths between the magnet and the duct are shown in fig . [ fig : exp](a ) depending on the average sodium velocity , which was determined using ultrasonic doppler velocimetry@xcite with the resolution of @xmath261@xcite although the rotation rate is seen to increase as the magnet is approached to the duct and to vary nearly linearly with the flow velocity in agreement with the theory , the linear fit shows a certain zero offset . obviously , there is some opposing force , which has to be overcome for the rotation to start . this requires the flow velocity of at least @xmath262 first , such an opposing force is caused by the dry friction in bearings . second , the magnet is oriented by the earth s magnetic field , which has a stronger effect than the friction and also has to be overcome for the rotation to start . third , even stronger effect is caused by the stainless steel walls of the duct , which are weakly ferromagnetic.@xcite the magnet , when approached to the duct , was observed to turn with the dipole moment perpendicular to the wall . it is also important to note that the magnet , when turned by hand away from the equilibrium orientation , returned to it monotonically without oscillations . this implies the inertia of magnet to be small relative to the electromagnetic drag force . thus , the inertia can not significantly contribute to the overcoming of the orienting torque due to the wall magnetization . however , when magnet rotates slowly , the orienting torque causes the rotation rate to fluctuate , which shows up as an increased scatter in the measured values seen in fig . [ fig : exp ] at low flow velocities . when the zero offset is removed and the remaining rotation rate is multiplied by the distance @xmath263 between the magnet axis and the liquid metal , where @xmath264 is the magnet radius and @xmath265 is the thickness of stainless steel wall , we obtain the relative rotation rate @xmath266 which represents a rescaled slope coefficient for fig . [ fig : exp](a ) and is plotted in fig . [ fig : exp](b ) versus the velocity . the short - magnet solution ( [ eq : omgb - trq ] ) in two limiting cases of a semi - infinite @xmath267 and a thin ( @xmath268 layer yields @xmath269 and @xmath270 respectively , which are smaller than the measured values @xmath271 seen in fig . [ fig : exp](b ) . on the other hand , the long - magnet solution ( [ eq : omg - trq ] ) , which appears more adequate for the experimental setup , yields @xmath272 for @xmath273 such quantitative differences between the experiment and theory are not surprising given the simplifications underlying the latter . first , theoretical model does not take into account the finite width of the duct . but this alone could hardly explain the high rotation rate of the magnet at which its field travels faster than the layer . this apparently being the case in the experiment implies the presence of significant velocity gradients , which are also ignored by the theory , but could be taken into account as outlined in sec . [ sub:2d ] . note that strong velocity gradients can be caused by the magnet itself , which due to its large size and strength may act as a magnetic obstacle partially blocking the flow and so increasing its local velocity . measured ( a ) and rescaled ( b ) magnet rotation rates along with the best linear fits versus the average velocity for @xmath274 and @xmath275 gaps between the magnet and the duct with the distance between the magnet axis and liquid metal @xmath276 where @xmath264 is the magnet radius and @xmath265 is the thickness of stainless steel wall.,title="fig:",scaledwidth=50.0%](-30,30)(a)measured ( a ) and rescaled ( b ) magnet rotation rates along with the best linear fits versus the average velocity for @xmath274 and @xmath275 gaps between the magnet and the duct with the distance between the magnet axis and liquid metal @xmath276 where @xmath264 is the magnet radius and @xmath265 is the thickness of stainless steel wall.,title="fig:",scaledwidth=50.0%](-30,30)(b ) we have presented a theory of single - magnet rotary flowmeter for two limiting cases of long and short magnets , which were modeled as linear and single dipoles , respectively . simple analytical solutions were obtained for the force and torque on slowly translating and rotating magnets due to eddy currents in layers of arbitrary depth and infinite horizontal extent . the velocity was assumed to be constant and the motion so slow that the induced magnetic field could be neglected . the latter assumption was not applicable to the long magnet rotating above a conducting half - space . in this case , to obtain a finite braking torque , the skin effect due to the induced magnetic field had to be taken into account however slow the rotation . for a single dipole of arbitrary orientation , compact analytical solutions were obtained in terms of both the electric potential and stream function induced by the layer translation and rotation in the co - rotating frame of reference . the electric stream function was used further to find the scalar potential of induced magnetic field at the dipole location , which resulted in simple expressions for the force and torque on the dipole . eventually , we found the equilibrium rotation rate at which the driving torque due to the layer translation is balanced by the braking torque due to the magnet rotation . an alternative approach was also considered , where the force on the magnet could be used to control its rotation rate so that the resulting force vanishes . in either case , the resulting rotation rate is directly proportional to the layer velocity and inversely proportional to the distance between the magnet and the liquid metal . these results were found in a qualitative agreement with the measurements on the liquid sodium flow . a more accurate quantitative agreement with experiment is limited due to the substantial approximations underlying the theoretical model , which neglects the finite lateral extension of the layer as well as the spatial and temporal variations of the velocity distribution . in conclusion , note that the resulting rotation rate is independent of the magnet strength and the electrical conductivity of the liquid metal provided that the mechanical friction or other external effects are negligible compared to the driving torque . this is the main advantage of rotary flowmeter over the lfv.@xcite x. fu , l. hu , k. m. lee , j. zou , x. d. ruan , and h. y. yang , dry calibration of electromagnetic flowmeters based on numerical models combining multiple physical phenomena ( multiphysics ) , _ j. appl . phys . _ * 108 * , 083908 - 11 ( 2010 ) . i. bucenieks , modelling of rotary inductive electromagnetic flowmeter for liquid metal flow control , in : _ proceedings of the 5th int . symposium on magnetic suspension technology , _ 204208 ( dresden , germany , 2005 ) . g. gerbeth , j. priede , i. bucenieks , a. bojarevics , j. gelfgat , and d. buchenau , verfahren und anordnung zur kontaklosen messung des durchflusses elektrisch leitfhiger medien , _ german patent application _ de 102007046881a1 ( 2009 ) . j. priede , d. buchenau , and g. gerbeth , force - free and contactless sensor for electromagnetic flowrate measurements , in : _ pamir conference _ , 815820 ( presqule de giens , france , 2008 ) ; _ magnetohydrodynamics _ * 45 * , 451458 ( 2009 ) . s. eckert , d. buchenau , g. gerbeth , f. stefani and f .- weiss , some recent developments in the field of measuring techniques and instrumentation for liquid metal flows , _ j. nucl . tech . _ * 48 * , 490498 ( 2011 ) .

rotational brownian motion of colloidal particles is a classical subject in statistical physics @xcite . contrary to its translational cousin , _ i.e. _ `` usual '' brownian motion , it does not result in changes of the particle location and is hence not as easily demonstrated in experiments . if , however , the suspended particles carry an electric or magnetic moment their orientation couples to external fields and the electric or magnetic relaxation properties of the suspension are direct consequences of the rotational diffusion of the constituting particles . in particular , in the well developed theory of ferrofluids the importance of rotational brownian motion for the theoretical description of the often complex and surprising hydrodynamic and magnetic properties was recognized already a long time ago @xcite . ferrofluids are suspensions of nano - size ferromagnetic grains in carrier liquids like water or oil . they combine the hydrodynamic properties of newtonian fluids with the magnetic behaviour of superparamagnets @xcite . due to the viscosity of the carrier liquid there is a coupling between the rotation of the magnetic grains and the local vorticity of the hydrodynamic flow . this coupling can , _ e.g. _ , be used to spin up ferromagnetic drops floating in a non - magnetic fluid of the same density with the help of a _ rotating magnetic field _ @xcite . complementary , a rotational ferrofluid flow exposed to a static magnetic field exhibits an enhanced shear viscosity @xcite . various related effects such as `` negative '' rotational viscosity @xcite , magneto - vortical resonance @xcite , and anomalously enhanced ac - response due to coherent particle rotation @xcite have been investigated . they rely on the exchange of angular momentum between rotating particles and an oscillating magnetic field . in these cases , the imposed non - zero _ flow vorticity _ of the ferrofluid is crucial for breaking the symmetry between clockwise and counter - clockwise particle rotation . in the present paper we investigate a much more indirect and subtle aspect of the interplay between rotational brownian motion of ferrofluid particles and their relaxation dynamics in an external magnetic field . we will show that a suitably designed time dependent external magnetic field _ without net rotating component _ may rectify the fluctuations of the particle orientation and set up a _ noise - induced _ rotation of the ferromagnetic grains . we will hence investigate how angular momentum can be transferred from an _ oscillating _ magnetic field to a ferrofluid _ at rest_. the effect was predicted theoretically and demonstrated experimentally in a previous short communication @xcite . in the present paper we provide much more details on the theoretical description and add several new results . the extraction of directed motion from random fluctuations is an old and controversial problem in statistical mechanics with a long and interesting history @xcite . although excluded by the second law of thermodynamics for equilibrium systems , rectification of fluctuations _ is _ possible in systems driven sufficiently far away from thermal equilibrium @xcite . the problem has gained renewed attention under the trademarks of `` thermal ratchets '' and `` brownian motors '' due to its possible relevance for biological transport @xcite and the prospects of nano - technology @xcite . ferrofluids are ideal systems to investigate such fluctuation driven transport phenomena and also to demonstrate them experimentally @xcite : as already discussed above , the rotational dynamics of the ferromagnetic grains is strongly influenced by thermal fluctuations . appropriate time - dependent potentials can be easily designed with the help of external magnetic fields . finally , directed rotational transport in ferrofluids should manifest itself as systematic rotation of the ferromagnetic nano - particles . this in turn can be easily detected from the resulting _ macroscopic _ torque on the carrier liquid . various somewhat related but still quite different phenomena in rotational dynamical systems are treated in @xcite . throughout the paper we will use two basic approximations which simplify the analysis considerably and which are rather common for ferrofluids @xcite . the first is to neglect neel - relaxation of the magnetization , _ i.e. _ the rotation of the magnetization vector with respect to the ferromagnetic particle . this is justified for particle sizes that are not too small and amounts to assuming that the magnetic moments are firmly attached to the geometry of the particles . any reorientation of the magnetic moment hence requires a rotation of the particle as a whole . the second approximation is to neglect dipole - dipole interactions between the particles . although these interactions may be important in concentrated ferrofluids , they are negligible in sufficiently diluted ones . the ratchet mechanism of central interest in the present investigation can operate without any particle - particle interaction . we will hence assume that our ferrofluids are sufficiently diluted and use a _ single particle _ model . for a more quantitative assessment of the role of the dipole - dipole interactions see _ e.g. _ @xcite and references therein . the paper is organized as follows . in section [ sec : model ] we introduce the general framework for describing the rotational brownian motion of a single suspended ferromagnetic particle in an external magnetic field . section [ sec : symm ] gives a detailed account of the symmetries characteristic of our system and discusses how these symmetries influence the possibility for a ratchet mechanism to operate . in section [ sec : fpe ] the central fokker - planck equation describing our system is analyzed . besides an investigation of the weak noise limit we describe the numerical solution of the fokker - planck equation and compare the results with two different approximate treatments yielding analytical expressions for the transferred angular momentum . finally , section [ sec : conc ] contains some conclusions . we consider the rotational motion of a spherical particle of volume @xmath0 and magnetic moment @xmath1 immersed in a liquid with dynamic viscosity @xmath2 and subject to a horizontal , time dependent , spatially homogeneous magnetic field @xmath3 . the field is composed of a constant part @xmath4 parallel to the @xmath5-axis and an oscillatory part @xmath6 with period @xmath7 along the @xmath8-direction @xmath9 different choices for the time dependence of @xmath10 are of interest . in the present paper we will mainly discuss two cases which are each representative for a whole class . our first standard choice is @xmath11 where the amplitudes @xmath12 , and the phase @xmath13 are control parameters . the main features of this time dependence are a zero average over one period and the presence of a higher harmonic of the basic frequency . as a second example , we will also discuss the form @xmath14 for which the average over one period is different from zero . it arises naturally if the constant field component is not perpendicular to the time dependent one . in any case , the magnetic field is of a pure oscillatory character , _ i.e. _ it _ does not contain a net rotating component_. the orientation of the particle at time @xmath15 is described by the unit vector @xmath16 where @xmath17 denotes the modulus of the magnetic moment . the time evolution of @xmath18 is given by @xmath19 where @xmath20 denotes the instantaneous angular velocity of the particle . changes of @xmath21 are due to torques on the particle . denoting by @xmath22 the angular part of the three dimensional nabla operator , the magnetic torque @xmath23 derives from the potential energy @xmath24 of a magnetic dipole in an external field @xcite . further , the viscosity @xmath2 of the carrier liquid gives rise to a viscous torque @xcite @xmath25 additionally , the interaction between the rotating particle and the surrounding liquid also causes thermal fluctuations which generate a stochastic torque @xcite @xmath26 here , @xmath27 is a vector of independent , @xmath13-correlated gaussian noise sources of zero mean , the noise intensity is related to the temperature @xmath28 and the dissipation ( [ defvisctor ] ) of the carrier liquid by the fluctuation - dissipation relation , and @xmath29 stands for boltzmann s constant . denoting the moment of inertia of the particle by @xmath30 , the equation of motion for @xmath21 acquires the form @xmath31 eqs.([evol1 ] ) and ( [ evol2 ] ) form a closed set of equations for the description of the rotational motion of the particle . using experimentally relevant parameter values ( density of the particle @xmath32 kg / m@xmath33 , particle radius @xmath34 nm , viscosity @xmath35 pas , time scale @xmath36 s ) we find that the first term on the l.h.s . ( [ evol2 ] ) is five to seven orders of magnitude smaller than the second one . we may hence safely neglect inertial effects @xcite and find for @xmath21 in the overdamped limit @xmath37 where we have introduced @xmath38 . using this result in ( [ evol1 ] ) yields a closed equation for the time evolution of @xmath18 @xmath39 it is convenient to introduce dimensionless units . to this end we measure time in units of the inverse of the external driving frequency , @xmath40 , use @xmath41 as unit for the magnetic field strength , @xmath42 , and rescale the noise intensity according to @xmath43 . the evolution equation for the orientation @xmath18 then reads @xmath44 introducing the brownian relaxation time @xmath45 we note that the rescaled , dimensionless noise intensity @xmath46 occurring in ( [ evol4 ] ) just gives the ratio between the relevant deterministic and stochastic time scales in the system : @xmath47 to proceed , we parametrize the orientation @xmath18 of the magnetic particle by two angles @xmath48 and @xmath49 according to @xmath50 from ( [ evol4 ] ) we then find the following langevin equations for the time evolution of these angles @xcite @xmath51 in the dimensionless units adopted , the noise intensity @xmath46 is given by ( [ defd ] ) and the potential ( [ defu ] ) takes the form @xmath52 the thermal fluctuations @xmath53 and @xmath54 are given by two independent , @xmath13-correlated gaussian noise sources of zero mean . note that in ( [ lan2 ] ) we are dealing with multiplicative white noise but since the multiplicative function is independent of @xmath55 , no ambiguity ( ito- vs. stratonovich - interpretation @xcite ) arises . it is instructive also to consider the simplified situation of a one - dimensional dynamics in which the vector @xmath18 is assumed to be constrained to the horizontal @xmath5-@xmath8-plane . accordingly , ( [ lan1 ] ) is replaced by @xmath56 and ( [ lan2 ] ) simplifies to @xmath57 where @xmath58 the observable of foremost interest in the present investigation is the time and ensemble averaged torque ( [ defmagtor ] ) exerted by the magnetic field upon the magnetic particle in the long time limit , _ i.e. _ after initial transient have died out . since the magnetic field ( [ deffield ] ) is constrained to the @xmath5-@xmath8 plane , only the @xmath59-component of this magnetic torque can be different from zero . suppressing the subscript @xmath60 in ( [ defmagtor ] ) from now on , we denote the averaged @xmath59-component of the magnetic torque by @xmath61 , where @xmath62 stands for the ensemble average over the different realizations of the noise terms in ( [ lan1 ] ) , ( [ lan2 ] ) and the overbar represents the time average over one period of the magnetic field . using ( [ deffield ] ) and ( [ defmagtor ] ) we get @xmath63 exploiting ( [ defpot ] ) and ( [ lan2 ] ) one readily finds the equivalent expressions @xmath64 for reasons of ergodicity , the ensemble average in ( [ torav ] ) is equivalent to a time average of a single realization over an infinite time interval . then the extra time average over one period of the external driving drops out and we are left with @xmath65 and similarly for the equivalent expressions in ( [ torav ] ) , ( [ torav ] ) . space - time plot of the potential @xmath66 from eq.([defpot ] ) for the time dependence ( [ foft1 ] ) with @xmath67 , and @xmath68 . dark and bright regions correspond to small and large values of @xmath69 , respectively . in the long - time limit , the deterministic dynamics ( [ lan1 ] ) , ( [ lan2 ] ) with @xmath70 approaches @xmath71 and a periodical @xmath55 , oscillating back and forth as represented by either of the full black lines . in the presence of a small amount of noise , occasional transitions across the unstable deterministic orbits shown as full white lines become possible which are schematically indicated by the dashed lines . the spatial asymmetry and temporal anharmonicity of the potential conspire to yield slightly different rates for noise induced increments and decrements of @xmath49 , respectively . as a result a noise driven rotation of the particles arises . for a detailed discussion see section [ inst1d ] . ] in the absence of thermal fluctuations , the particle orientation is governed by the overdamped deterministic relaxation dynamics given by ( [ lan1 ] ) , ( [ lan2 ] ) with @xmath70 . hence , at any given moment , the orientation @xmath72 tends to align with the instantaneous magnetic field @xmath73 which lies in the @xmath5-@xmath8-plane . one can then show @xcite that @xmath74 converges to @xmath75 for @xmath76 and that @xmath55 approaches a _ periodic _ long time behaviour . this implies via ( [ torav ] ) that @xmath77 _ i.e. _ , in the absence of thermal fluctuations no particle rotation will occur and no average torque can arise . an explicit example is displayed in fig . [ fig : fig1 ] . this scenario changes fundamentally in the presence of fluctuations , _ i.e. _ if @xmath78 . as shown qualitatively in fig . [ fig : fig1 ] , for small noise intensities , the time dependence of @xmath55 and @xmath74 will still closely follow the deterministic trajectories , except that now also rare , fluctuation induced _ transitions _ between different deterministic solutions may occur . if the rates of forward ( increasing @xmath49 by @xmath79 ) and backward transitions ( decreasing @xmath49 by @xmath79 ) are different from each other , then on the average a net rotation of the particle will occur , implying with ( [ eq24 ] ) that @xmath80 . this is a manifestation of the so - called ratchet effect @xcite in which an unbiased potential and undirected brownian fluctuations cooperate to produce directed transport . the detailed operation of the ratchet mechanism for the rotational motion of a colloidal particle is the main focus of the present paper . symmetry considerations admit conclusions from the equations of motion without actually solving them . for the investigation of ratchet systems , studies of symmetry turned out to be valuable since directed transport generically occurs if it is not forbidden by symmetries , a statement referred to as curie s principle @xcite . here we specifically address the question which transformations of the magnetic field components @xmath4 and @xmath6 in ( [ deffield ] ) leave the average @xmath59-component of the torque invariant , @xmath81 , and which lead to its inversion , @xmath82 . from eq . ( [ torav ] ) one readily concludes that the following transformations of @xmath49 and @xmath48 imply the indicated transformation behavior of @xmath61 : @xmath83 where @xmath84 and @xmath85 are arbitrary . observing that the values of both the original and the transformed @xmath74 are restricted to the interval @xmath86 $ ] and that the entire interval will indeed be sampled by some realization of the dynamics ( [ lan1 ] ) , ( [ lan2 ] ) , it follows that no transformations of the form @xmath87 with @xmath88 or @xmath89 with @xmath90 are possible . finally , in order to conclude @xmath82 in ( [ 1e ] ) one has to exploit that in ( [ torav ] ) the limit @xmath91 is not necessarily tantamount to @xmath92 with @xmath93 kept fixed , but can also be realized by letting @xmath94 with @xmath95 kept fixed . closer inspection of ( [ torav ] ) reveals that any further symmetry transformation that leaves @xmath61 invariant or changes its sign can be composed of the `` elementary '' transformations ( [ 1a])-([1e ] ) . next we ask ourselves what equations of motion govern the transformed @xmath96 , @xmath55 defined by ( [ 1a])-([1e ] ) . in the simplest case ( [ 1a ] ) one readily concludes from ( [ lan1 ] ) and ( [ lan2 ] ) that the transformed @xmath96 , @xmath55 satisfy the same equations of motion ( [ lan1 ] ) and ( [ lan2 ] ) but with a transformed potential @xmath97 with arbitrary @xmath98 . turning to the second transformation ( [ 1b ] ) one finds similarly that the transformed @xmath96 and @xmath55 satisfy the equations of motion ( [ lan1 ] ) and ( [ lan2 ] ) but now with a transformed potential @xmath99 and in addition with a transformed noise @xmath100 . since the statistical properties of the two noises @xmath101 and @xmath54 are equal , _ i.e. _ they are identical stochastic processes , the same follows for the corresponding processes given by ( [ lan1 ] ) and ( [ lan2 ] ) . in other words , to each realization of the original @xmath96 , @xmath55 corresponds a realization of the transformed @xmath96 , @xmath55 occurring with the same probability , and vice versa . since the average torque in ( [ torav ] ) is independent of the specific realization @xmath96 , @xmath55 with probability 1 ( _ i.e. _ up to a set of realizations of measure zero ) we can conclude that in this case @xmath82 . similar considerations can be done for the cases given by ( [ 1c ] ) and ( [ 1d ] ) . the last transformation ( [ 1e ] ) is insofar special as a mapping between the dynamics of the original and the transformed variables is possible only , if the term @xmath102 in ( [ lan1 ] ) identically vanishes , _ i.e. _ if @xmath103 . we therefore find the following implications of transformations in the potential @xmath104 : @xmath105 where @xmath98 is an arbitrary constant . finally , we have to determine all those transformations of the magnetic field components @xmath4 and @xmath6 in ( [ deffield ] ) which correspond via ( [ defpot ] ) to transformations of the potential @xmath106 given in ( [ 2a])-([2e ] ) . since changes in @xmath4 and @xmath6 always imply changes in the potential @xmath69 that depend on @xmath48 and @xmath49 , we can only induce transformations with @xmath107 . the only non - trivial way to realize ( [ 2a ] ) is @xmath108 combined with @xmath109 with an arbitrary integer @xmath110 . similarly , ( [ 2b ] ) and ( [ 2d ] ) are equivalent to @xmath111 and @xmath112 . on the other hand ( [ 2c ] ) only admits the trivial realization @xmath113 . finally , ( [ 2e ] ) can be implemented by @xmath114 . in summary , we find that @xmath115 where the trivial result ( [ 3c ] ) is listed only for the sake of completeness . while ( [ 3a])-([3d ] ) are intuitively more or less obvious , ( [ 3e ] ) respectively ( [ 2e ] ) is not . the additional condition @xmath116 shows that the constrained dynamics ( [ lan3 ] ) , ( [ deff ] ) has an extra symmetry which is lost if the full two - dimensional dynamics ( [ lan1 ] ) , ( [ lan2 ] ) is considered . we note that ( [ 3a])-([3e ] ) could also have been obtained in a more direct way . our somewhat more involved line of reasoning has the advantage that we can exclude that there are any transformations other than ( [ 3a])-([3e ] ) and combinations thereof which would leave @xmath61 invariant or change its sign . in order to see that the above considerations already allow some non - trivial conclusions about the possibility of angular momentum transfer from the oscillating magnetic field to the particle we combine ( [ 3a ] ) and ( [ 3b ] ) , ( [ 3b ] ) and ( [ 3d ] ) , and ( [ 3a ] ) and ( [ 3e ] ) to obtain @xmath117 respectively . from ( [ 3f ] ) we immediately can infer that the constant field in @xmath5-direction is indispensable for a non - zero average torque , @xmath118 similarly ( [ 3 g ] ) implies that for a time dependence obeying @xmath119 for some @xmath85 the average torque has to vanish as well @xmath120 choosing in particular @xmath121 , _ i.e. _ half the period of the external driving , we realize that an oscillating field @xmath6 with a fourier expansion containing only odd multiples of the basic frequency @xmath122 will result in a zero average torque @xmath61 , irrespective of the particular choices of the amplitudes @xmath123 and phases @xmath124 . eqs.([4a ] ) and ( [ 4c ] ) motivate our choices ( [ foft1 ] ) and ( [ foft2 ] ) as simple time dependencies of the oscillating field resulting in a non - zero average torque @xmath61 . for the special time dependence ( [ foft1 ] ) we find from ( [ 3 g ] ) @xmath125 as a consequence , upon continuously varying @xmath13 we can infer that there must exist a @xmath126 such that @xmath127 for @xmath128 or @xmath129 . it is important to note that this case of zero averaged torque is qualitatively different from the situations described by eqs . ( [ 4a ] ) and ( [ 4b ] ) since it is due to the _ fine tuning of a parameter _ rather than resulting from an underlying symmetry . consequently , upon variation of @xmath13 around @xmath130 the average torque @xmath61 changes sign , which is a realization of a so - called _ current inversion _ @xcite . furthermore , by fixing @xmath128 , a sign change of @xmath61 will generically also occur upon variation of any other parameter of the system . interesting examples for such a parameter include the driving frequency @xmath131 , the noise intensity @xmath46 , and the particle volume @xmath0 . in the latter case we hence face the intriguing possibility that in a polydisperse ferrofluid under the same experimental conditions the larger and smaller particles will rotate in _ opposite _ directions . we finally note that in the case where the dynamics is constrained to the @xmath5-@xmath8-axis , the additional symmetry ( [ 3h ] ) implies that @xmath132 . consequently the @xmath13-values where the torque changes sign are fixed by the additional symmetry and do not depend on the other parameters of the problem . hence no current reversal upon changes in these parameters is possible . this is the main difference between the full two - dimensional dynamics ( [ lan1 ] ) , ( [ lan2 ] ) and its simplified one - dimensional counterpart ( [ lan3 ] ) , see also @xcite . a quantitative analysis of the overdamped brownian rotation of a single ferrofluid particle in a time dependent external field can be built on the numerical simulation of the langevin equations ( [ lan1 ] ) , ( [ lan2 ] ) . such simulations unambiguously show the occurrence of noise induced rotation of the particle in an oscillating field @xcite . for a detailed study of the influence of the various parameters of the problem on the transferred torque @xmath61 it is , however , more convenient to start with the equivalent fokker - planck equation ( fpe ) for the probability density @xmath133 for the orientation ( [ defe ] ) of the particle @xmath134 here @xmath135 denotes the angular part of the three dimensional nabla operator . the equivalence of ( [ lan1 ] ) , ( [ lan2 ] ) and ( [ fpe ] ) is , _ e.g. _ , shown in @xcite . for a time - periodic potential @xmath104 , the solution @xmath136 of ( [ fpe ] ) will also be periodic after initial transients have died out . from this solution we can determine the average orientation of the particle @xmath137 with the help of which the average torque from ( [ torav])-([torav ] ) can be calculated according to ( [ defmagtor ] ) in yet another way , namely @xmath138_z .\ ] ] in the present section we discuss the numerical and approximate analytical solution of the fokker - planck equation ( [ fpe ] ) and the corresponding results ( [ eq48 ] ) , ( [ torfpe ] ) for the averaged torque @xmath61 . to first substantiate the intuitive understanding of the ratchet effect in our system advocated with fig . [ fig : fig1 ] , we start by considering the weak noise limit @xmath139 . to keep the analysis simple we will restrict ourselves in this part to the simplified one - dimensional model defined by ( [ lan3 ] ) , ( [ deff ] ) . in the second subsection we detail the numerical methods used to solve the fpe for the general case . finally we present some analytical results from the perturbative solution of the fpe . our starting point is the simplified one - dimensional model dynamics ( [ lan3 ] ) , ( [ deff ] ) with a @xmath13-correlated , unbiased gaussian noise source @xmath54 . the deterministic dynamics described by ( [ lan3 ] ) with @xmath70 has a family of stable and unstable periodic orbits @xmath141 and @xmath142 , respectively , where @xmath110 is an arbitrary integer ( cf . [ fig : fig1 ] ) @xcite . hence deterministically the particle simply follows the direction of the magnetic field with a certain time lag due to viscous damping and no particle rotation can occur . for any non - zero noise intensity @xmath143 , there will be few stochastic transitions between the deterministic trajectories and their detailed form in the limit @xmath139 can be determined . comparing the associated rates for forward ( increasing @xmath49 by @xmath79 ) and backward ( decreasing @xmath49 by @xmath79 ) phase slips we may directly obtain direction and magnitude of the noise - driven transport by means of ( [ eq24 ] ) . the analysis of this section builds on the path integral representation for the transition probability of a markovian stochastic process @xcite . the main complication is the _ time dependence _ of the potential @xmath104 . in notation and general strategy we follow the recent detailed analysis of noise - driven escape over oscillating barriers @xcite . we will only determine the dominating exponential term in the transition probabilities . a more detailed analysis including also the prefactor can be done along the lines of ref . @xcite . the transition probability @xmath144 of the markov process described by ( [ lan3 ] ) can be written in the form @xmath145}{d } ) \ , \ ] ] where the action functional @xmath146 $ ] is given by @xmath147=\frac 1 4 \int_{t_i}^{t_f } dt \big({\partial}_t{\varphi}(t)-f({\varphi}(t),t)\big)^2.\ ] ] with @xmath148 defined in ( [ deff ] ) . the intuitive understanding behind this representation is that _ all _ possible trajectories @xmath55 starting at initial time @xmath93 in @xmath149 and arriving at final time @xmath95 in @xmath150 contribute to the transition probability @xmath144 , each with a weight related to the value of the action @xmath151 evaluated along this trajectory . in the weak noise limit @xmath139 a laplace argument tells us that only those transitions @xmath152 contribute significantly to @xmath144 for which the action is minimal . all others have probabilities which are exponentially small for @xmath139 . in the weak noise limit we are hence able to determine the precise form of the ( dominating ) stochastic transitions by solving a variational problem . under rather general conditions which are satisfied in our particular setting one can show that the transition rate between two stable orbits of the dynamics is determined by the probability to reach from the starting stable orbit the unstable orbit separating the two stable ones @xcite . the subsequent relaxation to the final stable orbit is then purely deterministic and does not contribute to the transition probability @xmath144 in the weak - noise limit . we have hence to solve a variational problem for which the initial and final point of the trajectory are not fixed . instead they are required to lie on two known functions @xmath153 and @xmath154 , respectively . if we perform the optimization with respect to these locations we find @xmath155 hence the optimal transition trajectory @xmath152 must be _ parallel _ to the stable and unstable orbit at the contact points . this in turn implies @xmath156 and @xmath157 . the optimal transition trajectory @xmath152 hence starts out at @xmath156 at the stable orbit , moves for a very long time near to it , changes then in a rather short time interval to the immediate neighbourhood of the unstable orbit ( hence the name `` instanton '' ) which it finally reaches at @xmath157 @xcite . from the minimization of the action @xmath146 $ ] with respect to the functions @xmath55 we get the usual euler - lagrange equation for the minimizing instanton @xmath152 which for the action given by ( [ action ] ) takes the form @xmath158 choosing for the time dependence of @xmath6 the example specified by ( [ foft1 ] ) we have solved ( [ ele1d ] ) numerically as a boundary value problem for a system of ordinary differential equations . the initial and final points were chosen slightly off the stable and unstable orbit , respectively . if the difference between the initial and final time , @xmath159 , is larger than several periods of the external driving , the solution hardly depends on the precise value of these deviations from the deterministic orbits . having obtained @xmath152 , the corresponding value of the action @xmath151 can be calculated from ( [ action ] ) . in fig . [ fig : inst1d ] the resulting instanton trajectories @xmath152 corresponding to backward and forward transitions are displayed . they show the qualitative behaviour discussed above . note , however , that only the noise induced transition from the stable to the unstable orbit is shown . the complete transitions giving rise to @xmath160 comprises also the subsequent relaxations from the unstable orbit to the next stable one . consequently the stochastic transitions show substantially more structure than the rather poor caricatures used to represent them in fig . [ fig : fig1 ] . the corresponding values of the action are @xmath161 for the forward instanton increasing @xmath49 by @xmath79 and @xmath162 for the backward instanton decreasing @xmath49 by @xmath79 . hence , at least in the weak noise limit , forward transitions are more likely ( their action is smaller ) than backward transitions and correspondingly for the parameter set specified in the caption of fig . [ fig : inst1d ] we find a noise driven rotation in the direction of increasing @xmath49 . the mismatch between forward and backward transition clearly demonstrates the operation of the ratchet effect in our system . with the relevant transition probabilities being of order @xmath163 the effect is however very weak and in particular rapidly disappears with @xmath139 . numericaly determined instanton trajectories ( full lines ) describing the dominating weak noise stochastic transitions between stable ( dashed lines ) and unstable ( dashed - dotted lines ) deterministic periodic orbits of the one - dimensional rotational dynamics of a ferrofluid particle described by ( [ lan3 ] ) , ( [ deff ] ) , ( [ foft1 ] ) . the parameter values are @xmath164 and @xmath68 . for the forward instanton , increasing @xmath49 by @xmath79 ( right ) , the value of the action is @xmath161 , for the backward instanton , decreasing @xmath49 by @xmath79 ( left ) , it is @xmath162 . hence , a noise induced average net rotation in the `` forward direction '' results in the weak noise limit , implying @xmath165 according to ( [ eq24]).,title="fig : " ] 0.5 cm numericaly determined instanton trajectories ( full lines ) describing the dominating weak noise stochastic transitions between stable ( dashed lines ) and unstable ( dashed - dotted lines ) deterministic periodic orbits of the one - dimensional rotational dynamics of a ferrofluid particle described by ( [ lan3 ] ) , ( [ deff ] ) , ( [ foft1 ] ) . the parameter values are @xmath164 and @xmath68 . for the forward instanton , increasing @xmath49 by @xmath79 ( right ) , the value of the action is @xmath161 , for the backward instanton , decreasing @xmath49 by @xmath79 ( left ) , it is @xmath162 . hence , a noise induced average net rotation in the `` forward direction '' results in the weak noise limit , implying @xmath165 according to ( [ eq24]).,title="fig : " ] for the numerical solution of the fpe describing the complete two - dimensional dynamics for general values of the parameters it is convenient to expand the probability distribution @xmath136 in spherical harmonics @xmath166 @xcite @xmath167 with so far unknown time dependent complex expansion coefficients @xmath168 . since @xmath136 is real we have @xmath169 with the star denoting complex conjugation . moreover , from the normalization of @xmath136 we find @xmath170 plugging the ansatz ( [ pexp ] ) into the fpe ( [ fpe ] ) , exploiting ( [ defpot ] ) , and using well - known properties of the spherical harmonics we find for @xmath171 the following set of ordinary differential equations @xmath172 where we have introduced the auxiliary factors @xmath173 moreover , we find for the ensemble averages of the components of @xmath72 from ( [ defe ] ) @xmath174 where @xmath175 and @xmath176 denote the real and imaginary part of the complex number @xmath59 , respectively . for the @xmath59-component of the ensemble averaged but still time - dependent magnetic torque @xmath177 this implies via ( [ defmagtor ] ) @xmath178 we have used two somewhat different procedures to explicitly calculate this torque . the first is more general , the second is slightly more efficient for the special case of the time dependence ( [ foft1 ] ) . in the first method we solve the system ( [ odeset ] ) of ordinary differential equations for the expansion coefficients @xmath168 with @xmath179 by a standard routine . we have used values of @xmath180 up to 25 but found typically @xmath181 to be sufficient . starting with some initial condition , we integrate the equation until the periodic time dependence of the solution is reached , which was usually the case after at most two periods of the external field . we then integrate further for one period of the external driving , use the result for @xmath182 in ( [ torexp ] ) , and calculate the remaining time average numerically . some care is needed here since this time - averaged torque @xmath61 is typically 2 - 3 orders of magnitude smaller than the typical time - dependent values of @xmath183 . in the time average we are hence subtracting numbers of equal size from each other giving rise to well - known problems of numerical accuracy . the advantage of this method is that it works for any numerically sensible time dependence of the magnetic field @xmath6 in ( [ deffield ] ) . the second method builds on the fact that the long - time solution of the fpe will be periodic in time with the period @xmath79 of the external driving . it is hence useful to expand the time - dependent expansion coefficients @xmath168 in ( [ pexp ] ) into fourier modes with respect to time @xmath184 instead of a system of ode s for the coefficients @xmath168 we find now from the fpe ( [ fpe ] ) a system of _ algebraic _ equations for the coefficients @xmath185 . moreover , the time averaged torque @xmath61 can be readily expressed as function of some of these coefficients by means of ( [ torexp ] ) . the advantages of this method are that effectively we are directly dealing with the stationary solution of the fpe , _ i.e. _ no initial conditions are necessary and no equilibration process must be simulated , and that the time average of the torque needs not to be performed numerically . the disadvantage is that we have to specify @xmath6 in order to find the algebraic system for the @xmath185 and that for general @xmath6 this system will be rather dense with respect to the index @xmath186 . for the special time dependence ( [ foft1 ] ) the situation is somewhat more gratifying since @xmath6 involves only two fourier components . hence there are only couplings between coefficients @xmath185 differing in their @xmath186-value by at most 2 and the result for @xmath61 is a linear combination of just 5 different coefficients @xmath185 . this makes the numerical analysis rather fast . both methods must of course yield the same results when applied to the same @xmath6 . we have frequently used both to verify our numerical findings . using the numerical methods described above we have verified all the predictions derived in section [ sec : symm ] on the basis of symmetry arguments . in particular we find @xmath127 if @xmath187 in accordance with ( [ 4a ] ) and @xmath127 for @xmath188 , _ i.e. _ for ( [ foft1 ] ) with @xmath189 , or @xmath190 in accordance with ( [ 4b ] ) . in fig . [ fig : numres ] we give some examples for current inversions in our system . the left panel shows the averaged torque @xmath61 as function of the phase angle @xmath13 for a time dependent magnetic field @xmath6 of the form ( [ foft1 ] ) . the results are consistent with ( [ nofdel ] ) . moreover , the values of @xmath13 at which the torque changes sign are clearly different from @xmath75 and @xmath191 indicated by the squares on the horizontal axis . hence the numerical solution verifies the statement of section [ sec : symm ] that the current reversal in the full two dimensional model ( [ lan1 ] ) , ( [ lan2 ] ) occurs at values of @xmath13 which are not fixed by any symmetry but rather depend in a complicated manner on all the other parameters of the problem . this is further corroborated by the right panel of fig . [ fig : numres ] , depicting the torque @xmath61 as function of the relative particle size for @xmath128 . there are two different ways to scale the particle size that make sense in a ferrofluid . in the first , both the ferromagnetic core of the particle as well as the polymer coating are scaled with the same factor . this gives rise to a proportional change of the magnetic moment @xmath17 and the particle volume @xmath0 . in the dimensionless units adopted this implies unchanged fields @xmath4 and @xmath6 and a noise intensity @xmath46 rescaled according to ( [ defd ] ) . in the other case , only the polymer coating is scaled and therefore only the hydrodynamic radius of the particle changes . hence the magnetic moment remains the same whereas the volume @xmath0 changes . correspondingly the dimensionless fields @xmath4 and @xmath6 and the noise intensity @xmath46 change by the same factor . in view of ( [ fpe ] ) this can be absorbed in a rescaled time which in turn is equivalent to a changed frequency @xmath192 of the external field . as anticipated in section [ sec : symm ] , we find current inversions in both cases . left : averaged torque @xmath61 for the time modulation ( [ foft1 ] ) with @xmath164 and @xmath193 as function of the phase angle @xmath13 as obtained from the numerical solution of the fpe ( full line ) . symbols are results from simulations of the langevin equations ( [ lan1 ] ) and ( [ lan2 ] ) as reported in @xcite with statistical errors being smaller than the symbol size . the squares on the horizontal axis are the points @xmath194 and @xmath195 at which the current reversal would occur in the reduced model ( [ lan3 ] ) . right : dependence of @xmath61 on the relative particle size for the same values of the parameters and @xmath13 fixed to @xmath196 , the first point of current inversion in the left panel . the full line corresponds to a proportional change of the magnetic and hydrodynamic radii ( top sketch ) , the dashed line is for a change of the hydrodynamic radius only ( bottom sketch).,title="fig : " ] 0.5 cm left : averaged torque @xmath61 for the time modulation ( [ foft1 ] ) with @xmath164 and @xmath193 as function of the phase angle @xmath13 as obtained from the numerical solution of the fpe ( full line ) . symbols are results from simulations of the langevin equations ( [ lan1 ] ) and ( [ lan2 ] ) as reported in @xcite with statistical errors being smaller than the symbol size . the squares on the horizontal axis are the points @xmath194 and @xmath195 at which the current reversal would occur in the reduced model ( [ lan3 ] ) . right : dependence of @xmath61 on the relative particle size for the same values of the parameters and @xmath13 fixed to @xmath196 , the first point of current inversion in the left panel . the full line corresponds to a proportional change of the magnetic and hydrodynamic radii ( top sketch ) , the dashed line is for a change of the hydrodynamic radius only ( bottom sketch).,title="fig : " ] it is very helpful to have some analytical expression for the average torque @xmath61 since the dependencies on the various parameters of the system can then be identified much more directly . in the present and subsequent subsections we discuss two approximate methods to calculate the time averaged torque in our system . the first one employs the so - called effective field method to approximately solve the fpe . it has been a standard tool in the theory of ferrofluids for many years @xcite . two main ingredients are necessary . first , let us recall that the stationary ( equilibrium ) solution of the fpe ( [ fpe ] ) for a _ time independent _ homogeneous magnetic field @xmath3 is of the form @xmath197 where @xmath198 denotes the modulus of the magnetic field , @xmath199 . averages with this distribution are easily calculated , in particular we find the well - known result @xmath200 with the langevin function @xmath201 second , for general , time - dependent fields @xmath73 the fpe ( [ fpe ] ) can be used to derive the _ exact _ equation @xmath202 for the time evolution of the average @xmath203 . note that this equation for the first moment of @xmath204 has the usual flaw of being not closed but involving higher moments of the probability distribution . the central idea of the effective field approximation is now to assume that the solution @xmath205 of the fpe for a general time dependent magnetic field @xmath73 can be written as equilibrium distribution for some a - priori unknown _ effective field _ @xmath206 , _ i.e. _ @xmath207 . the average on the r.h.s . of ( [ exequ ] ) can then be performed and this equation gives rise to an evolution equation for the effective field @xmath208 of the form @xcite @xmath209 given @xmath73 , this is a closed equation for the time evolution of @xmath206 . having obtained @xmath206 we get the desired approximate result for @xmath203 from ( [ enot ] ) @xmath210 the solution of the partial differential equation ( [ fpe ] ) is hence replaced by the solution of the set of three coupled ordinary differential equations ( [ efff ] ) . there are other , equivalent forms of ( [ efff ] ) , see _ e.g. _ @xcite . exploiting ( [ efff ] ) , it is also possible to derive the following equivalent closed equation for @xmath203 @xcite @xmath211 where @xmath212 denotes the modulus of @xmath213 and the effective field @xmath208 is to be expressed as function of @xmath213 by the inverse of ( [ eefff ] ) . note the difference between @xmath214 and @xmath215 . in general , ( [ efff ] ) can only be solved numerically . in fig . [ fig : efff ] we give a comparison between results obtained in this way for the time dependence ( [ foft1 ] ) and the corresponding outcome of the numerical solution of the fpe . it is clearly seen that the effective field approximation yields rather accurate results for the time dependent orientation of the particle and , equivalently , for the time dependent magnetization of the ferrofluid . on the other hand the results for the time averaged torque @xmath61 differ from the numerically exact values by a factor between 2 and 3 . the reason for this discrepancy lies in the fact that @xmath61 is typically 2 - 3 orders of magnitude smaller than typical values of the time dependent torque @xmath177 . in calculating the time average of the torque we hence subtract quantities of comparable order and so amplify the inaccuracies of the effective field approximation from a few percent to a few hundred percent . horizontal components of the average orientation @xmath203 of the particle as function of time for one period of the external magnetic field . the lines show the results for the @xmath5-and @xmath8-component of @xmath203 ( full and dashed line , respectively ) from the numerical solution of the fokker - planck equation ( [ fpe ] ) with a time dependence of the external field as given in ( [ foft1 ] ) . the symbols are the corresponding results from the effective field theory . the parameter values are @xmath216 as well as @xmath217 ( left ) and @xmath193 ( right ) . the corresponding values for the time averaged torque @xmath61 are also displayed.,title="fig : " ] 0.5 cm horizontal components of the average orientation @xmath203 of the particle as function of time for one period of the external magnetic field . the lines show the results for the @xmath5-and @xmath8-component of @xmath203 ( full and dashed line , respectively ) from the numerical solution of the fokker - planck equation ( [ fpe ] ) with a time dependence of the external field as given in ( [ foft1 ] ) . the symbols are the corresponding results from the effective field theory . the parameter values are @xmath216 as well as @xmath217 ( left ) and @xmath193 ( right ) . the corresponding values for the time averaged torque @xmath61 are also displayed.,title="fig : " ] in order to obtain an analytical expression for @xmath61 within the framework of the effective field approximation , it is useful to consider the case of small values of @xmath218 in which the langevin function @xmath219 or equivalently its inverse may be approximated by the first terms of the respective power series expansion . using the lowest order approximation @xmath220 we have @xmath221 and from ( [ efff1 ] ) it then follows that @xmath222 for large @xmath15 we therefore find @xmath223 hence @xmath224 tends asymptotically to a constant whereas @xmath225 contains only oscillating parts . from ( [ defmagtor ] ) we may thus infer that the torque @xmath226 harmonically oscillates and therefore its time average vanishes , @xmath127 . to get a non - trivial result for @xmath61 we have to push the expansion of the langevin function further . in other words , the non - linearity of the magnetization curve is essential to get a non - zero average torque . if the next term in the expansion is included , @xmath227 , we already obtain an approximation for @xmath219 that is accurate to within one per cent for the experimentally relevant values of @xmath5 which are generally less than @xmath228 . calculating the corrections to ( [ efffpert1 ] ) and ( [ efffpert2 ] ) induced by these higher order terms we finally find after some algebra the following result for the average torque @xmath229 this result is in accordance with our symmetry considerations of section [ sec : symm ] since it shows that both a non - zero field component in @xmath5-direction and an even higher harmonic in the time dependence of @xmath6 are essential for a non - zero torque to occur , cf . ( [ 4a ] ) , ( [ 4b ] ) . in the last subsection we saw that the effective field approximation alone is not sufficient to get an explicit analytical expression for the time averaged torque @xmath61 in our system . additionally we had to expand the langevin function in its argument @xmath218 which amounts to a small field or equivalently large noise expansion . moreover , the approximate result for the torque was typically at variance with the numerically exact result by a factor of about 2 . it is therefore tempting to avoid the effective field approximation altogether and to use a perturbation expansion in the ratio of magnetic field strength and noise intensity right in the fokker - planck equation ( [ fpe ] ) . this is conveniently accomplished by making the formal substitution @xmath230 in ( [ fpe ] ) and by expanding the solution of the fpe in a power series in @xmath231 @xmath232 at the end of the calculation @xmath231 is put equal to one . it will turn out that we have to use the expansion up to third order in @xmath231 to get a non - zero result for the average torque @xmath61 . since the potential @xmath104 is of order @xmath231 we find in zeroth order @xmath233 describing the stationary distribution of pure rotational diffusion . using the expansion ( [ ppert ] ) in ( [ fpe ] ) we get a system of linear inhomogeneous partial differential equations for the various @xmath234 that can be solved in principle one after the other . this procedure is , however , rather cumbersome . it can be somewhat simplified by using the exact equation ( [ exequ ] ) , the r.h.s of which is of order @xmath231 . hence to find @xmath203 from this equation to order @xmath235 it is sufficient to calculate the average on the r.h.s . with an expression for @xmath136 accurate to order @xmath236 . in our case the lowest order non - zero result for @xmath61 is @xmath237 , hence we need @xmath203 to order @xmath238 ( cf . ( [ defmagtor ] ) , ( [ defe ] ) ) and therefore the second order result for @xmath136 would be sufficient . it is , however , more convenient to use again the expansion ( [ pexp ] ) of @xmath136 in terms of spherical harmonics . the perturbation ansatz ( [ ppert ] ) then translates into an expansion of the coefficients @xmath168 in powers of @xmath231 @xmath239 the ode s ( [ odeset ] ) couple only coefficients @xmath168 which differ in both @xmath240 and @xmath241 by 1 . moreover , we know from ( [ resa00 ] ) @xmath242 and hence @xmath243 . as shown by ( [ torexp ] ) the torque is determined by @xmath182 alone . we hence need an expression for @xmath182 correct up to order @xmath238 . to this end we determine from ( [ odeset ] ) and ( [ resa00 ] ) first the @xmath244 terms of @xmath245 . with these we calculate the necessary @xmath246 to order @xmath247 which in turn are used to determine the @xmath248 terms in @xmath182 . the procedure is sketched in fig . [ fig : pertdiag ] . note that it is sufficient to calculate coefficients @xmath168 with @xmath249 because of the symmetry property ( [ asym ] ) . diagrammatic sketch of the perturbative determination of the expansion coefficients @xmath250 in ( [ pexp ] ) . as shown by the arrows only @xmath250 differing in both @xmath240 and @xmath241 by 1 are coupled by equations ( [ odeset ] ) . in a first step we start from @xmath251 as given in ( [ resa00 ] ) and obtain the @xmath244 terms of @xmath252 ( dashed arrows ) . next @xmath253 and @xmath254 are determined to order @xmath247 ( dotted arrows ) . finally these are used to get the @xmath248 contributions to @xmath255 ( full arrows ) from which the torque can be extracted to order @xmath256 using ( [ torexp ] ) . all other @xmath250 are irrelevant for @xmath61 at this order in @xmath231 . note also that the left half of the diagram is redundant due to the symmetry property ( [ asym]).,height=241 ] in this way we first get from the @xmath244 equations @xmath257 since we are interested in the asymptotic solution valid for large @xmath15 we can put @xmath258 . next we use the @xmath259 equations @xmath260 to determine the relevant @xmath246 to the desired order . finally we obtain @xmath261 from @xmath262 the explicit expressions for the various coefficients @xmath263 are rather long , the above system of equations is therefore conveniently solved with the help of a computer algebra . focusing on the specific time dependent field @xmath6 from ( [ foft1 ] ) , exploiting ( [ torexp ] ) and performing the time average over one period of the external driving we finally get the following lowest order perturbative result for the average torque @xmath264 the dependencies on the parameters @xmath265 , and @xmath266 are identical to those of the effective field result ( [ effresn ] ) , whereas the dependence on the noise strength @xmath46 is more complicated . one readily verifies that the symmetry properties ( [ 3a])-([3d ] ) are satisfied by ( [ pertresn ] ) . in particular , these symmetries imply that the torque @xmath61 must to be an odd function of both the static field and the oscillating field . moreover , symmetry reasons also imply @xcite that in linear order of the oscillating field one can not expect a finite @xmath61 . rather , a coupling of several modes of the oscillating field ( harmonic mixing ) is required , see also @xcite . all together , these arguments explain that the first nontrivial contribution must be ( at least ) of fourth order in the magnetic field strength and that within a linear response theory one will always find @xmath267 @xcite . on the other hand this does of course not imply that the non - linear magnetization curve alone would `` explain '' the effect . to understand the microscopic origin of the angular momentum transfer from the external field to the particle one really has to go to the description in terms of the fpe as discussed above , see also @xcite . left : time averaged torque @xmath61 as a function of the noise intensity @xmath46 for the time dependence ( [ foft1 ] ) of the magnetic field . the parameter values are @xmath268 , and @xmath68 . the scaling of the magnetic field strength with @xmath46 ensures that the perturbation parameter @xmath231 stays constant . shown are the perturbative results ( [ pertresn ] ) ( circles ) and ( [ effresn ] ) ( squares ) together with the numerically exact result ( full line ) . right : same for the time dependence ( [ foft2 ] ) of the oscillating field with @xmath269 . the approximate results are given by ( [ pertresn2 ] ) and ( [ effresn2 ] ) , respectively.,title="fig : " ] left : time averaged torque @xmath61 as a function of the noise intensity @xmath46 for the time dependence ( [ foft1 ] ) of the magnetic field . the parameter values are @xmath268 , and @xmath68 . the scaling of the magnetic field strength with @xmath46 ensures that the perturbation parameter @xmath231 stays constant . shown are the perturbative results ( [ pertresn ] ) ( circles ) and ( [ effresn ] ) ( squares ) together with the numerically exact result ( full line ) . right : same for the time dependence ( [ foft2 ] ) of the oscillating field with @xmath269 . the approximate results are given by ( [ pertresn2 ] ) and ( [ effresn2 ] ) , respectively.,title="fig : " ] in the left panel of fig . [ fig : perttest ] we compare the two approximate expressions ( [ effresn ] ) and ( [ pertresn ] ) for the averaged torque with the results obtained from the numerical solution of the fpe . displayed is the dependence of the torque on the noise intensity @xmath46 which is different in the two approximations . in order to keep the expansion parameter @xmath270 roughly constant when changing @xmath46 we have scaled the magnetic fields strengths with @xmath46 . as can be seen , up to values of @xmath271 the accuracy of ( [ pertresn ] ) is rather good whereas ( [ effresn ] ) describes the torque only qualitatively . proportional change of magnetic field strength and noise intensity is equivalent to a change of the frequency @xmath192 of the external field in ( [ deffield ] ) . [ fig : perttest ] hence also demonstrates the resonance - like character of the investigated ratchet effect . @xmath61 is largest when the deterministic time scale of the external driving matches an intrinsic stochastic time scale of the system related to the brownian relaxation time @xmath272 defined in ( [ deftaub ] ) . the two approximate expressions for the torque ( [ effresn ] ) and ( [ pertresn ] ) differ in their dependence on the phase angle @xmath13 . when using ( [ pertresn ] ) to fit the experimental results reported in @xcite we find for the brownian relaxation time the value @xmath273s instead of @xmath274s as obtained on the basis of ( [ effresn ] ) in @xcite . the experiments were done with a ferrofluid with @xmath275 pas . using ( [ deftaub ] ) the two results for @xmath272 hence translate in fits for the particle diameter of @xmath276 nm and @xmath277 nm , respectively . these values exceed the typical diameter of roughly @xmath278 nm by factors between 2 and 3 , respectively . the main reason for this discrepancy is probably the polydispersedness of real ferrofluids , having a particle size distribution with a long tail ( see _ e.g. _ @xcite , ch.2 ) . hence , a whole spectrum of relaxation times is necessary to accurately describe the dynamics of the magnetization . however , this issue is not at the focus of the present investigation . we also note that the transferred torque typically increases with the particle size and hence the described ratchet effect is likely to be dominated by the larger grains in the ferrofluid . a similar approximate calculation of the average torque is also possible for the time dependence ( [ foft2 ] ) . the result of the effective field approximation is @xmath279 whereas we find from the perturbative solution of the fpe @xmath280 the two expression only differ significantly from each other if the noise intensity @xmath46 is very small . in the right panel of fig . [ fig : perttest ] they are compared with the result from the numerical solution of the fpe ( [ fpe ] ) . for @xmath271 the accuracy is again seen to be rather satisfactory in the present paper we have theoretically investigated the rotational brownian motion of colloidal ferromagnetic particles in an oscillating magnetic field . the central tool was the fokker - planck equation for the probability density @xmath204 of the particle orientation @xmath18 at time @xmath15 . solving this equation either numerically or approximately by using the effective field approximation as well as a perturbative expansion we have determined the time averaged torque @xmath61 exerted by the magnetic field on the particles . the main and a priori quite unexpected qualitative finding is the fact that a purely oscillating magnetic field without net rotating component can transfer angular momentum to a ferromagnetic grain . as basic mechanism behind this transfer , a ratchet effect was identified by which the magnetic field rectifies the thermal fluctuations of the particle orientation that arise due to random collisions with the molecules of the carrier liquid . the detailed operation of this ratchet effect in the present system was discussed on the basis of a weak - noise analysis of the fokker - planck equation for a closely related one - dimensional system in section [ inst1d ] . via the viscous coupling to the carrier liquid the torque on the particle is transmitted to the liquid . the combined _ microscopic _ torques of the huge number of individual nanoparticles then yields a _ macroscopic _ rotation of the ferrofluid as a whole , as observed experimentally in @xcite . in the absence of thermal noise , no average torque can arise , _ i.e. _ thermal fluctuations are an indispensable requirement for the rotation of the individual grains and hence of the ferrofluid as a whole to occur . the results from the fokker - planck equation are completely consistent with rather general symmetry considerations detailed in section [ sec : symm ] . moreover , they agree very well with simulations of the corresponding langevin equations as given in @xcite and quantitatively describe the experimental findings reported in @xcite . we remark that our present system puts forward a new type of thermal ratchet device which does not fit into any previously known specific class of ratchet systems . adopting the classification scheme from @xcite , our present system has some similarity with so - called asymmetrically tilting ratchets as well as with a so - called traveling potential ratchets , however in the generalized sense involving two counterpropagating traveling potentials . yet there remain significant differences with both these classes . in particular , we note that we are not dealing here with a periodic , asymmetric so - called `` ratchet''-potential . rather , at any fixed instance of time the relevant potential is perfectly symmetric about the instantaneous direction of the magnetic field and hence no preferential direction of rotation seems to exists . it is only via the time evolution that a symmetry breaking arise , which is sometimes called a dynamical symmetry breaking . in summary we hope that the present investigation has demonstrated that ferrofluids are very suitable systems to study various aspects of thermal ratchet behaviour , and that it may stimulate further theoretical , numerical and experimental work in this direction . special thanks is due to ubbo felderhoff for several interesting suggestions and to martin raible for providing the numerical simulation results in fig . discussions with hanns walter mller , konstantin morozov , and mark shliomis are gratefully acknowledged . part of this work was done during a sabbatical stay of a.e . at the laboratoire de physique thorique at the universit louis pasteur in strasbourg . he acknowledges the kind hospitality as well as financial support from the _ volkswagenstiftung_. this work was further supported by deutsche forschungsgemeinschaft under sfb 613 and re 1344/3 - 1 and by the esf - program stochdyn . 99 a. einstein , _ ann . d. physik _ * 19 * , 371 ( 1906 ) m. i. shliomis , _ sov * 17 * , 153 ( 1974 ) r. e. rosensweig , _ ferrohydrodynamics _ , ( cambridge university press , cambridge , 1985 ) j .- c . bacri , a. cebers , and r. perzynski , _ phys . lett . _ * 72 * , 2705 ( 1994 ) a. v. lebedev , a. engel , and k. i. morozov , _ new . _ * 5 * , 57 ( 2003 ) j. p. mctague , j. chem . phys . * 51 * , 133 ( 1969 ) m. i. shliomis and k. i. morozov , _ phys . fluids _ * 6 * , 2855 ( 1994 ) f. gazeau et al . , _ europhys . _ * 35 * , 609 ( 1996 ) h. w. mller , _ phys . lett . _ * 82 * , 3907 ( 1999 ) a. engel , h. w. mller , p. reimann , and a. jung , _ phys . lett . _ * 91 * , 060602 ( 2003 ) j. c. maxwell , _ theory of heat _ ( longmans , green , and co. , london , 1871 ) m. v. smoluchowski , _ z. physik _ * 13 * , 1069 ( 1912 ) r. p. feynman , r. b. leighton , and m. sands , _ the feynman lectures on physics - vol.1 _ , ch . 46 ( addison - wesley , reading , 1963 ) s. l. harvey and a. f. rex ( eds . ) , _ maxwell s demon : entropy , information , computing _ ( adam hilger , bristol , 1990 ) p. reimann , _ phys . rep . _ * 361 * , 57 ( 2002 ) r. d. astumian and p. hnggi , _ physics today _ * 55 * , 33 ( 2002 ) m. o. magnasco , _ phys . lett . _ * 71 * , 1477 ( 1993 ) f. jlicher , a. ajdari , and j. prost , _ rev . phys . _ * 69 * , 1269 ( 1997 ) j. rousselet , l. salome , a. ajdari , and j. prost , _ nature _ * 370 * , 446 ( 1994 ) h. linke ( ed . ) _ special issue on ratchets and brownian motors : basic experiments and applications _ phys . a _ * 75 * , 167 - 352 ( 2002 ) h. linke et al . , _ science _ * 286 * , 2314 ( 1999 ) i. turcu and c. m. lucaciu , j. phys . a : math . gen . * 22 * , 995 ( 1989 ) p. palffy - muhoray and e. weinan , _ mol . cryst . _ * 320 * , 193 ( 1998 ) s. flach , a. e. miroshnichenko , and a. a. ovchinikov , _ phys . rev . _ * e65 * , 104438 ( 2002 ) w. t. coffey , y. p. kalmykov , and j. t. waldron , _ the langevin equation with applications in physics , chemistry and electrical engineering _ , ( world scientific , singapore , 1996 ) f. ritort , _ phys . rev . lett . _ * 80 * , 6 ( 1998 ) z. wang , c. holm , and h. w. mller , _ phys . * e66 * , 021405 ( 2002 ) l. d. landau and e. m. lifshitz , _ electrodynamics of continuous media _ , 2nd ed . , 34 ( pergamon , new york , 1984 ) . l. d. landau and e. m. lifshitz , _ fluid mechanics _ , 2nd ed . , 20 , ( pergamon , new york , 1987 ) . m. raible and a. engel , _ langevin equation for the rotation of a magnetic particle _ , to appear in _ applied organometallic chemistry _ h. risken , _ the fokker - planck equation _ , springer , berlin 1984 . v. becker and a. engel , unpublished p. curie , j. phys . ( paris ) 3 . srie ( thorique et appliqu ) * t. iii * , 393 ( 1894 ) . a. ajdari , d. mukamel , l. peliti , and j. prost , j. phys . i france * 4 * , 1551 ( 1994 ) r. kanada and k. sasaki , j. phys . * 68 * , 3759 ( 1999 ) s. weiss , d. koelle , j. mller , k. barthel , and r. gross , europhys . lett . * 51 * , 499 ( 2000 ) . s. flach , o. yevtushenko , and y. zolotaryuk , phys . * 84 * , 2358 ( 2000 ) s. cilla and l. m. floria , phys . e * 63 * , 031110 ( 2001 ) e. neumann and a. pikovsky , eur . phys . j. b * 26 * , 219 ( 2002 ) o. yevtushenko , s. flach , y. zolotaryuk , and a. a. ovchinikov , europhys . lett . * 54 * , 141 ( 2001 ) b. yan , r. m. miura , and y .- d . chen , j. theor . 210 * , 141 ( 2001 ) p. reimann , phys . lett . * 86 * , 4992 ( 2001 ) s. denisov , s. flach , a. a. ovchinikov , o. yevtushenko , and y. zolotaryuk , phys . e * 66 * , 041104 ( 2002 ) k. seeger and w. maurer , _ solid state commun . _ * 27 * , 603 ( 1978 ) d. r. chialvo and m. m. millonas , _ phys . lett . _ * a209 * , 26 ( 1995 ) i. goychuk and p. hnggi , _ europhys . lett . _ * 43 * , 503 ( 1998 ) r. p. feynman , a. r. hibbs , _ quantum mechanics and path integrals _ ( mcgraw - hill , new york , 1965 ) l. s. schulman , _ techniques and applications of path integration _ ( wiley , new york , 1981 ) m. chaichian and a. demichev , _ path integrals in physics _ , vol.1 ( iop publishing , bristol , 2001 ) j. lehmann , p. reimann , and p. hnggi , _ phys . _ * 84 * , 1639 ( 2000 ) ; _ phys . rev . _ * e62 * , 6282 ( 2000 ) ; _ phys . ( b ) * 237 * , 53 ( 2003 ) and further references therein . m. abramowitz and i. stegun ( eds . ) , _ handbook of mathematical functions _ ( national bureau of standards , washington d.c . , 1964 ) m. a. martsenyuk , yu . l. raikher , and m. i. shliomis , _ sov . . jetp _ * 38 * , 412 ( 1974 ) h. w. mller and m. liu , _ phys . rev . _ * e64 * , 061405 ( 2001 ) t. alarcon , a. perez - madrid , j. m. rubi , _ phys . lett . _ * 85 * , 3995 ( 2000 ) m. shliomis , _ phys . lett . _ * 92 * , 188901 a. engel and p. reimann , _ phys . _ * 92 * , 188902

a central factor in the application of machine learning to a given task is the _ inductive bias _ , _ i.e. _ the choice of hypotheses space from which learned functions are taken . the restriction posed by the inductive bias is necessary for practical learning , and reflects prior knowledge regarding the task at hand . perhaps the most successful exemplar of inductive bias to date manifests itself in the use of convolutional networks ( @xcite ) for computer vision tasks . these hypotheses spaces are delivering unprecedented visual recognition results ( _ e.g. _ @xcite ) , largely responsible for the resurgence of deep learning ( @xcite ) . unfortunately , our formal understanding of the inductive bias behind convolutional networks is limited the assumptions encoded into these models , which seem to form an excellent prior knowledge for imagery data , are for the most part a mystery . existing works studying the inductive bias of deep networks ( not necessarily convolutional ) do so in the context of _ depth efficiency _ , essentially arguing that for a given amount of resources , more layers result in higher expressiveness . more precisely , depth efficiency refers to a situation where a function realized by a deep network of polynomial size , requires super - polynomial size in order to be realized ( or approximated ) by a shallower network . in recent years , a large body of research was devoted to proving existence of depth efficiency under different types of architectures ( see for example @xcite ) . nonetheless , despite the wide attention it is receiving , depth efficiency does not convey the complete story behind the inductive bias of deep networks . while it does suggest that depth brings forth functions that are otherwise unattainable , it does not explain why these functions are useful . loosely speaking , the hypotheses space of a polynomially sized deep network covers a small fraction of the space of all functions . we would like to understand why this small fraction is so successful in practice . a specific family of convolutional networks gaining increased attention is that of _ convolutional arithmetic circuits_. these models follow the standard paradigm of locality , weight sharing and pooling , yet differ from the most conventional convolutional networks in that their point - wise activations are linear , with non - linearity originating from product pooling . recently , analyzed the depth efficiency of convolutional arithmetic circuits , showing that besides a negligible ( zero measure ) set , all functions realizable by a deep network require exponential size in order to be realized ( or approximated ) by a shallow one . this result , termed _ complete depth efficiency _ , stands in contrast to previous depth efficiency results , which merely showed _ existence _ of functions efficiently realizable by deep networks but not by shallow ones . besides their analytic advantage , convolutional arithmetic circuits are also showing promising empirical performance . in particular , they are equivalent to simnets a deep learning architecture that excels in computationally constrained settings ( ) , and in addition , have recently been utilized for classification with missing data ( ) . motivated by these theoretical and practical merits , we focus our analysis in this paper on convolutional arithmetic circuits , viewing them as representative of the class of convolutional networks . we empirically validate our conclusions with both convolutional arithmetic circuits and _ convolutional rectifier networks _ convolutional networks with rectified linear ( relu , @xcite ) activation and max or average pooling . adaptation of the formal analysis to networks of the latter type , similarly to the adaptation of the analysis in carried out by , is left for future work . our analysis approaches the study of inductive bias from the direction of function inputs . specifically , we study the ability of convolutional arithmetic circuits to model correlation between regions of their input . to analyze the correlations of a function , we consider different partitions of input regions into disjoint sets , and ask how far the function is from being separable w.r.t . these partitions . distance from separability is measured through the notion of _ separation rank _ ( @xcite ) , which can be viewed as a surrogate of the @xmath0 distance from the closest separable function . for a given function and partition of its input , high separation rank implies that the function induces strong correlation between sides of the partition , and vice versa . we show that a deep network supports exponentially high separation ranks for certain input partitions , while being limited to polynomial or linear ( in network size ) separation ranks for others . the network s pooling geometry effectively determines which input partitions are favored in terms of separation rank , _ i.e. _ which partitions enjoy the possibility of exponentially high separation rank with polynomial network size , and which require network to be exponentially large . the standard choice of square contiguous pooling windows favors interleaved ( entangled ) partitions over coarse ones that divide the input into large distinct areas . other choices lead to different preferences , for example pooling windows that join together nodes with their spatial reflections lead to favoring partitions that split the input symmetrically . we conclude that in terms of modeled correlations , pooling geometry controls the inductive bias , and the particular design commonly employed in practice orients it towards the statistics of natural images ( nearby pixels more correlated than ones that are far apart ) . moreover , when processing data that departs from the usual domain of natural imagery , prior knowledge regarding its statistics can be used to derive respective pooling schemes , and accordingly tailor the inductive bias . with regards to depth efficiency , we show that separation ranks under favored input partitions are exponentially high for all but a negligible set of the functions realizable by a deep network . shallow networks on the other hand , treat all partitions equally , and support only linear ( in network size ) separation ranks . therefore , almost all functions that may be realized by a deep network require a replicating shallow network to have exponential size . by this we return to the complete depth efficiency result of , but with an added important insight into the benefit of functions brought forth by depth they are able to efficiently model strong correlation under favored partitions of the input . the remainder of the paper is organized as follows . [ sec : prelim ] provides a brief presentation of necessary background material from the field of tensor analysis . [ sec : cac ] describes the convolutional arithmetic circuits we analyze , and their relation to tensor decompositions . in sec . [ sec : sep_rank ] we convey the concept of separation rank , on which we base our analyses in sec . [ sec : analysis ] and [ sec : inductive_bias ] . the conclusions from our analyses are empirically validated in sec . [ sec : exp ] . finally , sec . [ sec : discussion ] concludes . the analyses carried out in this paper rely on concepts and results from the field of tensor analysis . in this section we establish the minimal background required in order to follow our arguments , referring the interested reader to @xcite for a broad and comprehensive introduction to the field . the core concept in tensor analysis is a _ tensor _ , which for our purposes may simply be thought of as a multi - dimensional array . the _ order _ of a tensor is defined to be the number of indexing entries in the array , which are referred to as _ modes_. the _ dimension _ of a tensor in a particular mode is defined as the number of values that may be taken by the index in that mode . for example , a @xmath1-by-@xmath2 matrix is a tensor of order @xmath3 , _ i.e. _ it has two modes , with dimension @xmath1 in mode @xmath4 and dimension @xmath2 in mode @xmath3 . if @xmath5 is a tensor of order @xmath6 and dimension @xmath7 in each mode @xmath8:=\{1,\ldots , n\}$ ] , the space of all configurations it can take is denoted , quite naturally , by @xmath9 . a fundamental operator in tensor analysis is the _ tensor product _ , which we denote by @xmath10 . it is an operator that intakes two tensors @xmath11 and @xmath12 ( orders @xmath13 and @xmath14 respectively ) , and returns a tensor @xmath15 ( order @xmath16 ) defined by : @xmath17 . notice that in the case @xmath18 , the tensor product reduces to the standard outer product between vectors , _ i.e. _ if @xmath19 and @xmath20 , then @xmath21 is no other than the rank-1 matrix @xmath22 . we now introduce the important concept of matricization , which is essentially the rearrangement of a tensor as a matrix . suppose @xmath5 is a tensor of order @xmath6 and dimension @xmath7 in each mode @xmath8 $ ] , and let @xmath23 be a partition of @xmath24 $ ] , _ i.e. _ @xmath25 and @xmath26 are disjoint subsets of @xmath24 $ ] whose union gives @xmath24 $ ] . we may write @xmath27 where @xmath28 , and similarly @xmath29 where @xmath30 . the _ matricization of @xmath5 w.r.t . the partition @xmath23 _ , denoted @xmath31 , is the @xmath32-by-@xmath33 matrix holding the entries of @xmath5 such that @xmath34 is placed in row index @xmath35 and column index @xmath36 . if @xmath37 or @xmath38 , then by definition @xmath31 is a row or column ( respectively ) vector of dimension @xmath39 holding @xmath34 in entry @xmath40 . a well known matrix operator is the _ kronecker product _ , which we denote by @xmath41 . for two matrices @xmath42 and @xmath43 , @xmath44 is the matrix in @xmath45 holding @xmath46 in row index @xmath47 and column index @xmath48 . let @xmath5 and @xmath49 be tensors of orders @xmath13 and @xmath14 respectively , and let @xmath23 be a partition of @xmath50 $ ] . the basic relation that binds together the tensor product , the matricization operator , and the kronecker product , is : _ i , j=_i , j_(i - p),(j - p ) [ eq : mat_tensor_prod ] where @xmath51 and @xmath52 are simply the sets obtained by subtracting @xmath13 from each of the elements in @xmath25 and @xmath26 respectively . in words , eq . [ eq : mat_tensor_prod ] implies that the matricization of the tensor product between @xmath5 and @xmath49 w.r.t . the partition @xmath23 of @xmath50 $ ] , is equal to the kronecker product between two matricizations : that of @xmath5 w.r.t . the partition of @xmath53 $ ] induced by the lower values of @xmath23 , and that of @xmath49 w.r.t . the partition of @xmath54 $ ] induced by the higher values of @xmath23 . the convolutional arithmetic circuit architecture on which we focus in this paper is the one considered in , portrayed in fig . [ fig : nets_patterns](a ) . instances processed by a network are represented as @xmath6-tuples of @xmath55-dimensional vectors . they are generally thought of as images , with the @xmath55-dimensional vectors corresponding to local patches . for example , instances could be @xmath56-by-@xmath56 rgb images , with local patches being @xmath57 regions crossing the three color bands . in this case , assuming a patch is taken around every pixel in an image ( boundaries padded ) , we have @xmath58 and @xmath59 . throughout the paper , we denote a general instance by @xmath60 , with @xmath61 standing for its patches . the first layer in a network is referred to as _ representation_. it consists of applying @xmath62 _ representation functions _ @xmath63 to all patches , thereby creating @xmath62 feature maps . in the case where representation functions are chosen as @xmath64 , with parameters @xmath65 and some point - wise activation @xmath66 , the representation layer reduces to a standard convolutional layer . more elaborate settings are also possible , for example modeling the representation as a cascade of convolutional layers with pooling in - between . following the representation , a network includes @xmath67 hidden layers indexed by @xmath68 . each hidden layer @xmath69 begins with a _ @xmath70 conv _ operator , which is simply a three - dimensional convolution with @xmath71 channels and filters of spatial dimensions @xmath4-by-@xmath4 . conv operator . the first , referred to as _ weight sharing _ , is the one described above , and corresponds to standard convolution . the second is more general , allowing filters that slide across the previous layer to have different weights at different spatial locations . it is shown in that without weight sharing , a convolutional arithmetic circuit with one hidden layer ( or more ) is universal , _ i.e. _ can realize any function if its size ( width ) is unbounded . this property is imperative for the study of depth efficiency , as that requires shallow networks to ultimately be able to replicate any function realized by a deep network . in this paper we limit the presentation to networks with weight sharing , which are not universal . we do so because they are more conventional , and since our entire analysis is oblivious to whether or not weights are shared ( applies as is to both settings ) . the only exception is where we reproduce the depth efficiency result of . there , we momentarily consider networks without weight sharing . ] this is followed by spatial pooling , that decimates feature maps by taking products of non - overlapping two - dimensional windows that cover the spatial extent . the last of the @xmath67 hidden layers ( @xmath72 ) reduces feature maps to singletons ( its pooling operator is global ) , creating a vector of dimension @xmath73 . this vector is mapped into @xmath74 network outputs through a final dense linear layer . altogether , the architectural parameters of a network are the type of representation functions ( @xmath75 ) , the pooling window shapes and sizes ( which in turn determine the number of hidden layers @xmath67 ) , and the number of channels in each layer ( @xmath62 for representation , @xmath76 for hidden layers , @xmath74 for output ) . given these architectural parameters , the learnable parameters of a network are the representation weights ( @xmath77 for channel @xmath78 ) , the conv weights ( @xmath79 for channel @xmath80 of hidden layer @xmath69 ) , and the output weights ( @xmath81 for output node @xmath82 ) . for a particular setting of weights , every node ( neuron ) in a given network realizes a function from @xmath83 to @xmath84 . the _ receptive field _ of a node refers to the indexes of input patches on which its function may depend . for example , the receptive field of node @xmath85 in channel @xmath80 of conv operator at hidden layer @xmath86 is @xmath87 , and that of an output node is @xmath24 $ ] , corresponding to the entire input . denote by @xmath88 the function realized by node @xmath85 of channel @xmath80 in conv operator at hidden layer @xmath69 , and let @xmath89 $ ] be its receptive field . by the structure of the network it is evident that @xmath90 does not depend on @xmath80 , so we may write @xmath91 instead . moreover , assuming pooling windows are uniform across channels ( as customary with convolutional networks ) , and taking into account the fact that they do not overlap , we conclude that @xmath92 and @xmath93 are necessarily disjoint if @xmath94 . a simple induction over @xmath68 then shows that @xmath88 may be expressed as @xmath95 , where @xmath96 stands for the receptive field @xmath91 , and @xmath97 is a tensor of order @xmath98 and dimension @xmath62 in each mode , with entries given by polynomials in the network s conv weights @xmath99 . taking the induction one step further ( from last hidden layer to network output ) , we obtain the following expression for functions realized by network outputs : h_y(x_1, ,x_n)=_d_1 d_n=1^ma_d_1 d_n^y_i=1^n f__d_i(x_i ) [ eq : h_y ] @xmath100 $ ] here is an output node index , and @xmath101 is the function realized by that node . @xmath102 is a tensor of order @xmath6 and dimension @xmath62 in each mode , with entries given by polynomials in the network s conv weights @xmath99 and output weights @xmath81 . hereafter , terms such as _ function realized by a network _ or _ coefficient tensor realized by a network _ , are to be understood as referring to @xmath101 or @xmath102 respectively . next , we present explicit expressions for @xmath102 under two canonical networks deep and shallow . [ [ deep - network . ] ] deep network . + + + + + + + + + + + + + consider a network as in fig . [ fig : nets_patterns](a ) , with pooling windows set to cover four entries each , resulting in @xmath103 hidden layers . the linear weights of such a network are @xmath104}$ ] for conv operator in hidden layer @xmath86 , @xmath105}$ ] for conv operator in hidden layer @xmath106 , and @xmath107}$ ] for dense output operator . they determine the coefficient tensor @xmath102 ( eq . [ eq : h_y ] ) through the following recursive decomposition : _ & = & _ = 1^r_0 a_^1 , ^4 a^0 , , + & & + _ & = & _ = 1^r_l-1 a_^l , ^4 ^l-1 , , l\{2 l-1 } , + & & + _ & = & _ = 1^r_l-1 a_^l , y ^4 ^l-1 , [ eq : hr_decomp ] @xmath108 and @xmath109 here are scalars representing entry @xmath110 in the vectors @xmath79 and @xmath81 respectively , and the symbol @xmath10 with a superscript stands for a repeated tensor product , _ e.g. _ @xmath111 . to verify that under pooling windows of size four @xmath102 is indeed given by eq . [ eq : hr_decomp ] , simply plug the rows of the decomposition into eq . [ eq : h_y ] , starting from bottom and continuing upwards . for context , eq . [ eq : hr_decomp ] describes what is known as a hierarchical tensor decomposition ( see chapter 11 in @xcite ) , with underlying tree over modes being a full quad - tree ( corresponding to the fact that the network s pooling windows cover four entries each ) . [ [ shallow - network . ] ] shallow network . + + + + + + + + + + + + + + + + the second network we pay special attention to is shallow , comprising a single hidden layer with global pooling see illustration in fig . [ fig : nets_patterns](b ) . the linear weights of such a network are @xmath104}$ ] for hidden conv operator and @xmath112}$ ] for dense output operator . they determine the coefficient tensor @xmath102 ( eq . [ eq : h_y ] ) as follows : a^y=_=1^r_0 a^1,y _ ^na^0 , [ eq : cp_decomp ] where @xmath113 stands for entry @xmath80 of @xmath114 , and again , the symbol @xmath10 with a superscript represents a repeated tensor product . the tensor decomposition in eq . [ eq : cp_decomp ] is an instance of the classic cp decomposition , also known as rank-1 decomposition ( see @xcite for a historic survey ) . to conclude this section , we relate the background material above , as well as our contribution described in the upcoming sections , to the work of . the latter shows that with arbitrary coefficient tensors @xmath102 , functions @xmath101 as in eq . [ eq : h_y ] form a universal hypotheses space . it is then shown that convolutional arithmetic circuits as in fig . [ fig : nets_patterns](a ) realize such functions by applying tensor decompositions to @xmath102 , with the type of decomposition determined by the structure of a network ( number of layers , number of channels in each layer _ etc . _ ) . the deep network ( fig . [ fig : nets_patterns](a ) with size-@xmath1 pooling windows and @xmath103 hidden layers ) and the shallow network ( fig . [ fig : nets_patterns](b ) ) presented hereinabove are two special cases , whose corresponding tensor decompositions are given in eq . [ eq : hr_decomp ] and [ eq : cp_decomp ] respectively . the central result in relates to inductive bias through the notion of depth efficiency it is shown that in the parameter space of a deep network , all weight settings but a set of ( lebesgue ) measure zero give rise to functions that can only be realized ( or approximated ) by a shallow network if the latter has exponential size . this result does not relate to the characteristics of instances @xmath60 , it only treats the ability of shallow networks to replicate functions realized by deep networks . in this paper we draw a line connecting the inductive bias to the nature of @xmath115 , by studying the relation between a network s architecture and its ability to model correlation among patches @xmath116 . specifically , in sec . [ sec : sep_rank ] we consider partitions @xmath23 of @xmath24 $ ] ( @xmath117 $ ] , where @xmath118 stands for disjoint union ) , and present the notion of separation rank as a measure of the correlation modeled between the patches indexed by @xmath25 and those indexed by @xmath26 . in sec . [ sec : analysis : sep2mat ] the separation rank of a network s function @xmath101 w.r.t . a partition @xmath23 is proven to be equal to the rank of @xmath119 the matricization of the coefficient tensor @xmath102 w.r.t . [ sec : analysis : deep ] derives lower and upper bounds on this rank for a deep network , showing that it supports exponential separation ranks with polynomial size for certain partitions , whereas for others it is required to be exponentially large . subsequently , sec . [ sec : analysis : shallow ] establishes an upper bound on @xmath120 for shallow networks , implying that these must be exponentially large in order to model exponential separation rank under any partition , and thus can not efficiently replicate a deep network s correlations . our analysis concludes in sec . [ sec : inductive_bias ] , where we discuss the pooling geometry of a deep network as a means for controlling the inductive bias by determining a correspondence between partitions @xmath23 and spatial partitions of the input . finally , we demonstrate experimentally in sec . [ sec : exp ] how different pooling geometries lead to superior performance in different tasks . our experiments include not only convolutional arithmetic circuits , but also convolutional rectifier networks , _ i.e. _ convolutional networks with relu activation and max or average pooling . in this section we define the concept of separation rank for functions realized by convolutional arithmetic circuits ( sec . [ sec : cac ] ) , _ i.e. _ real functions that take as input @xmath121 . the separation rank serves as a measure of the correlations such functions induce between different sets of input patches , _ i.e. _ different subsets of the variable set @xmath122 . let @xmath23 be a partition of input indexes , _ @xmath25 and @xmath26 are disjoint subsets of @xmath24 $ ] whose union gives @xmath24 $ ] . we may write @xmath27 where @xmath28 , and similarly @xmath29 where @xmath30 . for a function @xmath123 , the _ separation rank w.r.t . the partition @xmath23 _ is defined as follows : or @xmath38 then by definition @xmath124 ( unless @xmath125 , in which case @xmath126 ) . ] sep(h;i , j):=\{rn\{0 } : s.t .. [ eq : sep_rank ] + .h(x_1, ,x_n)=_=1^rg_(x_i_1, ,x_i_)g_(x_j_1, ,x_j _ ) } in words , it is the minimal number of summands that together give @xmath127 , where each summand is _ separable w.r.t . @xmath23 _ , _ i.e. _ is equal to a product of two functions one that intakes only patches indexed by @xmath25 , and another that intakes only patches indexed by @xmath26 . one may wonder if it is at all possible to express @xmath127 through such summands , _ i.e. _ if the separation rank of @xmath127 is finite . from the theory of tensor products between @xmath0 spaces ( see @xcite for a comprehensive coverage ) , we know that any @xmath128 , _ i.e. _ any @xmath127 that is measurable and square - integrable , may be approximated arbitrarily well by summations of the form @xmath129 . exact realization however is only guaranteed at the limit @xmath130 , thus in general the separation rank of @xmath127 need not be finite . nonetheless , as we show in sec . [ sec : analysis ] , for the class of functions we are interested in , namely functions realizable by convolutional arithmetic circuits , separation ranks are always finite . the concept of separation rank was introduced in @xcite for numerical treatment of high - dimensional functions , and has since been employed for various applications , _ e.g. _ quantum chemistry ( @xcite ) , particle engineering ( @xcite ) and machine learning ( @xcite ) . if the separation rank of a function w.r.t . a partition of its input is equal to @xmath4 , the function is separable , meaning it does not model any interaction between the sets of variables . specifically , if @xmath124 then there exist @xmath131 and @xmath132 such that @xmath133 , and the function @xmath127 can not take into account consistency between the values of @xmath134 and those of @xmath135 . in a statistical setting , if @xmath127 is a probability density function , this would mean that @xmath134 and @xmath135 are statistically independent . the higher @xmath136 is , the farther @xmath127 is from this situation , _ i.e. _ the more it models dependency between @xmath134 and @xmath135 , or equivalently , the stronger the correlation it induces between the patches indexed by @xmath25 and those indexed by @xmath26 . the interpretation of separation rank as a measure of deviation from separability is formalized in app . [ app : sep_rank_l2 ] , where it is shown that @xmath136 is closely related to the @xmath0 distance of @xmath127 from the set of separable functions w.r.t . specifically , we define @xmath137 as the latter distance divided by the @xmath0 norm of @xmath127 would accordingly rescale @xmath137 , rendering the latter uninformative in terms of deviation from separability . ] , and show that @xmath136 provides an upper bound on @xmath137 . while it is not possible to lay out a general lower bound on @xmath137 in terms of @xmath136 , we show that the specific lower bounds on @xmath136 underlying our analyses can be translated into lower bounds on @xmath137 . this implies that our results , facilitated by upper and lower bounds on separation ranks of convolutional arithmetic circuits , may equivalently be framed in terms of @xmath0 distances from separable functions . in this section we analyze convolutional arithmetic circuits ( sec . [ sec : cac ] ) in terms of the correlations they can model between sides of different input partitions , _ i.e. _ in terms of the separation ranks ( sec . [ sec : sep_rank ] ) they support under different partitions @xmath23 of @xmath24 $ ] . we begin in sec . [ sec : analysis : sep2mat ] , establishing a correspondence between separation ranks and coefficient tensor matricization ranks . this correspondence is then used in sec . [ sec : analysis : deep ] and [ sec : analysis : shallow ] to analyze the deep and shallow networks ( respectively ) presented in sec . [ sec : cac ] . we note that we focus on these particular networks merely for simplicity of presentation the analysis can easily be adapted to account for alternative networks with different depths and pooling schemes . let @xmath101 be a function realized by a convolutional arithmetic circuit , with corresponding coefficient tensor @xmath102 ( eq . [ eq : h_y ] ) . denote by @xmath23 an arbitrary partition of @xmath24 $ ] , _ i.e. _ @xmath117 $ ] . we are interested in studying @xmath138 the separation rank of @xmath101 w.r.t . @xmath23 ( eq . [ eq : sep_rank ] ) . as claim [ claim : sep_mat_ranks_equal ] below states , assuming representation functions @xmath139}$ ] are linearly independent ( if they are not , we drop dependent functions and modify @xmath102 accordingly is dependent , _ i.e. _ there exist @xmath140 such that @xmath141 . we may then plug this into eq . [ eq : h_y ] , and obtain an expression for @xmath101 that has @xmath142 as representation functions , and a coefficient tensor with dimension @xmath143 in each mode . continuing in this fashion , one arrives at an expression for @xmath101 whose representation functions are linearly independent . ] ) , this separation rank is equal to the rank of @xmath119 the matricization of the coefficient tensor @xmath102 w.r.t . the partition @xmath23 . our problem thus translates to studying ranks of matricized coefficient tensors . [ claim : sep_mat_ranks_equal ] let @xmath101 be a function realized by a convolutional arithmetic circuit ( fig . [ fig : nets_patterns](a ) ) , with corresponding coefficient tensor @xmath102 ( eq . [ eq : h_y ] ) . assume that the network s representation functions @xmath75 are linearly independent , and that they , as well as the functions @xmath144 in the definition of separation rank ( eq . [ eq : sep_rank ] ) , are measurable and square - integrable . may seem as a limitation at first glance , as for example neurons @xmath64 , with parameters @xmath65 and sigmoid or relu activation @xmath66 , do not meet this condition . however , since in practice our inputs are bounded ( _ e.g. _ they represent image pixels by holding intensity values ) , we may view functions as having compact support , which , as long as they are continuous ( holds in all cases of interest ) , ensures square - integrability . ] then , for any partition @xmath23 of @xmath24 $ ] , it holds that @xmath145 . see app . [ app : proofs : sep_mat_ranks_equal ] . as the linear weights of a network vary , so do the coefficient tensors ( @xmath102 ) it gives rise to . accordingly , for a particular partition @xmath23 , a network does not correspond to a single value of @xmath120 , but rather supports a range of values . we analyze this range by quantifying its maximum , which reflects the strongest correlation that the network can model between the input patches indexed by @xmath25 and those indexed by @xmath26 . one may wonder if the maximal value of @xmath120 is the appropriate statistic to measure , as a - priori , it may be that @xmath120 is maximal for very few of the network s weight settings , and much lower for all the rest . apparently , as claim [ claim : mat_rank_max_everywhere ] below states , this is not the case , and in fact @xmath120 is maximal under almost all of the network s weight settings . [ claim : mat_rank_max_everywhere ] consider a convolutional arithmetic circuit ( fig . [ fig : nets_patterns](a ) ) with corresponding coefficient tensor @xmath102 ( eq . [ eq : h_y ] ) . @xmath102 depends on the network s linear weights @xmath99 and @xmath81 , thus for a given partition @xmath23 of @xmath24 $ ] , @xmath120 is a function of these weights . this function obtains its maximum almost everywhere ( w.r.t . lebesgue measure ) . see app . [ app : proofs : mat_rank_max_everywhere ] . in this subsection we study correlations modeled by the deep network presented in sec . [ sec : cac ] ( fig . [ fig : nets_patterns](a ) with size-@xmath1 pooling windows and @xmath103 hidden layers ) . in accordance with sec . [ sec : analysis : sep2mat ] , we do so by characterizing the maximal ranks of coefficient tensor matricizations under different partitions . recall from eq . [ eq : hr_decomp ] the hierarchical decomposition expressing a coefficient tensor @xmath102 realized by the deep network . we are interested in matricizations of this tensor under different partitions of @xmath24 $ ] . let @xmath23 be an arbitrary partition , _ i.e. _ @xmath117 $ ] . matricizing the last level of eq . [ eq : hr_decomp ] w.r.t . @xmath23 , while applying the relation in eq . [ eq : mat_tensor_prod ] , gives : ^y_i , j & = & _ = 1^r_l-1a_^l , y^l-1,^l-1,^l-1,^l-1,_i , j + & = & _ = 1^r_l-1a_^l , y^l-1,^l-1,_i , j + & & ^l-1,^l-1,_(i-24^l-1),(j-24^l-1 ) + applying eq . [ eq : mat_tensor_prod ] again , this time to matricizations of the tensor @xmath146 , we obtain : ^y_i , j & = & _ = 1^r_l-1a_^l , y^l-1,_i , j + & & ^l-1,_(i-4^l-1),(j-4^l-1 ) + & & ^l-1,_(i-24^l-1),(j-24^l-1 ) + & & ^l-1,_(i-34^l-1),(j-34^l-1 ) for every @xmath147 $ ] define @xmath148 $ ] and @xmath149 $ ] . in words , @xmath150 represents the partition induced by @xmath23 on the @xmath151th quadrant of @xmath24 $ ] , _ i.e. _ on the @xmath151th size-@xmath152 group of input patches . we now have the following matricized version of the last level in eq . [ eq : hr_decomp ] : @xmath153 where the symbol @xmath41 with a running index stands for an iterative kronecker product . to derive analogous matricized versions for the upper levels of eq . [ eq : hr_decomp ] , we define for @xmath154 $ ] : i_l , k:=(i-(k-1)4^l ) j_l , k:=(j-(k-1)4^l)[eq:_i_lk_j_lk ] that is to say , @xmath155 represents the partition induced by @xmath23 on the set of indexes @xmath156 , _ i.e. _ on the @xmath151th size-@xmath157 group of input patches . with this notation in hand , traversing upwards through the levels of eq . [ eq : hr_decomp ] , with repeated application of the relation in eq . [ eq : mat_tensor_prod ] , one arrives at the following matrix decomposition for @xmath119 : _ & = & _ = 1^r_0 a_^1 , ^0,_i_0,4(k-1)+t , j_0,4(k-1)+t , + & & + _ & = & _ = 1^r_l-1 a_^l , ^l-1,_i_l-1,4(k-1)+t , j_l-1,4(k-1)+t , l\{2 l-1 } , + & & + _ & = & _ = 1^r_l-1 a_^l , y ^l-1,_i_l-1,t , j_l-1,t [ eq : mat_hr_decomp ] eq . [ eq : mat_hr_decomp ] expresses @xmath119 the matricization w.r.t . the partition @xmath23 of a coefficient tensor @xmath102 realized by the deep network , in terms of the network s conv weights @xmath99 and output weights @xmath81 . as discussed above , our interest lies in the maximal rank that this matricization can take . theorem [ theorem : deep_mat_rank_bounds ] below provides lower and upper bounds on this maximal rank , by making use of eq . [ eq : mat_hr_decomp ] , and of the rank - multiplicative property of the kronecker product ( @xmath158 ) . [ theorem : deep_mat_rank_bounds ] let @xmath23 be a partition of @xmath24 $ ] , and @xmath119 be the matricization w.r.t . @xmath23 of a coefficient tensor @xmath102 ( eq . [ eq : h_y ] ) realized by the deep network ( fig . [ fig : nets_patterns](a ) with size-@xmath1 pooling windows ) . for every @xmath159 and @xmath160 $ ] , define @xmath161 and @xmath162 as in eq . [ eq:_i_lk_j_lk ] . then , the maximal rank that @xmath119 can take ( when network weights vary ) is : * no smaller than @xmath163 , where @xmath164 : i_{1,k}\neq\emptyset \wedge j_{1,k}\neq\emptyset\ } \right\rvert}$ ] . * no greater than @xmath165 , where @xmath166 for @xmath167 $ ] , and @xmath168 for @xmath169,k\in[n/4^l]$ ] . see app . [ app : proofs : deep_mat_rank_bounds ] . the lower bound in theorem [ theorem : deep_mat_rank_bounds ] is exponential in @xmath170 , the latter defined to be the number of size-@xmath1 patch groups that are split by the partition @xmath23 , _ i.e. _ whose indexes are divided between @xmath25 and @xmath26 . partitions that split many of the size-@xmath1 patch groups will thus lead to a large lower bound . for example , consider the partition @xmath171 defined as follows : i^odd=\{1,3, ,n-1 } j^even=\{2,4, ,n } [ eq : part_odd_even ] this partition splits all size-@xmath1 patch groups ( @xmath172 ) , leading to a lower bound that is exponential in the number of patches ( @xmath6 ) . the upper bound in theorem [ theorem : deep_mat_rank_bounds ] is expressed via constants @xmath173 , defined recursively over levels @xmath68 , with @xmath151 ranging over @xmath174 for each level @xmath69 . what prevents @xmath173 from growing double - exponentially fast ( w.r.t . @xmath69 ) is the minimization with @xmath175 . specifically , if @xmath176 is small , _ i.e. _ if the partition induced by @xmath23 on the @xmath151th size-@xmath157 group of patches is unbalanced ( most of the patches belong to one side of the partition , and only a few belong to the other ) , @xmath173 will be of reasonable size . the higher this takes place in the hierarchy ( _ i.e. _ the larger @xmath69 is ) , the lower our eventual upper bound will be . in other words , if partitions induced by @xmath23 on size-@xmath157 patch groups are unbalanced for large values of @xmath69 , the upper bound in theorem [ theorem : deep_mat_rank_bounds ] will be small . for example , consider the partition @xmath177 defined by : i^low=\{1, ,n/2 } j^high=\{n/2 + 1, ,n } [ eq : part_low_high ] under @xmath177 , all partitions induced on size-@xmath152 patch groups ( quadrants of @xmath24 $ ] ) are completely one - sided ( @xmath178 for all @xmath147 $ ] ) , resulting in the upper bound being no greater than @xmath73 linear in network size . to summarize this discussion , theorem [ theorem : deep_mat_rank_bounds ] states that with the deep network , the maximal rank of a coefficient tensor matricization w.r.t . @xmath23 , highly depends on the nature of the partition @xmath23 it will be exponentially high for partitions such as @xmath171 , that split many size-@xmath1 patch groups , while being only polynomial ( or linear ) for partitions like @xmath177 , under which size-@xmath157 patch groups are unevenly divided for large values of @xmath69 . since the rank of a coefficient tensor matricization w.r.t . @xmath23 corresponds to the strength of correlation modeled between input patches indexed by @xmath25 and those indexed by @xmath26 ( sec . [ sec : analysis : sep2mat ] ) , we conclude that the ability of a polynomially sized deep network to model correlation between sets of input patches highly depends on the nature of these sets . we now turn to study correlations modeled by the shallow network presented in sec . [ sec : cac ] ( fig . [ fig : nets_patterns](b ) ) . in line with sec . [ sec : analysis : sep2mat ] , this is achieved by characterizing the maximal ranks of coefficient tensor matricizations under different partitions . recall from eq . [ eq : cp_decomp ] the cp decomposition expressing a coefficient tensor @xmath102 realized by the shallow network . for an arbitrary partition @xmath23 of @xmath24 $ ] , _ i.e. _ @xmath117 $ ] , matricizing this decomposition with repeated application of the relation in eq . [ eq : mat_tensor_prod ] , gives the following expression for @xmath119 the matricization w.r.t . @xmath23 of a coefficient tensor realized by the shallow network : ^y_i , j=_=1^r_0 a^1,y _ ( ^a^0 , ) ( ^a^0,)^[eq : mat_cp_decomp ] @xmath179 and @xmath180 here are column vectors of dimensions @xmath181 and @xmath182 respectively , standing for the kronecker products of @xmath183 with itself @xmath184 and @xmath185 times ( respectively ) . [ eq : mat_cp_decomp ] immediately leads to two observations regarding the ranks that may be taken by @xmath119 . first , they depend on the partition @xmath23 only through its division size , _ i.e. _ through @xmath184 and @xmath185 . second , they are no greater than @xmath186 , meaning that the maximal rank is linear ( or less ) in network size . in light of sec . [ sec : analysis : sep2mat ] and [ sec : analysis : deep ] , these findings imply that in contrast to the deep network , which with polynomial size supports exponential separation ranks under favored partitions , the shallow network treats all partitions ( of a given division size ) equally , and can only give rise to an exponential separation rank if its size is exponential . suppose now that we would like to use the shallow network to replicate a function realized by a polynomially sized deep network . so long as the deep network s function admits an exponential separation rank under at least one of the favored partitions ( _ e.g. _ @xmath171 eq . [ eq : part_odd_even ] ) , the shallow network would have to be exponentially large in order to replicate it , _ i.e. _ depth efficiency takes place . ) are not universal . in particular , it may very well be that a function realized by a polynomially sized deep network can not be replicated by the shallow network , no matter how large ( wide ) we allow it to be . in such scenarios depth efficiency does not provide insight into the complexity of functions brought forth by depth . to obtain a shallow network that is universal , thus an appropriate gauge for depth efficiency , we may remove the constraint of weight sharing , _ i.e. _ allow the filters in the hidden conv operator to hold different weights at different spatial locations ( see for proof that this indeed leads to universality ) . all results we have established for the original shallow network remain valid when weight sharing is removed . in particular , the separation ranks of the network are still linear in its size . this implies that as suggested , depth efficiency indeed holds . ] since all but a negligible set of the functions realizable by the deep network give rise to maximal separation ranks ( sec [ sec : analysis : sep2mat ] ) , we obtain the complete depth efficiency result of . however , unlike , which did not provide any explanation for the usefulness of functions brought forth by depth , we obtain an insight into their utility they are able to efficiently model strong correlation under favored partitions of the input . the deep network presented in sec . [ sec : cac ] , whose correlations we analyzed in sec . [ sec : analysis : deep ] , was defined as having size-@xmath1 pooling windows , _ i.e. _ pooling windows covering four entries each . we have yet to specify the shapes of these windows , or equivalently , the spatial ( two - dimensional ) locations of nodes grouped together in the process of pooling . in compliance with standard convolutional network design , we now assume that the network s ( size-@xmath1 ) pooling windows are contiguous square blocks , _ i.e. _ have shape @xmath187 . under this configuration , the network s functional description ( eq . [ eq : h_y ] with @xmath102 given by eq . [ eq : hr_decomp ] ) induces a spatial ordering of input patches ) , the nodes corresponding to patches @xmath188 are pooled into one group , those corresponding to @xmath189 are pooled into another , and so forth . similar assumptions hold for the deeper layers . for example , in the second pooling operation ( hidden layer @xmath4 ) , the node with receptive field @xmath190 , _ i.e. _ the one corresponding to the quadruple of patches @xmath191 , is assumed to be pooled together with the nodes whose receptive fields are @xmath192 , @xmath193 and @xmath194 . ] , which may be described by the following recursive process : * set the index of the top - left patch to @xmath4 . * for @xmath195 : replicate the already - assigned top - left @xmath196-by-@xmath196 block of indexes , and place copies on its right , bottom - right and bottom . then , add a @xmath197 offset to all indexes in the right copy , a @xmath198 offset to all indexes in the bottom - right copy , and a @xmath199 offset to all indexes in the bottom copy . with this spatial ordering ( illustrated in fig . [ fig : nets_patterns](c ) ) , partitions @xmath23 of @xmath24 $ ] convey a spatial pattern . for example , the partition @xmath171 ( eq . [ eq : part_odd_even ] ) corresponds to the pattern illustrated on the left of fig . [ fig : nets_patterns](c ) , whereas @xmath177 ( eq . [ eq : part_low_high ] ) corresponds to the pattern illustrated on the right . our analysis ( sec . [ sec : analysis : deep ] ) shows that the deep network is able to model strong correlation under @xmath171 , while being inefficient for modeling correlation under @xmath177 . more generally , partitions for which @xmath170 , defined in theorem [ theorem : deep_mat_rank_bounds ] , is high , convey patterns that split many @xmath187 patch blocks , _ i.e. _ are highly entangled . these partitions enjoy the possibility of strong correlation . on the other hand , partitions for which @xmath176 is small for large values of @xmath69 ( see eq . [ eq:_i_lk_j_lk ] for definition of @xmath161 and @xmath162 ) convey patterns that divide large @xmath200 patch blocks unevenly , _ i.e. _ separate the input to distinct contiguous regions . these partitions , as we have seen , suffer from limited low correlations . we conclude that with @xmath187 pooling , the deep network is able to model strong correlation between input regions that are highly entangled , at the expense of being inefficient for modeling correlation between input regions that are far apart . had we selected a different pooling regime , the preference of input partition patterns in terms of modeled correlation would change . for example , if pooling windows were set to group nodes with their spatial reflections ( horizontal , vertical and horizontal - vertical ) , coarse patterns that divide the input symmetrically , such as the one illustrated on the right of fig . [ fig : nets_patterns](c ) , would enjoy the possibility of strong correlation , whereas many entangled patterns would now suffer from limited low correlation . the choice of pooling shapes thus serves as a means for controlling the inductive bias in terms of correlations modeled between input regions . square contiguous windows , as commonly employed in practice , lead to a preference that complies with our intuition regarding the statistics of natural images ( nearby pixels more correlated than distant ones ) . other pooling schemes lead to different preferences , and this allows tailoring a network to data that departs from the usual domain of natural imagery . we demonstrate this experimentally in the next section , where it is shown how different pooling geometries lead to superior performance in different tasks . the main conclusion from our analyses ( sec . [ sec : analysis ] and [ sec : inductive_bias ] ) is that the pooling geometry of a deep convolutional network controls its inductive bias by determining which correlations between input regions can be modeled efficiently . we have also seen that shallow networks can not model correlations efficiently , regardless of the considered input regions . in this section we validate these assertions empirically , not only with convolutional arithmetic circuits ( subject of our analyses ) , but also with convolutional rectifier networks convolutional networks with relu activation and max or average pooling . for conciseness , we defer to app . [ app : impl ] some details regarding our implementation . the latter is fully available online at https://github.com/huji-deep/inductive-pooling . our experiments are based on a synthetic classification benchmark inspired by medical imaging tasks . instances to be classified are @xmath56-by-@xmath56 binary images , each displaying a random distorted oval shape ( _ blob _ ) with missing pixels in its interior ( _ holes _ ) . for each image , two continuous scores in range @xmath201 $ ] are computed . the first , referred to as _ closedness _ , reflects how morphologically closed a blob is , and is defined to be the ratio between the number of pixels in the blob , and the number of pixels in its closure ( see app . [ app : morph_closure ] for exact definition of the latter ) . the second score , named _ symmetry _ , reflects the degree to which a blob is left - right symmetric about its center . it is measured by cropping the bounding box around a blob , applying a left - right flip to the latter , and computing the ratio between the number of pixels in the intersection of the blob and its reflection , and the number of pixels in the blob . to generate labeled sets for classification ( train and test ) , we render multiple images , sort them according to their closedness and symmetry , and for each of the two scores , assign the label `` high '' to the top 40% and the label `` low '' to the bottom 40% ( the mid 20% are considered ill - defined ) . this creates two binary ( two - class ) classification tasks one for closedness and one for symmetry ( see fig . [ fig : exp_data ] for a sample of images participating in both tasks ) . given that closedness is a property of a local nature , we expect its classification task to require a predictor to be able to model strong correlations between neighboring pixels . symmetry on the other hand is a property that relates pixels to their reflections , thus we expect its classification task to demand that a predictor be able to model correlations across distances . we evaluated the deep convolutional arithmetic circuit considered throughout the paper ( fig . [ fig : nets_patterns](a ) with size-@xmath1 pooling windows ) under two different pooling geometries . the first , referred to as _ , comprises standard @xmath187 pooling windows . the second , dubbed _ mirror _ , pools together nodes with their horizontal , vertical and horizontal - vertical reflections . in both cases , input patches ( @xmath116 ) were set as individual pixels , resulting in @xmath58 patches and @xmath202 hidden layers . @xmath203 representation functions ( @xmath75 ) were fixed , the first realizing the identity on binary inputs ( @xmath204 for @xmath205 ) , and the second realizing negation ( @xmath206 for @xmath205 ) . classification was realized through @xmath207 network outputs , with prediction following the stronger activation . the number of channels across all hidden layers was uniform , and varied between @xmath208 and @xmath209 . [ fig : exp_results_cac_pool4 ] shows the results of applying the deep network with both square and mirror pooling , to both closedness and symmetry tasks , where each of the latter has @xmath210 images for training and @xmath211 images for testing . as can be seen in the figure , square pooling significantly outperforms mirror pooling in closedness classification , whereas the opposite occurs in symmetry classification . this complies with our discussion in sec . [ sec : inductive_bias ] , according to which square pooling supports modeling correlations between entangled ( neighboring ) regions of the input , whereas mirror pooling puts focus on correlations between input regions that are symmetric w.r.t . one another . we thus obtain a demonstration of how prior knowledge regarding a task at hand may be used to tailor the inductive bias of a deep convolutional network by designing an appropriate pooling geometry . in addition to the deep network , we also evaluated the shallow convolutional arithmetic circuit analyzed in the paper ( fig . [ fig : nets_patterns](b ) ) . the architectural choices for this network were the same as those described above for the deep network besides the number of hidden channels , which in this case applied to the network s single hidden layer , and varied between @xmath212 and @xmath213 . the highest train and test accuracies delivered by this network ( with @xmath213 hidden channels ) were roughly @xmath214 on closedness task , and @xmath215 on symmetry task . the fact that these accuracies are inferior to those of the deep network , even when the latter s pooling geometry is not optimal for the task at hand , complies with our analysis in sec . [ sec : analysis ] . namely , it complies with the observation that separation ranks ( correlations ) are sometimes exponential and sometimes polynomial with the deep network , whereas with the shallow one they are never more than linear in network size . finally , to assess the validity of our findings for convolutional networks in general , not just convolutional arithmetic circuits , we repeated the above experiments with convolutional rectifier networks . namely , we placed relu activations after every conv operator , switched the pooling operation from product to average , and re - evaluated the deep ( square and mirror pooling geometries ) and shallow networks . we then reiterated this process once more , with pooling operation set to max instead of average . the results obtained by the deep networks are presented in fig . [ fig : exp_results_crn_pool4 ] . the shallow network with average pooling reached train / test accuracies of roughly @xmath216 on closedness task , and @xmath217 on symmetry task . with max pooling , performance of the shallow network did not exceed chance . altogether , convolutional rectifier networks exhibit the same phenomena observed with convolutional arithmetic circuits , indicating that the conclusions from our analyses likely apply to such networks as well . formal adaptation of the analyses to convolutional rectifier networks , similarly to the adaptation of carried out in , is left for future work . through the notion of separation rank , we studied the relation between the architecture of a convolutional network , and its ability to model correlations among input regions . for a given input partition , the separation rank quantifies how far a function is from separability , which in a probabilistic setting , corresponds to statistical independence between sides of the partition . our analysis shows that a polynomially sized deep convolutional arithmetic circuit supports exponentially high separation ranks for certain input partitions , while being limited to polynomial or linear ( in network size ) separation ranks for others . the network s pooling window shapes effectively determine which input partitions are favored in terms of separation rank , _ i.e. _ which partitions enjoy the possibility of exponentially high separation ranks with polynomial network size , and which require network to be exponentially large . pooling geometry thus serves as a means for controlling the inductive bias . the particular pooling scheme commonly employed in practice square contiguous windows , favors interleaved partitions over ones that divide the input to distinct areas , thus orients the inductive bias towards the statistics of natural images ( nearby pixels more correlated than distant ones ) . other pooling schemes lead to different preferences , and this allows tailoring the network to data that departs from the usual domain of natural imagery . as opposed to deep convolutional arithmetic circuits , shallow ones support only linear ( in network size ) separation ranks . therefore , in order to replicate a function realized by a deep network ( exponential separation rank ) , a shallow network must be exponentially large . by this we derive the depth efficiency result of , but in addition , provide an insight into the benefit of functions brought forth by depth they are able to efficiently model strong correlation under favored partitions of the input . we validated our conclusions empirically , with convolutional arithmetic circuits as well as convolutional rectifier networks convolutional networks with relu activation and max or average pooling . our experiments demonstrate how different pooling geometries lead to superior performance in different tasks . specifically , we evaluate deep networks in the measurement of shape continuity , a task of a local nature , and show that standard square pooling windows outperform ones that join together nodes with their spatial reflections . in contrast , when measuring shape symmetry , modeling correlations across distances is of vital importance , and the latter pooling geometry is superior to the conventional one . shallow networks are inefficient at modeling correlations of any kind , and indeed lead to poor performance on both tasks . finally , our analyses and results bring forth the possibility of expanding the coverage of correlations efficiently modeled by a deep convolutional network . specifically , by blending together multiple pooling geometries in the hidden layers of a network , it is possible to facilitate simultaneous support for a wide variety of correlations suiting data of different types . investigation of this direction , from both theoretical and empirical perspectives , is viewed as a promising avenue for future research . this work is supported by intel grant icri - ci # 9 - 2012 - 6133 , by isf center grant 1790/12 and by the european research council ( theorydl project ) . nadav cohen is supported by a google doctoral fellowship in machine learning . we prove the equality in two steps , first showing that @xmath218 , and then establishing the converse . the first step is elementary , and does not make use of the representation functions ( @xmath75 ) linear independence , or of measurability / square - integrability . the second step does rely on these assumptions , and employs slightly more advanced mathematical machinery . throughout the proof , we assume without loss of generality that the partition @xmath23 of @xmath24 $ ] is such that @xmath25 takes on lower values , while @xmath26 takes on higher ones . that is to say , we assume that @xmath219 and @xmath220 . and @xmath29 , and define an auxiliary function @xmath221 by permuting the entries of @xmath101 such that those indexed by @xmath25 are on the left and those indexed by @xmath26 on the right , _ i.e. _ @xmath222 . obviously @xmath223 , where the partition @xmath224 is defined by @xmath225 and @xmath226 . analogously to the definition of @xmath221 , let @xmath227 be the tensor obtained by permuting the modes of @xmath102 such that those indexed by @xmath25 are on the left and those indexed by @xmath26 on the right , _ it is not difficult to see that matricizing @xmath229 w.r.t . @xmath224 is equivalent to matricizing @xmath230 w.r.t . @xmath23 , _ i.e. _ @xmath231 , and in particular @xmath232 . moreover , since by definition @xmath230 is a coefficient tensor corresponding to @xmath101 ( eq . [ eq : h_y ] ) , @xmath227 will be a coefficient tensor that corresponds to @xmath221 . now , our proof will show that @xmath233 , which , in light of the equalities above , implies @xmath145 , as required . ] to prove that @xmath218 , denote by @xmath234 the rank of @xmath119 . the latter is an @xmath181-by-@xmath182 matrix , thus there exist vectors @xmath235 and @xmath236 such that @xmath237 . for every @xmath238 $ ] , let @xmath239 be the tensor of order @xmath184 and dimension @xmath62 in each mode whose arrangement as a column vector gives @xmath240 , _ i.e. _ whose matricization w.r.t . the partition @xmath241,\emptyset)$ ] is equal to @xmath240 . similarly , let @xmath242 , @xmath238 $ ] , be the tensor of order @xmath243 and dimension @xmath62 in each mode whose matricization w.r.t . the partition @xmath244)$ ] ( arrangement as a row vector ) is equal to @xmath245 . it holds that : ^y_i , j & = & _ = 1^ru_v_^ + & = & _ = 1^r^_[],^ _ , [ ] + & = & _ = 1^r^_i , j^_(i-),(j- ) + & = & _ = 1^r^^_i , j + & = & _ = 1^rb^^_i , j where the third equality relies on the assumption @xmath246 , the fourth equality makes use of the relation in eq . [ eq : mat_tensor_prod ] , and the last equality is based on the linearity of the matricization operator . since matricizations are merely rearrangements of tensors , the fact that @xmath247 implies @xmath248 , or equivalently , @xmath249 for every @xmath250 $ ] . plugging this into eq . [ eq : h_y ] gives : h_y(x_1, ,x_n ) & = & _ d_1 d_n=1^ma_d_1 d_n^y_i=1^n f__d_i(x_i ) + & = & _ d_1 d_n=1^m_=1^rb^_d_1 d_c^_d_+1 d_n_i=1^n f__d_i(x_i ) + & = & _ = 1^r(_d_1 d_=1^mb^_d_1 d__i=1^ f__d_i(x_i ) ) + & & ( _ d_+1 d_n=1^mc^_d_+1 d_n_i=+1^n f__d_i(x_i ) ) [ eq : mat_rank_over_sep_rank ] for every @xmath238 $ ] , define the functions @xmath251 and @xmath252 as follows : g_(x_1, ,x _ ) & : = & _ d_1 d_=1^mb^_d_1 d__i=1^ f__d_i(x_i ) + g_(x_1, ,x _ ) & : = & _ d_1 d_=1^mc^_d_1 d__i=1^ f__d_i(x_i ) substituting these into eq . [ eq : mat_rank_over_sep_rank ] leads to : @xmath253 which by definition of the separation rank ( eq . [ eq : sep_rank ] ) , implies @xmath254 . by this we have shown that @xmath218 , as required . for proving the converse inequality , _ i.e. _ @xmath255 , we rely on basic concepts and results from functional analysis , or more specifically , from the topic of @xmath0 spaces . while a full introduction to this topic is beyond our scope ( the interested reader is referred to @xcite ) , we briefly lay out here the minimal background required in order to follow our proof . for any @xmath256 , @xmath257 is formally defined as the hilbert space of lebesgue measurable square - integrable real functions over @xmath258 on which they differ has measure zero . ] , equipped with standard ( point - wise ) addition and scalar multiplication , as well as the inner product defined by integration over point - wise multiplication . for our purposes , @xmath257 may simply be thought of as the ( infinite - dimensional ) vector space of functions @xmath259 satisfying @xmath260 , with inner product defined by @xmath261 . our proof will make use of the following basic facts related to @xmath0 spaces : [ fact : orth_decomp ] if @xmath262 is a finite - dimensional subspace of @xmath257 , then any @xmath263 may be expressed as @xmath264 , with @xmath265 and @xmath266 ( _ i.e. _ @xmath267 is orthogonal to all elements in @xmath262 ) . moreover , such a representation is unique , so in the case where @xmath268 , we necessarily have @xmath269 and @xmath270 . [ fact : prod ] if @xmath271 , then the function @xmath272 belongs to @xmath273 . [ fact : prod_orth ] let @xmath262 and @xmath274 be finite - dimensional subspaces of @xmath275 and @xmath276 respectively , and define @xmath277 to be the subspace spanned by @xmath278 . given @xmath271 , consider the function @xmath272 in @xmath273 . this function belongs to @xmath279 if @xmath280 or @xmath281 . [ fact : prod_ind ] if @xmath282 are linearly independent , then for any @xmath283 , the set of functions @xmath284}$ ] is linearly independent in @xmath285 . to facilitate application of the theory of @xmath0 spaces , we now make use of the assumption that the network s representation functions @xmath75 , as well as the functions @xmath144 in the definition of separation rank ( eq . [ eq : sep_rank ] ) , are measurable and square - integrable . taking into account the expression given in eq . [ eq : h_y ] for @xmath101 , as well as fact [ fact : prod ] above , one readily sees that @xmath286 implies @xmath287 . the separation rank @xmath138 will be the minimal non - negative integer @xmath234 such that there exist @xmath288 and @xmath289 for which : h_y(x_1, ,x_n)=_=1^rg_(x_1, ,x_)g_(x_+1, ,x_n ) [ eq : h_y_sep ] we would like to show that @xmath255 . our strategy for achieving this will be to start from eq . [ eq : h_y_sep ] , and derive an expression for @xmath119 comprising a sum of @xmath234 rank-@xmath4 matrices . as an initial step along this path , define the following finite - dimensional subspaces : v&:=&span\{(x_1, ,x_)_i=1^f__d_i(x_i)}_d_1 d _ l^2((r^s)^ ) [ eq : v ] + v&:=&span\{(x_1, ,x_)_i=1^f__d_i(x_i)}_d_1 d _ l^2((r^s)^ ) [ eq : vpr ] + u&:=&span\{(x_1, ,x_n)_i=1^nf__d_i(x_i)}_d_1 d_n l^2((r^s)^n ) [ eq : u ] notice that @xmath290 ( eq . [ eq : h_y ] ) , and that @xmath291 is the span of products from @xmath262 and @xmath274 , _ i.e. _ : u = span\{(x_1, ,x_n)p(x_1, ,x_)p(x_+1, ,x_n):pv , pv } [ eq : u_prod_v_vpr ] returning to eq . [ eq : h_y_sep ] , we apply fact [ fact : orth_decomp ] to obtain orthogonal decompositions of @xmath292 w.r.t . @xmath262 , and of @xmath293 w.r.t . this gives @xmath294 , @xmath295 , @xmath296 and @xmath297 , such that @xmath298 and @xmath299 for every @xmath238 $ ] . plug this into eq . [ eq : h_y_sep ] : h_y(x_1, ,x_n ) & = & _ = 1^rg_(x_1, ,x_)g_(x_+1, ,x_n ) + & = & _ = 1^r(p_(x_1, ,x_)+_(x_1, ,x _ ) ) + & & ( p_(x_+1, ,x_n)+_(x_+1, ,x_n ) ) + & = & _ = 1^rp_(x_1, ,x_)p_(x_+1, ,x_n ) + & & + _ = 1^rp_(x_1, ,x_)_(x_+1, ,x_n ) + & & + _ = 1^r_(x_1, ,x_)p_(x_+1, ,x_n ) + & & + _ = 1^r_(x_1, ,x_)_(x_+1, ,x_n ) given that @xmath291 is the span of products from @xmath262 and @xmath274 ( eq . [ eq : u_prod_v_vpr ] ) , and that @xmath300 , one readily sees that the first term in the latter expression belongs to @xmath291 , while , according to fact [ fact : prod_orth ] , the second , third and fourth terms are orthogonal to @xmath291 . we thus obtained an orthogonal decomposition of @xmath101 w.r.t . @xmath291 . since @xmath101 is contained in @xmath291 , the orthogonal component must vanish ( fact [ fact : orth_decomp ] ) , and we amount at : h_y(x_1, ,x_n)=_=1^rp_(x_1, ,x_)p_(x_+1, ,x_n ) [ eq : h_y_sep_subspace ] for every @xmath238 $ ] , let @xmath239 and @xmath242 be coefficient tensors of @xmath301 and @xmath302 w.r.t . the functions that span @xmath262 and @xmath274 ( eq . [ eq : v ] and [ eq : vpr ] ) , respectively . put formally , @xmath239 and @xmath242 are tensors of orders @xmath184 and @xmath185 ( respectively ) , with dimension @xmath62 in each mode , meeting : p_(x_1, ,x _ ) & = & _ d_1 d_=1^mb^_d_1 d__i=1^ f__d_i(x_i ) + p_(x_1, ,x _ ) & = & _ d_1 d_=1^mc^_d_1 d__i=1^ f__d_i(x_i ) substitute into eq . [ eq : h_y_sep_subspace ] : h_y(x_1, ,x_n ) & = & _ = 1^r(_d_1 d_=1^mb^_d_1 d__i=1^ f__d_i(x_i ) ) + & & ( _ d_+1 d_n=1^mc^_d_+1 d_n_i=+1^n f__d_i(x_i ) ) + & = & _ = 1^r_d_1 d_n=1^mb^_d_1 d_n_i=1^n f__d_i(x_i ) + & = & _ d_1 d_n=1^m(_=1^rb^_d_1 d_c^_d_+1 d_n)_i=1^n f__d_i(x_i ) compare this expression for @xmath101 to that given in eq . [ eq : h_y ] : _ d_1 d_n=1^m(_=1^rb^_d_1 d_c^_d_+1 d_n)_i=1^n f__d_i(x_i ) = _ d_1 d_n=1^ma_d_1 d_n^y_i=1^n f__d_i(x_i ) [ eq : h_y_compare ] at this point we utilize the given linear independence of @xmath286 , from which it follows ( fact [ fact : prod_ind ] ) that the functions spanning @xmath291 ( eq . [ eq : u ] ) are linearly independent in @xmath303 . both sides of eq . [ eq : h_y_compare ] are linear combinations of these functions , thus their coefficients must coincide : @xmath304 } ~~\longleftrightarrow~~ { { \mathcal a}}^y=\sum\nolimits_{\nu=1}^r{{\mathcal b}}^{\nu}\otimes{{\mathcal c}}^{\nu}\ ] ] matricizing the tensor equation on the right w.r.t . @xmath23 gives : ^y_i , j & = & _ = 1^rb^^_i , j + & = & _ = 1^r^^_i , j + & = & _ = 1^r^_i , j^_(i-),(j- ) + & = & _ = 1^r^_[],^ _ , [ ] where the second equality is based on the linearity of the matricization operator , the third equality relies on the relation in eq . [ eq : mat_tensor_prod ] , and the last equality makes use of the assumption @xmath246 . for every @xmath238 $ ] , @xmath305,\emptyset}$ ] is a column vector of dimension @xmath181 and @xmath306}$ ] is a row vector of dimension @xmath182 . denoting these by @xmath240 and @xmath245 respectively , we may write : @xmath307 this shows that @xmath308 . since @xmath234 is a general non - negative integer that admits eq . [ eq : h_y_sep ] , we may take it to be minimal , _ i.e. _ to be equal to @xmath138 the separation rank of @xmath101 w.r.t . @xmath23 . by this we obtain @xmath309 , which is what we set out to prove . the claim is framed in measure theoretical terms , and in accordance , so will its proof be . while a complete introduction to measure theory is beyond our scope ( the interested reader is referred to @xcite ) , we briefly convey here the intuition behind the concepts we will be using , as well as facts we rely upon . the _ lebesgue measure _ is defined over sets in a euclidean space , and may be interpreted as quantifying their `` volume '' . for example , the lebesgue measure of a unit hypercube is one , of the entire space is infinity , and of a finite set of points is zero . in this context , when a phenomenon is said to occur _ almost everywhere _ , it means that the set of points in which it does not occur has lebesgue measure zero , _ i.e. _ is negligible . an important result we will make use of ( proven in @xcite for example ) is the following . given a polynomial defined over @xmath310 real variables , the set of points in @xmath258 on which it vanishes is either the entire space ( when the polynomial in question is the zero polynomial ) , or it must have lebesgue measure zero . in other words , if a polynomial is not identically zero , it must be different from zero almost everywhere . heading on to the proof , we recall from sec . [ sec : cac ] that the entries of the coefficient tensor @xmath102 ( eq . [ eq : h_y ] ) are given by polynomials in the network s conv weights @xmath99 and output weights @xmath81 . since @xmath119 the matricization of @xmath102 w.r.t . the partition @xmath23 , is merely a rearrangement of the tensor as a matrix , this matrix too has entries given by polynomials in the network s linear weights . now , denote by @xmath311 the maximal rank taken by @xmath119 as network weights vary , and consider a specific setting of weights for which this rank is attained . we may assume without loss of generality that under this setting , the top - left @xmath311-by-@xmath311 block of @xmath119 is non - singular . the corresponding minor , _ i.e. _ the determinant of the sub - matrix @xmath312 , is thus a polynomial defined over @xmath99 and @xmath81 which is not identically zero . in light of the above , this polynomial is different from zero almost everywhere , implying that @xmath313 almost everywhere . since @xmath314 , and since by definition @xmath311 is the maximal rank that @xmath119 can take , we have that @xmath120 is maximal almost everywhere . the matrix decomposition in eq . [ eq : mat_hr_decomp ] expresses @xmath31 in terms of the network s linear weights @xmath104}$ ] for conv operator in hidden layer @xmath86 , @xmath105}$ ] for conv operator in hidden layer @xmath106 , and @xmath315 for node @xmath82 of dense output operator . we prove lower and upper bounds on the maximal rank that @xmath31 can take as these weights vary . our proof relies on the rank - multiplicative property of the kronecker product ( @xmath158 for any real matrices @xmath316 and @xmath317 see @xcite for proof ) , but is otherwise elementary . beginning with the lower bound , consider the following weight setting ( @xmath318 here stands for a vector holding @xmath4 in entry @xmath80 and @xmath86 at all other entries , @xmath319 stands for a vector holding @xmath86 at all entries , and @xmath320 stands for a vector holding @xmath4 at all entries , with the dimension of a vector to be understood by context ) : a^0 , & = & \ { ll e _ & , \{r_0,m } + 0 & , . [ eq : weight_setting ] + a^1 , & = & \ { ll 1 & , = 1 + 0 & , . + a^l , & = & \ { ll e_1 & , = 1 + 0 & , . + a^l , y & = & e_1 let @xmath321 $ ] . recalling the definition of @xmath161 and @xmath162 from eq . [ eq:_i_lk_j_lk ] , consider the sets @xmath322 and @xmath323 , as well as @xmath324 and @xmath325 for @xmath326 $ ] . @xmath327 is a partition of @xmath328 $ ] , _ i.e. _ @xmath329 $ ] , and for every @xmath326 $ ] we have @xmath330 and @xmath331 if @xmath332 belongs to @xmath322 , and otherwise @xmath333 and @xmath334 if @xmath332 belongs to @xmath323 . this implies that for an arbitrary vector @xmath335 , the matricization @xmath336 is equal to @xmath335 if @xmath337 , and to @xmath338 if @xmath339 . accordingly , for any @xmath340 $ ] : @xmath341 assume that @xmath342 . by our setting @xmath343 , so the above matrix holds @xmath4 in a single entry and @xmath86 in all the rest . moreover , if the matrix is not a row or column vector , _ i.e. _ if both @xmath322 and @xmath323 are non - empty , the column index and row index of the entry holding @xmath4 are both unique w.r.t . @xmath80 , _ i.e. _ they do not repeat as @xmath80 ranges over @xmath344 . we thus have : @xmath345 since we set @xmath346 and @xmath347 for @xmath348 , we may write : @xmath349 the latter matrix is by definition equal to @xmath350 ( see top row of eq . [ eq : mat_hr_decomp ] ) , and so for every @xmath321 $ ] : rank^1,1_i_1,n , j_1,n=\ { ll \{r_0,m } & , i_1,n j_1,n + 1 & , i_1,n= j_1,n= . [ eq : phi_1_1_rank ] now , the fact that we set @xmath351 and @xmath352 for @xmath353 , implies that the second to last levels of the decomposition in eq . [ eq : mat_hr_decomp ] collapse to : @xmath354 applying the rank - multiplicative property of the kronecker product , and plugging in eq . [ eq : phi_1_1_rank ] , we obtain : @xmath355 where @xmath356 : i_{1,t}\neq\emptyset \wedge j_{1,t}\neq\emptyset\ } \right\rvert}$ ] . this equality holds for the specific weight setting we defined in eq . [ eq : weight_setting ] . maximizing over all weight settings gives the sought after lower bound : @xmath357 moving on to the upper bound , we show by induction over @xmath106 that for any @xmath160 $ ] and @xmath358 $ ] , the rank of @xmath359 is no greater than @xmath173 , regardless of the chosen weight setting . for the base case @xmath360 we have : @xmath361 the @xmath362-by-@xmath363 matrix @xmath364 is given here as a sum of @xmath365 rank-1 terms , thus obviously its rank is no greater than @xmath366 . since by definition @xmath367 for all @xmath368 $ ] , we may write : @xmath369 @xmath370 is defined by the right hand side of this inequality , so our inductive hypotheses holds for @xmath360 . for @xmath371 : @xmath372 taking ranks : rank^l,_i_l , k , j_l , k & = & rank(_=1^r_l-1 a_^l , ^l-1,_i_l-1,4(k-1)+t , j_l-1,4(k-1)+t ) + & & _ = 1^r_l-1 rank ( ^l-1,_i_l-1,4(k-1)+t , j_l-1,4(k-1)+t ) + & = & _ = 1^r_l-1_t=1 ^ 4 rank^l-1,_i_l-1,4(k-1)+t , j_l-1,4(k-1)+t + & & _ = 1^r_l-1_t=1 ^ 4 c^l-1,4(k-1)+t + & = & r_l-1_t=1 ^ 4 c^l-1,4(k-1)+t where we used rank sub - additivity in the second line , the rank - multiplicative property of the kronecker product in the third line , and our inductive hypotheses for @xmath373 in the fourth line . since the number rows and columns in @xmath359 is @xmath374 and @xmath375 respectively , we may incorporate these terms into the inequality , obtaining : @xmath376 the right hand side here is equal to @xmath173 by definition , so our inductive hypotheses indeed holds for all @xmath106 . to establish the sought after upper bound on the rank of @xmath119 , we recall that the latter is given by : @xmath153 carry out a series of steps similar to before , while making use of our inductive hypotheses for @xmath72 : rank^y_i , j & = & rank(_=1^r_l-1 a_^l , y ^l-1,_i_l-1,t , j_l-1,t ) + & & _ = 1^r_l-1 rank ( ^l-1,_i_l-1,t , j_l-1,t ) + & = & _ = 1^r_l-1_t=1 ^ 4 rank^l-1,_i_l-1,t , j_l-1,t + & & _ = 1^r_l-1_t=1 ^ 4 c^l-1,t + & = & r_l-1_t=1 ^ 4 c^l-1,t since @xmath119 has @xmath377 rows and @xmath378 columns , we may include these terms in the inequality , thus reaching the upper bound we set out to prove . our analysis of correlations modeled by convolutional networks is based on the concept of separation rank , conveyed in sec . [ sec : sep_rank ] . when the separation rank of a function w.r.t . a partition of its input is equal to @xmath4 , the function is separable , meaning it does not model any interaction between sides of the partition . we argued that the higher the separation rank , the farther the function is from this situation , _ i.e. _ the stronger the correlation it induces between sides of the partition . in the current appendix we formalize this argument , by relating separation rank to the @xmath0 distance from the set of separable functions . we begin by defining and characterizing a normalized ( scale invariant ) version of this distance ( app . [ app : sep_rank_l2:normalized_dist ] ) . it is then shown ( app . [ app : sep_rank_l2:ub_sep_rank ] ) that separation rank provides an upper bound on the normalized distance . finally , a lower bound that applies to deep convolutional arithmetic circuits is derived ( app . [ app : sep_rank_l2:lb_cac ] ) , based on the lower bound for their separation ranks established in sec . [ sec : analysis : deep ] . together , these steps imply that our entire analysis , facilitated by upper and lower bounds on separation ranks of convolutional arithmetic circuits , can be interpreted as based on upper and lower bounds on ( normalized ) @xmath0 distances from separable functions . in the text hereafter , we assume familiarity of the reader with the contents of sec . [ sec : prelim ] , [ sec : cac ] , [ sec : sep_rank ] , [ sec : analysis ] and the proofs given in app . [ app : proofs ] . we also rely on basic knowledge in the topic of @xmath0 spaces ( see discussion in app . [ app : proofs : sep_mat_ranks_equal ] for minimal background required in order to follow our arguments ) , as well as several results concerning singular values of matrices . in line with sec . [ sec : analysis ] , an assumption throughout this appendix is that all functions in question are measurable and square - integrable ( _ i.e. _ belong to @xmath0 over the respective euclidean space ) , and in app . [ app : sep_rank_l2:lb_cac ] , we also make use of the fact that representation functions ( @xmath75 ) of a convolutional arithmetic circuit can be regarded as linearly independent ( see sec . [ sec : analysis : sep2mat ] ) . finally , for convenience , we now fix @xmath23 an arbitrary partition of @xmath24 $ ] . specifically , @xmath25 and @xmath26 are disjoint subsets of @xmath24 $ ] whose union gives @xmath24 $ ] , denoted by @xmath27 with @xmath28 , and @xmath29 with @xmath30 . for a function @xmath128 ( which is not identically zero ) , the _ normalized @xmath0 distance from the set of separable functions w.r.t . @xmath23 _ , is defined as follows : d(h;i , j):=_h(x_1, ,x_n)-g(x_i_1, ,x_i_)g(x_j_1, ,x_j _ ) [ eq : normalized_dist ] where @xmath379 refers to the norm of @xmath0 space , _ e.g. _ @xmath380 . in words , @xmath137 is defined as the minimal @xmath0 distance between @xmath127 and a function that is separable w.r.t . @xmath23 , divided by the norm of @xmath127 . the normalization ( division by @xmath381 ) admits scale invariance to @xmath137 , and is of critical importance without it , rescaling @xmath127 would accordingly rescale the distance measure , rendering the latter uninformative in terms of deviation from separability . it is worthwhile noting the resemblance between @xmath137 and the concept of mutual information ( see @xcite for a comprehensive introduction ) . both measures quantify the interaction that a normalized function is the minimal @xmath0 distance between @xmath382 and a function separable w.r.t . accordingly , we may view @xmath137 as operating on normalized functions . ] induces between input variables , by measuring distance from separable functions . the difference between the measures is threefold . first , mutual information considers probability density functions ( non - negative and in @xmath383 ) , while @xmath137 applies to functions in @xmath0 . second , the notion of distance in mutual information is quantified through the kullback - leibler divergence , whereas in @xmath137 it is simply the @xmath0 metric . third , while mutual information evaluates the distance from a specific separable function product of marginal distributions , @xmath137 evaluates the minimal distance across all separable functions . we now turn to establish a spectral characterization of @xmath137 , which will be used in app . [ app : sep_rank_l2:ub_sep_rank ] and [ app : sep_rank_l2:lb_cac ] for deriving upper and lower bounds ( respectively ) . assume we have the following expression for @xmath127 : h(x_1, ,x_n)=_=1^m_=1^ma_,_(x_i_1, ,x_i_)_(x_j_1, ,x_j _ ) [ eq : orth_sep_decomp ] where @xmath384 and @xmath385 are positive integers , @xmath316 is an @xmath384-by-@xmath385 real matrix , and @xmath386 are orthonormal sets of functions in @xmath387 respectively . we refer to such expression as an _ orthonormal separable decomposition _ of @xmath127 , with @xmath316 being its _ coefficient matrix_. we will show that for any orthonormal separable decomposition , @xmath137 is given by the following formula : d(h;i , j)= [ eq : normalized_dist_formula ] where @xmath388 are the singular values of the coefficient matrix @xmath316 . this implies that if the largest singular value of @xmath316 accounts for a significant portion of the spectral energy , the normalized @xmath0 distance of @xmath127 from separable functions is small . on the other hand , if all but a fraction of the spectral energy is attributed to trailing singular values , @xmath127 is far from being separable ( @xmath137 is close to @xmath4 ) . as a first step in deriving eq . [ eq : normalized_dist_formula ] , we show that @xmath389 : h ^2 & & + & & ( _ = 1^m_=1^ma_,_(x_i_1, ,x_i_)_(x_j_1, ,x_j_))^2dx_1dx_n + & & _ , |=1^m_,|=1^ma_,a_|,|_(x_i_1, ,x_i_)_(x_j_1, ,x_j _ ) + & & _ |(x_i_1, ,x_i_)_|(x_j_1, ,x_j_)dx_1dx_n + & & _ , |=1^m_,|=1^ma_,a_|,|_(x_i_1, ,x_i_)_(x_j_1, ,x_j _ ) + & & _ |(x_i_1, ,x_i_)_|(x_j_1, ,x_j_)dx_1dx_n + & & _ , |=1^m_,|=1^ma_,a_|,|_(x_i_1, ,x_i_)_|(x_i_1, ,x_i_)dx_i_1dx_i _ + & & _(x_j_1, ,x_j_)_|(x_j_1, ,x_j_)dx_j_1dx_j _ + & & _ , |=1^m_,|=1^ma_,a_|,| \ { ll 1 & , =| + 0 & , } \ { ll 1 & , =| + 0 & , } + & & _ = 1^m_=1^ma_,^2 + & & _ 1 ^ 2(a)++_\{m , m}^2(a ) [ eq : orth_sep_decomp_norm ] equality @xmath390 here originates from the definition of @xmath0 norm . @xmath391 is obtained by plugging in the expression in eq . [ eq : orth_sep_decomp ] . @xmath392 is merely an arithmetic manipulation . @xmath393 follows from the linearity of integration . @xmath394 makes use of fubini s theorem ( see @xcite ) . @xmath395 results from the orthonormality of @xmath396 and @xmath397 . @xmath398 is a trivial computation . finally , @xmath399 is an outcome of the fact that the squared frobenius norm of a matrix , _ i.e. _ the sum of squares over its entries , is equal to the sum of squares over its singular values ( see @xcite for proof ) . let @xmath400 . by fact [ fact : orth_decomp ] in app . [ app : proofs : sep_mat_ranks_equal ] , there exist scalars @xmath401 , and a function @xmath402 orthogonal to @xmath403 , such that @xmath404 . similarly , for any @xmath405 there exist @xmath406 and @xmath407 such that @xmath408 . fact [ fact : prod ] in app . [ app : proofs : sep_mat_ranks_equal ] indicates that the function given by @xmath409 belongs to @xmath303 . we may express it as follows : g(x_i_1, ,x_i_)g(x_j_1, ,x_j _ ) & = & _ = 1^m_=1^m___(x_i_1, ,x_i_)_(x_j_1, ,x_j _ ) + & & + ( _ = 1^m__(x_i_1, ,x_i_))(x_j_1, ,x_j _ ) + & & + ( x_i_1, ,x_i_)(_=1^m__(x_j_1, ,x_j _ ) ) + & & + ( x_i_1, ,x_i_)(x_j_1, ,x_j _ ) according to fact [ fact : prod_orth ] in app . [ app : proofs : sep_mat_ranks_equal ] , the second , third and fourth terms on the right hand side of the above are orthogonal to @xmath410,\mu'\in[m']}$ ] . denote their summation by @xmath411 , and subtract the overall function from @xmath127 ( given by eq . [ eq : orth_sep_decomp ] ) : & & h(x_1, ,x_n)-g(x_i_1, ,x_i_)g(x_j_1, ,x_j _ ) + & & = _ = 1^m_=1^ma_,_(x_i_1, ,x_i_)_(x_j_1, ,x_j _ ) + & & -_=1^m_=1^m___(x_i_1, ,x_i_)_(x_j_1, ,x_j_)-e(x_1, ,x_n ) + & & = _ = 1^m_=1^m(a_,-__)_(x_i_1, ,x_i_)_(x_j_1, ,x_j_)-e(x_1, ,x_n ) since the two terms in the latter expression are orthogonal to one another , we have : h(x_1, ,x_n)-g(x_i_1, ,x_i_)g(x_j_1, ,x_j _ ) ^2 + = _ = 1^m_=1^m(a_,-__)_(x_i_1, ,x_i_)_(x_j_1, ,x_j _ ) ^2+(x_1, ,x_n ) ^2 applying a sequence of steps as in eq . [ eq : orth_sep_decomp_norm ] to the first term in the second line of the above , we obtain : @xmath412 @xmath413 if @xmath267 and @xmath414 are the zero functions , implying that : @xmath415 with equality holding if @xmath416 and @xmath417 . now , @xmath418 is the squared frobenius distance between the matrix @xmath316 and the rank-@xmath4 matrix @xmath419 , where @xmath420 and @xmath421 are column vectors holding @xmath422 and @xmath423 respectively . this squared distance is greater than or equal to the sum of squares over the second to last singular values of @xmath316 , and moreover , the inequality holds with equality for proper choices of @xmath420 and @xmath421 ( @xcite ) . from this we conclude that : @xmath424 with equality holding if @xmath425 and @xmath426 are set to @xmath427 and @xmath428 ( respectively ) for proper choices of @xmath422 and @xmath423 . we thus have the infimum over all possible @xmath429 : _ h(x_1, ,x_n)-g(x_i_1, ,x_i_)g(x_j_1, ,x_j _ ) ^2=_2 ^ 2(a)++_\{m , m}^2(a ) [ eq : unnormalized_dist_sq_formula ] recall that we would like to derive the formula in eq . [ eq : normalized_dist_formula ] for @xmath137 , assuming @xmath127 is given by the orthonormal separable decomposition in eq . [ eq : orth_sep_decomp ] . taking square root of the equalities established in eq . [ eq : orth_sep_decomp_norm ] and [ eq : unnormalized_dist_sq_formula ] , and plugging them into the definition of @xmath137 ( eq . [ eq : normalized_dist ] ) , we obtain the sought after result . we now relate @xmath137 the normalized @xmath0 distance of @xmath128 from the set of separable functions w.r.t . @xmath23 ( eq . [ eq : normalized_dist ] ) , to @xmath136 the separation rank of @xmath127 w.r.t . @xmath23 ( eq . [ eq : sep_rank ] ) . specifically , we make use of the formula in eq . [ eq : normalized_dist_formula ] to derive an upper bound on @xmath137 in terms of @xmath136 . assuming @xmath127 has finite separation rank ( otherwise the bound we derive is trivial ) , we may express it as : h(x_1, ,x_n)=_=1^rg_(x_i_1, ,x_i_)g_(x_j_1, ,x_j _ ) [ eq : sep_decomp ] where @xmath234 is some positive integer ( necessarily greater than or equal to @xmath136 ) , and @xmath288 , @xmath289 . let @xmath430 and @xmath431 be two sets of orthonormal functions spanning @xmath432 and @xmath433 respectively . by definition , for every @xmath434 there exist @xmath435 and @xmath436 such that @xmath437 and @xmath438 . plugging this into eq . [ eq : sep_decomp ] , we obtain : @xmath439 this is an orthonormal separable decomposition of @xmath127 ( eq . [ eq : orth_sep_decomp ] ) , with coefficient matrix @xmath440 , where @xmath441^{\top}$ ] and @xmath442^{\top}$ ] for every @xmath434 . obviously the rank of @xmath316 is no greater than @xmath234 , implying : @xmath443 where as in app . [ app : sep_rank_l2:normalized_dist ] , @xmath388 stand for the singular values of @xmath316 . introducing this inequality into eq . [ eq : normalized_dist_formula ] gives : @xmath444 the latter holds for any @xmath445 that admits eq . [ eq : sep_decomp ] , so in particular we may take it to be minimal , _ i.e. _ to be equal to @xmath136 ( @xmath127 is identically zero ) . ] , bringing forth the sought after upper bound : d(h;i , j ) [ eq : normalized_dist_ub ] by eq . [ eq : normalized_dist_ub ] , low separation rank implies proximity ( in normalized @xmath0 sense ) to a separable function . we may use the inequality to translate the upper bounds on separation ranks established for deep and shallow convolutional arithmetic circuits ( sec . [ sec : analysis : deep ] and [ sec : analysis : shallow ] respectively ) , into upper bounds on normalized @xmath0 distances from separable functions . to completely frame our analysis in terms of the latter measure , a translation of the lower bound on separation ranks of deep convolutional arithmetic circuits ( sec . [ sec : analysis : deep ] ) is also required . eq . [ eq : normalized_dist_ub ] does not facilitate such translation , and in fact , it is easy to construct functions @xmath127 whose separation ranks are high yet are very close ( in normalized @xmath0 sense ) to separable functions . is given by an orthonormal separable decomposition ( eq . [ eq : orth_sep_decomp ] ) , with coefficient matrix @xmath316 that has high rank but whose spectral energy is highly concentrated on one singular value . ] however , as we show in app . [ app : sep_rank_l2:lb_cac ] below , the specific lower bound of interest can indeed be translated , and our analysis may entirely be framed in terms of normalized @xmath0 distance from separable functions . let @xmath287 be a function realized by a deep convolutional arithmetic circuit ( fig . [ fig : nets_patterns](a ) with size-@xmath1 pooling windows and @xmath103 hidden layers ) , _ i.e. _ @xmath101 is given by eq . [ eq : h_y ] , where @xmath286 are linearly independent representation functions , and @xmath102 is a coefficient tensor of order @xmath6 and dimension @xmath62 in each mode , determined by the linear weights of the network ( @xmath446 ) through the hierarchical decomposition in eq . [ eq : hr_decomp ] . rearrange eq . [ eq : h_y ] by grouping indexes @xmath447 in accordance with the partition @xmath23 : h_y(x_1, ,x_n)= _ d_i_1 d_i_=1^m _ d_j_1 d_j_=1^m a_d_1 d_n^y(_t=1^ f__d_i_t(x_i_t ) ) ( _ t=1^ f__d_j_t(x_j_t ) ) [ eq : h_y_i_j ] let @xmath448 , and define the following mapping : @xmath449^{{\left\lverti \right\rvert}}\to[m]\quad,\quad \mu(d_{i_1},\ldots , d_{i_{{\left\lverti \right\rvert}}})=1+\sum\nolimits_{t=1}^{{\left\lverti \right\rvert}}(d_{i_t}-1){\cdot}m^{{\left\lverti \right\rvert}-t}\ ] ] @xmath450 is a one - to - one correspondence between the index sets @xmath451^{{\left\lverti \right\rvert}}$ ] and @xmath452 $ ] . we slightly abuse notation , and denote by @xmath453 the tuple in @xmath451^{{\left\lverti \right\rvert}}$ ] that maps to @xmath454 $ ] . additionally , we denote the function @xmath455 , which according to fact [ fact : prod ] in app . [ app : proofs : sep_mat_ranks_equal ] belongs to @xmath456 , by @xmath457 . in the exact same manner , we let @xmath458 , and define the bijective mapping : @xmath459^{{\left\lvertj \right\rvert}}\to[m']\quad,\quad \mu'(d_{j_1},\ldots , d_{j_{{\left\lvertj \right\rvert}}})=1+\sum\nolimits_{t=1}^{{\left\lvertj \right\rvert}}(d_{j_t}-1){\cdot}m^{{\left\lvertj \right\rvert}-t}\ ] ] as before , @xmath460 stands for the tuple in @xmath451^{{\left\lvertj \right\rvert}}$ ] that maps to @xmath461 $ ] , and the function @xmath462 is denoted by @xmath463 . now , recall the definition of matricization given in sec . [ sec : prelim ] , and consider @xmath119 the matricization of the coefficient tensor @xmath102 w.r.t . this is a matrix of size @xmath384-by-@xmath385 , holding @xmath34 in row index @xmath464 and column index @xmath465 . rewriting eq . [ eq : h_y_i_j ] with the indexes @xmath450 and @xmath466 instead of @xmath467 and @xmath468 , we obtain : h_y(x_1, ,x_n)= _ = 1^m _ =1^m ( ^y_i , j)_,_(x_i_1, ,x_i _ ) _(x_j_1, ,x_j _ ) [ eq : h_y_i_j_mat ] this equation has the form of eq . [ eq : orth_sep_decomp ] . however , for it to qualify as an orthonormal separable decomposition , the sets of functions @xmath430 and @xmath431 must be orthonormal . if the latter holds eq . [ eq : normalized_dist_formula ] may be applied , giving an expression for @xmath469 the normalized @xmath0 distance of @xmath101 from the set of separable functions w.r.t . @xmath23 , in terms of the singular values of @xmath119 . we now direct our attention to the special case where @xmath286 the network s representation functions , are known to be orthonormal . the general setting , in which only linear independence is known , will be treated thereafter . orthonormality of representation functions implies that @xmath470 are orthonormal as well : , _ | & & _ ( x_i_1, ,x_i_)_|(x_i_1, ,x_i_)dx_i_1dx_i _ + & & _ t=1^f__d_i_t()(x_i_t)_t=1^f__d_i_t(|)(x_i_t)dx_i_1dx_i _ + & & _ t=1^f__d_i_t()(x_i_t)f__d_i_t(|)(x_i_t)dx_i_t + & & _ t=1^,f__d_i_t(| ) + & & _ t=1^ \ { ll 1 & , d_i_t()=d_i_t(| ) + 0 & , } + & & \ { ll 1 & , d_i_t()=d_i_t(| ) + 0 & , . + & & \ { ll 1 & , =| + 0 & , . @xmath390 and @xmath393 here follow from the definition of inner product in @xmath0 space , @xmath391 replaces @xmath471 and @xmath472 by their definitions , @xmath392 makes use of fubini s theorem ( see @xcite ) , @xmath394 relies on the ( temporary ) assumption that representation functions are orthonormal , @xmath395 is a trivial step , and @xmath398 owes to the fact that @xmath473 is an injective mapping . a similar sequence of steps ( applied to @xmath474 ) shows that in addition to @xmath475 , the functions @xmath476 will also be orthonormal if @xmath477 are . we conclude that if representation functions are orthonormal , eq . [ eq : h_y_i_j_mat ] indeed provides an orthonormal separable decomposition of @xmath101 , and the formula in eq . [ eq : normalized_dist_formula ] may be applied : d(h_y;i , j)= [ eq : normalized_dist_formula_cac ] where @xmath478 are the singular values of the coefficient tensor matricization @xmath119 . in sec . [ sec : analysis : deep ] we showed that the maximal separation rank realizable by a deep network is greater than or equal to @xmath163 , where @xmath479 are the number of channels in the representation and first hidden layers ( respectively ) , and @xmath170 stands for the number of index quadruplets ( sets of the form @xmath480 for some @xmath481 $ ] ) that are split by the partition @xmath23 . to prove this lower bound , we presented in app . [ app : proofs : deep_mat_rank_bounds ] a specific setting for the linear weights of the network ( @xmath446 ) under which @xmath482 . careful examination of the proof shows that with this particular weight setting , not only is the rank of @xmath119 equal to @xmath163 , but also , all of its non - zero singular values are equal to one another.$ ] , @xmath350 has one of two forms : it is either a non - zero ( row / column ) vector , or it is a matrix holding @xmath4 in several entries and @xmath86 in all the rest , where any two entries holding @xmath4 reside in different rows and different columns . the first of the two forms admits a single non - zero singular value . the second brings forth several singular values equal to @xmath4 , possibly accompanied by null singular values . in both cases , all non - zero singular values of @xmath350 are equal to one another . now , since @xmath483 , and since the kronecker product multiplies singular values ( see @xcite ) , we have that all non - zero singular values of @xmath119 are equal , as required . ] this implies that @xmath484 , and since we currently assume that @xmath477 are orthonormal , eq . [ eq : normalized_dist_formula_cac ] applies and we obtain @xmath485 . maximizing over all possible weight settings , we arrive at the following lower bound for the normalized @xmath0 distance from separable functions brought forth by a deep convolutional arithmetic circuit : _ \{a^l,}_l , , a^l , yd(h_y|_\{a^l,}_l,,a^l , y ; i , j ) [ eq : normalized_dist_max_lb ] turning to the general case , we omit the assumption that representation functions @xmath286 are orthonormal , and merely rely on their linear independence . the latter implies that the dimension of @xmath486 is @xmath62 , thus there exist orthonormal functions @xmath487 that span it . let @xmath488 be a transition matrix between the bases the matrix defined by @xmath489}$ ] . suppose now that we replace the original representation functions @xmath477 by the orthonormal ones @xmath490 . using the latter , the lower bound in eq . [ eq : normalized_dist_max_lb ] applies , and there exists a setting for the linear weights of the network @xmath446 , such that @xmath491 . recalling the structure of convolutional arithmetic circuits ( fig . [ fig : nets_patterns](a ) ) , one readily sees that if we return to the original representation functions @xmath477 , while multiplying conv weights in hidden layer @xmath86 by @xmath492 ( _ i.e. _ mapping @xmath493 ) , the overall function @xmath101 remains unchanged , and in particular @xmath491 still holds . we conclude that the lower bound in eq . [ eq : normalized_dist_max_lb ] applies , even if representation functions are not orthonormal . to summarize , we translated the lower bound from sec . [ sec : analysis : deep ] on the maximal separation rank realized by a deep convolutional arithmetic circuit , into a lower bound on the maximal normalized @xmath0 distance from separable functions ( eq . [ eq : normalized_dist_max_lb ] ) . this , along with the translation of upper bounds facilitated in app . [ app : sep_rank_l2:ub_sep_rank ] , implies that the analysis carried out in the paper , which studies correlations modeled by convolutional networks through the notion of separation rank , may equivalently be framed in terms of normalized @xmath0 distance from separable functions . we note however that there is one particular aspect in our original analysis that does not carry through the translation . namely , in sec . [ sec : analysis : sep2mat ] it was shown that separation ranks realized by convolutional arithmetic circuits are maximal almost always , _ i.e. _ for all linear weight settings but a set of ( lebesgue ) measure zero . put differently , for a given partition @xmath23 , the maximal separation rank brought forth by a network characterizes almost all functions realized by it . an equivalent statement does not hold with the continuous measure of normalized @xmath0 distance from separable functions . the behavior of this measure across the hypotheses space of a network is non - trivial , and forms a subject for future research . in this appendix we provide implementation details omitted from the description of our experiments in sec . [ sec : exp ] . our implementation , available online at https://github.com/huji-deep/inductive-pooling , is based on the simnets branch ( ) of caffe toolbox ( @xcite ) . the latter realizes convolutional arithmetic circuits in log - space for numerical stability . when training convolutional arithmetic circuits , we followed the hyper - parameter choices made by . in particular , our objective function was the cross - entropy loss with no @xmath0 regularization ( _ i.e. _ with weight decay set to @xmath86 ) , optimized using adam ( @xcite ) with step - size @xmath494 and moment decay rates @xmath495 . @xmath496 iterations with batch size @xmath212 ( @xmath497 epochs ) were run , with the step - size @xmath110 decreasing by a factor of @xmath498 after @xmath499 iterations ( @xmath500 epochs ) . we did not use dropout ( @xcite ) , as the limiting factor in terms of accuracies was the difficulty of fitting training data ( as opposed to overfitting ) see fig . [ fig : exp_results_cac_pool4 ] . for training the conventional convolutional rectifier networks , we merely switched the hyper - parameters of adam to the recommended settings specified in @xcite ( @xmath501 ) , and set weight decay to the standard value of @xmath502 . the synthetic dataset used in our experiments ( sec . [ sec : exp ] ) consists of binary images displaying different shapes ( blobs ) . one of the tasks facilitated by this dataset is the detection of morphologically closed blobs , _ i.e. _ of images that are relatively similar to their morphological closure . the procedure we followed for computing the morphological closure of a binary image is : 1 . pad the given image with background ( @xmath86 value ) pixels 2 . morphological dilation : simultaneously turn on ( set to @xmath4 ) all pixels that have a ( left , right , top or bottom ) neighbor originally active ( holding @xmath4 ) 3 . morphological erosion : simultaneously turn off ( set to @xmath86 ) all pixels that have a ( left , right , top or bottom ) neighbor currently inactive ( holding @xmath86 ) 4 . remove pixels introduced in padding it is not difficult to see that any pixel active in the original image is necessarily active in its closure . moreover , pixels that are originally inactive yet are surrounded by active ones will also be turned on in the closure , hence the effect of `` gap filling '' . finally , we note that the particular sequence of steps described above represents the most basic form of morphological closure . the interested reader is referred to @xcite for a much more comprehensive introduction .

it has long been recognized that when an atom is situated in a material medium , the local electromagnetic field acting on it differs from the macroscopic one due to the gaps between the atom and the surrounding medium atoms , which are ignored on a coarse - grained macroscopic scale @xcite . accounting for the difference between the two fields hence requires a correction the local - field correction . classical calculations of local - field effects , which typically have their origin in near dipole - dipole interactions , can be found in textbooks ( see , e.g. , ref . @xcite ) . in quantum theory , investigations of local - field effects are often related to the problem of the spontaneous decay of an excited guest atom embedded in a ( dielectric ) host . local - field effects in spontaneous decay have been studied , e.g. , for crystals @xcite and disordered dielectrics @xcite on the basis on microscopic models for coupled atomic dipoles . in macroscopic descriptions , local - field effects are frequently taken into account by regarding the guest atom as being enclosed in a virtual @xcite or real ( spherical ) cavity @xcite surrounded by the medium , with the cavity size being small compared to the relevant transition wavelength . the cavity in the former model is virtual in the sense that it does not perturb the macroscopic field . it is filled by the atoms comprising the medium , which produce no net effect at the central position in important special cases such as cubic or random structures . in the latter model , the field is modified by the presence of the cavity , which is an empty region containing only the guest atom . microscopic models often tend to agree with the virtual - cavity results @xcite , while many recent experiments on spontaneous emission in dielectrics support the real - cavity model @xcite . it has been presumed that while the virtual - cavity model applies to interstitial atoms , the real - cavity model is specific to substitutional atoms , and that the case of substitutional atoms occurs prevalently for impurity atoms in disordered dielectrics @xcite . since in the approaches to the local - field effects , the host medium has typically been assumed to be a bulk medium that extends homogeneously to infinity , the question of the effect of the size and shape of the host medium on the local field has arisen . in a macroscopic approach , the spontaneous - decay rate of an excited atom in some free - space region can be given in terms of the imaginary part of the green tensor of the macroscopic maxwell equations , which characterizes the ( macroscopic ) environment of the atom . this relation in principle allows for including local - field corrections for atoms embedded in arbitrary material configurations by assuming a real ( spherical ) cavity surrounding the atom and calculating the corresponding green tensor . using the real - cavity model , toma @xcite studied the local - field correction to the spontaneous - decay rate of an excited atom , which is located at the center of a dispersing and absorbing dielectric sphere . reformulating the result by representing it in a form , which does not explicitly refer to the highly symmetric system considered , he made the conjecture that it may also remain valid beyond the specific example and hence also apply to other locations of the atom and other shapes of the host body . based on a numerical computation of the respective green tensors , rahmani and bryant @xcite considered the case of an atom at an arbitrary location within a dielectric sphere or a dielectric cube . comparing their results for the dielectric sphere including the local - field correction with earlier results disregarding the local - field correction @xcite , they suggested a rate formula , which in the case of weakly absorbing material corresponds to toma s formula . however , since their approach relies heavily on numerical calculations , it can not produce explicit expressions for the quantities that are related to the local - field correction . the exact analytical evaluations of the green tensors of realistic systems which have finite sizes and include a cavity can be very cumbersome . in this paper , we present an attempt to overcome this difficulty by writing the green tensor as a born series in terms of the susceptibility , where in many situations one can restrict oneself to several leading - order terms . in particular , we show that to linear order the spontaneous - decay rate of a guest atom in a dielectric host body can be separated into a term representing the local - field correction to the decay rate in free space and a term related to the scattering green tensor of the body without the atom a result which exactly corresponds to toma s conjecture mentioned above . furthermore , we show that for atoms which are situated at the interior of a macroscopic body toma s conjecture remains valid beyond the linear order . we illustrate the theory by discussing in detail the case of an atom embedded in a spherical dielectric body . the paper is organized as follows . the basic equations for the spontaneous - decay rate and the born expansion of the green tensor determining the rate are given in sec . [ sec_be ] . they are used in sec . [ subsec_rc ] to study the problem of the local - field corrected decay rate within the frame of the real - cavity model , and a proof of toma s formula is given . the examples of an atom embedded in a bulk dielectric medium or in a dielectric sphere are examined in sec . [ sec_ex ] , followed by a summary ( sec . [ sum ] ) . consider an excited two - level electric - dipole emitter , henceforth referred to as an atom , which is positioned at @xmath0 and surrounded by dispersing and absorbing dielectric bodies . the spontaneous - decay rate can be given in the form of @xcite @xmath1 where @xmath2 and @xmath3 are the ( real ) dipole matrix element and ( shifted ) frequency of the relevant atomic transition , respectively , and @xmath4 @xmath5 @xmath6 . the green tensor of the bodies , @xmath7 , satisfies the equation @xmath8 ( @xmath9 , unit tensor ) together with the boundary condition @xmath10 where @xmath11 is the frequency- and space - dependent complex permittivity which satisfies the kramers kronig relations . note that satisfaction of the boundary condition ( [ e4b ] ) is ensured by assuming @xmath12 @xmath13 @xmath14 . equation ( [ e3 ] ) always applies when the atom is placed in some free - space region . separating the green tensor into bulk and scattering parts @xmath15 and @xmath16 , respectively , @xmath17 and taking into account that in the case where the bulk part refers to free space , the relation @xmath18 holds ( see , e.g. , ref . @xcite ) , we may rewrite eq . ( [ e3 ] ) as @xmath19 where @xmath20 is the spontaneous - decay rate in free space . if the atom is embedded in a body , application of eq . ( [ e3 ] ) requires special care in two respects . firstly , the coincidence limit of the bulk part of the green tensor diverges when the permittivity of the body is complex , as is the case in general . only if material absorption can be neglected so that the permittivity can be regarded as being real , @xmath21 @xmath22 @xmath23 , this limit exists , @xmath24 ( see , e.g. , ref . @xcite ) , leading to @xmath25 secondly , the green tensor of the macroscopic maxwell equations does not account for the fact that the local field felt by the atom is different from the macroscopic one in general . that is , even if absorption is neglected , the rate formula ( [ eq.10 ] ) is not complete because it lacks the local - field corrections . to calculate the ( scattering part of the ) green tensor for an arbitrary arrangement of dielectric bodies , it may be helpful to use an appropriate born expansion . decomposing the permittivity as @xmath26 and assuming that the solution @xmath27 to the equation @xmath28 is known [ where @xmath29 is defined as in eq . ( [ e5 ] ) with @xmath30 instead of @xmath31 , the green tensor can be written in the form of a born series , @xmath32\nonumber\\ & \quad\times \overline{\bm{g}}(\mathbf{r},\mathbf{s}_1,\omega)\!\cdot\ ! \overline{\bm{g}}(\mathbf{s}_1,\mathbf{s}_2,\omega)\!\cdots\ ! \overline{\bm{g}}(\mathbf{s}_k,\mathbf{r}',\omega),\end{aligned}\ ] ] as can be verified using the relationships @xmath33,\\ \label{e11 } \hat{h}\,\delta_k\bm{g}(\mathbf{r},\mathbf{r}',\omega ) = & \;\frac{\omega^2}{c^2}\chi(\mathbf{r},\omega ) \bigl[\delta_{k-1}{\bm{g}}(\mathbf{r},\mathbf{r}',\omega ) \nonumber\\ & \ ; -\delta_k\bm{g}(\mathbf{r},\mathbf{r}',\omega)\bigr ] \qquad\mbox{for } k>1.\end{aligned}\ ] ] the expansion ( [ e8 ] ) of the green tensor is valid for arbitrarily spatially varying @xmath34 and @xmath35 . obviously , it may be very useful when @xmath36 can be regarded as being a ( small ) perturbation to @xmath30 such that one makes only a small error by disregarding the higher - order terms . in particular , this is the case if the bodies are weakly polarizable , as we shall assume in the following . for weakly polarizable bodies , it is natural to regard the susceptibilities of the bodies as a small perturbation to the free - space permittivity @xmath34 @xmath5 @xmath37 @xmath38 @xmath39 , i.e. , @xmath40 @xmath41 @xmath42 . this means that we focus on frequencies that are sufficiently far from a resonance frequency of the dielectric material . note that the small ( positive ) imaginary part of @xmath30 ensures that @xmath43 fulfills the boundary condition according to eq . ( [ e4b ] ) so that the spatial integrals in eq . ( [ e9 ] ) converge . we have ( see , e.g. , ref . @xcite ) @xmath44 where @xmath45 @xmath46 @xmath5 @xmath47 @xmath48 @xmath49 , @xmath50 @xmath5 @xmath51 , @xmath52 @xmath5 @xmath53 , @xmath54 @xmath5 @xmath55 . separating the green tensor into bulk and scattering parts in accordance with eq . ( [ e14.2 ] ) , assuming the atom to be located in a free - space region such that @xmath56 [ cf . ( [ eq.6 ] ) ] , and applying eq . ( [ e8 ] ) , we may represent the scattering part of the green tensor in the rate formula ( [ e14.4 ] ) as @xmath57 for @xmath58 and @xmath59 belonging to a free - space region , substitution of eq . ( [ e13 ] ) into eq . ( [ e9 ] ) yields the first- and second - order terms in the born expansion , @xmath60 and @xmath61 , respectively , as follows : @xmath62 e^{i\sqrt{1+i\eta}\,(q+q')}\end{aligned}\ ] ] [ @xmath46 @xmath5 @xmath47 @xmath48 @xmath63 , @xmath54 @xmath5 @xmath55 , @xmath64 @xmath5 @xmath65 , @xmath66 @xmath5 @xmath67 ; @xmath68 @xmath5 @xmath63 @xmath48 @xmath49 , @xmath69 @xmath5 @xmath70 , @xmath71 @xmath5 @xmath72 , @xmath73 @xmath5 @xmath74 , @xmath75 e^{i\sqrt{1+i\eta}\,(q+q')}\nonumber\\ & + \frac{k^7}{64\pi^3}\!\int\!\mathrm{d}^3s_1\ , \chi(\mathbf{s}_1,\omega)\ ! \int\!\mathrm{d}^3s_2\,\chi(\mathbf{s}_2,\omega ) e^{i\sqrt{1+i\eta}\,(q_1+q_{12}+q_2 ) } \nonumber\\ & \quad\times\bigl[a_1a_{12}a_2\bm{i } -a_1a_{12}b_2\tilde{\mathbf{u}}_2\tilde{\mathbf{u}}_2 -a_1b_{12}a_2\tilde{\mathbf{u}}_{12}\tilde{\mathbf{u}}_{12 } \nonumber\\ & \qquad -b_1a_{12}a_2\tilde{\mathbf{u}}_1\tilde{\bm{u}}_1 + a_1b_{12}b_2(\tilde{\mathbf{u}}_{12 } \!\cdot\!\tilde{\mathbf{u}}_2 ) \tilde{\mathbf{u}}_{12}\tilde{\mathbf{u}}_2\nonumber\\ & \qquad + b_1a_{12}b_2(\tilde{\mathbf{u}}_1 \!\cdot\!\tilde{\mathbf{u}}_2 ) \tilde{\mathbf{u}}_1\tilde{\mathbf{u}}_2 + b_1b_{12}a_2(\tilde{\mathbf{u}}_1 \!\cdot\!\tilde{\mathbf{u}}_{12 } ) \tilde{\mathbf{u}}_1\tilde{\mathbf{u}}_{12 } \nonumber\\ & \qquad -b_1b_{12}b_2(\tilde{\mathbf{u}}_1 \!\cdot\!\tilde{\mathbf{u}}_{12 } ) ( \tilde{\mathbf{u}}_{12 } \!\cdot\!\tilde{\mathbf{u}}_2 ) \tilde{\mathbf{u}}_1\tilde{\mathbf{u}}_2\bigr]\end{aligned}\ ] ] [ @xmath76 @xmath5 @xmath47 @xmath48 @xmath77 , @xmath78 @xmath5 @xmath77 @xmath48 @xmath79 , @xmath80 @xmath5 @xmath79 @xmath48 @xmath49 ; @xmath81 @xmath5 @xmath82 , @xmath83 @xmath5 @xmath84 , @xmath85 @xmath5 @xmath86 for @xmath87 @xmath88 @xmath89 . consider an excited two - level atom embedded in an arbitrary dispersing and absorbing dielectric body characterized by @xmath11 . in order to find the spontaneous - decay rate including local - field corrections , we employ the real - cavity model , that is to say , we assume that the atom is located at the center of an empty - space region of the form of a spherical cavity of radius @xmath90 , as sketched in fig . [ schematic](a ) , where we have denoted the cavity volume by @xmath91 , the volume of the host body without the cavity by @xmath92 ( the overall volume of the host body being @xmath93 @xmath5 @xmath94 ) , and all the remaining space by @xmath95 . hence , the permittivity of the system changes to @xmath96 into ( b ) @xmath97 and ( c ) @xmath98 ; the dot indicates the position of the guest atom . ] the cavity radius @xmath90 is a model parameter representing an average distance from the atom to the nearest neighboring atoms constituting the host body ; it has to be determined from other ( preferably microscopic ) calculations or experiments . note that the real - cavity model is applicable provided that the unperturbed host body is homogeneous ( and isotropic ) in the region where the guest atom is implanted , @xmath99 restricting our attention to the first - order term in the born expansion ( [ e14.6 ] ) , the ( scattering ) green tensor corresponding to @xmath100 can be calculated from eq . ( [ e14.1 ] ) , where after some manipulations one obtains @xmath101\nonumber\\ & \qquad = \bm{g}_\mathrm{c}^{(1 ) } ( \mathbf{r}_\mathrm{a},\mathbf{r}_\mathrm{a},\omega_\mathrm{a } ) + \bm{g}_\mathrm{b}^{(1 ) } ( \mathbf{r}_\mathrm{a},\mathbf{r}_\mathrm{a},\omega_\mathrm{a})\end{aligned}\ ] ] [ @xmath46 @xmath5 @xmath102 @xmath48 @xmath63 , @xmath54 @xmath5 @xmath103 , where , again to linear order in the susceptibility , @xmath104\end{gathered}\ ] ] is the scattering green tensor of a spherical cavity embedded in a bulk medium of susceptibility @xmath105 @xmath106 @xmath107 [ cf . [ schematic](b ) ] , and @xmath108 \nonumber\\ & -\frac{k_\mathrm{a}^4\chi(\omega_\mathrm{a})}{16\pi^2 } \int_{b\cup v } \mathrm{d}^3s\ , e^{(2i-\eta)q}\bigl[a^2\bm{i}+(b^2 - 2ab ) \tilde{\mathbf{u}}\tilde{\mathbf{u}}\bigr]\end{aligned}\ ] ] is the scattering part of the green tensor of the host body without the cavity [ cf . [ schematic](c ) ] . the result can be verified by applying eq . ( [ e14.2 ] ) [ together with eqs . ( [ e8 ] ) and ( [ e14.1 ] ) ] to the two systems mentioned and using eq . ( [ e15.2 ] ) . equations ( [ e15.3])([e15.5 ] ) show that @xmath16 in the rate formula ( [ e14.4 ] ) can be written as the sum of two terms , where the first term , @xmath97 , only depends on the cavity radius and the local permittivity of the host body at the position of the atom , whereas the second term , @xmath98 , is determined by the properties of the host body in the absence of the atom . hence , the term in the decay rate which is proportional to @xmath109 can be regarded as being the local - field correction to the uncorrected term proportional to @xmath110 . the fact that the local - field correction additively enters the rate formula is due to the linear expansion in @xmath36 . inspection of the second - order term in the born expansion , eq . ( [ e16 ] ) , indicates that in general , terms depending on both the cavity and the ( unperturbed ) host body will appear which may lead to a breakdown of the additivity . however , as shown in sec . [ subsec_con ] , there are situations where a generalization beyond the linear approximation is possible . equations ( [ e15.4 ] ) and ( [ e15.5 ] ) can be further evaluated by introducing a spherical coordinate system whose origin coincides with the position of the atom , @xmath111 where @xmath112 and @xmath113 , respectively , refer to the inner and outer boundary areas of the integration volumes sketched in figs . [ schematic](b ) and [ schematic](c ) . performing the radial integral in eq . ( [ e15.4 ] ) , we find , on recalling eq . ( [ e14 ] ) , @xmath114 ( @xmath115 @xmath5 @xmath116 ) with @xmath117 = -\frac{k_\mathrm{a}\chi(\omega_\mathrm{a})}{16\pi^2 } \int_0^{2\pi}\mathrm{d}\phi\int_0^\pi\mathrm{d}\theta\sin\theta \biggl\{e^{(2i-\eta)q } \nonumber\\ & \times\biggl[\biggl(\frac{1}{3q^3}-\frac{2i}{3q^2 } -\frac{5}{3q}+\frac{i}{2}\biggr)\bm{i } + \biggl(\frac{1}{q^3}-\frac{2i}{q^2}+\frac{3}{q } -\frac{i}{2}\biggr ) \tilde{\mathbf{s}}\tilde{\mathbf{s}}\biggr ] \nonumber\\ & \hspace{9ex } + 4i\,\mathrm{ei}[(2i\!-\!\eta)q]\biggl(\frac{1}{3}\bm{i } -\tilde{\mathbf{s}}\tilde{\mathbf{s}}\biggr)\biggr\}\end{aligned}\ ] ] [ @xmath118 @xmath5 @xmath119 ; @xmath120 @xmath5 @xmath121 ; @xmath122 , exponential integral ] . using the fact that @xmath115 is independent of @xmath123 and @xmath124 as well as the relation @xmath125 from eq . ( [ e19 ] ) we obtain @xmath126 in particular , when the radius of the cavity is much smaller than the atomic transition wavelength , @xmath127 then eq . ( [ e19.1 ] ) reduces to @xmath128\bm{i}.\ ] ] substitution of this result together with eq . ( [ e19b1 ] ) into eq . ( [ e15.3 ] ) reveals that to linear order in @xmath36 , @xmath129\bm{i } \\ + \bm{g}_\mathrm{b}^{(1 ) } ( \mathbf{r}_\mathrm{a},\mathbf{r}_\mathrm{a},\omega_\mathrm{a } ) + o(k_\mathrm{a}r_\mathrm{c}).\end{gathered}\ ] ] for a homogeneous host body which is star - shaped w.r.t . the position of the atom ( i.e. , every point on the outer boundary area can be connected to the atomic position by a straight line that lies entirely within the body ) , @xmath98 as given by eq . ( [ e15.5 ] ) can be evaluated in a similar manner . using again spherical coordinates and recalling eq . ( [ e14 ] ) , we may evaluate the radial integrals in eq . ( [ e15.5 ] ) to obtain @xmath130 -\bm{f}_0(0)\nonumber\\ & \quad -\bigl[\lim_{\eta\to 0_+}\lim_{q\to\infty } \bm{f}_\eta(q)-\bm{f}_0(0)\bigr ] = \lim_{\eta\to 0_+}\bm{f}_\eta[q_\mathrm{o}(\phi,\theta)]\end{aligned}\ ] ] [ @xmath131 @xmath5 @xmath132 @xmath5 @xmath133 , where henceforth @xmath113 denotes the outer boundaries of the host body . substituting eq . ( [ e18 ] ) into eq . ( [ e14.4 ] ) , we find that to linear order in @xmath36 the local - field corrected spontaneous - decay rate of an atom within a body can be given as follows : @xmath134 where @xmath135 is the free - space decay rate as given in eq . ( [ e14.5 ] ) , @xmath136 and @xmath137 equations ( [ e24])([e24.2 ] ) are valid for an atom embedded in a weakly polarizable , dispersing and absorbing body of arbitrary size and shape . note that only @xmath138 takes into account the effect of local - field correction , while @xmath139 , which is simply determined by the scattering part of the green tensor of the host body without the cavity , reflects the uncorrected influence of the size and shape of the body on the decay rate . in particular , the first two terms in eq . ( [ e24.1 ] ) obviously result from irreversible energy transfer from the atom to the surrounding matter . if these two terms are dominant over the last one , effectively no radiation is emitted . we conclude this section by making two remarks concerning the necessary conditions under which eq . ( [ e24 ] ) provides a good approximation to the spontaneous - decay rate . ( i ) the permittivity often appears in the green tensor as a common factor @xmath140 [ cf . ( [ e46 ] ) in sec . [ subsec_oc ] ] . to be on the conservative side , one should then take @xmath141 [ with @xmath142 @xmath5 @xmath143 rather than @xmath36 as the small parameter . ( ii ) equation ( [ e24.1 ] ) already contains terms of the order of @xmath144 . thus @xmath144 , not @xmath145 , should be much smaller than unity for the linear approximation to yield a good estimation of the decay rate . the cavity radius @xmath90 represents the average distance between the atom and the constituents of the body . for , say , @xmath146 @xmath147 @xmath148 , it is sufficient to require that , so that . toma @xcite has found that in the special case of an atom situated at the center of a homogenous , dielectric sphere , the real - cavity model ( fig . [ sphere ] in the case where ) leads to the relation @xmath149\biggr\}\bm{i}\nonumber\\ & \quad+\biggl(\frac{3\varepsilon}{2\varepsilon+1}\biggr)^2 \bm{g}_\mathrm{b}^{(1 ) } ( \mathbf{r}_\mathrm{a},\mathbf{r}_\mathrm{a},\omega_\mathrm{a } ) + o(k_\mathrm{a}r_\mathrm{c}),\end{aligned}\ ] ] @xmath150 @xmath5 @xmath151 , where @xmath152 is the scattering green tensor of the homogeneous dielectric sphere , whose geometry dependence is for @xmath58 @xmath5 @xmath49 @xmath5 @xmath102 at the sphere center entirely given by its dependence on the sphere radius @xmath153 . as already mentioned in sec . [ sec_intro ] , he made the conjecture that this relation might be more generally valid for ( i ) bodies of arbitrary sizes and shapes and ( ii ) arbitrary positions of the atom , provided that the atom is not on the body surface . let us study the validity of this conjecture in more detail . it can easily be seen that to linear order in @xmath36 , eq . ( [ e21 ] ) obviously reduces to eq . ( [ e18 ] ) , so the results of sec . [ linear ] show that in this order the conjecture is true even under the more general conditions of inhomogeneous host bodies , provided that the requirement ( [ e15.2 ] ) for the applicability of the real - cavity model is valid . moreover , we will demonstrate in the following that eq . ( [ e21 ] ) remains valid beyond the linear order in @xmath36 , provided that the respective host body can be regarded as being homogeneous in the vicinity of the atom , i.e. , @xmath154 with @xmath155 being some small positive number . we begin with the case of a homogeneous body , in which case satisfaction of the condition ( [ e15.2b ] ) simply ensures that the entire cavity lies inside the body . we first recall that @xmath156 determines the electric field reaching the point @xmath0 , in which it has originated , after being scattered at the surfaces of inhomogeneity [ cf . [ schematic](a ) ] . at the cavity surface the electric field can be separated into two parts , namely , one part that is ( multiply ) reflected at this surface and eventually returns to the point @xmath0 , and one part that is eventually transmitted to the exterior of the cavity . the contribution due to the reflected part is obviously given by @xmath157 [ cf . [ schematic](b ) ] , which reads @xcite @xmath158\biggr\}\bm{i}\nonumber\\ & \quad+o(k_\mathrm{a}r_\mathrm{c}).\end{aligned}\ ] ] in an infinitely extended body [ cf . fig . [ schematic](b ) ] , the transmitted part at a point @xmath58 outside the cavity is determined by the green tensor @xcite @xmath159 [ @xmath46 @xmath5 @xmath47 @xmath48 @xmath102 , @xmath54 @xmath5 @xmath160 , @xmath161 @xmath5 @xmath162 , and @xmath64 and @xmath66 according to eq . ( [ e14 ] ) ] . the coefficient @xmath163 ( which corresponds to @xmath164 in ref . @xcite ) is given by @xmath165 ' -[z_0j_1(z_0)]'h_1^{(1)}(z_0)}{j_1(z_0)[z_1h_1^{(1)}(z_1 ) ] ' -\varepsilon[z_0j_1(z_0)]'h_1^{(1)}(z_1)}\,,\ ] ] where @xmath166 @xmath5 @xmath116 , @xmath167 @xmath5 @xmath168 ( the primes denote derivatives w.r.t . @xmath166 and @xmath167 ) , and @xmath169 and @xmath170 are the spherical bessel and hankel functions , respectively . inspection of eq . ( [ e21e ] ) shows that @xmath171 substitution of this result into eq . ( [ e21d ] ) leads to @xmath172 where @xmath173 is the green tensor of the infinite body without the cavity [ according to eqs . ( [ e13 ] ) and ( [ e14 ] ) with @xmath174 . in other words , the electric field transmitted through the cavity surface to a point outside the cavity is equal to the field that would be transmitted to the same point in the absence of the cavity , multiplied by a global factor . by means of the general symmetry property ( see , e.g. , ref . @xcite ) @xmath175 we also have @xmath176 i.e. , the electric field transmitted through the cavity surface from an arbitrary point outside the cavity is also equal to the corresponding result in the absence of the cavity , multiplied by the same factor . for a finite body , the electric field will be ( multiply ) reflected from the body s outer surface , eventually giving rise to a field at @xmath0 . without the cavity , processes of this kind are taken into account by replacing the infinite - body green tensor @xmath177 with its finite - body counterpart @xmath178 . combining this observation with eqs . ( [ e21 g ] ) and ( [ e21i ] ) , we conclude , on recalling the linearity of maxwell s equations , that the electric field which is transmitted though the cavity surface , scattered at the outer body surface , and finally retransmitted into the cavity is given by @xmath179 so combining eqs . ( [ e21c ] ) and ( [ e21j ] ) , we arrive at eq . ( [ e21 ] ) . in this derivation , we have disregarded processes involving scattering of the field at the cavity surface from the outside . processes of this kind can indeed be neglected , because their contributions are of orders higher than @xmath180 . to see this , note that for a cavity in bulk material [ fig . [ schematic](b ) ] the field at a point @xmath58 outside the cavity originating in a point @xmath59 outside the cavity via scattering at the cavity surface , is characterized by @xcite @xmath181,\end{gathered}\ ] ] where @xmath182 ' -[z_0j_m(z_0)]'j_m(z_1)}{j_m(z_0)[z_1h_m^{(1)}(z_1 ) ] ' -[z_0j_m(z_0)]'h_m^{(1)}(z_1)}\,,\\ & b_m^n=-\frac{j_m(z_0)[z_1j_m(z_1 ) ] ' -\varepsilon[z_0j_m(z_0)]'j_m(z_1)}{j_m(z_0)[z_1h_m^{(1)}(z_1 ) ] ' -\varepsilon[z_0j_m(z_0)]'h_m^{(1)}(z_1)},\end{aligned}\ ] ] and the tensors @xmath183 and @xmath184 do not depend on @xmath90 . using the relations @xcite @xmath185,\\ \label{e21n } & h_m^{(1)}(z ) = -\frac{i(2m-1)!!}{z^{m+1 } } \bigl[1+o(z^2)\bigr],\end{aligned}\ ] ] one can easily show that @xmath186,\quad b_m^n = o\bigl[(k_\mathrm{a}r_\mathrm{c})^{2m+1}\bigr],\ ] ] so eq . ( [ e21k ] ) leads to @xmath187.\ ] ] processes involving reflections of the field at the cavity surface from the outside hence start to contribute in third order of @xmath146 and therefore do not need to be included in eq . ( [ e21 ] ) . so far we have demonstrated the validity of the relation ( [ e21 ] ) for homogeneous dielectric bodies of arbitrary shapes provided that the atom is situated at the interior of the body , such that the condition ( [ e15.2b ] ) is satisfied . note that this condition practically coincides with the condition ( [ e15.2 ] ) for the applicability of the real - cavity model . the arguments given above can be extended to inhomogeneous bodies . if the condition ( [ e15.2b ] ) is satisfied , one can divide such a body into a more or less small homogeneous part containing the cavity plus an inhomogeneous rest . equations ( [ e21 g ] ) , ( [ e21i ] ) , and ( [ e21p ] ) then still describe the propagation of the electric field inside the homogeneous part of the body , and the effect of the inhomogeneous part can be taken into account by the scattering at the ( fictitious ) surface dividing the two parts . consequently , we are again left with eq . ( [ e21 ] ) . substituting eq . ( [ e21 ] ) into eq . ( [ e14.4 ] ) and recalling eq . ( [ e14.5 ] ) , we can again cast the local - field corrected spontaneous - decay rate in the form of of eq . ( [ e24 ] ) , where now @xmath188\biggr\},\end{aligned}\ ] ] and @xmath189 \!\cdot\mathbf{d}_\mathrm{a}\ ] ] [ @xmath150 @xmath5 @xmath190 . recall that @xmath98 is the ( unperturbed ) scattering part of the green tensor of the host body without the cavity . needless to say that to linear order in @xmath36 , eqs . ( [ e22.2 ] ) and ( [ e22.2b ] ) reduce to eqs . ( [ e24.1 ] ) and ( [ e24.2 ] ) , respectively . in particular in the case of weakly absorbing matter , it may be sufficient to retain in eqs . ( [ e22.2 ] ) and ( [ e22.2b ] ) only terms to linear order in @xmath191 . from eqs . ( [ e24 ] ) , ( [ e22.2 ] ) , and ( [ e22.2b ] ) it then follows that when @xmath192 then @xmath193 can be given in the form @xmath194 where @xmath195 and @xmath196 is the uncorrected decay rate as given by eq . ( [ eq.10 ] ) with @xmath156 @xmath106 @xmath197 . rahmani and bryant @xcite concluded from a numerical computation of the green tensor of a dielectric sphere that contains a ( small ) empty sphere at an arbitrary position inside the material that the local - field corrected spontaneous decay rate has the form @xmath193 @xmath5 @xmath198 @xmath38 @xmath199 , where the shift term @xmath199 is due to absorption . comparing with eq . ( [ eq.61 ] ) [ together with eq . ( [ eq.62 ] ) ] , we see that this is indeed the case when the effect of absorption is sufficiently weak and in particular , the inequality ( [ eq.60 ] ) holds . however , from eqs . ( [ e24 ] ) , ( [ e22.2 ] ) , and ( [ e22.2b ] ) it is clearly seen that in general , @xmath193 can not be given in the form assumed in ref . note that already the analytical solution for the special case considered in ref . @xcite implies that this form can not be true in general . in the case of bulk material , we have @xmath98 @xmath5 @xmath14 , so eq . ( [ e24 ] ) simplifies to @xmath200 where @xmath201 is given by eq . ( [ e22.2 ] ) @xcite , which to linear order in @xmath36 reduces to eq . ( [ e24.1 ] ) . in particular , when material absorption can be neglected , @xmath150 @xmath22 @xmath202 , we simply have @xcite @xmath203 [ which follows directly from eq . ( [ eq.61 ] ) with @xmath196 @xmath5 @xmath204 , cf . ( [ eq.10 ] ) ] . note that in this case the virtual - cavity model leads to @xmath205 ( see , e.g. , ref . it is not difficult to see that to linear order in @xmath36 both eq . ( [ e0 ] ) and eq . ( [ e0a ] ) lead to @xmath206\gamma_0,\ ] ] i.e. , the first - order theory for the spontaneous decay rate does not distinguish between the virtual- and the real - cavity model @xcite , provided that absorption can be disregarded . let us consider an atom in a homogeneous dielectric sphere as sketched in fig . [ sphere ] and first calculate the spontaneous - decay rate to linear order in @xmath36 . in particular , we need to calculate @xmath139 as given by eq . ( [ e24.2 ] ) with @xmath98 from eq . ( [ e19b2 ] ) together with eq . ( [ e19 ] ) . noting that @xmath207\ ] ] ( cf . [ sphere ] ) is independent of @xmath124 , performing the @xmath124-integral , and using the fact that @xmath208 @xmath5 @xmath209 for a radially ( @xmath210 ) oriented dipole and @xmath208 @xmath5 @xmath211 for a tangentially ( @xmath212 ) oriented dipole , we find @xmath213,\end{aligned}\ ] ] where @xmath214 e^{(2i-\eta)q_\mathrm{o}}\nonumber\\ & + 4i\biggl(\frac{1}{3}-z\biggr ) \mathrm{ei}[(2i-\eta)q_\mathrm{o}],\\[1ex ] \label{e28 } f_\eta^\parallel(q_\mathrm{o},z ) = & \ ; f_\eta^\perp[q_\mathrm{o},(1-z)/2],\end{aligned}\ ] ] and @xmath131 according to eq . ( [ e25 ] ) , with @xmath215 being replaced by @xmath216 . in order to exactly calculate @xmath139 [ eq . ( [ e22.2b ] ) ] , we make use of the exact green tensor for a dielectric sphere as given in ref . @xcite , leading to @xmath217 ^ 2 \biggr\},\end{gathered}\ ] ] @xmath218 } { ( nk_\mathrm{a}l_\mathrm{a})\mathrm{d}l_\mathrm{a}}\biggr)^2 \biggr]\biggr\},\end{gathered}\ ] ] where @xmath219 ' -[z_1h_m^{(1)}(z_1)]'h_m^{(1)}(z_0 ) } { \varepsilon j_m^{(1)}(z_1)[z_0h_m^{(1)}(z_0 ) ] ' -[z_1j_m^{(1)}(z_1)]'h_m^{(1)}(z_0)}\,,\\ \label{e37 } & c^m_m=-\frac{h_m^{(1)}(z_1)[z_0h_m^{(1)}(z_0 ) ] ' -[z_1h_m^{(1)}(z_1)]'h_m^{(1)}(z_0 ) } { j_m^{(1)}(z_1)[z_0h_m^{(1)}(z_0 ) ] ' -[z_1j_m^{(1)}(z_1)]'h_m^{(1)}(z_0)}\,,\end{aligned}\ ] ] with @xmath166 @xmath106 @xmath220 , @xmath167 @xmath106 @xmath221 . to compare the exact decay rate [ eq . ( [ e24 ] ) together with eqs . ( [ e14.5 ] ) , and ( [ e22.2 ] ) as well as eqs . ( [ e34 ] ) and/or ( [ e35 ] ) ] with the decay rate obtained in the linear born approximation [ eq . ( [ e24 ] ) together with eqs . ( [ e14.5 ] ) , ( [ e24.1 ] ) , and ( [ e26 ] ) ] , let us consider , for simplicity , the case where the atom is positioned at the center of the sphere . setting @xmath222 @xmath5 @xmath14 and hence , in eqs . ( [ e26])([e28 ] ) , and carrying out the @xmath216 integral in eq . ( [ e26 ] ) , we derive @xmath223 with @xmath224e^{(2i-\eta)k_\mathrm{a}r}\biggr\},\end{gathered}\ ] ] giving @xmath225e^{2ik_\mathrm{a}r}\biggr\}\end{gathered}\ ] ] for any finite radius @xmath153 , and approaching zero in the limit @xmath153 @xmath226 @xmath227 ( which has to be taken before performing the limit @xmath228 @xmath226 @xmath14 ) . when material absorption is small enough such that @xmath229 @xmath41 @xmath230 , and the radius of the sphere is large @xmath231 @xmath232 @xmath42 , then eq . ( [ e24 ] ) together with eqs . ( [ e24.1 ] ) and ( [ e44.1 ] ) reduces to [ @xmath105 @xmath22 @xmath233 @xmath234\gamma_0 -\frac{\chi(\omega_\mathrm{a})}{2}\cos(2k_\mathrm{a}r)\gamma_0.\ ] ] note that eq . ( [ e2 ] ) differs from eq . ( [ e1 ] ) in the second term , which reflects the finite size of the sphere . when the atom is on center , only the terms in eqs . ( [ e34 ] ) and ( [ e35 ] ) contribute to the exact @xmath235 , cf . ( [ e21 m ] ) and ( [ e21n ] ) , and hence we find @xmath236.\ ] ] combining eqs . ( [ e24 ] ) , ( [ e22.2 ] ) , and ( [ e45 ] ) , we arrive at @xmath237 which can be shown to agree with eq . ( [ e44.1 ] ) to linear order in @xmath36 . in a more sophisticated linearization of eq . ( [ eq.80 ] ) with respect to @xmath36 , it may be advantageous to leave the @xmath36-dependence in the exponentials appearing in the coefficient @xmath238 unchanged . in particular , for sufficiently small material absorption , @xmath229 @xmath41 @xmath230 , and for a sufficiently large sphere , @xmath231 @xmath232 @xmath42 , this kind of linear approximation leads to [ @xmath105 @xmath22 @xmath233 @xmath239\gamma_0\\ -\frac{\chi(\omega_\mathrm{a})}{2}\ , \cos\bigl[2\mathrm{re}(n)k_\mathrm{a}r\bigr ] e^{-2\mathrm{im}(n)k_\mathrm{a}r}\gamma_0,\end{gathered}\ ] ] recall eqs . ( [ e2.1 ] ) and ( [ e47 ] ) . @xmath5 @xmath240 and ( b ) @xmath150 @xmath5 @xmath241 as a function of the sphere radius @xmath153 . the radius of the real ( empty - space ) cavity is @xmath90 @xmath5 @xmath242 . for comparison , the decay rate in a bulk medium of the same permittivity ( dotted lines ) is also plotted . ] comparison of eq . ( [ e46 ] ) with eq . ( [ e2 ] ) indicates that with increasing size of the sphere , the validity of the linear born approximation to the decay rate becomes less satisfying in two respects . ( i ) the oscillating term is not damped , and ( ii ) there is an accumulated error in the phase . the former effect is insignificant as long as @xmath243 @xmath41 @xmath42 . this restriction still allows for a large range of sphere sizes ; for example , for @xmath145 @xmath22 @xmath244 and @xmath245 @xmath5 @xmath246 @xmath22 @xmath247 , the condition follows . figure [ cnt ] shows the dependence on the sphere radius of the local - field corrected spontaneous - decay rates according to the exact equation ( [ eq.80 ] ) and the linear born expansion [ eq . ( [ e24 ] ) together with eqs . ( [ e24.1 ] ) and ( [ e44.1 ] ) ] for two values of the permittivity . for the parameters used , we have @xmath144 @xmath5 @xmath248 , so that the first two terms in the curly brackets in eq . ( [ eq.80 ] ) , which arise from absorption , are negligibly small compared to the last one . as expected from eqs . ( [ e2 ] ) and ( [ e46 ] ) , the decay rate oscillates with the sphere radius around the bulk value . for small spheres , the linear born expansion is seen to be in good agreement with the exact result . it is further seen that , as the sphere radius increases , an increasing phase shift develops between the exact rate and the rate in the linear born approximation . a comparison of figs . [ cnt](a ) and ( b ) reveals that this phase shift is larger for larger values @xmath249 . @xmath5 @xmath250 as a function of @xmath145 ( @xmath251 @xmath5 @xmath252 ) , for real - cavity radii @xmath90 @xmath5 @xmath242 and @xmath253 . ] the dependence of the decay rate on the imaginary part of the susceptibility is illustrated in fig . [ ci ] for two values of the real - cavity radius . it can be seen that the disagreement between the curves representing the linear born expansion and the exact result increases with increasing material absorption or decreasing cavity radius , i.e. , with an increase in the combined factor @xmath144 . in particular , if in the case where @xmath90 @xmath5 @xmath242 , @xmath145 changes from @xmath254 to @xmath255 , then @xmath256 changes from @xmath257 to @xmath37 , and the agreement worsens from being very good to being moderately good . for the larger cavity radius @xmath90 @xmath5 @xmath258 , which physically means a more dilute medium , @xmath256 varies from @xmath259 to @xmath260 for the same variation of @xmath145 . throughout this range , the combined factor remains much smaller than one and a good agreement is observed . basing on a numerical computation of @xmath261 as given by eq . ( [ e26 ] ) in the linear born approximation , we have also studied the case when the atom is localized at an arbitrary position inside the sphere . @xmath5 @xmath240 and radius and ( b ) @xmath153 @xmath5 @xmath262 . the real - cavity radius is @xmath90 @xmath5 @xmath242 . the solid ( dashed ) curves refer to radially ( tangentially ) oriented transition dipole moments . ] figure [ offcnt ] compares the position dependences of the local - field corrected decay rate , as given by eq . ( [ e24 ] ) together with eqs . ( [ e24.1 ] ) and ( [ e26 ] ) , and the uncorrected rate according to eq . ( [ eq.10 ] ) ( which is valid if absorption is neglected ) , with the second term being given by eq . ( [ e26 ] ) , for two sphere radii ( for the uncorrected rate beyond the linear born approximation , see also ref . @xcite ) . from the figure it is seen that when @xmath222 @xmath13 @xmath14 , i.e. , when the atom is not localized at the center of the sphere , radial and tangential dipole orientations have to be distinguished , especially when the sphere radius exceeds the wavelength of the radiation emitted by the atom , such that interference effects begin to play a role . note that the whispering gallery resonances which may give rise to strong enhancement near the rim of the sphere are not manifested here . the existence of these resonances requires larger values of the real part of the permittivity or larger sphere radii . unfortunately , in such cases the linear born expansion is too rough an approximation to the decay rate . from the figure it is further seen that for the parameters used , the local - field correction increases the decay rate ; the amount of increase is determined essentially by the last term in eq . ( [ e24.1 ] ) . in fig . [ dp ] the dependence of the local - field corrected decay rate on the sphere radius is illustrated . it can be seen that the further the atom is displaced from the sphere center , the smaller the amplitudes of oscillation of the decay rate become . @xmath5 @xmath263 and ( b ) @xmath222 @xmath5 @xmath264 off center of a dielectric sphere of permittivity as a function of the sphere radius in linear born approximation . the cavity radius is @xmath146 @xmath5 @xmath148 . the solid ( dashed ) curves refer to radially ( tangentially ) oriented transition dipole moments . for comparison , the curves for an atom at the center are also shown . ] expressing the spontaneous - decay rate of an excited atom in the presence of dielectric bodies in terms of the scattering part of the associated green tensor of the macroscopic maxwell equations and expanding the green tensor in a born series enables one to systematically approach arbitrary geometries . in particular , in the case of weakly polarizable bodies it may be possible to neglect higher - order terms in the born expansion and approximate the scattering part of the green tensor by the term linear in the susceptibility , and the decay rate accordingly . using the real - cavity model of the local - field correction , we have applied the theory to the problem of the local - field correction to the spontaneous - decay rate of an excited atom embedded in a dispersing and absorbing dielectric body of arbitrary size and shape . in this way , we have derived a rate formula , which , within the linear born approximation , applies to atoms in arbitrary dielectric bodies , and have given explicit conditions of its validity . to illustrate the results , we have considered the case of an atom at an arbitrary position inside spherical body in more detail . it has surprisingly turned out that the rate formula found in linear born approximation agrees , to linear order in the susceptibility , with a rate formula suggested by toma @xcite from his study of the real - cavity model in the special and analytically solvable case of a spherically symmetric system . we have then shown that this quite general formula indeed applies to atoms in dielectric bodies of arbitrary sizes and shapes , provided that the atoms are not in the very vicinity of the surfaces of the bodies . so it can be shown that the scattering part of the green tensor that enters the basic formula for the decay rate can always be decomposed into a term that only depends on the properties of the local environment of the guest atom and the local - field corrected scattering part of the green tensor of the host body without the guest atom , where the correction simply appears in form of a factor . in particular , the former term can be substantially determined by the absorptive properties of the host body , thereby giving rise to a shift of the decay rate . finally , we note that in the same spirit as in the treatment of the spontaneous decay , the born expansion can also be employed in studying other phenomena of the atom field interaction in realistic systems whose green tensors are not known or analytically too involved . typical examples may be the casimir - polder interaction of a ground - state atom with an inhomogeneous body @xcite , the resonance fluorescence of an atom near such a body , or the resonant energy transfer between atoms embedded in realistic bodies . we would like to thank m. s. toma for discussions . thanks the alexander von humboldt stiftung and the national program for basic research of vietnam for support . this work was supported by the deutsche forschungsgemeinschaft .

two - level emitters interacting with a cavity photon mode are widely studied in quantum optics with respect to spontaneous emission and superradiance @xcite , cooperativity and lasing @xcite , as well as the emission of nonclassical light @xcite . for sufficiently weak light - matter coupling , when the photon - dressing of emitter states is negligibly small , the emitter - cavity system can be studied with the quantum optical master equation , usually in combination with the rotating - wave approximation @xcite . the quantum optical master equation fails at strong light - matter coupling , to the extent that it predicts unphysical emission at zero temperature if the number of photons in the ground state is finite @xcite . the correct theoretical description of systems with ( ultra-)strong light - matter coupling @xcite has attracted increasing interest recently @xcite . essentially , the quantum optical master equation has to be replaced by a master equation expressed in the photon - dressed emitter eigenstates @xcite . while the master equation remains markovian , which is justified because of the weak emitter - environment and cavity - environment couplings @xcite , it now requires full diagonalization of the interacting emitter - cavity hamiltonian . such an equation was used in recent studies of photon blockade effects @xcite , spontaneous conversion of virtual to real photons @xcite , and the emission of nonclassical light from a single emitter @xcite . in this paper we study the emission of few emitters in a cavity , with particular focus on the photon statistics of the emitted light . our goal is the characterization of temperature and coupling regimes where nonclassical light @xcite is generated . a major result will be the identification of two clearly distinguished neighboring regimes with pronounced sub - poissonian and super - poissonian photon statistics at strong coupling . our results are obtained with the full input - output formalism @xcite and master equation @xcite without further approximations . to understand the relevance of the different approximations involved in traditional quantum optics treatments we make two comparisons . first , we compare the results that are obtained when the counter - rotating light - matter interaction terms are included in the hamiltonian to those when they are dropped . second , we contrast the results obtained with the full master equation with results from the quantum optical master equation . the latter comparison will clearly show the necessity of using the correct master equation already at weak coupling if the photon statistics is of interest . this issue has been studied conclusively for a single emitter in ref . @xcite , which also contains glauber function plots for few emitters in the supplemental material but omits the further analysis of the situation that we give here . the paper is organized as follows . in sec . [ sec : model ] we introduce the physical situation under study together with the master equation used for its analysis . in sec . [ sec : spectra ] we discuss the emission spectra in relation to the energy spectra of the emitter - cavity hamiltonian , while the statistics of the emitted photons is studied in sec . [ sec : statistics ] . we conclude in sec . [ sec : concl ] . the appendices collect further information on the theoretical approach . [ app : inout ] gives details of the input - output formalism . in app . [ app : me ] we derive the master equation , and give a few analytical results for the photon statistics in app . [ app : analyt ] . the interaction of @xmath1 two - level emitters with a single cavity photon mode is described by the dicke model @xcite @xmath2 where the operator @xmath3 annihilates ( creates ) a cavity photon with frequency @xmath4 and @xmath5 ( @xmath6 ) is the corresponding lowering ( raising ) operator for the @xmath7th emitter with transition energy @xmath8 . throughout this work , we consider the resonant case @xmath9 . we allow for different emitter - photon coupling strengths for the co - rotating ( @xmath10 ) and counter - rotating ( @xmath11 ) interaction terms . changing @xmath11 relative to @xmath10 interpolates between the tavis - cummings ( tc ) limit ( @xmath12 ) without and the dicke limit ( @xmath13 ) with counter - rotating terms . both situations can be realized experimentally @xcite . the rotating - wave approximation consists in replacing the dicke by the tc limit . dissipation arises from the coupling of the emitters and the cavity to the environment . for a bosonic environment the coupling terms are of the form @xmath14 where @xmath15 is a ( hermitian ) emitter or cavity operator and the @xmath16 are bosonic operators for the environment photons ( at frequencies @xmath17 with coupling constants @xmath18 ) . as the operator @xmath15 we choose the field operator @xmath19 for the coupling of the cavity and the transition operator @xmath20 for the coupling of the @xmath7th emitter to the environment . at sufficiently weak coupling to the environment , the emitter - cavity system density matrix @xmath21 obeys a markovian master equation @xcite @xmath22 - { \mathrm i}\frac{1}{2 } \sum_\omega \xi(\omega ) \big [ s_\omega^\dagger s_\omega , \rho(t ) \big ] \nonumber\\ & & + \frac{1}{2 } \sum_\omega \chi(\omega ) \big ( \big [ s_\omega \rho(t ) , s_\omega^\dagger \big ] + \big [ s_\omega , \rho(t ) s_\omega^\dagger \big ] \big ) \ , , \nonumber\\\end{aligned}\ ] ] where @xmath23 is the projection of @xmath15 onto transitions between eigenstates @xmath24 , @xmath25 of @xmath26 with energy difference @xmath27 ( see app . [ app : me ] for a derivation ) . for the sake of notational simplicity we state the master equation for a single coupling term . multiple coupling terms lead to additional contributions of the same form . the functions @xmath28 and @xmath29 in eq . follow from the environment spectral function @xmath30 and its analytical continuation @xmath31 into the upper half plane , with @xmath32 . for a thermal environment with inverse temperature @xmath33 we get @xmath34 \quad & \text { if } \omega > 0 \ ; , \\[0.25ex ] \gamma(-\omega ) n(-\omega , t ) & \text { if } \omega < 0 \end{cases}\ ] ] and @xmath35 \quad & \text { if } \omega > 0 \ ; , \\[0.25ex ] - { \mathop{{\rm re}}}\gamma(-\omega + { \mathrm i}0^+ ) n(-\omega , t ) & \text { if } \omega < 0 \ ; , \end{cases}\ ] ] with the bose - einstein distribution function @xmath36 note that in the zero temperature limit @xmath37 such that the master equation contains only dissipative terms for transitions @xmath38 with positive energy @xmath39 , i.e. , dissipation correctly leads to energy decrease . in particular , the problem of unphysical emission from the ground state encountered for the quantum optical master equations is resolved . in the present work we assume an ohmic spectral function @xmath40 for the cavity - environment coupling , and use @xmath41 in all numerical computations . to reduce the number of free parameters we assume the same spectral function @xmath42 for the emitter - environment couplings . the respective environment temperatures are also identical . as we show in app . [ app : me ] , the master equation splits into two equations of motion @xmath43 for the matrix elements @xmath44 of the density operator . in these equations , @xmath45 and @xmath46 s_{k , n } + { \mathrm i}e_n \,.\ ] ] the general solution of eq . is @xmath47 because @xmath48 for all @xmath49 , the off - diagonal elements of @xmath50 decay exponentially . hence , the stationary state fulfills @xmath51 the diagonal elements @xmath52 are determined by the stationary solution of the pauli master equation . if the system is coupled to a thermal environment as in eqs . and , the stationary solution of eq . is the thermal state @xmath53 of the system corresponding to the temperature @xmath54 of the environment . the emission spectrum and photon statistics can now be computed through a standard input - output formalism ( see app . [ app : inout ] ) , which leads to the projected cavity - environment coupling operator @xmath55 describing the emission . the correlation functions of @xmath56 and @xmath57 characterize the properties of the emitted light . the emission spectrum of the cavity is @xmath58 and the second - order glauber function @xcite reads @xmath59 note that evaluation of eqs . , requires diagonalization of the hamiltonian @xmath26 . because the stationary state @xmath60 from eq . is diagonal in the eigenbasis @xmath25 of @xmath26 , we can evaluate the @xmath61-integration in eq . analytically as @xmath62 ^ 2 + [ { \mathop{{\rm re}}}(z_m+z_n)]^2 } \ , . \nonumber\end{aligned}\ ] ] the emission spectrum @xmath63 is the sum of lorentz peaks with width @xmath64 at the respective transition energies @xmath65 , which according to eq . are shifted relative to the transition energies @xmath66 of the closed system by a lamb shift that results from coupling to the environment . it is instructive to compare the master equation to the quantum optical master equation @xcite @xmath67 - { \mathrm i}\sum_\pm \frac{\xi_\pm}{2 } [ s_\pm^\dagger s_\pm , \rho(t ) ] \\ & & + \sum_\pm \frac{\chi_\pm}{2 } \big ( \big [ s_\pm \rho(t ) , s_\pm^\dagger \big ] + \big [ s_\pm , \rho(t ) s_\pm^\dagger \big ] \big ) \ ; , \nonumber\end{aligned}\ ] ] which is obtained by replacing the projected operators @xmath68 with the ` bare ' operators @xmath69 , and by assuming @xmath70 , @xmath71 in the vicinity of a typical transition energy @xmath72 . note that @xmath73 for the cavity - environment coupling and @xmath74 for the emitter - environment coupling . evidently this approximation can be valid only for weak light - matter coupling @xmath75 , when the dressing of emitter states by cavity photons can be neglected . because the quantum optical master equation does not distinguish between energy - increasing and energy - decreasing transitions , which are equally contained in the unprojected operator @xmath15 because of hermiticity , it can lead to unphysical predictions such as emission out of the ground state . furthermore , because failure to observe the above distinction is tantamount to a high - temperature approximation , one will expect that the quantum optical master equation fails at the prediction of non - thermal photon statistics at low temperatures . therefore , we use the more general master equation . the first characterization of the light generated in the cavity is provided by the emission spectrum . because the emission spectrum depends on the ( lamb - shifted ) energy spectrum of the dicke hamiltonian we start with a discussion of the eigenvalues of @xmath26 for few emitters , before we turn to the actual function @xmath63 obtained from numerical solution of the master equation . to construct the energy spectrum of @xmath26 we notice first that the eigenstates of @xmath1 uncoupled two - level emitters can be classified as angular momentum eigenstates with total angular momentum @xmath76 . since @xmath26 commutes with the total angular momentum operator , states with different @xmath77 do not mix even at finite coupling @xmath78 . for fixed @xmath77 , the @xmath79 quantum number @xmath80 can assume the values @xmath81 , and a corresponding emitter eigenstate has energy @xmath82 . note that for @xmath83 the classification in terms of @xmath77 , @xmath80 is not exhaustive , since different emitter states can have identical values . however , these states give the same contribution to the emission spectrum . the cavity photon eigenstates are fock states @xmath25 with energies @xmath84 . schematic energy level pattern for the construction of the spectrum of the dicke model . horizontal arrows depict the co - rotating interaction terms in eq . ( coupling constant @xmath10 ) , diagonal arrows depict the counter - rotating terms ( @xmath11 ) . ] for given @xmath77 we can arrange the eigenstates of the uncoupled emitter - cavity system as the rungs of a ladder diagram as in fig . [ fig : ladder ] . working at resonance @xmath85 , the energy @xmath86 of each state is given by the total number of emitter and cavity excitations . the co - rotating light - matter interaction terms in @xmath26 preserve the number of excitations and connect states at the same energy level ( horizontal arrows in fig . [ fig : ladder ] ) . the counter - rotating terms change the number of excitations by two ( diagonal arrows in fig . [ fig : ladder ] ) . this simple scheme explains many properties of the energy spectra of @xmath26 shown in fig . [ fig : e ] . eigenvalues of @xmath26 in eq . for @xmath13 ( panels a - c ) and @xmath12 ( panels d - f ) as functions of the coupling strength @xmath10 . panels ( a , b ) show the results for one emitter , whereas the number of emitters is @xmath87 in panels ( b , e ) and @xmath88 in panels ( c , f ) . ] for @xmath89 it is @xmath90 , and we recover the jaynes - cummings ladder @xcite for @xmath12 . the lowest level ( @xmath91 , @xmath92 ) does not couple to any other state and hence leads to the @xmath10-independent eigenvalue zero of @xmath26 [ see fig . [ fig : e](d ) ] . every other level consists of two ladder rungs . they lead to the eigenvalues @xmath93 , for @xmath94 . for @xmath13 corrections arise from coupling between states at different height in the ladder but the energy level pattern remains discernible [ fig . [ fig : e](a ) ] . for @xmath87 [ see fig . [ fig : e](b , e ) ] we have either triplet ( @xmath95 ) or singlet ( @xmath96 ) emitter states . for @xmath12 , the triplet states lead to the eigenvalue zero ( @xmath92 ) , the two eigenvalues @xmath97 ( @xmath98 ) , and the three eigenvalues @xmath99 , @xmath100 for @xmath101 . the singlet states do not couple with each other and lead to the @xmath10-independent eigenvalues @xmath102 for @xmath103 . it follows that the eigenvalues @xmath99 for @xmath101 are twofold degenerate ( one triplet , one singlet state ) . this degeneracy is lifted for @xmath13 , but the energies of the singlet states remain fixed . for @xmath88 we have quadruplet ( @xmath104 ) and doublet ( @xmath105 ) emitter states . the ladder scheme for the doublet is equal to that for @xmath89 and hence leads to the same energy spectrum , apart from the fact that all energies are shifted up by @xmath106 when going from @xmath107 to @xmath108 . notice that the doublet states are two - fold degenerate , because the angular momentum classification of the emitter states is not unique in this case . the quadruplet states lead to one ( starting at zero for @xmath109 ) , two ( at @xmath106 ) , three ( at @xmath110 ) , and four ( at @xmath99 with @xmath111 ) additional eigenvalues in figs . [ fig : e](f ) . because of the close vicinity of many states in the energy spectrum the corrections resulting from the counter - rotating terms for @xmath13 are large . this trend continues if @xmath1 is increased further . in fig . [ fig : spectra ] we show the emission spectrum @xmath63 for @xmath112 emitters at different coupling strength @xmath10 and environment temperature @xmath113 . these data , as well as those for the glauber function @xmath114 shown later , have been computed with a maximal number of @xmath115 cavity photons in the numerical diagonalization of @xmath26 , which is sufficient for the given parameter combinations . emission spectra @xmath63 for @xmath112 emitters . panels ( i - iii ) show results for @xmath13 , panels ( iv - vi ) for @xmath12 . the emitter - cavity coupling strength and the environment temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] emission spectra @xmath63 for @xmath112 emitters . panels ( i - iii ) show results for @xmath13 , panels ( iv - vi ) for @xmath12 . the emitter - cavity coupling strength and the environment temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] + emission spectra @xmath63 for @xmath112 emitters . panels ( i - iii ) show results for @xmath13 , panels ( iv - vi ) for @xmath12 . the emitter - cavity coupling strength and the environment temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] emission spectra @xmath63 for @xmath112 emitters . panels ( i - iii ) show results for @xmath13 , panels ( iv - vi ) for @xmath12 . the emitter - cavity coupling strength and the environment temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] + emission spectra @xmath63 for @xmath112 emitters . panels ( i - iii ) show results for @xmath13 , panels ( iv - vi ) for @xmath12 . the emitter - cavity coupling strength and the environment temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] emission spectra @xmath63 for @xmath112 emitters . panels ( i - iii ) show results for @xmath13 , panels ( iv - vi ) for @xmath12 . the emitter - cavity coupling strength and the environment temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] for low temperatures @xmath122 , only the first possible transition into the ground state contributes to the emission spectrum . it leads to the single peak in panels ( i ) and ( iii ) of fig . [ fig : spectra ] . with increasing temperature transitions involving higher excited states begin to contribute . for example the two peaks in panels ( ii ) , ( iv ) correspond to the transition from the 2nd to the 1st excited state and from the 3rd excited to the ground state . as could be deduced already from panels ( b ) , ( e ) in fig . [ fig : e ] , the transitions tend to have smaller energies in the tc limit than in the dicke limit , which leads to the red shift of the emission peaks in panel ( v ) relative to those in panel ( ii ) . however , at not too strong coupling the low - lying states still have comparable energies , and the emission spectra look similar . the situation changes at ultrastrong coupling when the co - rotating and counter - rotating terms are of equal magnitude ( panels ( iii ) , ( vi ) ) . in addition to the markedly different peak energies the peak height has now decreased by two orders of magnitude in the dicke limit , but not in the tc limit . the decrease of peak height can be recognized in the @xmath72-integrated emission spectrum @xmath123 shown in fig . [ fig : intspectra ] . the equality with the given expectation value follows directly from eq . . only in the dicke limit , but not in the tc limit , the total emission becomes small again at ultrastrong coupling and low temperatures . still , one sees that both plots agree nicely for not too strong coupling ( @xmath124 ) . this observation sets the upper limit of the coupling strength ( here , for @xmath112 emitters ) below which the presence or absence of counter - rotating interaction terms does not affect the light emission significantly . we will find the same behavior for the glauber function . ( color online ) expectation value @xmath125 for @xmath112 emitters , as a function of temperature @xmath113 and coupling strength @xmath10 . panel ( a ) shows the result for @xmath13 , panel ( b ) for @xmath12 . crosses mark the parameters used in fig . [ fig : spectra ] . , title="fig : " ] ( color online ) expectation value @xmath125 for @xmath112 emitters , as a function of temperature @xmath113 and coupling strength @xmath10 . panel ( a ) shows the result for @xmath13 , panel ( b ) for @xmath12 . crosses mark the parameters used in fig . [ fig : spectra ] . , title="fig : " ] a basic decision on the possible generation of nonclassical light is possible with the glauber function @xmath126 at zero time delay . for @xmath127 the emitted photons have a poissonian distribution , while @xmath128 indicates super - poissonian statistics . thermal light has @xmath129 . by contrast , @xmath130 indicates nonclassical light with sub - poissonian photon statistics . further information on photon ( anti-)bunching is provided by the full time - dependent function @xmath114 . the glauber function @xmath126 for one emitter ( @xmath89 ) is shown in fig . [ fig : g2_chart_ext ] . two distinct regions can be identified in the dicke limit in panel ( a ) ( where @xmath13 ) . a triangular region with @xmath130 , which stretches out along the vertical axis , indicates the emission of nonclassical light with sub - poissonian photon statistics at low temperatures and moderate - to - strong light - matter coupling . it lies below an elongated region with strongly super - poissonian photon statistics ( @xmath131 ) at larger coupling , which extends diagonally towards higher temperatures . both regions are embedded in the background of thermal light with @xmath132 . the situation is distinctly different in the tc limit ( @xmath12 ) in panel ( b ) , where the super - poissonian region is pushed back in favor of a second sub - poissonian region that continues towards ultrastrong coupling . note , however , that the emission of nonclassical light in the first sub - poissonian region is observed equally in both limits . ( color online ) glauber function @xmath126 at zero time - delay for one emitter ( @xmath107 ) , as a function of temperature @xmath113 and coupling strength @xmath10 . panel ( a ) shows the result for @xmath13 , panel ( b ) for @xmath12 . note that all values @xmath133 are assigned the same ( dark - red ) color in the density plots.,title="fig : " ] ( color online ) glauber function @xmath126 at zero time - delay for one emitter ( @xmath107 ) , as a function of temperature @xmath113 and coupling strength @xmath10 . panel ( a ) shows the result for @xmath13 , panel ( b ) for @xmath12 . note that all values @xmath133 are assigned the same ( dark - red ) color in the density plots.,title="fig : " ] ( color online ) glauber function @xmath126 at zero time - delay as a function of temperature @xmath113 and coupling strength @xmath10 , for @xmath112 emitters ( top panels a , c ) and @xmath108 emitters ( bottom panels b , d ) , the left panels ( a , b ) show results for @xmath13 , whereas @xmath12 in panels ( c , d ) . note that all values @xmath133 are assigned the same ( dark - red ) color in the density plots . , title="fig : " ] ( color online ) glauber function @xmath126 at zero time - delay as a function of temperature @xmath113 and coupling strength @xmath10 , for @xmath112 emitters ( top panels a , c ) and @xmath108 emitters ( bottom panels b , d ) , the left panels ( a , b ) show results for @xmath13 , whereas @xmath12 in panels ( c , d ) . note that all values @xmath133 are assigned the same ( dark - red ) color in the density plots . , title="fig : " ] + ( color online ) glauber function @xmath126 at zero time - delay as a function of temperature @xmath113 and coupling strength @xmath10 , for @xmath112 emitters ( top panels a , c ) and @xmath108 emitters ( bottom panels b , d ) , the left panels ( a , b ) show results for @xmath13 , whereas @xmath12 in panels ( c , d ) . note that all values @xmath133 are assigned the same ( dark - red ) color in the density plots . , title="fig : " ] ( color online ) glauber function @xmath126 at zero time - delay as a function of temperature @xmath113 and coupling strength @xmath10 , for @xmath112 emitters ( top panels a , c ) and @xmath108 emitters ( bottom panels b , d ) , the left panels ( a , b ) show results for @xmath13 , whereas @xmath12 in panels ( c , d ) . note that all values @xmath133 are assigned the same ( dark - red ) color in the density plots . , title="fig : " ] the distinctive features of the glauber function persist for multiple emitters ( see fig . [ fig : g2_chart ] ) , but the regions are shifted to smaller couplings @xmath10 as the number of emitters increases from one to three . the obvious similarity between @xmath126 for @xmath134 emitters visible in figs . [ fig : g2_chart_ext ] , [ fig : g2_chart ] can be expressed as an approximate relation between the respective emitter - cavity coupling @xmath10 . in the dicke limit ( @xmath13 ) we find that the features of @xmath126 are closely reproduced under the scaling @xmath135 . in the tc limit ( @xmath12 ) features are reproduced under the scaling @xmath136 . interestingly , the proper scaling of @xmath10 depends on the presence of counter - rotating interaction terms in the hamiltonian . this difference is in contrast to the semiclassical theory where the mean cavity photon number in the steady state scales @xmath137 both in the dicke and tc limit . not surprisingly , the glauber function @xmath126 is more sensitive to the details of light - matter coupling than the semiclassical theory that neglects quantum correlations in favor of a mean - field approximation . our arguments in favor of the above scaling relations depend on several observations , which we now develop for the tc limit ( @xmath12 ) . without counter - rotating interaction terms the hamiltonian @xmath26 commutes with the operator @xmath138 , which counts the total number of excitations . hence , @xmath26 is block diagonal with blocks of the form @xmath139 , where @xmath140 denotes the eigenvalue of @xmath141 , @xmath142 is the identity matrix , and the matrix block @xmath143 contains the @xmath10-independent matrix elements of the co - rotating interaction terms in @xmath26 . from this form of the blocks it is evident that the eigenvectors of @xmath26 do not depend on @xmath10 , i.e. , the matrix elements of @xmath144 that enter eq . are constant . the dependence of @xmath126 on @xmath10 results from the eigenvalues only , which determine the occupation of the states in the stationary ( thermal ) state and the prefactors of @xmath144 . if we can show that the eigenvalues scale approximately as @xmath145 the above relation follows . let us focus on the low lying states that give the dominant contribution in the interesting temperature regimes . these states can be found in the ladder diagram of states in fig . [ fig : ladder_rwa_a ] . they must be connected to the ground state at energy zero by a diagonal arrow that gives the action of the operator @xmath146 , i.e. , of @xmath147 . schematic energy level pattern . horizontal arrows depict the co - rotating interaction terms in eq . ( coupling constant @xmath10 ) , diagonal arrows illustrate the action of the cavity photon annihilation operator @xmath146 . ] for the denominator @xmath125 of @xmath126 from eq . states contribute which are separated by one vertical step in the ladder diagram . the energy of the most relevant first excited state is given by @xmath148 , which has the postulated scaling . this scaling of the first excited state for few emitters has been verified experimentally in ref . @xcite . for the numerator @xmath149 of @xmath126 , where each operator appears twice , states contribute which are separated by two vertical steps on the ladder . now the second excited state is most relevant , which is the linear combination of the two ( @xmath107 ) or three ( @xmath150 ) vertical rungs that occur for @xmath151 excitations . the corresponding @xmath152 or @xmath153 matrix from the above block decomposition of @xmath26 is @xmath154 diagonalization gives the energies @xmath155 for @xmath89 , while @xmath156 for @xmath157 . with the approximation @xmath158 , which is good enough for a rule of thumb , this is again the postulated scaling . put together , the energies that enter the computation of @xmath126 scale roughly as @xmath145 , which concludes our argument in favor of the observed relation `` @xmath136 '' in the tc limit . in the dicke limit @xmath13 the block decomposition of @xmath26 is not possible because of the counter - rotating interaction terms . the eigenvectors of @xmath26 now depend on @xmath10 , and the previous argument can not be easily translated . however , inspection of the energy spectra in fig . [ fig : e ] strongly suggests that the observed relation is still related to an approximate relation between the eigenvalues of @xmath26 for different @xmath1 , now with the scaling @xmath135 . results for the glauber function obtained with the quantum optical master equation are shown in fig . [ fig : g2_chart_qo ] in the dicke limit @xmath13 . in stark contrast to the results from figs . [ fig : g2_chart_ext ] , [ fig : g2_chart ] the quantum optical master equation does not predict the emission of nonclassical light with sub - poissonian photon statistics in any part of the parameter space . the situation does not improve in the tc limit @xmath12 where @xmath159 = 0 $ ] and the quantum optical master equation gives the stationary ( thermal ) state @xmath160 leading to @xmath161 independent of the number of emitters @xmath1 , the coupling strength @xmath10 , or the temperature @xmath113 , thus always predicting the emission of thermal light . while it may not be surprising that the quantum optical master equation fails , because the weak coupling condition @xmath162 is not satisfied , it is remarkable that it fails to capture any features from the previous glauber function plots in figs . [ fig : g2_chart_ext ] , [ fig : g2_chart ] . this failure highlights the importance of using the correct master equation not only for strong light - matter coupling but also if one is interested in properties following from higher - order correlation functions , such as the photon statistics obtained from the second order glauber function . ( color online ) glauber function @xmath126 computed with the quantum optical master equation , shown as a function of temperature @xmath113 and coupling strength @xmath10 for @xmath13 . panel ( a ) gives the result for @xmath107 emitter , whereas @xmath112 in panel ( b ) . , title="fig : " ] ( color online ) glauber function @xmath126 computed with the quantum optical master equation , shown as a function of temperature @xmath113 and coupling strength @xmath10 for @xmath13 . panel ( a ) gives the result for @xmath107 emitter , whereas @xmath112 in panel ( b ) . , title="fig : " ] a further property to distinguish classical and nonclassical light is the time - coincidence statistics of the emitted photons , which can be deduced from the time - dependent glauber function @xmath114 . for classical light , @xmath114 has a non - positive initial slope at @xmath163 . this indicates photon bunching , i.e. , that the probability of observing two photons at equal times is larger than the probability of observing them at different times . conversely , a positive slope indicates photon antibunching , which is possible only for nonclassical light . in the long - time limit , @xmath164 in all cases . in fig . [ fig : g2_time ] we plot @xmath114 for the parameter combinations marked in the two upper panels in fig . [ fig : g2_chart ] . we see that @xmath114 is always a strictly monotonic function of @xmath165 . therefore , in the present situation photon bunching and antibunching coincide precisely with super - poissonian and sub - poissonian photon statistics . only if @xmath166 in panel ( ii ) the function @xmath114 oscillates slightly , but the overall decay is still indicative of photon bunching . glauber function @xmath114 as a function of time for @xmath112 emitters . panels ( i - iii ) show the results for @xmath13 , whereas in panels ( iv - vi ) @xmath12 . the emitter - cavity coupling strength and the bath temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] glauber function @xmath114 as a function of time for @xmath112 emitters . panels ( i - iii ) show the results for @xmath13 , whereas in panels ( iv - vi ) @xmath12 . the emitter - cavity coupling strength and the bath temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] + glauber function @xmath114 as a function of time for @xmath112 emitters . panels ( i - iii ) show the results for @xmath13 , whereas in panels ( iv - vi ) @xmath12 . the emitter - cavity coupling strength and the bath temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] glauber function @xmath114 as a function of time for @xmath112 emitters . panels ( i - iii ) show the results for @xmath13 , whereas in panels ( iv - vi ) @xmath12 . the emitter - cavity coupling strength and the bath temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] + glauber function @xmath114 as a function of time for @xmath112 emitters . panels ( i - iii ) show the results for @xmath13 , whereas in panels ( iv - vi ) @xmath12 . the emitter - cavity coupling strength and the bath temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] glauber function @xmath114 as a function of time for @xmath112 emitters . panels ( i - iii ) show the results for @xmath13 , whereas in panels ( iv - vi ) @xmath12 . the emitter - cavity coupling strength and the bath temperature are @xmath116 , @xmath117 in panels ( i , iv ) , @xmath118 , @xmath119 in panels ( ii , v ) , and @xmath120 , @xmath121 in panels ( iii , vi).,title="fig : " ] our analysis of the light generated by few emitters in a cavity reveals a non - trivial dependence of the photon statistics on the light - matter coupling and temperature . clearly identifiable parameters regimes with sub- and super - poissonian photon statistics appear at strong and ultrastrong coupling , and lie immediately next to each other . tuning of the light - matter coupling or change of the temperature can thus have a tremendous effect on the photon statistics . as a general trend we find strong signatures of nonclassical light at strong coupling . thermal photon statistics , on the other hand , requires weak coupling or high temperatures : it is the exception rather than the rule at low temperatures . the photon statistics , and to a lesser degree also the total emission , is strongly influenced by the presence of counter - rotating light - matter interaction terms in the hamiltonian . these terms are responsible for the prevalence of super - poissonian over sub - poissonian light at ultrastrong coupling . not surprisingly , the convenient rotating - wave approximation ( i.e. , identification of the dicke by the tc limit ) gives the wrong prediction when the coupling becomes too large . nevertheless , the scenarios with and without counter - rotating terms are surprisingly similar at not too strong coupling , which shows that generation of nonclassical light is not a peculiar effect arising from the fine - tuning of interaction terms in the hamiltonian but a rather robust feature . we have provided an approximate rule to relate the emission of few emitters to the emission of a single emitter , under appropriate scaling of the coupling constant . in accordance with this rule , the features of the glauber function observed for one emitter occur at comparably smaller values of the individual emitter - cavity coupling in the case of a few emitters . the reason is that all emitters interact with the same cavity mode , which magnifies the effects of resonant emission and ( re-)absorption of cavity photons . broadly speaking , generation of nonclassical light is easier with more emitters because the required coupling of each individual emitter to the cavity mode can be reduced . our analysis of strong light - matter coupling required use of the full input - output formalism and of the full master equation , which carefully distinguishes between transitions at different energies . if this correct treatment is replaced by the standard quantum optical master equation results change completely . especially , the prediction of nonclassical light does not survive the additional approximations made in the replacement . while the quantum optical master equation could not be expected to work at strong coupling , its outright failure at describing any of the distinctive features observed in the photon statistics shows that use of the right master equation is essential in all situations , perhaps apart from extremely weak coupling . the price one has to pay is full diagonalization of the hamiltonian . we here focus on the system at thermal equilibrium . future work should address emission if the system is driven coherently through external photon sources . this will require addition of explicitly time - dependent periodic terms to the hamiltonian , and thus combination of the present master equation with the floquet formalism . by contrast , a perturbative expansion in the driving strength is sufficient only for weak off - resonant driving , but then the possible new effects would be weak too . this work was supported by deutsche forschungsgemeinschaft through sonderforschungsbereich 652 project b5 . we follow standard input - output theory @xcite . the interaction hamiltonian in eq . for the cavity - environment coupling in the continuum limit is @xmath167 where @xmath168 is the environment density of states and @xmath169 the cavity - environment coupling function , i.e. , the environment spectral function is @xmath170 . @xmath171 together with the free hamiltonian @xmath172 of the environment photons and the commutator @xmath173 = \delta(\omega - \omega')$ ] lead to the equation of motion @xmath174 for the field quadratures of the environment . for @xmath175 , the formal solution of eq . is @xmath176 we define input ( output ) field operators @xmath177 and make use of the spectral function @xmath40 to obtain the input - output relation @xmath178 where @xmath56 denotes the positive frequency component of @xmath179 , i.e. , @xmath56 acts as a lowering operator . the explicit definition of @xmath56 in the system - energy eigenbasis is given in eq . . we consider the dissipative dynamics of the system density matrix in the weak system - environment coupling limit . for strong coupling within the system the quantum optical master equation predicts unphysical emission from the ground state @xcite . going one step back in the derivation of the quantum optical master equation , the second - order time - convolutionless projection operator method @xcite gives a time - local master equation leading to consistent results including the counter - rotating terms @xcite . nevertheless , this master equation does in general not generate positive dynamics @xcite . this problem was resolved by a recently derived master equation in the system eigenbasis @xcite and we here recapitulate its derivation . the total hamiltonian is the sum of the contribution of the system , @xmath26 , the contribution of the reservoir , @xmath180 , and the interaction @xmath171 . we note that the interaction hamiltonian in eq . is of the general form @xmath181 , where @xmath15 ( @xmath182 ) is a hermitian system ( reservoir ) operator . a more general coupling @xmath183 can also be considered , but leads to the same qualitative results . the dynamics of the density operator @xmath184 of the total system in the interaction picture is described by the von neumann equation @xmath185 \,.\ ] ] as a notational convenience , we mark operators in the interaction picture with a hat . the interaction hamiltonian and the density operator in the interaction picture are defined as @xmath186 where the time evolution operator of the uncoupled system and reservoir is @xmath187 in the limit of weak system - reservoir coupling several approximations are performed . first of all , within the born approximation initial factorization of the density operator is assumed , @xmath188 , and the back - action of the system onto the reservoir is neglected , @xmath189 . secondly , the markov approximation is performed by replacing @xmath190 at retarded times @xmath61 with @xmath50 at the local time @xmath165 . in the third place , assuming that the reservoir correlation time is small compared to the relaxation time of the system , the time integration is extended to infinity to arrive at the born - markov equation of motion @xmath191 \big ] \big\ } { { \mathrm d}}\tau \,,\ ] ] where @xmath192 denotes the partial trace over the reservoir degrees of freedom and @xmath193 is assumed . we further assume a thermal reservoir state @xmath194 and define the reservoir correlation function @xmath195 to evaluate the traces in eq . . this yields the master equation @xmath196 c(\tau ) { { \mathrm d}}\tau + \text{h.c . } \,,\ ] ] where h.c . denotes the hermitian conjugate . we introduce the transition operators in eq . that are the discrete fourier components of the interaction picture @xmath197 , i.e. , @xmath198 equivalently , @xmath199 = -\omega s_\omega$ ] . in addition , we introduce the even and odd fourier transforms of the reservoir correlation function @xmath200 for a thermal photon reservoir with spectral function @xmath201 the functions @xmath28 and @xmath29 are given in eqs . and . with these definitions we find @xmath202 \nonumber\\ & & + \text{h.c . } \,.\end{aligned}\ ] ] eq . is the standard born - markov master equation in the system energy - eigenbasis . it contains the dissipative parts proportional to @xmath28 and the lamb - shift terms proportional to @xmath29 . because eq . is not of lindblad type , it does , in general , not preserve the positivity of the density operator . inspecting eq . we recognize that it contains oscillating terms proportional to @xmath203 . if we assume that the relaxation of the system is slow compared with all oscillations @xmath203 we can neglect the contribution from terms with @xmath204 . this approximation is called secular or rotating - wave approximation and the master equation in the schrdinger picture simplifies to the result given in eq . . this equation is the lindblad master equation that includes the lamb shift of the unperturbed system energies @xmath205 as well as reservoir induced dissipation effects to lowest order in the system - reservoir interaction strength . as is already known in the literature , special care has to be taken if the spectrum of @xmath26 is degenerate @xcite . but even if the eigenvalues @xmath205 are non - degenerate we may have situations where energy differences are degenerate , i.e. @xmath206 for @xmath207 . the consequences of these two different types of degeneracy can be understood when we decompose the density matrix into blocks . in particular , we write @xmath208 ( @xmath209 ) for the matrix containing the elements @xmath210 ( @xmath211 ) with @xmath212 and @xmath213 . the master equation in this block notation reads @xmath214 where the summations are over different system energies only , and the complex function @xmath215 is introduced . we see that the last two lines in this equation are block - diagonal . for @xmath216 , the kronecker - delta in the first line evaluates to @xmath217 such that diagonal blocks couple to diagonal blocks , only . for @xmath218 , the first line contains terms with @xmath219 only , such that non - diagonal blocks do not couple to diagonal ones . nevertheless , a non - diagonal block @xmath220 couples to another non - diagonal block @xmath221 with @xmath222 if the respective transition energies are degenerate . thus , energy level degeneracy introduces a block structure implying that a diagonal density matrix element couples to non - diagonal elements within diagonal blocks whereas energy transition degeneracy leads to a coupling of non - diagonal blocks to different non - diagonal blocks . we remark that both subtleties have their origin in the rotating wave approximation . on the one hand , this approximation leads to the lindblad structure of eq . . on the other hand , it results in strict kronecker delta s between the two transition energies @xmath223 and @xmath72 . consider a situation where each degeneracy in the spectrum of @xmath26 as well as in their differences is lifted by a small @xmath224 parameter . then , each block contains a single element only , implying that the equations for the diagonal density matrix elements no longer couple to non - diagonal elements . in addition , any non - diagonal element of the density matrix evolves independently from all other elements . this behavior does not change when we let each @xmath225 . in this limit , the equations become independent of the @xmath224 parameters , but are different from the @xmath226 case . in particular , for every non - zero @xmath225 we get the two equations and for the diagonal and non - diagonal density matrix elements . we remark that in real physical systems one will never have perfectly equal or equidistant energies because each small perturbation will lift the degeneracies . in the theoretical description we may argue that the lamb shift lifts degeneracies . nevertheless , we have to keep in mind , that with eqs . to we can not study effects that rely on degenerate energies or degenerate transitions , e.g. the perfectly harmonic oscillator or a system composed of completely uncoupled identical subsystems are not correctly described . in this section we derive analytical results for the glauber @xmath126-function in the tc limit ( @xmath227 ) for a single emitter ( @xmath89 ) . according to the argumentation in sec . [ sec : stats_scale ] , the dominant contribution to the denominator of @xmath126 in eq . at low temperatures is that of the first excited state with energy @xmath228 . specifically , the denominator @xmath125 is approximated by @xmath229 in this expression the exponential @xmath230 is the thermal population of the first excited state and the prefactor @xmath231 is the squared transition matrix element of @xmath56 between the first excited state and the ground state . the most relevant state for the numerator of @xmath126 at low temperatures is the lowest eigenstate with energy @xmath232 of the @xmath233 matrix given in eq . . to evaluate the matrix elements of the operators @xmath234 we have to consider the four possible transition sequences @xmath235 , where @xmath236 denotes the ground state and @xmath237 are the two eigenstates with energies @xmath238 . this yields the expression @xmath239 ^ 2 ( \omega_0 - g)^2 \nonumber\\ & & + \frac{1}{4 } \big [ \omega_0 - ( \sqrt{2 } - 1 ) g \big ] ( \omega_0 ^ 2 - g^2 ) \big [ \omega_0 - ( \sqrt{2 } + 1 ) g \big ] \nonumber\\ & & + \frac{3 - \sqrt{8}}{8 } \big [ \omega_0 - ( \sqrt{2 } + 1 ) g \big]^2 ( \omega_0 + g)^2 \bigg\ } { \mathrm e}^{-\beta ( 2 \omega_0 - \sqrt{2 } g ) } \nonumber\\\end{aligned}\ ] ] approximating the numerator of @xmath126 . ( color online ) analytical results for the glauber function @xmath126 for one emitter ( @xmath107 ) , as a function of temperature @xmath113 and coupling strength @xmath10 in the tc limit . panel ( a ) shows the result following from eqs . and , whereas in panel ( b ) the result is refined for @xmath240 as explained in the text.,title="fig : " ] ( color online ) analytical results for the glauber function @xmath126 for one emitter ( @xmath107 ) , as a function of temperature @xmath113 and coupling strength @xmath10 in the tc limit . panel ( a ) shows the result following from eqs . and , whereas in panel ( b ) the result is refined for @xmath240 as explained in the text.,title="fig : " ] the results belonging to eqs . and are plotted in fig . [ fig : g2_chart_ana](a ) . compared to the exact numerical results in fig . [ fig : g2_chart_ext](b ) a good agreement appears for @xmath241 and low temperatures . for high values of @xmath242 the first excited state becomes closer and closer to the ground state such that the finite temperature leads to significant contributions from transitions not involving the ground state . for this reason , the upper part of fig . [ fig : g2_chart_ana](a ) is not well reproduced . in contrast , the lower part of fig . [ fig : g2_chart_ana](a ) is not in accordance with the exact numerical results because our assumption of low temperatures @xmath243 does not include the limit @xmath244 . to improve the results in regions with @xmath240 we additionally have to take into account the transition sequences @xmath245 for the denominator and @xmath246 for the numerator . the result is shown in fig . [ fig : g2_chart_ana](b ) , where the agreement to the exact results in fig . [ fig : g2_chart_ext](b ) is now very good for all temperatures and @xmath247 . note that at @xmath248 a crossing of the eigenvalues of the second and third excited state occurs , as can be seen in fig . [ fig : e](d ) . this indicates that the role of these states in the calculation of @xmath126 is interchanged . the impact of these eigenvalue - crossings would analytically be reproduced if we include contributions to @xmath126 from higher excited states . 47ifxundefined [ 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 _ _ , , vol . ( , , ) @noop _ _ ( , , ) @noop * * , ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( )

the number of astrophysical sources thought to possess surface magnetic fields above the quantum critical field strength of g has been steadily growing in recent years . the five presently known soft gamma - ray repeaters ( sgrs ) and six or seven anomalous x - ray pulsars ( axps ) are believed to be magnetars , neutron stars having surface fields in the range @xmath0 g ( duncan & thompson 1992 ) . in addition , the parkes multibeam survey ( manchester et al . 2001 ) has discovered at least six new radio pulsars that have surface fields near or above critical , and several with fields comparable to those of some of the magnetars . the radiation processes that determine the observed x - ray and gamma - ray spectra in these sources are operating in the extreme relativistic and quantum regimes and thus require treatment that is accurate in such environments . one process that is particularly important in the source emission models is resonant compton scattering , in which electrons scatter photons at the cyclotron resonance with a cross section much larger than at continuum energies . relativistic electrons can blue - shift low energy photons into the resonance , upscattering the photons into a high - energy continuum ( daugherty & harding 1989 , dermer 1990 ) . the quantum electrodynamical ( qed ) cross section for compton scattering at the cyclotron fundamental ( herold 1979 ) and higher harmonics ( daugherty & harding 1986 , bussard , alexander & mszros 1986 ) , accurate in arbitrarily high magnetic fields , has been known for some time . however , these derivations do not treat the natural line widths that render the cross section finite at the cyclotron resonances . in order to use such rates in spectral calculations in astrophysical models , the line widths that originate from the lifetimes of the excited landau states must be included in the cross section ( wasserman & salpeter 1980 ) . the width of the @xmath1th resonance is equal to the cyclotron decay rate from that state ( pavlov et al . 1991 ) , and the prescription for incorporating the widths in the qed cross section in the high - field regime has been discussed by harding & daugherty ( 1991 ) and graziani ( 1993 ) . including resonant line widths in the scattering cross section necessarily requires spin - dependent decay rates , which appear in an infinite sum over landau state @xmath1 and spin of the intermediate virtual states , even in the case of ground state - to - ground state scattering in the fundamental . different spin states have different decay rates , and thus different resonant energy denominators . since the electrons and positrons in intermediate states have non - zero momentum parallel to the magnetic field , it is important to use basis states that yield a spin dependence that is lorentz invariant , i.e. boosts along the magnetic field do not lead to a mixing between the spin states . as is true for all quantum processes , the spin - dependent rates and cross sections depend on the choice of electron wavefunctions in a uniform magnetic field . historically , several choices of wavefunctions have been used in calculations of the scattering cross section and cyclotron decay rates . the two most widely used wavefunctions are those of johnson & lippman ( 1949 ) and sokolov & ternov ( 1968 ) . the johnson & lippman ( jl ) wavefunctions are derived in cartesian coordinates and are eigenstates of the kinetic momentum operator . the sokolov & ternov ( st ) wavefunctions , specifically their `` transverse polarization '' states , are derived in cylindrical coordinates and are eigenfunctons of the magnetic moment operator . given the different spin dependence of the st and jl eigenstates , one must use caution in making the appropriate choice when treating spin - dependent processes . herold , ruder & wunner ( 1982 ) and melrose and parle ( 1983 ) have noted that the st eigenstates have desirable properties that the jl states do not possess , such as being eigenfunctions of the hamiltonian including radiation corrections , having symmetry between positron and electron states , and diagonalization of the self - energy shift operator . as found by graziani ( 1993 ) , the st wavefunctions also diagonalize the landau - dirac operator , and are the physically correct choices for spin - dependent treatments and for incorporating widths in the scattering cross section . although the spin - averaged st and jl cyclotron decay rates are equal , the spin - dependent decay rates are not , except in the special case in which the initial momentum of the electron parallel to the field vanishes . perhaps the most fundamental argument against use of the jl wavefunctions is that radiative corrections cause excited jl landau levels to become unstable to spin - flip transition within landau states on timescales comparable to the timescale for decay to a lower landau state . these radiative corrections to the magnetic moment break the spin degeneracy of both the st and jl excited landau states ( herold , ruder & wunner 1982 ) , causing spin - dependent energy level shifts , and implying a dependence of the rates for spin - flip transitions within a landau state on the choice of wavefunctions . the rates of such spin - flip transitions within split st landau states have been evaluated ( geprgs et al . 1994 ; see also parle 1987 ) , and are found to be of order of for , which is negligible compared to decay rates between landau states . the existence of such relatively long - lived states is a premise of the s - matrix formalism and is also essential to any astrophysical calculations involving spin - dependent transitions between landau states . in contrast , it is anticipated that since the jl eigenstates do not diagonalize the self - interaction hamiltonian including radiation corrections ( i.e. the mass operator ) , the associated mixing incurred in s - matrix evaluations will render spin - flip transition rates with fixed comparable to ordinary cyclotronic rates with changes in , a situation that is unphysical . in this paper we discuss the lorentz tranformation characteristics of cyclotron decay rates for both st and jl formulations , and derive simplified expressions for the decay rates from an arbritrary excited landau state to the ground state . we show that the st eigenstates preserve separability of the spin dependence under lorentz boosts along the local magnetic field , a desirable property that does not extend to the jl formalism , for which such lorentz boosts mix the corresponding spin states . in the st formulation , by taking advantage of its spin - state preserving characteristics , our analytic simplifications can be elegantly applied to the spin - dependent decay rates ; in the jl formalism the simplifications can be compactly applied only to the spin - averaged rates . the resulting expressions replace integrals over emergent photon angle with series of legendre functions of the second kind , which correspond to sums of elementary functions , easily yielding simple asymptotic forms . the expressions derived here should have wide applicability to modeling cyclotron emission , compton scattering and other qed processes in super - critical fields . for such magnetar - type fields , resonant scattering takes place primarily at the fundamental , since the cyclotron energy exceeds 1 mev . although the resonant scattering line widths formally involve infinite sums over landau states , in the case of the fundamental resonance the sum is dominated by the @xmath2 state , whose width is equal to the @xmath3 cyclotron decay rate . for cyclotron scattering in higher harmonics @xmath4 , the intermediate sums in the vertex functions have the largest contributions from the @xmath5 state . however , for , cyclotron transitions to the ground landau state dominate ( sokolov , zhukovskii & nikitina 1973 , white 1974 , harding & preece 1987 ) , so that cyclotron decay rates for @xmath3 transitions treated in this paper should be good approximations to the widths of excited states , a circumstance pertinent to astrophysical models of magnetars . this presentation of spin - dependent cyclotron rates appropriately focuses first on the formulation generated using the _ transverse polarization _ eigenstates of the dirac equation derived by sokolov & ternov ( 1968 ) for electrons in a uniform magnetic field * b*. explicit forms for them can also be found in herold , ruder & wunner ( 1982 ) and harding & preece ( 1987 ) , whose exposition on cyclotron emission essentially forms the basis of results developed here . herold et al . observed that these states diagonalize the operator that describes the electronic self - energy shift in an external magnetic field . another attractive feature of the sokolov & ternov eigenstates is that they possess charge conjugation symmetries , i.e. between electron ( positive energy ) and positron ( negative energy ) : see eq . ( 12 ) of herold , ruder & wunner ( 1982 ) . an additional asset of these wavefunctions that is enunciated here is that they yield cyclotron rates whose spin dependence is effectively separable from lorentz transformations along the field : i.e. such boosts do not mix spin states . this is a very useful characteristic that is not present in the johnson & lippmann formalism addressed in section [ sec : jlformalism ] below . herold , ruder & wunner ( 1982 ) obtained general expressions in their eq . ( 17 ) for the spin - dependent cyclotron rate for transitions between arbitrary landau levels , but for the case of zero initial momentum parallel to the field . latal ( 1986 ) independently obtained similar results for cyclotron transitions to the ground state , but retained arbitrary initial momenta along * b * ; his eq . ( 22 ) forms the starting point for the exposition here . denoting as the spin quantum number of the initial electron , ground state transitions are characterized by a single final spin . the energy level quantum numbers are initially , and after the transition . latal s spin - dependent _ total _ rates can be written in the form @xmath6 where is the magnetic field strength , expressed dimensionlessly hereafter in units of the quantum critical field gauss , and the _ spin - averaged _ , total cyclotron transition rate for is given by @xmath7 \label{eq : gammave_st}\\[-5.5pt ] & \times & \ ; \dover{(1-\xi)^{n-1}}{(1+\xi)^{n+1}}\biggl ( 1 + 2nb - \dover{1}{\xi } \biggr ) \ ; \exp \biggl ( - n\ , \dover{1-\xi}{1+\xi } \biggr ) \ ; .\nonumber\end{aligned}\ ] ] here is the compton wavelength of the electron over , and other quantities in this equation are defined as follows . for initial and final dimensionless momenta ( i.e. , in units of ) , and respectively , parallel to the field , the electron s initial energy and final energy , both dimensionless ( i.e. , in units of ) , are @xmath8 this convention of using dimensionless energies and momenta will be adopted throughout the paper . the integration variable is the angle cosine of the emitted photon with respect the magnetic field direction , i.e. for photon wavenumber vector . no integration by parts has been performed when obtaining eq . ( [ eq : gammave_st ] ) , so that the cyclotron rate , differential in photon angles , , again averaged over initial electron spins , corresponds directly to the integrand of eq . ( [ eq : gammave_st ] ) : @xmath9 corresponding spin - dependent differential cyclotron rates can be deduced from eq . ( [ eq : gamma_st ] ) in similar fashion . the remaining variable , , emerges naturally from energy - momentum conservation in the interaction . only momentum parallel to the field is conserved , since the system is not translationally - invariant orthogonal to * b*. hence cyclotron emission satisfies two such conservation relations : @xmath10 \label{eq : e_p_conserve}\\[-5.5pt ] p & = & q + \omega\mu\nonumber\end{aligned}\ ] ] for . here is the dimensionless photon energy , having been scaled by . the elimination of from the system in eq . ( [ eq : e_p_conserve ] ) leads to the identities @xmath11 \label{eq : omega_ident}\\[-5.5pt ] \omega ( 1-\mu^2 ) & = & ( 1-\xi ) \ , ( e_n - p\mu ) \nonumber\end{aligned}\ ] ] that are simultaneously satisfied , with @xmath12 being the specialization of the definition in eq . ( 11b ) of latal ( 1986 ) . the angle integration in eq . ( [ eq : gammave_st ] ) for the cyclotron rate is non - trivial , due largely to the complicated dependence of on , and the presence of the exponential . this completes the definitions relevant to eq . ( [ eq : gammave_st ] ) . however , we also note that the identities in eq . ( [ eq : omega_ident ] ) combine to yield @xmath13 which is precisely the argument of the exponential in eq . ( [ eq : gammave_st ] ) . this factor results from the specialization of associated laguerre functions that are formed from the spatial integration ( fourier transform ) of the eigenfunctions . since is lorentz invariant under boosts along * b * , so also is . furthermore , forming using eq . ( [ eq : e_p_conserve ] ) , and then eliminating using the second identity in eq . ( [ eq : omega_ident ] ) , quickly leads to the equivalence @xmath14 when weighted by top and bottom , this is a ratio of invariant products of four - momenta associated with the interaction ; i.e. is also a lorentz invariant for boosts along the field , which can be inferred directly from eq . ( [ eq : kappa_def ] ) . there is no need to compute eq . ( [ eq : gammave_st ] ) as it stands for arbitrary . instead , one can appeal to a simple lorentz transformation protocol . consider boosts along the field between the inertial frame where the initial electron possesses momentum along the field , and that frame where the initial parallel momentum is zero . in this latter `` rest '' frame let be the photon angle cosine with respect to * b * , be the initial parallel momentum , and the initial electron energy be denoted @xmath15 then , for a dimensionless boost velocity and lorentz factor along * b * , the initial electron quantities in eq . ( [ eq : gammave_st ] ) satisfy @xmath16 \label{eq : lorentz_boost}\\[-5.5pt ] p & = & \gamma ( p_0 - \beta \erg_n ) \;\equiv\ ; -\gamma\beta\erg_n\quad , \nonumber\end{aligned}\ ] ] from which is established . the photon angle cosine satisfies the aberration formula @xmath17 this latter relation identifies a suitable change of variables for the angle integration , for which . other transformation identities include and , from which the invariance in form of is established via the definition in eq . ( [ eq : xi_def ] ) . it then follows , after a modicum of algebra , that this pure change of variables leads to an alternative form for the rate in eq . ( [ eq : gammave_st ] ) : @xmath18 with @xmath19 where , now is obtained from eq . ( [ eq : xi_def ] ) using the correspondences , and . accordingly , this identifies attractive lorentz transformation behavior , with the rate reduced ( i.e. lifetime dilated ) by just the lorentz factor of the boost . moreover , such boosts along the field keep the spin states `` separated , '' i.e. there is no implied mixing of states incurred by such transformations . this inherent simplicity is an appealing characteristic of the sokolov & ternov states , and was identified by graziani ( 1993 ) ; it is not exhibited in the johnson & lippman formalism explored in sec . [ sec : jlformalism ] . clearly , this extraction of cases via a simple modification factor outside the integral is expedient for the subsequent analytic developments . while the integral expressions for the cyclotron rates can be routinely evaluated numerically , they can also be represented by compact analytic series in terms of elementary functions that are readily amenable to computation . the integration variable for in eq ( [ eq : inb_def ] ) is not the most convenient ; a more expedient choice for the purposes of analytic reduction is @xmath20 the integration is even in , and since , the integration range maps over to @xmath21 after a modest amount of algebra , the change of variables then leads to the form @xmath22 this integral can be evaluated in terms of appell functions , the degenerate , two - dimensional hypergeometric functions , using eq . 3.385 of gradshteyn & ryzhik ( 1980 ) . such a step does not facilitate evaluation , since appell function expositions usually develop double , infinite power series representations ( e.g. see burchnall & chaundy 1941 ; exton 1976 ) . analytic progress is quickly made via taylor series expansion of the exponential around . the order of integration and the infinite summation can then be interchanged because the integration is uniformly convergent on the interval , since is strictly less than unity . another change of variables , with the definition @xmath23 then establishes @xmath24 with @xmath25 using the identity 8.715.2 of gradshteyn & ryzhik ( 1980 ) gives an integral representation of the legendre function of the second kind : @xmath26 these special functions are finite sums of elementary logarithmic and polynomial functions of ( e.g. see abramowitz & stegun 1965 ) . note that the are generalizable to associated legendre functions , defined in 8.703 of gradshteyn & ryzhik ( 1980 ) , from which . it follows from eqs . ( [ eq : jzn_def ] ) and ( [ eq : qnu_def ] ) that @xmath27 and relevant properties of the and functions are listed in the appendix . computational issues for the are discussed there also , and an expedient approach for integer is to compute the using the recurrence relation in eq . ( [ eq : jrecurrence ] ) . the final result of these analytic reductions is @xmath28 which expresses the spin - averaged cyclotron rate as a comparatively simple series of elementary functions . for low values of , this series is rapidly convergent for all , typically requiring 34 terms to achieve 0.01% accuracy . such a rate of convergence renders the series evaluation computationally much more efficient than a numerical integration of eq . ( [ eq : gammave_st ] ) by simpson s rule or a quadrature technique . the appearance of the legendre functions of the second kind , , in the series expansion for the rate is a result of the cylindrical symmetry imposed by the presence of the external magnetic field . evaluations of eq . ( [ eq : gammave_st_red ] ) are presented in fig . [ fig : cyc_st_ave ] for different values of and for , demonstrating the well - known rapid increase of the rate with when . the dependence on is relatively weak for supercritical fields . the figure also clearly exhibits the property , which can , with extensive effort , be deduced from the integral form for the spin - averaged rate in eq . ( [ eq : gammave_st_alt ] ) . this decline of with increasing is evident in the asymptotic approximations for and derived shortly . [ fig : cyc_st_ave ] also exhibits the logarithm of the ratio @xmath29 of the spin - dependent rates for ; higher cases display similar behavior . this curve illustrates how the rate is approximately independent of initial spin when , whereas initial states lead to much faster transition rates ( by of order ) when , reflecting the well - known dominance of non - spin - flip cyclotron transitions by non - relativistic electrons ( e.g. see melrose & zheleznyakov 1981 ) . now consider the two pertinent asymptotic limits for the spin - averaged rate in eq . ( [ eq : gammave_st_red ] ) . when , and . using the first identity in eq . ( [ eq : j_nu_ident ] ) , eq . 8.771.2 of gradshteyn & ryzhik ( 1980 ) can then be invoked to obtain the leading order dependence of , noting the specialization of the common hypergeometric function . the result , for general arguments , is @xmath30 here , represents the gamma function . this is the leading order ( ) term in the series evaluation of , and hence this result is readily checked by taking the limit of eq . ( [ eq : inb_alt ] ) . the doubling formula for the gamma function , in eq . 8.335.1 of gradshteyn & ryzhik ( 1980 ) , can now be used to yield an asymptotic form @xmath31 which is just eq . ( 31d ) of latal ( 1986 ) . this form illustrates the strong dependence of the cyclotron rates on both and . a sample case of the limit in eq . ( [ eq : gammave_lownb ] ) is displayed in fig . [ fig : cyc_st_ave ] . this limit can only be compared with the well - studied classical cyclotron limit in a restricted fashion , since it is still an essentially quantum result . moreover , it incorporates the intrinsically relativistic effect of spin - orbit coupling . observe that eq . ( [ eq : gammave_lownb ] ) , in conjunction with eq . ( [ eq : gamma_st ] ) , reproduces both the non - spin - flip ( ) and spin - flip ( ) st rates encapsulated in eqs . ( 18 ) and ( 21 ) , respectively , of melrose & russell ( 2002 ) , or equivalently , the specialization of eq . ( 18 ) of herold , ruder & wunner ( 1982 ) . classical formulations of cyclotron emissivities ( e.g. see eq . ( 6.19 ) of bekefi 1966 ) yield rates that depend on components of the electron velocity perpendicular to the field : . here the classical rate effectively represents the limit of a sum over various transitions , with . by invoking the correspondence , it is quickly established that , the same dependence as in eq . ( [ eq : gammave_lownb ] ) . yet in general , the classical does not equal the quantum , since here contributions are omitted from consideration . however , in the particular case of , there is only a single final state , namely , and with the prescription , eq . ( 6.19 ) of bekefi ( 1966 ) specializes to yield , which is precisely the value obtained in eq . ( [ eq : gammave_lownb ] ) . derivation of the , ultra - quantum limit is slightly more involved . in this situation , and so that the entire series in eq . ( [ eq : gammave_st_red ] ) is retained . the key useful result is derived in the appendix , in eq . ( [ eq : j_n_taylor_zeq1 ] ) , which establishes as . including the next order logarithmic contribution then leads to an approximate form for in eq . ( [ eq : inb_taylor_series ] ) , so that it then follows that @xmath32 here , is the incomplete gamma function in 8.350.1 of gradshteyn & ryzhik ( 1980 ) . note that the same result can be derived directly from the limit of eq . ( [ eq : inb_alt ] ) using the integral definition of the incomplete gamma function . the weak dependence of the rate on in this limit is now apparent , and is evinced in fig . [ fig : cyc_st_ave ] by the crowding of curves for supercritical fields . in the limit of , further specialization is possible using the limit as , derived in the appendix , to yield the approximation @xmath33 this ultra - quantum result corresponds to the limit given in eq . ( 32a ) of latal ( 1986 ) , though the numerical coefficient of latal s does not agree exactly with the factor in eq . ( [ eq : gammave_highnb_too ] ) , the difference being around 10% . however , the expressions for quantum cyclotron transition rates derived in eq . ( 9 ) of sokolov , zhukovskii & nikitina ( 1973 ) and eq . ( 2.56 ) of white ( 1974 ) reduce exactly to eq . ( [ eq : gammave_highnb_too ] ) in the limit , confirming the asymptotic analysis here . a sample case of this limit is displayed in fig . [ fig : cyc_st_ave ] , the precision of which is around 2% relative to the exact result , and better than eq . ( 32a ) of latal ( 1986 ) . the first order correction to this asymptotic limit that would follow from the logarithmic term in eq . ( [ eq : gammave_highnb ] ) is relative to the leading order contribution in eq . ( [ eq : gammave_highnb_too ] ) . in the frame , the spin - averaged differential cyclotron rates can be immediately extracted from eq . ( [ eq : gammave_st_alt ] ) using the prescription implied by eq . ( [ eq : gammave_diff_def ] ) . however , given that eqs . ( [ eq : omega_ident ] ) and ( [ eq : xi_def ] ) define a one - to - one correspondence between the angle cosine and the energy of the emitted cyclotron photon , the differential rates can suitably be presented as differential photon spectra , . the photon energy is , so that the kinematic bound translates to @xmath34 the change of variables is then defined by two branches , @xmath35 with the factors inside the square root reflecting the kinematics associated with both the cyclotron radiation and the one - photon pair creation interactions . this has a jacobian , leading to the spin - averaged differential spectrum @xmath36 \label{eq : dgammave_domega_st}\\[-5.5pt ] & \times & \dover{(\erg_n-\omega ) nb-\omega } { \sqrt{(\erg_n-1-\omega ) ( \erg_n+1-\omega ) } } \ ; \exp \bigl ( - n\ , \dover{\omega}{\omega_- } \bigr ) \ ; .\nonumber\end{aligned}\ ] ] this spectral rate is summed over the polarizations of the radiated photons . note that a factor of has been included to account for the two branches of the relation between and . observe also that this spectrum can be obtained directly from eq . ( [ eq : gammave_st_alt ] ) and the form for in eq . ( [ eq : inb_alt ] ) by using the correspondence . key characteristics of the differential spectrum include its broadening and shifting to higher energies as increases , behavior that is largely governed by the kinematic result in eq . ( [ eq : omega_bounds ] ) . an exponential decline is present in cases until a sharp kinematic peaking at overtakes it . in cases , this exponential contribution is dominated by the power - law factor , leading to a steeply - rising spectrum . when , nearly all of the radiative power in the spectrum is confined to energies . in the opposite domain , , the spectra are extremely narrow , and little insight is gained by attempting an involved comparison with the classical limit ; most classical results are rendered for sums over a multitude of transitions . to conclude the formalism based on sokolov & ternov states , expressions for the cyclotron energy loss rate , or power , can be expeditiously presented . the development of such rates parallels that for , with the inclusion of an extra ( dimensionless ) weighting factor to the differential decay rate . hence , one can quickly write down @xmath37 & = & \dover{\fsc m_ec^3}{\lambar}\ , \dover{n^n}{(n-1)!}\ , \dover{b}{e_n } \int_{-1}^1 \dover{\omega \ , d\mu}{(e_n - p\mu)^2 } \label{eq : powerave_st}\\[2pt ] & \times & \ ; \dover{(1-\xi)^{n-1}}{(1+\xi)^{n+1}}\biggl ( 1 + 2nb - \dover{1}{\xi } \biggr ) \ ; \exp \biggl ( - n\ , \dover{1-\xi}{1+\xi } \biggr ) \ ; .\nonumber\end{aligned}\ ] ] as the spin - averaged power , using eq . ( [ eq : gammave_st ] ) . this can be conveniently transformed to the integration variable , representing the frame , using an analysis similar to that presented in sec . [ sec : st_lorentz ] . the only change is the manipulation of the additional factor inside the integration , for which the lorentz transformation relations in eqs . ( [ eq : lorentz_boost ] ) and ( [ eq : aberration ] ) quickly yield . since is even in , it follows that the term proportional to in the transformed integrand of eq . ( [ eq : powerave_st ] ) does not contribute . therefore , comparison with eqs . ( [ eq : gammave_st_alt ] ) and ( [ eq : inb_def ] ) then indicates that @xmath38 is the form for the cyclotron power for general , with @xmath39 where is given by eq . ( [ eq : phi_def ] ) . this expression for the power is identical to that in eq . ( 28 ) of latal ( 1986 ) , who illustrated the magnetic field and dependence in his fig . 1 . the spin - dependent powers are , of course , just given by additional multiplicative factors : for . observe that the dependence has now disappeared from the analysis , highlighting the fact that is a lorentz invariant for boosts along * b*. this follows from the exact compensation of time dilation and photon redshifting effects when integrating over all angles ; such an attractive invariance property for the total power clearly does not extend to its differential counterpart . the integral for can be manipulated in similar fashion to that for . since the only difference is an extra factor , which translates to a factor in the light of eq . ( [ eq : phi_def ] ) , it follows that can be worked into the form of eq . ( [ eq : inb_alt ] ) , but with the extra factor . the subsequent developments in sec . [ sec : st_analytic ] rapidly lead to the identity @xmath40 \label{eq : knb_alt_too}\\[-5.5pt ] & \equiv & \sum_{k=0}^{\infty } \dover{(-n)^k}{k!}\ , \biggl ( 1 - \dover{k}{n } \biggr ) j_{n+k}(z_n ) \;\ ; , \nonumber\end{aligned}\ ] ] for insertion into eq . ( [ eq : powerave_st_alt ] ) . given this form , asymptotic limits can be routinely obtained by replicating the approaches of sec . [ sec : stasymptotics ] . when , it is evident that , since then , as can be inferred for from eq . ( [ eq : j_n_lownb ] ) . furthermore , for , the result in eq . ( [ eq : j_n_taylor_zeq1 ] ) quickly yields a leading order contribution . then the identity and eq . ( [ eq : gamma_inc_limit ] ) can be used to derive the limiting form as . the results are @xmath41 while the case here replicates that inferred from eq . ( 31 ) of latal ( 1986 ) , the result differs from the numerical result in eq . ( 32a ) of latal ( 1986 ) by around 10% . the mean cyclotron photon energy is just the ratio of to , so that one quickly deduces from a comparison of eq . ( [ eq : powerave_asymptotics ] ) with eqs . ( [ eq : gammave_lownb ] ) and ( [ eq : gammave_highnb_too ] ) that when , as expected from classical cyclotron theory , and that when . note that latal ( 1986 ) misses the factor ( see his eq . ( 32b ) for the domain ) due to the imprecise nature of his estimates of the pertinent numerical factors in and . these behaviors of are reflected by the differential spectral form in eq . ( [ eq : dgammave_domega_st ] ) , where for the spectrum is essentially a delta function pinned at the cyclotron harmonic energy , and where for and the spectrum is somewhat broad but markedly asymmetric , skewed strongly towards . a complementary formulation of the cyclotron problem was provided by daugherty & ventura ( 1978 , hereafter dv78 ) using wavefunctions derived by johnson & lippmann ( 1949 ) . this is essentially the foremost presentation in the literature of cyclotron radiation calculations that specifically uses the johnson & lippmann eigenstates , hereafter referred to as jl states . these states are eigenvalues simultaneously of the dirac hamiltonian and the kinetic momentum operator , where is the vector potential , and therefore are not eigenstates of any physically meaningful spin operator ( see melrose & parle 1983 ) . the relative lack of symmetry in the jl states between the electron and positron wavefunctions leads to more complicated lorentz transformation characteristics for the cyclotron rates for boosts along the magnetic field , as shall become evident below . the spin - averaged rate should ( and does ) , of course , reduce to the sokolov & ternov formalism result in eq . ( [ eq : gammave_st ] ) . for direct comparison with the sokolov & ternov formalism results above , it is necessary to generalize the cyclotron emission results in the appendix of dv78 to treat arbitrary transitions , specifically their eqs . ( a2 ) and ( a3 ) . this is a routine s - matrix calculation in quantum electrodynamics , and essentially parallels the cyclotron absorption calculations presented in sec . ii of dv78 . the reader is therefore referred to dv78 for relevant background and some definitional material , and here only a detailed outline of the derivation is provided . wherever possible , the same notation as used in the st formulation above is adopted . the starting point is eq . ( a1 ) of dv78 , which can be integrated over photon solid angles to yield the spin - dependent cyclotron rate @xmath42 where is the density of photon states in a box of size ; this size establishes the normalization of the wavefunctions . note that is dimensionless because of the adopted convention of dimensionless photon energies . a sum over the photon polarization states has been effected , with the standard convention for the labelling of the photon polarizations being adopted : 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 . also , @xmath43 is the relevant integration over the final parallel momentum of the electron and the coordinate defining the mean location of the final electron wavefunction perpendicular to * b*. here is the normalizing time for the qed perturbation calculation . the central contribution to the integrand comes from @xmath44 the s - matrix element for cyclotron emission that is analogous to eq . ( 14 ) of dv78 that describes cyclotron absorption . here @xmath45 is the fourier transform of the spatial portion of the interaction amplitude , basically the vertex function of melrose and parle ( 1983 ) , but with the temporal phase factors already extracted . the initial electron wavefunction is , and can possess either spin designation ( note that this spin notation differs from the notation adopted by dv78 ) ; the final electron wavefunction , , is an , ground state , so that an explicit spin label is suppressed . expressions for these spatial parts of the johnson & lippmann eigenfunctions of the dirac equation are given eq . ( 10 ) of dv78 ; these do not exhibit the same charge conjugation symmetry that the sokolov & ternov `` transverse polarization '' states do . in eq . ( [ eq : vfi_def ] ) , is one of the two `` polarization matrices '' in eq . ( 15 ) of dv78 , representing contracted products of polarization vectors and the dirac matrices . since the and functions possess the dimensions of , it follows that and consequently and are dimensionless . observe also that the temporal integration over the wavefunction products has already been performed leading to the function in eq . ( [ eq : sfi_def ] ) that expresses energy conservation . finally , note that defines the mean location of the initial electron wavefunction perpendicular to * b * , i.e. is a counterpart to . the squaring of the matrix elements in eq . ( [ eq : phi_sigma_def ] ) is performed using the approach of , for example , bjorken & drell ( 1964 ) for handling the function . the result is @xmath46 in parallel with the developments leading to eq . ( 19 ) of dv78 . integrations analogous to those in eq . ( [ eq : vfi_def ] ) have been performed analytically in dv78 for the cyclotron absorption case , resulting in the appearance of associated laguerre functions that are characteristic of magnetized qed calculations involving either free electrons or electron propagators . here , since specialization to the case is made , these functions reduce to simple combinations of power - laws and exponentials , viz . , @xmath47 in the notation of eq . ( 18 ) of dv78 . as these spatial integrations embody translational invariance along the field , they yield a function expressing conservation of momentum parallel to * b * , i.e. establishing , and completing the conservation laws in eq . ( [ eq : e_p_conserve ] ) . a simple crossing symmetry manipulation of eq . ( 16 ) of dv78 facilitates the evaluation of eq . ( [ eq : vfi_def ] ) , and also eq . ( [ eq : phi_sigma_def ] ) , which can be compared closely with eq . ( 19 ) of dv78 . using the identity in eq . ( [ eq : omegasq ] ) , and reverting to the notation , a modest amount of algebra then yields relatively compact expressions for the spin - dependent cyclotron emission rate in the johnson & lippmann formalism : @xmath48 \label{eq : gamma_jl_too}\\[-5.5pt ] & \times & \ ; ( e_n+1)\ , ( e_0 + 1)\ , \bigl\ { \lambda^{\zeta}_{\parallel } + \lambda^{\zeta}_{\perp } \bigr\}\quad .\nonumber\end{aligned}\ ] ] the different photon polarization and electron spin combinations yield the following contributions : @xmath49 \label{eq : lambda_par}\\[-3.5pt ] \lambda^{-1}_{\parallel } & = & 2nb\ ; \biggl\ { \biggl ( \dover{q}{e_0 + 1 } + \dover{p}{e_n+1 } \biggr)\ , \dover{\omega ( 1-\mu^2)}{2nb } - \dover{\mu}{e_n+1}\ , \biggr\}^2 \nonumber \end{aligned}\ ] ] for the polarization , and @xmath50 \label{eq : lambda_perp}\\[-5.5pt ] \lambda^{-1}_{\perp } & = & \dover{2nb}{(e_n+1)^2 } \nonumber \end{aligned}\ ] ] for photons of polarization . it is now almost trivial to demonstrate that for the case , eq . ( [ eq : gamma_jl_too ] ) in concert with eqs . ( [ eq : lambda_par ] ) and ( [ eq : lambda_perp ] ) reduces to eqs . ( a2 ) and ( a3 ) of daugherty & ventura ( 1978 ) , when and , respectively . accordingly , eqs . ( [ eq : gamma_jl_too][eq : lambda_perp ] ) serve as a generalization of the cyclotron emission rates presented in dv78 to arbitrary transitions . the isolation of the photon polarization contributions in this development is an attribute that is absent from the exposition in the appendix of daugherty & ventura ( 1978 ) . such an isolation expedites the algebraic developments that reduce this form into expressions that much more closely resemble those in the sokolov & ternov formulation , i.e. eqs . ( [ eq : gamma_st ] ) and ( [ eq : gammave_st ] ) . the analytic reduction is routine , but lengthy , involving the expedient use of several identities . among these are eqs . ( [ eq : e_p_conserve][eq : xi_def ] ) , which can be used to quickly derive @xmath51 the latter of which is just a rearrangement of eq . ( [ eq : xi_altern ] ) . for the case , it is just a short path to yield the equivalences @xmath52 \label{eq : lambda_1_alt}\\[-5.5pt ] \lambda^{+1}_{\perp } & = & \biggl\ { -\dover{\omega}{e_n+1}\ ; \dover{(e_0 + 1)\mu -q}{e_0 + 1 } \biggr\}^2\quad , \nonumber\end{aligned}\ ] ] and for , a moderate amount of work to yield @xmath53 with this assembly of tools , and various algebraic manipulations , one is lead to the following final forms for the spin - dependent cyclotron rates in the johnson & lippmann formulation : @xmath54 where eq . ( [ eq : kappa_def ] ) gives in terms of , and the spin - dependent factors in the integrand are @xmath55 \label{eq : x_def}\\[-5.5pt ] x_{-1 } & = & 2 \bigl ( 1 + 2nb - \dover{1}{\xi } \ , \bigr ) - x_{+1}\quad .\nonumber\end{aligned}\ ] ] these are clearly forms resembling the integral that latal ( 1986 ) derived , though indicating a slightly more complicated differential distribution than for the st case . the average of the factors is obviously @xmath56 from which it is evident that the spin - averaged cyclotron rate for this johnson & lippmann formulation is identical to that of the sokolov & ternov formalism , i.e. eqs . ( [ eq : gamma_st ] ) and ( [ eq : gammave_st ] ) , as should be the situation . such degeneracy between formalisms also prevails for the power radiated when averaged over the initial electron spins . note that while the spin - dependent rates are generally different for the two sets of wavefunctions , in the limit , it is clear from eq . ( [ eq : x_def ] ) that so that the two formalisms become degenerate in this limit . this is a consequence of the mathematical identity of the sokolov & ternov and johnson & lippmann eigenstates in the special case of . as with the analysis in sec . [ sec : stformalism ] , it is instructive to discern the lorentz transformation behavior of the johnson & lippmann rates . such a development closely parallels that in sec . [ sec : st_lorentz ] , again writing the relevant integrals in terms of variables in the frame in which . with the identities and , it can be quickly established that @xmath57 which can be used also for . as for the cyclotron power considerations of sec . [ sec : power_st ] , since the spin - averaged integrand or is even in , the term proportional to in eq . ( [ eq : x_1_lorentz ] ) contributes exactly zero to the integral in eq . ( [ eq : gamma_jl_fin ] ) . this simplification clearly also applies to . it then follows that an alternative form for the rate in eq . ( [ eq : gamma_jl_fin ] ) is @xmath58 this form conveniently avails itself of the analytic developments of sec . [ sec : st_analytic ] , so that no further reduction of integrals is necessary : the evaluation of as a series of legendre functions of the second kind in eq . ( [ eq : inb_alt_too ] ) is immediately applicable . such a simplification can not be applied to the spin - dependent cyclotron powers , a point that is addressed below . the lorentz - transformed expression for the cyclotron rates in eq . ( [ eq : gamma_jl_final ] ) is attractively simple , but is not as elegant as its equivalent for the sokolov & ternov formulation , as embodied in eqs . ( [ eq : gamma_st ] ) and ( [ eq : gammave_st_alt ] ) . here , boosts along the field do not introduce simple time dilation factors of , essentially amounting to a mixing of jl spin states induced by such transformations . this less than ideal characteristic is a consequence of the fact that the jl wavefunctions are eigenstates of a somewhat convoluted spin operator that is not symmetric under charge conjugation ( e.g. , see melrose & parle 1983 ) , nor is it manifestly covariant under boosts along the field . in contrast , the st `` transverse polarization '' wavefunctions are eigenstates of the component along * b * of the magnetic moment operator ( here and are matrix vectors for the dirac algebra ) , and such eigenstates exhibit both appropriate symmetry between positrons and electrons , and simple transformation properties for boosts parallel to the field . the sensitivity of the spin - dependent cyclotron rates , and in particular their lorentz transformation characteristics , to the choice of spin eigenfunctions is naturally expected , since the inherently relativistic effect of spin - orbit coupling is explicitly incorporated in the dirac equation in an external field , thereby emphasizing the interpretative importance of the choice of wavefunctions . the greater elegance of the results in sec . [ sec : stformalism ] is yet another argument in favor of usage of sokolov & ternov states for the cyclotron problem , and therefore for other qed processes where spin - dependent cyclotron rates are required . note that the differential spectrum for spin - dependent considerations can be routinely obtained in a manner similar to the derivation of eq . ( [ eq : dgammave_domega_st ] ) , observing that can be replaced by in eq . ( [ eq : x_1_lorentz ] ) , or , as desired , and then using the substitution . the ratios of the jl rates in eq . ( [ eq : gamma_jl_final ] ) to their st counterparts are plotted in fig . [ fig : cyc_jl_st_ratio ] , being specifically given by @xmath59 in general , this ratio is less than unity for and greater than unity for ; it is also clearly a monotonically increasing function of for , and decreases monotonically with for . this ratio approaches unity for either ( dotted line ) or for , and in the case of also for . in the limit , the ratio approaches when , the maximum departure from unity in this case . in the domain , as , for the minimum in the ratio is and occurs for . the significant deviations from unity underline the importance of the choice of wavefunctions when retaining spin dependence . this becomes a profound issue for resonant compton scattering problems ( e.g. see graziani 1993 ) , where , as discussed above , the spin - dependent cyclotron lifetime is required to render the intermediate electron states metastable and thereby truncate the divergent resonance at the cyclotron frequency . it is also pertinent to handling resonances in magnetic coulomb scattering ( e.g. langer 1981 ) , two - photon pair creation in strong fields ( e.g. kozlenkov & mitrofanov 1986 ) , magnetic photon splitting above pair creation threshold ( discussed briefly in baring 2000 ) , and other general external field problems involving electrons either in free states or as propagators ( graziani , harding & sina , 1995 ) . to conclude this section , a brief exposition of the spin - dependent powers is offered . the spin - averaged power is , of course , that in eq . ( [ eq : powerave_st_alt ] ) . to retain spin dependence in the jl formalism , one proceeds by weighting the integrand of eq . ( [ eq : gamma_jl_fin ] ) with a factor . guided by the analysis in sec . [ sec : power_st ] , terms that are odd in contribute zero , and the by now routine manipulations yield the form @xmath60 \label{eq : power_jl_spins}\\[-3.5pt ] & \times & \hskip -5pt \biggl\ { \biggl\lbrack 1 - \dover{\zeta ( \gamma + \erg_n)}{\erg_n ( \gamma\erg_n+1 ) } \ , \biggr\rbrack\ , k_n(b ) + \zeta\ ; \dover{2\gamma\beta^2\ , ( nb)^3}{\erg_n^3 ( \gamma\erg_n+1)}\ , l_n(b ) \biggr\ } \ , , \nonumber\end{aligned}\ ] ] where is the integral in eq . ( [ eq : knb_def ] ) , or equivalently , the series in eq . ( [ eq : knb_alt_too ] ) , and the term emerges from a term in the integrand resulting from the product , and corresponds to @xmath61 eq . ( [ eq : power_jl_spins ] ) is clearly not as elegant as the corresponding st result , namely , but does reduce to this simpler form in the limit , as expected . furthermore , averaging eq . ( [ eq : power_jl_spins ] ) over spins then yields exactly , so that degeneracy between the jl and st formalisms is realized when spin information is not explicitly retained . note that asymptotic limits of for small and large can be readily obtained but are not particularly elucidating for the purposes of this exposition . this paper presents useful analytic developments of spin - dependent cyclotron decay rates / widths for general transitions between landau states , expressing these in terms of various series of legendre functions of the second kind . the rates and the radiated cyclotron powers are derived using two popular eigenstate formalisms appropriate for the dirac equation in a uniform magnetic field , namely that due to sokolov & ternov ( 1968 ) , and that due to johnson & lippmann ( 1949 ) . the resulting expressions are eq . ( [ eq : gammave_st_red ] ) for the cyclotron rate and eq . ( [ eq : powerave_st_alt ] ) in conjunction with eq . ( [ eq : knb_alt_too ] ) for the sokolov & ternov formulation , and eq . ( [ eq : gamma_jl_final ] ) for the rate and eq . ( [ eq : power_jl_spins ] ) for the power for the johnson & lippmann case . these compact forms can be expediently used in astrophysical models , since they generally amount to rapidly convergent series of elementary functions . these spin - dependent results are particularly important for addressing resonant cyclotronic divergences that appear in various higher - order qed mechanisms in external magnetic fields . the johnson & lippmann developments extend the work of daugherty & ventura ( 1978 ) to arbitrary transitions , while the series development permits a more accurate determination of asymptotics of the sokolov & ternov rates than that offered in latal ( 1986 ) . in addition , the lorentz transformation characteristics of the rates are derived for boosts along the magnetic field , yielding a simple time dilation multiplicative factor in the st case , but a slightly more complicated and less ideal dependence on boost lorentz factor for the jl analysis , where spin - state mixing is implied by such boosts . such comparitive complexity of the jl formalism is more pronounced for the radiated power . this comparison again underlines the superior attributes of sokolov & ternov eigenstates for use in evaluating rates and cross sections of magnetized processes in quantum electrodynamics . we thank the referee , carlo graziani , for incisive comments that were helpful to the polishing of the manuscript . we also thank joe daugherty for the provision of materials pertaining to the johnson & lippmann formalism , and plg acknowledges the assistance of stefan coltisor during the early stages of the development of the decay rates using the johnson & lippmann wavefunctions . in performing this research , we are grateful for the generous support of the nasa astrophysics theory program , the research corporation ( grant cc5813 ) , and the national science foundation ( reu and grant ast-0307365 ) . this appendix provides several identities germane to the developments in the text , principally in relation to the legendre functions of the _ second kind _ , , which form the basis of the series expansions for the cyclotron rates . they can be defined in terms of hypergeometric functions ( e.g. see eq . 8.820.2 of gradshteyn & ryzhik 1980 ) , or alternatively the integral representation in eq . ( [ eq : qnu_def ] ) . for integer indices , the specialization of interest here , they are simple combinations of logarithmic and polynomial functions ( e.g. see abramowitz & stegun 1965 ) , that lead to similar simplicity for the using eq . ( [ eq : jnu_ident ] ) . observe that , where the associated legendre functions are defined in 8.703 of gradshteyn & ryzhik ( 1980 ) . note also the alternative identities @xmath62 embodied in the results 8.734.3 and 8.752.5 in gradshteyn & ryzhik ( 1980 ) . explicit functional forms for some of the are given in abramowitz & stegun ( 1965 ) and eq . 8.827 of gradshteyn & ryzhik ( 1980 ) , from which the following identities for the first few can be formed : @xmath63 j_2(z ) & = & -1 + \dover{3z^2}{2 } - \dover{3z\ , ( z^2 - 1)}{4 } \ , \log\bigl\vert \dover{z + 1}{z - 1 } \bigr\vert\quad , \label{eq : j_n_ident}\\[2pt ] j_3(z ) & = & -\dover{13z}{6 } + \dover{5z^3}{2 } - \dover{(5z^2 - 1)\ , ( z^2 - 1)}{4 } \ , \log\bigl\vert \dover{z + 1}{z - 1 } \bigr\vert\quad .\nonumber\end{aligned}\ ] ] an effective way to derive expressions for the in terms of elementary functions , at least for low , is to use the rodrigues formula in 8.836.1 of gradshteyn & ryzhik ( 1980 ) , and then invoke eq . ( [ eq : jnu_ident ] ) . an alternative approach is to use identities 8.831.2 and 8.831.3 of gradshteyn & ryzhik ( 1980 ) , to arrive at a relation between and a sum of legendre polynomials and a logarithmic term . this quickly yields @xmath64 \label{eq : j_n_series}\\[-5.5pt ] & = & \dover{1}{2}\,\bigl [ p_{n-1}(z ) - z\ , p_n(z)\bigr ] \ , \log\bigl\vert \dover{z + 1}{z - 1 } \bigr\vert - \sum_{k=1}^{n-1 } \dover{1}{k}\ , p_{k-1}(z)\ , p_{n-1-k}(z ) + z \sum_{k=1}^{n } \dover{1}{k}\ , p_{k-1}(z)\ , p_{n - k}(z)\quad , \nonumber\end{aligned}\ ] ] a form that clearly displays the combined logarithmic and polynomial character of these functions , and which is useful for asymptotic results , explored just below . there are various possible approaches to numerical calculations of the functions for integer indices . options include direct use of eq . ( [ eq : j_n_series ] ) , combined with an efficient algorithm for computing the legendre polynomials for arguments , such as invoking recurrence relations ( e.g. see eq . 8.5.3 of abramowitz & stegun 1965 ) , which are numerically accurate for upward recurrence in the polynomial index . alternatively , direct recurrence of the functions , which satisfy the same relations as the , is viable , though this is only stable for downward iterations in when . this approach therefore requires evaluation of the for two particular initial values , perhaps via the hypergeometric function series using the identity in eq . 8.1.5 of abramowitz & stegun ( 1965 ) . however , probably the most expedient numerical technique is to use the recurrence relation for the functions directly : @xmath65 this identity can be deduced after a modicum of algebra from eq . 8.5.3 of abramowitz & stegun ( 1965 ) . note that strong cancellation often arises between the last two terms of this equation , so that for large and , the values of decline exponentially with according to eq . ( [ eq : j_n_lownb ] ) . starting with initial values prescribed by eq . ( [ eq : j_n_ident ] ) , this technique is stable to upward recurrence in for any , the range of arguments appropriate for the cyclotron problem . this quickly yields suitably precise values ( numerically tested for ) for a range of indices required in the series summations that appear in the cyclotron rates and powers , for example in eqs . ( [ eq : gammave_st_red ] ) and ( [ eq : knb_alt_too ] ) . the use of this recurrence becomes superfluous typically for , since such indices are generally required only in the classical cyclotron limit when , and the asymptotic limit in eq . ( [ eq : j_n_lownb ] ) is then the preferred computational tool . now focusing on asymptotics issues , two limiting values of are required for the analysis in sec . [ sec : st_analytic ] . the limit is routinely derived , and is listed in eq . ( [ eq : j_n_lownb ] ) . since , the other limiting domain is . this is a somewhat more involved case . by inspection of eq . ( [ eq : j_n_series ] ) , as , the logarithmic terms approach zero since . the legendre polynomials thus simplify to approximately unity , to leading order , and it follows that in the neighborhood of , @xmath66 defines the leading order terms to the taylor series expansion . insertion in eq . ( [ eq : inb_alt_too ] ) then yields @xmath67 correct to order . here , is the incomplete gamma function , and can be introduced using the series identity 8.354.1 of gradshteyn & ryzhik ( 1980 ) . another identity required for the development of eq . ( [ eq : gammave_highnb_too ] ) pertains to the evaluation of for large . for low values of , say , one can use 8.352.1 of gradshteyn & ryzhik ( 1980 ) , namely @xmath68 as an efficient series evaluation for either the or incomplete gamma functions . for larger , one can appeal to the identity , together with stirling s expansion and identity 6.5.25 of abramowitz & stegun ( 1965 ) , namely , to establish the results @xmath69 retaining the next order contributions to and then yields corrections of approximately to the two ratios in eq . ( [ eq : gamma_inc_limit ] ) . abramowitz , m. & stegun , i. a. 1965 , _ handbook of mathematical functions _ , ( dover , new york ) . baring , m. g. 2000 , , @xmath7062 , 016003 . bekefi , g. 1966 , _ radiation processes in plasmas _ , ( wiley and sons , new york ) bussard , r. w. , alexander , s. b. & mszros , p 1986 , , @xmath7034 , 440 . bjorken , j. d. & drell , s. d. 1964 , _ relativistic quantum mechanics _ , ( mcgraw - hill , new york ) burchnall , j. l. & chaundy , t. w. 1941 , quarterly j. math.,@xmath7012 , 112 . daugherty , j. k. & ventura , j. 1978 , , @xmath7018 , 1053 . daugherty , j. k. & harding , a. k. 1986 , , @xmath70309 , 362 . daugherty , j. k. & harding , a. k. 1989 , , @xmath70336 , 861 . dermer , c. d. 1990 , , @xmath70360 , 197 . duncan , r. c. & thompson , c. 1992 , , @xmath70392 , 9 . exton , h. 1976 , _ multiple hypergeometric functions and applications _ , ( ellis horwood ltd . , chichester ) . geprgs , r. , riffert , h. , herold , h. , ruder , h. & wunner , g. 1994 , , @xmath7049 , 5582 . gradshteyn , i. s. and ryzhik , i. m. 1980 , _ table of integrals , series and products _ , ( academic press , new york ) . graziani , c. 1993 , , @xmath70412 , 351 . graziani , c. , harding , a. k. & sina , r. 1995 , , @xmath7051 , 7097 . harding , a. k. & preece , r. d. 1987 , , @xmath70319 , 939 . harding , a. k. & daugherty , j. k. , 1991 , , @xmath70374 , 687 . herold , h. , ruder , h. & wunner , g. 1982 , , @xmath70115 , 90 . johnson , m. h. , & lippmann , b. a. 1949 , phys . rev.,@xmath7076 , 828 . kozlenkov , a. a. & mitrofanov , i. g. 1986 , , @xmath7064 , 1173 . langer , s. h. 1981 , , @xmath7023 , 328 . latal , h. g. 1986 , , @xmath70309 , 372 . manchester , r. n. et al . 2001 , , @xmath70328 , 17 . melrose , d. b. & parle , a. j. 1983 , aust . j. phys . , @xmath7036 , 755 . melrose , d. b. & russell , k. 2002 , j. phys . a : math . gen.,@xmath7035 , 135 . melrose , d. b. & zheleznyakov , v. v. 1981 , , @xmath7095 , 86 . parle , a. j. 1987 , aust . j. phys . , @xmath7040 , 1 . pavlov , g. g. , bezchastnov , v. g. , mszros , p. & alexander , s. g. , 1991 , , @xmath70380 , 541 . sokolov , a. a. , zhukovskii , v. ch . & nikitina , n. s. 1973 , phys . lett.,@xmath7043a , 85 . sokolov , a. a. & ternov , i. m. 1968 , synchrotron radiation , ( pergamon press , oxford ) . wasserman , i. & salpeter , e. 1980 , , @xmath70241 , 1107 . white , d. 1974 , , @xmath709 , 868 .

the unprecedented development of laser technologies in recent two decades , mainly due to chirped pulse amplification technique @xcite , has allowed increase of intensity of laser sources in the first stage a six orders of magnitude @xcite reaching the relativistic intensities @xmath0@xmath1 in the infrared and optical domains , respectively , then exceeding these values up to @xmath2 for current superpower laser beams of ultrarelativistic intensities @xcite . in many laboratories @xcite compact lasers can deliver @xmath3 petawatt short pulses , with focused intensities as high as @xmath4 . next generation multipetawatt and exawatt optical laser systems @xcite will be capable to deliver ultrahigh intensities exceeding @xmath5 , which are well above the ultrarelativistic regime of electron - laser interaction . at such intensities the radiation - matter interaction enters the phase of ultrafast dynamics in supershort time scales , with formation of such important fields in condensed - matter - physics that the attoscience at1,at2,atto1,atto2,atto3,atto4 , and relativistic optics mourou , rel - rep2,br .- kr.,rel - rep1,pukh are . the appearance of laser sources with ultrarelativistic intensities has opened real possibilities for revelation of many nonlinear electrodynamic phenomena at the interaction of superpower radiation fields with high brightness accelerator beams @xcite , relativistic plasma targets of solid densities @xcite , and qed vacuum @xcite.laser sources of such enormous intensities predetermine , in particular , the future of laser accelerators @xcite of superperhigh energies , including laser - plasma accelerators @xcite . thus , currently available optical lasers provide @xmath6 up to @xmath7 , meanwhile with the next generation of laser systems one can manipulate with beams at @xmath8 @xmath9 . in such ultrastrong laser fields , electrons can reach to ultrarelativistic energies . however to obtain ultrarelativistic net energies for field - free electrons we need the third body for satisfaction of conservation laws for energy - momentum of a free electron ( after the interaction ) with absorption / radiation of photons in the field of a plane monochromatic wave @xcite , or take advantage of nonplane character of tightly focused supershort laser pulses @xcite . nevertheless , the actual way to get net energy exchange of relativistic electrons is using of coherent processes of laser - particle interaction with the additional resonances . among those the induced cherenkov , compton , and undulator processes are especially of interest @xcite . these are coherent induced processes where a critical intensity of stimulated wave exists ( because of coherent character of corresponding spontaneous process ) above which the induced process proceeds only in one direction - coherent radiation ( wave amplification ) , or absorption ( particle acceleration ) depending on the particle initial velocity with respect to resonance value ( satisfying the condition of coherency ) of each process . the latter is the most important feature of nonlinear resonance inherent in these processes to get high net energy change of particles , specifically for laser acceleration @xcite , or free electron laser ( fel ) @xcite , as well as to generate relativistic electron beams of low energy spreads and emittances due to the threshold nature of nonlinear resonance in these induced coherent processes . in principle , the realization of acceleration or inverse problem of fel in any laser - induced process requires the possible largest coherent length of particle - laser stimulated interaction at which the accumulation of particle energy in the acceleration regime , or coherent radiation -in the wave amplification regime take place on the whole coherent length of certain process . in other cases , the existence of an additional resonance ( e.g. , in a static magnetic field @xcite ) is necessary where the particle - wave interaction cross section enhances by many orders in magnitude . these are the linear resonances well enough investigated during the past decades . here we will consider laser - induced processes on free electrons with the large effective gain of interaction due to nonlinear resonance achieved in the given field of a strong laser radiation . in the induced cherenkov , compton , and undulator processes such resonance is adequate to fulfilment of condition of coherency between the particle and wave achieved in the field at the aforementioned critical intensity of the total wave - field . in the result of such interaction , at the laser intensities larger than critical value a nonlinear threshold phenomenon of particles reflection or capture take place -accelerating or decelerating electrons on the shortest interaction lengths , even shorter than a laser wavelength @xcite . in contrast to induced cherenkov process @xcite where the intensities of applied laser fields are confined by ionization threshold of dielectric media @xcite in vacuum processes in the induced compton @xcite and undulator @xcite ones there is no restriction on the electromagnetic ( em ) field strengths that allows to apply current superpower laser beams of ultrarelativistic intensities mourou for acceleration of particles up to superhigh energies . the other practical difference between these induced processes is in the relativistic character of initial resonance width of particles to reach the threshold value of corresponding nonlinear resonance in the field for reflection / capture phenomenon . thus , in contrast to cherenkov and undulator processes where practically initial relativistic / ultrarelativistic particles are needed for realization of nonlinear cherenkov resonance in gaseous media @xcite , or nonlinear resonance in a magnetic undulator with the step of the order of a few @xmath10 @xcite , induced compton mechanism for particles reflection / capture and , consequently , acceleration regime practically may be realized for arbitrary initial energies of the particles @xcite . in particular , using laser pulses of near frequencies one can accelerate initially nonrelativistic beams , or even particles in rest . second , at the same frequencies of counterpropagating laser pulses in the induced compton process one can achieve the cancellation of particles transverse momenta acquired in the field to obtain quasi - colinear particles bunches from the emerging relativistic electron targets ( elimination of a bunch angular divergence ) , as it has been proposed in the paper @xcite for generation of uniform relativistic electron layers at the coherent compton backscattering on the ultra - thin solid foils ( see , also , @xcite ) . a promising way for achieving of laser - driven electron acceleration is the use of a plasma medium @xcite . however , the laser - plasma accelerator schemes face problems connected with the inherent instabilities in laser - plasma interaction processes . on the other hand , the spectrum of direct acceleration mechanisms of charged particles by a single laser pulse is very restricted , since one should use the laser beams focused to subwavelength waist radii , or use subcycle laser pulses , or use radially polarized lasers , violating the plane character of a coherent laser pulse for particles acceleration by a single wave - pulse in vacuum without the third body . all these scenarios with different field configurations have been investigated both theoretically and experimentally in the works 22_exp , a1,a2,a3,a4,a5,a6,a7,a8,a9,a10,a11,a12,a13,a14,a15,a16,a17 . note that beside the focusing of a laser pulse to subwavelength waist radii we can get nonplane laser beam by terminating the field , either by reflection , absorption , or diffraction @xcite . the proof of principle experiment of this type has been reported in ref . @xcite . here it is used initially relativistic electron beam and a moderately strong laser pulse . to obtain dense enough electron bunches it is reasonable to consider electrons acceleration from nanoscale - solid plasma - targets meyer - ter - vehn , mirror1,mirror2,mirror3 at intensities high enough to separate all electrons from ions @xcite . thus , combining these two schemes one can obtain ultrarelativistic solid density electron bunches 11b,12b,13b,14b,15b,16b,17b,18b . such bunches can be used to obtain high - flux of positrons , @xmath11-quanta @xcite with possible applications in material science , medicine , and nuclear physics . the recent achievements in the laser technology related to implementation of supershort femtosecond laser sources and subsycle pulses of relativistic intensities exceeding the intra - atomic fields open real opportunities for realization of laser - plasma accelerators of ultrahigh energies mey .- ter,1p,2p,3p,4p,5p,6p,7p,8p . these schemes are based on the laser - plasma - wake - field and laser - beat - wave mechanisms of laser - assisted acceleration in plasma . the interaction of such powerful radiation fields with the plasma has led to generation of quasi - monoenergetic relativistic electron beams in laser - driven plasma accelerators a9,a10,a11,a12,a13,a14,a15,a16,a17 . concerning the interaction of ultrashort laser pulses of relativistic intensities with the free electrons in vacuum , i.e. the implementation of laser accelerators , one should note that experiments in this area are gathering power at present . one of the first laser accelerator concepts is the inverse cherenkov accelerator ( ica ) . the ica scheme has been investigated since the advent of lasers [ see , for example , @xcite ] . the appearance of powerful laser sources already in the next decade after the first lasers has stimulated comprehensive theoretical @xcite and experimental p1,p2,p3,p4,p5,kim1,kim2,kim3,kim4 investigations towards the implementation of both induced cherenkov problems ica and cherenkov laser realization in diverse schemes of multiphoton interaction that makes a phase velocity matching of em wave with a particle by controlling the refraction index of gaseous medium so as to be a wave phase velocity in a dielectric medium less than light speed in vacuum . various interaction schemes of electrons with stimulating radiation have been considered as with ultrarelativistic electron beams in a gaseous medium for the optical region @xcite , as well as with mildly relativistic beams in dielectric waveguides for the microwave region w1,w2,w3,w4,w5,w6,w7,w8,w9,unmag . in the latter case , the electron beam passes over the dielectric in the vacuum surface cherenkov process that enables one to avoid the impeding factors of the medium ( the multiple scattering of electrons on atoms , the ionization losses in the medium etc . ) . besides , the usage of solid - state dielectric as a slow - wave structure ( with rather high dielectric constant @xmath12 , @xmath13 ) enables the achievement of electron - wave synchronism by mildly or moderately relativistic electron beams . it is important to note that the effect of critical field exists even in a very week wave field if the interaction occurs in the region close to the initial cherenkov resonance . therefore one can not escape these peculiarities restricting only the intensity of the external wave field and solve the problem in the scope of the linear theory even for very week wave fields @xcite . note that regarding the induced cherenkov process a series of systematical wrong works have been made during the last two decades of past century , which were discussed in special review @xcite devoted to consideration of those results . in the mentioned review the all principal mistakes have been shown in detail , therefore in the current review article we will not repeat the analysis of those papers which reader can get in the review @xcite . the particle reflection phenomenon from a slowed plane wave pulse has been applied also for cases where the group velocity of a wave pulse due to the dispersion can be less than light speed in vacuum @xcite : in ref . @xcite for the case of a plasma , in ref . @xcite at the focusing of an ultrashort laser pulse in vacuum , in the so - called by authors dispersion - dominated propagation regime . nonlinear resonance of threshold nature requiring shortest acceleration length also takes place at the charged particle interaction with strong laser radiation _ _ _ _ and static magnetic field along the wave propagation direction _ _ _ _ ( when the resonant effect of the wave on the particle rotational motion in the uniform magnetic field is possible ) in a medium @xcite . in vacuum , as a result of the interaction of a charged particle with a monochromatic em wave and uniform magnetic field the resonance created at the initial moment for the free - particle velocity automatically holds throughout the whole interaction process due to the equal doppler shifts of the larmor and wave frequencies in the field . this phenomenon is known as autoresonance @xcite . _ _ _ _ this property of cyclotron resonance in vacuum makes possible as particles acceleration @xcite , as well as the creation of a generator of coherent microwave radiation / radioemission by an electron beam , namely , a cyclotron resonance maser ( crm ) ginz1,ginz2,mcn , coo , nus , ai1,ai2,brat . from the point of view of quantum theory the relativistic nonequidistant landau levels of the particle in the wave field become equidistant in the autoresonance with the quantum recoil at the absorption / emission of photons by the particle . _ _ _ _ in addition , the dynamic stark effect of the wave electric field on the transversal bound states of the particle does not violate the equidistance of landau levels in the autoresonance . _ _ then the inverse process , that is , multiphoton resonant excitation of landau levels by strong em wave and , consequently , the particle acceleration in vacuum due to cyclotron resonance , in principle , is possible . _ _ in a medium with arbitrary refraction properties ( dielectric or plasma ) because of the different doppler shifts of the larmor and wave frequencies in the interaction process the autoresonance is violated . however , the threshold ( by the wave intensity ) phenomenon of electron hysteresis in a medium due to the nonlinear cyclotron resonance in the field of strong monochromatic em wave takes place @xcite ( this process has been treated in the earlier paper @xcite , where , however , only an oscillating solution has been obtained ) . _ _ _ in contrast to autoresonance , the nonlinear cyclotron resonance in a medium proceeds with a large enough resonant width . _ _ _ this so - called phenomenon of electron hysteresis leads to significant acceleration of particles , especially in the plasmalike media where the superstrong laser fields of ultrarelativistic intensities can be applied @xcite . note that the use of dielectriclike ( gaseous ) media in this process makes it possible to realize cyclotron resonance in the optical domain ( with laser radiation ) due to the possibility of arbitrarily small doppler shift of a wave frequency close to the cherenkov cone , in contrast to the vacuum case where the cyclotron resonance for the existing maximal powerful static magnetic fields is possible only in the microwave domain @xcite . the particle reflection phenomenon has a significant property : at the above - threshold interaction the induced process occurs only in one direction either direct ( laser scheme ) or inverse ( accelerator scheme ) @xcite meanwhile at the below - threshold interaction the both direct and inverse processes take place simultaneously that reduces in general the net gain for each problem . therefore this phenomenon gives in principal new opportunity for implementation of nonlinear schemes of fel or laser accelerators in the induced coherent processes at the above - threshold interaction regime . due to this feature the reflection phenomenon may also be used for the monochromatization of particle beams @xcite . on the other hand , because of reflection phenomenon the nonlinear compton effect in media with @xmath14 can proceed only at the wave@xmath15intensities below the critical one @xcite . for the nonplane em wave pulses of small space sizes ( tightly focused ) and of short duration , such as current superpower laser pulses are , it is impossible to solve the problem analytically . hence , in the papers last , wigler,2_beams the reflectionand capture phenomena in case of actual nonplane - short laser pulses has been investigated by numerical integration of relativistic equations of motion . the phenomenon of particle reflectionand capture by em wave being of pure classical nature , nevertheless leads to quantum effects of probability density modulation of a free particle state on the hard x - ray frequencies because of interference of incident and reflected ( inelastic ) de broglie waves after the reflection from a wave barrier @xcite and of zone structure of particle states in the wave potential well capture regime like the particle states in a crystal lattice @xcite . on the other hand , it is evident that the role of particle spin in this process _ _ _ _ is important since in dielectriclike media the wave periodic em field in the intrinsic frame of reference becomes a static magnetic field and spin interaction _ _ _ _ with such a field should resemble the zeeman effect . besides , the critical intensity depends on the particle spin projection and , consequently , it is differ for different initial spin orientations . choosing the certain value for critical intensity corresponding to the certain spin orientation of particles in the beam , with the appropriate laser intensity ( above this critical value ) one can achieve the particle beam polarization.these are the quantum effects in the above critical regime of induced cherenkov process . below the critical intensity the coherent effects of particles classical bunching on optical and shorter wavelengths @xcite , quantum effects of modulation of particle probability density on the wave harmonics mod , mod.2,mod.3 , and inelastic diffraction scattering of the particles on the traveling em wave in a dielectric medium takes place dif.2,dif.3,dif.4 like the kapitza dirac effect on a standing wave lattice @xcite or bragg diffraction on a crystal lattice @xcite . the interest to kapitza dirac effect @xcite has been increased especially after the successful realization of experiment with high - intensity and strongly coherent laser beams - gratings @xcite . it follows to note that observation of diffraction effect during the next 70 years after its prediction had not been succeeded . for the acquaintance with the experimental situation in this area since the advent of this effect , or for detailed references on earlier work we refer the reader to review papers @xcite . the significance of kapitza dirac effect , apart from its quantum - mechanical meaning as a best example of demonstration of electron matter wave diffracted by light and , moreover , as an unique sample of a diffraction system with reversed properties of the matter and light , is also conditioned by important applications , since electron beams difracted from highly coherent laser gratings are coherent with each other . hence , the kapitza dirac effect is a very convenient , even maybe an irreplaceable means to realize coherent electron beams . such beams can serve as a basis for construction of new important tools of diverse species , e.g. , coherent beam splitters , new type electron interferometers which would operate at rather low electron energies ( typical for existing now in atomic physics ) etc . kapitza dirac effect on standing wave lattice is a particular case of the induced compton process in the field of two counterpropagating em waves of the same frequencies at which electron moves in perpendicular direction to wavevectors of counterpropagating waves ( classical condition of resonance ) , to exclude the doppler shifting of waves frequencies because of longitudinal component of electron velocity . at the quantum condition of resonance -taking into account the quantum recoil as well- the exact condition of resonance is satisfied at the small angle to perpendicular direction that corresponds to elastic bragg diffraction on the phase lattice of a standing wave . nevertheless , the phase matching between the electrons and counterpropagating waves in the induced compton process can also be fulfilled in general case of bichromatic em waves of different frequencies if electron moves at the certain angle with respect to wavevectors of counterpropagating waves at which the condition of coherency between the electron and waves of different frequencies is satisfied dif.2 . however , in contrast to kapitza dirac effect , diffraction of electrons in this case is inelastic . thus , due to the induced compton effect in the two wave fields an electron absorbs @xmath16 photons from the one wave and coherently radiates @xmath16 photons into the other wave of different frequency and vice versa . this is the condition of coherency in the induced compton process corresponding to the resonance between the doppler - shifted frequencies in the intrinsic frame of reference of an electron in the bichromatic counterpropagating waves at which the conservation of the number of photons in the induced compton process takes place fed.,dif.2,fed.mci ( see , also the paper @xcite where theoretical analysis of the scattering of electrons by a strong standing wave with a slowly varying amplitude has been made ) , in contrast to spontaneous compton effect in the strong wave field where after the multiphoton absorption a single photon is emitted @xcite so , the scope of the kapitza dirac effect has been extended since 1975 in the works @xcite for inelastic diffraction scattering of electrons on strong travelling wave in the induced compton , cherenkov , and undulator / wiggler processes , taking into account the the mentioned peculiarity in these processes . note that the term `` strong wave '' here is relative , since the mentioned nonlinear resonance takes place even in the very weak wave - fields if the electrons initially are close to the resonance state , i.e. the electrons initial longitudinal velocity is close to the phase velocity of the slowed wave , at which the critical field is also very small and , respectively , very low wave intensities may be above critical @xcite . at this background , the paper @xcite is completely misunderstanding where authors , citing the papers @xcite but bypassing the existence of critical field with aforementioned peculiarity , they obtain the same formula of the paper @xcite for multiphoton probability of diffraction scattering on the travelling wave in a dielectric medium ( by other way -on the base of the helmholtz kirchhoff diffraction theory- following to papers @xcite ) , and after four decades claim about the possibility of diffraction effect of electrons on the travelling wave in a dielectric medium . the organization of the paper is as follows . in sec . ii the exact nonlinear theory of induced cherenkov process in the strong laser fields is presented . the peculiarity inherent to nonlinear resonance in coherent processes of threshold character and existence of critical intensity leading to particle reflection and capture phenomenon is described in subsec . ii 1 . in subsec . laser acceleration of charged particles in gaseous media cherenkov accelerator with the variable refraction index of a gaseous medium is considered . in subsec . the nonlinear cyclotron resonance of threshold nature in a medium electron hysteresis phenomenon is presented . in subsec . ii 4 . the theory of coherent radiation of charged particle beams in capture regime cherenkov amplifier is presented . in sec . the quantum theory of nonlinear induced cherenkov process is considered . in subsec . the spin - effects at the electron reflection phenomenon and the possibility to achieve the polarized particle beams is considered . in subsec . quantum coherent effect of electrons reflection from the traveling - wave - phase - lattice in the exact resonance is investigated . in subsec . the quantum regime of spinor particles capture by slowed traveling wave at the exact bragg resonance is considered . in subsec . the quantum modulation effect of particle state on the optical harmonics at the pump wave intensities below the critical value is presented . in sec . iv the vacuum versions of induced coherent processes induced compton and undulator / wiggler processes are investigated revealing the particle reflection nonlinear threshold phenomena in vacuum , as well as inelastic diffraction effect of electrons like to kapitza dirac effect in vacuum are considered . in subsec . induced compton process in the field of two bichromatic counterpropagating waves is investigated . in subsec . the reflection phenomenon in the undulator / wiggler is considered . in subsec . iv 3 . the general theory of inelastic diffraction effect of electrons on the slowing travelling wave is presented . in subsec . diffraction regime of electron coherent scattering on the traveling wave phase - lattice is considered . in subsec . the bragg regime of exact resonance on travelling wave lattice is presented . finally , conclusions are given in sec . v. relativistic classical equations of motion for a charged particle energy in the field of a plane em wave give the exact solution @xcite : @xmath17 @xmath18\biggr]^{1/2}\biggr\ } \label{1}\ ] ] where @xmath19 is the refraction index of a medium ( @xmath20 and @xmath21 are the dielectric and magnetic permittivities of the medium , respectively ) , @xmath22 , @xmath23 , @xmath24 are respectively particle mass , initial energy , momentum and velocity -directed at the angle @xmath25 to the wave propagation direction @xmath26 ( @xmath27 ) , @xmath28 is the light speed in vacuum , and @xmath29 denotes the dimensionless relativistic invariant intensity parameter of the wave : @xmath30 with the vector potential @xmath31 ; @xmath32 is the retarding wave coordinate of the plane em wave in a medium , and @xmath33 , @xmath34 slowly varying amplitudes of the vector potential and electric field of a quasimonochromatic wave amplitudes ( the wave is linearly polarized along the axis @xmath35 : @xmath36 ) . hereafter we will assume that the wave frequency @xmath37 is far from the main resonance transitions between the atomic levels of the medium to prohibit the wave absorption and nonlinear optical effects in the medium and consequently @xmath38 will correspond to the linear refraction index of the medium . as is seen from eq . ( [ 1 ] ) the expression determining the particle energy in the wave field is , first , not single - valued and , second , may become imaginary depending on particle and wave parameters . the peculiarity arising in the induced cherenkov process because of particle strong wave nonlinear interaction is connected with this fact . hence , treatment of the particle dynamics in this process should start by clarification of these questions . to consider the behavior of a particle upon nonlinear interaction with a strong wave in a medium on the basis of eq . ( [ 1 ] ) we will analyze the case where the initial velocity of the particle at @xmath39 ( @xmath40 ) is directed along the wave propagation direction for which the picture of the particle nonlinear dynamics is physically more evident . in this case ( [ 1 ] ) becomes : @xmath41 , \label{5}\]]@xmath42 as is seen , eq . ( [ 5 ] ) is twovalence and , at first , we shall provide the unique definition of the particle energy in accordance with the initial condition . in the case of plasma ( @xmath43 ) or vacuum ( @xmath44 ) the term under the root is always positive , hence , in these cases one has to take before the root only the upper sign ( @xmath45 ) to satisfy the initial condition @xmath46 when @xmath47 . in the case @xmath44 , eq . ( [ 5 ] ) yields known vacuum result . further investigation is devoted to the case of a medium with refraction index @xmath48 . in this case the nature of the particle motion essentially depends on the initial conditions and the value of the parameter @xmath29 as far as the expression under the root in eq . ( [ 5 ] ) may become negative . to solve this problem one needs to pass the complex plane were eq . ( [ 5 ] ) has the form of known inverse jukowski function . if @xmath49 ( @xmath50 is the maximum value of the parameter @xmath51 ) the expression under the root in eq . ( [ 5 ] ) is always positive and in front of the root one has to take the upper sign ( @xmath45 ) according to the initial condition . then @xmath52 after the interaction ( @xmath53 ) and the particle energy remains unchanged . if @xmath54 the particle is unable to penetrate into the wave , i.e. , into the region @xmath55 since at @xmath55 the root in eq . ( [ 5 ] ) becomes a complex one . this complexity now is bypassed via continuously passing from one riemann sheet to another , which corresponds to changing the inverse jukowski function from @xmath45 to @xmath56 before the root . hence , the upper sign ( @xmath45 ) in this case stands up to the value of the wave intensity @xmath57 , then at @xmath58 the root changes its sign from @xmath45 to @xmath56 , providing continuous value for the particle energy in the field . the intensity value @xmath59 of the wave is a turn point for the particle motion , so that we call it the critical value . thus , when the maximum value of the wave intensity exceeds the critical value a transverse plane em wave in the medium becomes a potential barrier and the reflection of the particle from the wave envelope ( @xmath51 ) takes place . if now @xmath60 , we obtain after the reflection for the particle energy @xmath61 . \label{7}\ ] ] if the initial conditions are such that the wave pulse overtakes the particle ( @xmath62 ) , then after the reflection @xmath63 and the particle is accelerated . but if the particle overtakes the wave ( @xmath64 ) , then @xmath65 and particle deceleration takes place . this threshold phenomenon of the particle reflection can be more clearly presented in the frame of reference connected with the wave . in this frame the electric field of the wave vanishes ( @xmath66 ) and there is only the static magnetic field ( @xmath67 ) . the phenomenon of charged particle reflection from a plane em wave may also be used for the monochromatization of particle beams . the fact that above the critical intensity value the induced cherenkov process occurs in only one direction either emission or absorption and for the initial cherenkov velocity @xmath68 the energy of the particle after the reflection does not change , in principle enables conversion of the energetic or angular spreads of charged particle beams . in this case the energy of the particle is given by eq . ( [ 1 ] ) , which at the actual values of the parameters for induced cherenkov process the second term under the root is much smaller than the third one , that is , @xmath69 and for the critical field in this case ( at the wave linear polarization - electric field strength along the axis @xmath35 ) we have @xmath70(in the case @xmath71 , @xmath72 is determined by eq . ( [ 6 ] ) ) . if the maximal value of the wave intensity @xmath73 , then the particle energy after the reflection is @xmath74 . \label{10}\ ] ] let the charged particle beam with an initial energetic ( @xmath75 ) and angular ( @xmath76 ) spread interact with a plane transverse em wave of intensity @xmath73 in a gaseous medium . to keep the mean energy @xmath77 of the beam unchanged after the interaction ( at the adiabatic turning on and turning off of the wave ) the axis of the beam with mean velocity @xmath78 must be pointed at the cherenkov angle ( @xmath79 ) to the laser beam , i.e. , @xmath80 . under this condition the particles with velocities @xmath81 will acquire an energy and the other particles for which the longitudinal velocities exceed the phase velocity of the wave ( @xmath82 ) will loss an energy according to eq . ( [ 10 ] ) . as a result the energies of the particles @xmath83 will approach close to the mean energy @xmath77 of the beam ( @xmath84 @xmath85 ) and the final energetic width of the beam will become less than the initial one . as there is one free parameter ( for a specified velocity @xmath86 the parameters @xmath79 and @xmath87 are related by cherenkov condition ) it is possible to use it to control the exchange in the energy of the particles after the reflection ( [ 10 ] ) and to reach the minimal final energy spread of the beam @xmath88 monochromatization . depending on the relation between the initial energetic and angular spreads and mean energy of the beam , the opposite process may occur , namely angular narrowing of the beam . physically it is clear that with the monochromatization the angular divergence of the beam will increase and the opposite the angular narrowing of the beam leads to demonochromatization ( in accordance with liouville s theorem ) . more detailed consideration of this effect with the quantitative results can be found in the papers @xcite ( for suppression of a beam spread by vacuum versions of reflection phenomenon see the papers @xcite ) . to illustrate the typical picture of nonlinear interaction of a charged particle with a strong em wave in a medium we present the graphics of numerical solutions for the laser pulse of finite duration , showing the behavior of particle dynamics below and above critical intensity , with the effect of acceleration . at first we will not take into account the dependence of the slowly varying intensity envelope of a laser beam from the transversal coordinates . thus , a laser beam may be modeled as @xmath89where @xmath90 characterizes the pulse duration . is plotted when the wave intensity is above the critical point . ] the particle initial energy is taken to be @xmath91 @xmath92 and the initial velocity is directed at the angle @xmath93 @xmath94 to the wave propagation direction ( @xmath95 ) . the refraction index of the gaseous medium for this calculation has been chosen to be @xmath96 . figure 1 illustrates the evolution of the particle energy : the energy versus the position @xmath97 is plotted for a neodymium laser ( @xmath98 @xmath99 ) with electric field strength @xmath100 @xmath101 and @xmath102 ( @xmath103 is the wave period ) . for these parameter values the wave intensity is above the critical point and , as we see from this figure , the particle energy is abruptly changed corresponding to the reflectionphenomenon . figure 2a illustrates the evolution of the energies of particles with different initial interaction angles . the initial energies for all particles are @xmath104 @xmath105 . figure 2b illustrates the role of initial conditions : the final energy versus the interaction angle is plotted . as follows from eq . ( [ 9 ] ) the critical intensity , as well as the final energy ( [ 10 ] ) , depend on the initial interaction angle and as a consequence we have this picture . note that the acceleration rate neither depends on the field magnitude ( only should be above threshold field ) nor on the interaction length . , ( b ) @xmath106 , and ( c ) @xmath107 . ] to demonstrate the dependence of the considered process on transversal profile of the laser intensity for actual beams in fig . 3 the evolution of the energies of particles with various initial phases ( with initial energies @xmath104 @xmath105 ) is illustrated . the laser beam transversal profile is modeled by the gaussian function @xmath108with @xmath109 @xmath110 . as we see from this figure the acceleration picture is essentially changed depending on the entrance coordinates of the particles . this is the manifestation of the threshold nature of the reflection phenomenon . if for the intensity exceeding the critical value a plane em wave becomes a potential barrier for the external particle ( with respect to the wave ) , then for the particle initially situated in the wave it may become a potential well and particle capture by the wave will take place @xcite . the phenomenon of charged particle reflection and capture by a transverse em wave can be used for particle acceleration in laser fields . as the application of large intensities in this process is restricted because of the medium ionization the acceleration owing to reflection in the medium with refraction index @xmath111 single reflection is relatively small . however , if the refraction index decreases along the wave propagation direction in such a way that the condition of particle synchronous motion with the wave @xmath112 takes place continuously , the phase velocity of the wave will increase all the time and the particle being in front of the wave barrier ( at @xmath55 ) will continuously be reflected , i.e. , continuously accelerated . the law @xmath113 must have an adiabatic character not to allow the particle to leave the wave after the single reflection . such variation law of the refraction index can be realized in a gaseous medium adiabatically decreasing the pressure . for particle acceleration one can also use the capture regime . in this case in the medium with @xmath111 the particle energy does not change on average ( particle makes stable oscillations around the equilibrium phases in the wave moving with average velocity @xmath114 ) . if one decreases the refraction index along the propagation direction of the wave , so that the particle does not leave the equilibrium phases , then the wave will continuously accelerate the particle . then , to realize the capture regime one needs @xmath115 . for not very strong fields this is sufficiently strict confinement on the transverse momentum of the particle . for particle acceleration by laser fields one can use the capture regime corresponding to large transverse momenta of the particle @xmath116 so , we will consider the general case of particle capture with arbitrary initial momentum @xmath117 and laser acceleration in gaseous medium with varying refraction index @xmath118 . the variation law of the medium refraction index for acceleration is determined by equation : @xmath119 . \label{62}\ ] ] it is seen from this equation that for the particle acceleration in the capture regime via decreasing refraction index of the medium ( @xmath120 ) one needs @xmath121 ( equilibrium transverse momentum of the particle must be directed along the vector of the wave electric field ) . in the opposite case the continuous deceleration of the particle will take place accompanied by induced cherenkov radiation ( regime of continuous amplification of the wave by the particle beam at @xmath122 ) . the energy of equilibrium particle acquired on the distance @xmath97 is defined by @xmath123 equation ( [ 63 ] ) in the general case defines the particle acceleration in the capture regime when the medium refraction index falls along the wave propagation . it defines the longitudinal dimension of such cherenkov accelerator as well . the transverse dimension of the latter is defined by @xmath124 here @xmath125 and @xmath126 are the initial equilibrium values of the energy and transverse coordinate of the particle ( @xmath127 is the transverse dimension of cherenkov accelerator ) . as is seen from eq . ( [ 65 ] ) the particle acceleration takes place if @xmath128 , and in the opposite case deceleration occurs ( @xmath129 ) in accordance with what was mentioned above . for relativistic particles , when @xmath130 and @xmath131 , from eq . ( [ 62 ] ) we have @xmath132 for energy of the equilibrium particle at large acceleration we have @xmath133 the estimations show that , for example , at electric field strengths of laser radiation @xmath134 @xmath101 an electron with initial energy @xmath135 @xmath105 acquires energy @xmath136 @xmath105 already at the distance @xmath137 @xmath138 . the transverse dimension of acceleration @xmath139is of the order of a few millimeters and the longitudinal dimension of the system is of the order of the transverse one ( a few times larger ) . at the distance @xmath137 @xmath140 the particle energy gain is of the order of @xmath141 @xmath142 . note that because of multiple scattering on the atoms of the medium the particles can leave the regime of stable motion as a result of change of @xmath143 . the analysis shows that the multiple scattering essentially falls in the above - mentioned gaseous media for laser field strengths @xmath144 @xmath145 . to illustrate the particle acceleration in the capture regime we will represent the results of numerical solution of relativistic equations of motion in the field of an actual laser beam with the electric field strength @xmath146@xmath147where @xmath90 characterizes the pulse duration and @xmath148 is the initial phase . simulations have been made for neodymium laser ( @xmath149 @xmath99 ) with electric field strength @xmath100 @xmath101 and @xmath150 @xmath151 . the variation law for the refraction index of the medium is defined in self - consistent manner that may be approximated by the function @xmath152where @xmath153 , @xmath154 are the initial and final values of the refraction index and @xmath155 characterizes the decreasing rate . . in ( b ) the final energy versus the initial phase is plotted . ] @xmath156 . in ( b ) the final energy versus the initial phase is plotted . ] figure 4 illustrates the evolution of the particle energy in the capture regime . the particle initial energy is taken to be @xmath157 @xmath158 and the initial velocity is directed at the angle @xmath159 @xmath94 to the wave propagation direction ( @xmath160 ) . the initial value of the refraction index has been chosen to be @xmath161 . as we see in the capture regime with variable refraction index , one can achieve considerable acceleration . to show the role of initial conditions in fig . 5a the evolution of the energies of particles with the same initial energies @xmath157 @xmath158 ( @xmath162 @xmath94 ) and various initial phases is illustrated . the initial entrance coordinate is @xmath163 . figure 5b displays the role of initial conditions : the final energy versus the initial phase is plotted . in fig . 6 the parameters are the same as in fig . 5a except the initial entrance coordinate , which is taken to be @xmath164 @xmath156 . as we see from these figures the captured particles are accelerated , while the particles situated in the unstable phases ( or if the conditions for capture are not fulfilled ) after the interaction remain with the initial energy . consider now the dynamics of a charged particle in the field of a strong em wave in a medium at the presence of a static uniform magnetic field @xmath165 along the wave propagation direction nonlinear cyclotron resonance in a medium . in this case the problem can be solved analytically only for the circular polarization of monochromatic wave and if the initial velocity of the particle is directed along the axis of the wave propagation . as far as the equation for the particle longitudinal momentum is not changed in the presence of a uniform magnetic field with respect to equation in the field of a plane em wave in a medium , and the equation for the particle energy change in the field remains unchanged , then one can represent the particle longitudinal velocity @xmath166 and energy @xmath167 ^{1/2}\right\ } \label{4.54}\]]via the transversal momentum @xmath168 in the field . here the parameter @xmath169 is @xmath170note that @xmath171 is the critical value of the wave intensity ( [ 6 ] ) ( at @xmath48 ) for the particle reflectionphenomenon in the absence of a static magnetic field ( @xmath172 ) . the sign @xmath45under the root in ( [ 4.54 ] ) corresponds to the case of the interaction in dielectriclike media with @xmath48 and the sign @xmath56 , plasmalike media with @xmath43 . note that in contrast to the case @xmath172 ( induced cherenkov process ) in eq . ( [ 4.54 ] ) before the root , only the sign @xmath45 is taken ( in accordance with the initial conditions @xmath173 and @xmath174 of the free particle ) since , as will be shown below , in this case the expression under the root is always positive and consequently the root can not change its sign . now the considered problem reduces to definition of the particle transversal momentum @xmath175 which in the wave coordinate @xmath176 @xmath177 obey the following equation : @xmath178 ^{1/2}}z\left ( \tau \right ) , \label{4.57}\ ] ] where the complex quantity @xmath179 related to the dimensionless parameter of the particle transversal momentum@xmath180@xmath181 and @xmath182 is the larmor frequency for the initial velocity of the particle . for an arbitrary plane em wave eq . ( [ 4.57 ] ) is a nonlinear equation the exact solution of which can not be found . however , for the monochromatic wave of circular polarization when @xmath183 one can find the exact solution of eq . ( [ 4.57 ] ) . the latter is sought in the form @xmath184 and for the transversal momentum of the particle we obtain the following algebraic equation : @xmath185where the quantities @xmath186 , @xmath187 are expressed in the scale of the parameter @xmath171 : @xmath188 we will not represent here the exact solution of eq . ( [ 4.60 ] ) for @xmath189 . an interesting nonlinear phenomenon exists in this process @xcite which can be found out through the graphical solution of eq . ( [ 4.60 ] ) . thus , depending on the ratio of the larmor and wave frequencies as well as on the initial velocity of the particle ( in the case of dielectriclike medium where @xmath190 ) the solution of eq . ( [ 4.60 ] ) is a single - valued or multivalent that essentially changes the interaction behavior of the particle with a strong em wave at the nonlinear cyclotron resonance in a medium . hence , we will consider separately the cases @xmath191 and @xmath192 at@xmath193 where @xmath194is the doppler - shifted frequency of the wave for the initial velocity of the particle . if @xmath64 the effects considered here will take place with the opposite circular polarization of the wave ( @xmath195 ) or in the opposite direction of the uniform magnetic field ( @xmath196 ) . on the normalized em wave amplitude @xmath197 at @xmath198 . ] consider first the case of a medium with refractive index @xmath48 ( sign @xmath45under the root ) in eq . ( [ 4.60 ] ) . we will turn on the em wave adiabatically and draw the graphic of dependence of the particle transversal momentum on the wave intensity @xmath199 . for the case @xmath200 the latter is illustrated in fig . as is seen from this graphic with the increase of the wave intensity the transversal momentum of the particle increases in the field ( consequently the energy as well ) and vice versa : with the decrease of the wave intensity it decreases in the field and after the passing of the wave ( @xmath201 ) the transversal momentum becomes zero ( @xmath202 ) , i.e. , the particle momentum - energy remain unchanged : @xmath203 and @xmath174 . with the increase of the transversal momentum the longitudinal velocity of the particle increases as well , but in contrast to the case @xmath204 it always remains smaller than the wave phase velocity if initially the wave overtakes the particle ( @xmath62 ) and larger if the particle overtakes the wave ( @xmath64 ) . for this reason the particle reflection phenomenon vanishes in the presence of a uniform magnetic field . indeed , as is seen from eq . ( [ 4.60 ] ) for an arbitrary finite value of @xmath197 we have @xmath205 and longitudinal velocity of the particle in the field @xmath206 if @xmath207 and @xmath208 if @xmath64 . the value @xmath209 may be reached only at @xmath210 when the root in eq . ( [ 4.60 ] ) becomes zero and @xmath211 . so , the expression under the root in eq . ( [ 4.60 ] ) can not become zero for finite intensities of the em wave and , consequently , the root can not change its sign . consider now the case @xmath192 . the graphic of dependence of the particle transversal momentum on the wave intensity @xmath199 in this case is illustrated in fig . 7b . as is seen from this graphic @xmath199 is already a multivalent function : for wave intensities smaller than the value corresponding to the maximum point of the curve @xmath199 three values of the particle transversal momentum exist for each value of the wave intensity . at the maximum point , which will be called a critical one , the wave intensity has the value @xmath212 ^{3/2}. \label{4.63}\]]there are two values @xmath213 and @xmath214 which correspond to critical intensity ( [ 4.63 ] ) . the first one , @xmath215 , is the value of the parameter @xmath216 corresponding to particle transversal momentum at the maximum point of the curve @xmath217 . from the extremum condition of eq . ( [ 4.60 ] ) for @xmath218 we have @xmath219 ^{1/2}. \label{4.64}\]]the second critical value for the parameter @xmath216 corresponding to critical intensity @xmath220 is situated on the left - hand side branch of the curve @xmath199 . to determine its value one needs the analytic solution @xmath221 of eq . ( [ 4.60 ] ) , but there is no necessity here to present the bulk expression for @xmath222 . we shall decide on that branch of the curve @xmath199 which corresponds to real motion of the particle . up to the critical point the particle transversal momentum can be changed on that branch which corresponds to initial condition @xmath202 at @xmath223 . on this branch the particle momentum increases with the increase of the wave intensity and vice versa . it is evident that with further increase of the field the particle can not be situated on the right - hand side from the critical point . hence , it should pass to the left - hand side branch of the curve @xmath199 . indeed , it is easy to see that the critical point is an unstable state for the particle , while all states on the left - hand side branch of the curve @xmath199 are stable and at the critical point the particle changes instantaneously its transversal momentum and passes by jumping to that branch . the further variation of the particle transversal momentum occurs already on this branch . note that the instantaneity here is related to the fact that the solution of eq . ( [ 4.57 ] ) has been found for the monochromatic wave . it is clear that the momentum change actually occurs during finite time . this jump variation of the particle momentum ( energy ) is due to the induced resonant absorption of energy from the wave at the critical point because of which the particle state at this point becomes unstable and it leaves the resonance point for a stable state that corresponds to the transversal momentum @xmath222 on the left - hand side branch of the curve @xmath199 . indeed , if one draws a graphic of the dependence of the particle transversal momentum on the ratio of the larmor and wave frequencies @xmath224 for a certain intensity of the wave ( fig . 8) , then it will be seen from the graphic @xmath225 that the cyclotron resonance in the strong em wave field takes place at the critical point with the satisfaction of the condition @xmath226 . the latter means that to reach the cyclotron resonance in a medium , in contrast to vacuum autoresonance it is necessary to be initially under the resonance condition , since due to the effect of the strong wave field in a medium with refractive index @xmath48 the larmor frequency increases in the field and then reaches the resonance value . in vacuum the cyclotron resonance proceeds at @xmath227 which survives infinitely , because of which the energy of the particle turns to infinity . thus , from eq . ( [ 4.60 ] ) in this case ( @xmath44 ) for the particle transversal momentum we have at @xmath228 @xmath229 which diverges ( consequently the energy as well ) at @xmath230 . as is seen from fig . 8 this divergence vanishes in a medium . with the further increase of the field ( @xmath231 ) the transversal momentum of the particle will continuously increase on the left - hand side branch of the curve @xmath199 and tend to value @xmath232 at @xmath233 . with the decrease of the field the transversal momentum decreases on this branch and at @xmath234 already has only the value @xmath222 since the value @xmath213 corresponds to the unstable state at the resonance point and now there is no reason for inverse transition from the stable state to the unstable one . with the further decrease of the field the transversal momentum decreases , but as is seen from fig . 7 after the interaction ( @xmath223 ) the particle does not return to the initial state ( @xmath235 at @xmath223 ) and remains with the final transversal momentum @xmath236 ^{1/2}. \label{4.66}\]]this is a nonlinear phenomenon of charged particle hysteresis @xcite in the cyclotron resonance with a strong em wave in a medium at intensities exceeding the threshold value ( [ 4.63 ] ) . the energy acquired by the particle due to hysteresis is given by @xmath237 . \label{4.68}\]]if the wave intensity is smaller than the critical value ( [ 4.63 ] ) the energy of the particle oscillates in the field and after the interaction remains unchanged ( such solution only has been obtained in the paper buchb ) . equation ( [ 4.68 ] ) determines the particle acceleration due to a strong transversal em wave at the cyclotron resonance with the powerful static magnetic field in a gaseous medium ( @xmath238 ) . because of the latter one can achieve the cyclotron resonance using optical ( laser ) radiation in a medium with the refractive index @xmath48 , since the doppler shift for a wave frequency @xmath239 ( see eq . ( [ 4.62 ] ) ) in this case may be arbitrarily small in contrast to vacuum , where the cyclotron resonance for the existing powerful static magnetic fields is possible only in the radio - frequency domain . on the other hand , the application of powerful laser radiation for large acceleration of the particles in gaseous media is confined by the ionization threshold of the medium . consider now the case of a plasmous medium ( @xmath43 ) . in eq . ( [ 4.60 ] ) this case should take the sign @xmath56under the root at which the confinement for the particle transversal momentum , existing in a dielectriclike medium , vanishes . in addition , the above - considered behavior of the cyclotron resonance in a plasmous medium takes place with the inverse relation between the initial larmor and wave frequencies @xmath224 . thus , at @xmath240 with the increase of the wave intensity the transversal momentum of the particle increases in the field and vice versa : with the decrease of the wave intensity it decreases in the field and after the passing of the wave ( @xmath223 ) the transversal momentum becomes zero ( @xmath202 ) , i.e. , the particle momentum - energy remain unchanged : @xmath203 and @xmath174 . the nonlinear phenomenon of particle hysteresis in a plasmous medium takes place at @xmath241 , since in a medium with refractive index @xmath43 the larmor frequency decreases in the field and then becomes equal to the resonance value . on the normalized em wave amplitude @xmath197 at @xmath242 . ] the graphic of dependence of the particle transversal momentum on the wave intensity @xmath199 in this case is illustrated in fig . 9 . as is seen from this graphic , in contrast to the case of dielectriclike media the parameter @xmath243 in the plasmas increases with no limit at the increase of the field . the latter allows the large acceleration of the particles achieved by the current superstrong laser fields of ultrarelativistic intensities ( @xmath244 ) due to this phenomenon of hysteresis in the plasmas . the final transversal momentum of the particle as a result of the hysteresis in this case is @xmath245 ^{1/2 } , \label{4.69}\]]the final energy of which will be determined by the same equation ( [ 4.68 ] ) since both the numerator and denominator of the fraction in the expression analogous to eq . ( [ 4.68 ] ) for the particle energy in a plasma change sign . note an interesting effect at the cyclotron resonance in a medium as well . at @xmath227 no matter how weak the em wave field is @xmath246 @xmath171 ( that is , @xmath247 even for @xmath248 ) from eq . ( [ 4.60 ] ) it follows that @xmath249that is , an essential nonlinearity ( @xmath250 ) arises in a case where one would expect a linear dependence on the field according to linear theory . it is the consequence of nonlinear cyclotron resonance the width of which is large enough in this case : @xmath251 the particle capture phenomenon may in principle serve as a fel mechanism ( cherenkov amplifier ) . for the latter one needs to solve the self - consistent problem on the basis of the set of maxwell vlasov equations . this problem can be solved analytically if the particle initial velocity in the capture regime is directed along the wave propagation and has a value close to the cherenkov one : @xmath252 ; parameter @xmath253 . the amplitude of the initial wave field should be a slowly varying function of the space - time coordinates ( @xmath97 , @xmath254 ) with respect to the phase . the problem is sensitive to the wave polarization , therefore it will be investigated for both circular and linear polarizations . consider at first the wave cherenkov amplification in the capture regime for the circular polarization of stimulated wave with the boundary conditions @xmath255 , @xmath256related to particle we assume that it crosses the boundary of the medium @xmath257 at the moment @xmath258 . to define the electric current of the particle stream we assume that the space is continuously filled with the charged particles . then at the moment @xmath259 in the point @xmath97 will be situated only the particles for which @xmath260 @xmath261 ( with accuracy of a small parameter @xmath253 ) . then for the self - consistent field we obtain the equation @xcite equation for self - consistent field has a simpler form over wave coordinates @xmath262 , @xmath263 . for the field amplitude in these coordinates @xmath264 we have @xmath265 . \label{37}\]]where @xmath266 is the mean density of the particles in the initial stream , which will be assumed constant ( since @xmath253 the variation @xmath267 is small and can be neglected).the simple analytic solution can be received at the incident monochromatic wave : @xmath268 @xmath186 . in this case @xmath269 does not depend on @xmath270 , i.e. , @xmath271 @xmath272 , and for the quantity @xmath273we have the nonlinear equation of anharmonic oscillator @xmath274the general solution of which is the incomplete elliptic integral of the first kind @xmath275@xmath276 in the linear case when @xmath277 from eq . ( [ 40 ] ) we have capture @xmath278hence , for @xmath279 , which corresponds to particles initial velocity @xmath280 , exponential amplification of the incident wave occurs . for @xmath281 , that is , @xmath282 , the amplification vanishes on average . the quantity in eq . ( [ 41 ] ) @xmath283is the coherent length of amplification . equation ( [ 37 ] ) is an analogue of the equation of the quantum amplifier . the role of inverse population in atomic systems here performs detuning of the cherenkov resonance @xmath284 ( parameter @xmath21 ) . consider now the case of linear polarization of incident wave @xmath285 . by analogy with the previous case for the slowly varying amplitude of the self - consistent field we have the equation @xmath286@xmath287according to which all harmonics are radiated in contrast to circular polarization of the wave . consider eq . ( [ 49 ] ) with regard to the presence and absence of synchronism . in the last case , when @xmath288 @xmath153 taking into account the slow variation of the field amplitude from eq . ( [ 49 ] ) we obtain capture @xmath289as is seen from this formula in the absence of synchronism , there is a weak dependence of radiation field on harmonics number . in the case of synchronism ( @xmath290 @xmath153 ) , ( [ 49 ] ) becomes @xmath291 for the first harmonic ( fundamental coherent radiation ) the results repeat almost exactly the case of wave circular polarization ( eqs . ( [ 40])(42 ) ) , the only difference being that the coherence length in this case is @xmath292 . to determine the radiation on the other harmonics in the case of synchronism consider the problem in the given field . then , for large @xmath97 when @xmath293for the harmonics amplitudes we have @xmath294hence , the radiation intensity on the harmonics @xmath295 as in the linear regime the coherence length increases as energy squared , and the losses of the particles in the medium depend on energy logarithmically , then the energy increase for amplification of weak signals does not give an essential advantage . the optimal energy is @xmath296 . then @xmath297 , where @xmath298 is the electron classical radius . the estimations show that for the amplification of optical radiation in the capture regime with @xmath299 , electron beams of large densities are necessary . the situation considerably will be improved if media with varying refraction index @xmath300 are used . then along the direction of increase of @xmath300 the particles will be continuously decelerated , and the wave continuously amplified ( a regime inverse to cherenkov accelerator ) . though the phenomenon of particle reflection from the front of a plane em wave is of classical nature , which means that quantum effects of above - barrier reflection and tunnel passage should be small enough , nevertheless the quantum consideration of this phenomenon is worthy of note in relation to the appearance of an important coherent quantum effect as a result of classical reflection of particles . the influence of spin interaction is not essential here ; on the other hand , it is quantitatively small enough in the induced cherenkov process ( for optical frequencies ) and may be neglected . the qualitative aspect of spin effects in the induced cherenkov process will be considered below . neglecting the spin interaction , the dirac equation in quadratic form becomes the klein gordon equation , the exact solution of which can be obtained when the particle initial velocity is parallel to the wave propagation direction ( @xmath301 ) and the latter is monochromatic of circular polarization ( @xmath302 ) . to calculate the probability of reflection from the wave barrier one needs to consider an em pulse with the envelope of intensity damped asymptotically at infinity . for a laser pulse of form @xmath303(@xmath304 is the maximal value of intensity and @xmath305 is the half - width of the pulse ) at @xmath306 for the coefficient of reflection we have @xmath307 } { 1+\exp \left [ \pi \widetilde{\omega } \tau _ { 0}\left ( \frac{\xi _ { 0}}{\xi _ { cr}}-1\right ) \right ] } . \label{85}\]]where@xmath308is the quantum frequency corresponding to particle classical energy change due to reflection ( see eq . ( [ 7 ] ) ) and @xmath72 is the classical value of critical intensity ( [ 6 ] ) . the major quantity @xmath309 in eq . ( [ 85 ] ) @xmath310 ( for actual parameters of electron and laser beams in a medium with refractive index @xmath311 the parameter @xmath312 for laser pulse duration @xmath313 @xmath314 ) , hence eq . ( 85 ) corresponds to above reflection regime @xmath306 . this equation shows that @xmath315 with great accuracy ( the coefficient of tunnel passage in this case is of the order @xmath316 $ ] ) . if @xmath317 then the coefficient of reflection @xmath318 with the same accuracy , i.e. the above barrier reflection is negligibly small in this case . thus , the quantum effects of tunnel passage and above barrier reflection do not impact on the classical phenomenon of particle reflection from the plane em wave . this is physically clear since the compton wavelength of a particle ( electron ) is much smaller than the space size of actual em pulses . nevertheless , due to the particle quantum feature as a result of classical reflection the coherent effect of quantum modulation of the free particle probability density and , consequently , electric current density occurs because of superposition of an incident and reflected particle s matter waves . the density of electric current of the particle beam is modulated at frequency @xmath319 @xmath320 \right\ } , \label{88}\]]where @xmath321 is the electric current density of the initially homogeneous and monochromatic particle beam . the modulation frequency @xmath319 in actual cases lies in the x - ray domain as follows from the estimation of particle classical energy change @xmath322 due to reflection ( @xmath323 @xmath324 /@xmath325 ) . note that quantum modulation in contrast to classical modulation is exceptionally the feature of a single particle and so is conserved after the interaction . consider now the role of spin effects in the nonlinear quantum dynamics of a spinor particle in the field of a plane monochromatic em wave in a medium , i.e. , in the reflection and capture phenomenon of an electron . the exact solution of the dirac equation for external electron ( with respect to wave ) can be found for the above - considered case when the particle initial velocity is parallel to the wave propagation direction and the latter is of circular polarization ( @xmath326 ) . in the result of spin interaction two critical values of the wave intensity appear corresponding to different initial spin projections @xmath327 along the direction of particle motion : @xmath328 ^{2}}{\left ( n^{2}-1\right ) } . \label{92}\]]and for particles energies we have respectively : @xmath329for the reflected particles energies with the spin projections @xmath330 and @xmath331 , respectively , we have : @xmath332@xmath333 in particular , from this equation it follows that in eq . ( [ 92 ] ) @xmath334 corresponds to a particle with the spin directed along the axis @xmath335 , while @xmath336 corresponds to the opposite one . the expressions of particle wave functions show that the degeneration of particle states over the spin projection that takes place in vacuum ( volkov states ) vanishes in a dielectriclike medium . in that case the wave function @xmath337 corresponds to superposition state with energies @xmath338 and @xmath339 , while @xmath340 corresponds to energies @xmath341 and @xmath342 . the removal of degeneration of volkov states is related to the fact that in a medium with refractive index @xmath48 in the intrinsic frame of reference of the wave there is only a static magnetic field and the spin interaction with such a field results in the splitting of the particle states as by the zeeman effect . the splitting value is : @xmath343 . in vacuum this splitting vanishes as it follows from eqs . ( [ 92]),([94 ] ) and dirac wave function in a medium passes to the volkov wave function ( @xcite ) . the spin interaction in a medium within the nonlinear threshold phenomenon of particle reflection may lead to particle beam polarization since the critical intensity ( [ 92 ] ) depends on spin projection along the direction of particle motion . thus , if the condition @xmath344 @xmath345 holds , then only the particles with certain direction of the spin ( along the axis @xmath335 ) will be reflected . since the velocities of reflected particles are different from the nonreflected ones , then by separating the particles after the interaction a polarized beam may be obtained . as was shown above , the exact solution of the dirac equation can be achieved only for the particular case when the particle initial velocity is parallel to the wave propagation direction , which is monochromatic and is of circular polarization . in other cases the quantum equations of motions ( both nonrelativistic and relativistic ) are reduced to ordinary differential equations of the second order of hill or mathieu type , the exact solution of which are unknown . in these cases one needs to develop adequate approximations for the quantum description of particle wave nonlinear interaction at the intensities close to critical value ( indeed below the threshold of classical reflectionphenomenon ) when the probabilities of multiphoton absorption / emission become maximal @xcite . the expected quantum effect in this case is electron reflection from the phase lattice of slowed em wave in a dielectric medium @xcite . note that on the base of this effect a nonlinear scheme of coherent x - ray source ( induced fel ) has been proposed @xcite . to solve dirac equation it is more straightforward to pass to the frame of reference of the rest of the wave ( @xmath346 frame moving with velocity @xmath347 ) . in the @xmath346 frame there is only the static magnetic field that will be described by the following vector potential @xmath348 , where the wavenumber in this frame @xmath349 . the wave function of a particle in the @xmath346 frame is connected with the wave function in the laboratory frame @xmath350 by the lorentz transformation of the bispinors @xmath351(@xmath352 ) . since the interaction hamiltonian does not depend on the time and transverse coordinates the eigenvalues of the hamiltonian @xmath353 and momentum operators @xmath354 , @xmath355 are conserved : @xmath356 , @xmath357 , @xmath358the solution of dirac equation for wave function @xmath359 can be represented in the form of a linear combination of free solutions of the dirac equation @xmath360 with amplitudes @xmath361 depending only on @xmath362 : @xmath363the solution in the form ( [ 72 ] ) corresponds to the expansion of the wave function in a complete set of the wave functions of a particle with certain energy and transverse momentum @xmath364 ( with longitudinal momenta @xmath365 and spin projections @xmath366 ) . the latter are normalized to one particle per unit volume . since there is symmetry with respect to the direction @xmath367 ( the @xmath35 axis ) , we have taken , without loss of generality , the vector @xmath368 in the @xmath369 plane @xmath370 . according to expansion ( [ 72 ] ) the induced cherenkov effect in the @xmath346 frame corresponds to elastic scattering process by which the reflection of the particle from the wave field occurs : @xmath371 . however , in contrast to classical reflection when the periodic wave field becomes a potential barrier for the particle at the intensity @xmath55 , this quantum above - barrier reflection takes place regardless of how weak the wave field is . hence , the probability of multiphoton absorption / radiation of the incident wave photons by the particle in the @xmath350 frame , that is , induced cherenkov effect , will be determined by the probability of particle elastic reflection in the @xmath346 frame . substituting eq . ( [ 72 ] ) into dirac equation and then multiplying by the hermitian conjugate functions we obtain a set of differential equations for the unknown functions @xmath372 . for simplicity we shall assume that before the interaction there are only particles with specified longitudinal momentum and spin state , i.e. , @xmath373from the condition of conservation of the norm we have @xmath374and the probability of reflection is @xmath375 . the equations for amplitudes @xmath376 , @xmath377 and @xmath378 , @xmath379 are separated and after unitarian transformation @xmath380 \label{74}\]]the problem simplifies and we obtain set of equations for the amplitudes @xmath381 and @xmath382 : @xmath383 b_{3,1}(x^{\prime } ) , \label{74a}\]]in unitarian transformation @xmath384 is the angle between the particle momentum and the direction of the wave propagation in the @xmath346 frame . the new amplitudes @xmath385 and @xmath382 satisfy the same initial conditions : @xmath386 @xmath387 according to eq . ( [ 73a ] ) . a similar set of equations is also obtained for the amplitudes @xmath388 and @xmath389 , then after unitarian transformation -for the amplitudes @xmath390 and @xmath391 . the new amplitudes @xmath385 and @xmath382 satisfy the same initial conditions : @xmath392 @xmath393 according to eq . ( [ 73a ] ) and @xmath394 because of conservation of particle energy and transverse momentum ( in @xmath346 frame ) the real transitions in the field will occur from a @xmath395 state to the @xmath396 one and , consequently , the probabilities of multiphoton scattering will have maximal values for the resonant transitions @xmath397the latter expresses the condition of exact resonance between the particle de broglie wave and the incident wave lattice . in the @xmath350 frame the inelastic scattering of the particle on the moving phase lattice takes place and eq . ( [ 75 ] ) corresponds to the known cherenkov conservation law @xmath398where @xmath25 is the angle between the particle momentum and the wave propagation direction ( the cherenkov angle ) , and @xmath399 and @xmath400 are the particle initial velocity and energy in the @xmath350 frame . at the exact resonance ( @xmath401 ) , according to the boundary conditions ( [ 73b ] ) for the reflection coefficient we have @xmath402 , \label{77a}\]]where @xmath403 is the coherent interaction length . the reflection coefficient in the laboratory frame of reference is the probability of absorption at @xmath62 or emission at @xmath404 . the latter can be obtained expressing the quantities @xmath405 and @xmath406 by the quantities in this frame since the reflection coefficient is lorentz invariant . so @xmath407 , \label{77b}\]]where @xmath408 ^{1/2}\]]@xmath409and @xmath410 for actual cases is the laser pulse duration in the @xmath350 frame . the condition of applicability of resonant approximation is equivalent to the condition @xmath411 which restricts the intensity of the wave as well as the cherenkov angle . besides , we must take into account the very sensitivity of the parameter @xmath412 toward the argument of bessel function , according to eq . ( [ 77c ] ) . for the wave intensities when @xmath413 the reflection coefficient is of the order of one that can occur for a large number of photons @xmath414 for the argument of the bessel function @xmath415 in eq . ( [ 77c ] ) ( according to the asymptotic behavior of bessel function @xmath416 at @xmath417 ) . for the off resonant solution , when @xmath418 , but @xmath419 we have the following expression for the reflection coefficient : @xmath420}{1+\frac{f_{s}^{2}}{\omega _ { s}^{2}}\sinh ^{2}[\omega _ { s}\triangle x^{\prime } ] } ; \text { \quad } \omega _ { s}=\sqrt{% f_{s}^{2}-\delta _ { s}^{2}/4 } , \label{78}\]]which has the same behavior as in the case of exact resonance . in the opposite case when @xmath421 the reflection coefficient is an oscillating function of interaction length . the multiphoton induced cherenkov interaction in the capture regime corresponding to transitions between the particle bound states occurs at the nonzero initial angles of particle motion with respect to the wave propagation direction , at which , as mentioned above , the dirac or klein gordon equations are of hill or mathieu type and unable to solve it exactly . however , as was shown in the quantum description of reflection phenomenon ( free free transitions ) , the interaction at the arbitrary initial angle resonantly connects two states of the particle ( in the intrinsic frame of reference of the wave the states with longitudinal momenta @xmath422 of the incident particle and @xmath423 of the scattered particle ; @xmath16 is the number of absorbed or radiated photons with a wave vector @xmath424 ) , which makes available the application of resonant approximation to determine the multiphoton probabilities of free free transitions in induced nonlinear cherenkov process . concerning the quantum description of the particle s bound states in the capture regime one must take into account the degeneration of initial states of free particles in the longitudinal momentum . therefore , regardless of how weak the field of the wave is , the usual perturbation theory in stimulated cherenkov process is not applicable because of such degeneration of the states , and the interaction near the resonance is needed for description by the secular equation . the latter , in particular , reveals the zone structure of the particle states in the field of a transverse em wave in a dielectriclike medium . note that in contrast to the zone structure for the energy of electron states in a crystal lattice , the zone structure in this process holds for the conserved quantity@xmath425 as the energy could not be quantum characteristic of the state in the nonstationary field of the wave @xcite . at first we will consider the case of scalar particles . according to floquet s theorem the solution of klein gordon equation in the wave coordinate @xmath270 may be represented in the form @xmath426where @xmath427 \label{100a}\]]is the major quantity in the induced nonlinear cherenkov process , which is the renormalized ( because of intensity effect ) generalized momentum of the particle in the laboratory frame conjugate to wave coordinate @xmath270 . it connects the width of initial cherenkov resonance @xmath428 @xmath429 and wave intensity ( @xmath430 ) as the main relation between the physical quantities of this process determining also the condition of nonlinear resonance ( @xmath431 ) . in the intrinsic frame of reference of the wave @xmath432 corresponds to longitudinal momentum @xmath422 of the particle on which the degeneration exists the recurrent equation for the coefficients @xmath433 can be solved in approximation of the perturbation theory by the wave function which is valid at the satisfaction of condition@xmath434 regarding those values @xmath432 for which condition ( [ 101a ] ) does not hold , the usual perturbation theory is already not applicable . in particular , if the expression on the left - hand side of this condition is zero , i.e. , at @xmath435 and @xmath436 ( @xmath437 ) , when @xmath438 it is evident that we already have two states @xmath439 and @xmath440 , which are degenerated in the longitudinal momentum @xmath432 , since @xmath441 . because of this double degeneration in the state parameter @xmath432 for the definite @xmath442 of the initial unperturbed system it is necessary to use perturbation theory for the degenerated states on the basis of the secular equation . thus , within secular perturbation theory from the compatibility of equations for @xmath443 and @xmath440 we have @xmath444 , where @xmath445 is the correction to the value @xmath446 at the fulfillment of condition ( [ 102 ] ) for @xmath447 : @xmath448by the standard method from eq . ( [ 104a ] ) one can obtain the following set of equations for the amplitudes @xmath439 and @xmath449 : . the signs @xmath56 and @xmath45 relate to @xmath450 and @xmath451 @xmath452 , respectively . thus , at the fulfillment of condition ( [ 102 ] ) we have a jump in the value of @xmath453 , which is equal to @xmath454 , i.e. , @xmath455 @xmath456 @xmath457 @xmath458 for @xmath447 the matrix element of transition from state @xmath439 to state @xmath449 ( here we note the state without a phase ) is equal to @xmath459 . for large @xmath460 ( @xmath461 ) the matrix element of transition @xmath462 is equal to zero in the first order of perturbation theory . in this case it makes sense to take into account the transitions to the states with other energies in higher order . for example , for @xmath463 it is necessary to consider the transitions @xmath464 and @xmath465 . for arbitrary @xmath460 the matrix element of transition is defined by @xmath466it should be noted that here it is also necessary to take into account the corrections to the energy eigenvalue of state @xmath439 in the appropriate order , however , the latter are only of quantitative character , unlike the qualitative corrections ( [ 107 ] ) , and will be omitted . as is seen from eq . ( [ 106 ] ) , the permitted and forbidden zones arise for the particle states in the wave . the widths of permitted zones in the general case of @xmath460-photon resonance are defined from the condition @xmath467 @xmath468 @xmath469 such zone structure for the particle states in the wave arises in dielectriclike media because of particle capture by the wave and periodic character of the field quantum influence of infinite potential wells on the particle states similar to zone structure of electron states in a crystal lattice @xcite . consider now the case of spinor particles . proceeding from the dirac equation for the components of the spinor @xmath470 : @xmath471 repeating the procedure as in the case of scalar particles , we obtain the bragg condition ( [ 102 ] ) , at which it is necessary to use the secular perturbation theory for degenerated states . at @xmath447 we obtain the following system of equations for coefficients @xmath472 and @xmath473 : @xmath474where @xmath475from eq . ( [ 117 ] ) for the correction to @xmath446 we obtain @xmath476it is easy to see that @xmath477 , where @xmath478 . hence , each spinor component of particle wave function has two values corresponding to the top and bottom borders of the first forbidden zone : @xmath479for @xmath480 we have the same expressions as ( [ 118 ] ) , where it is only necessary to replace @xmath481 by @xmath482 . at @xmath463 we have already two channels for the transition from state @xmath483 to state @xmath484 . the first is the result of the interaction described by a term quadratic in the field ( @xmath485 ) , the matrix element of which at @xmath463 is equal to @xmath486 , and the second channel proceeds both in the case of scalar particles via transitions @xmath487 and @xmath488 , stipulated by the charge interaction @xmath489 , as well as for the spin interaction , the matrix elements of which at each transition are equal to @xmath490 and @xmath491 , respectively . therefore , for two - photon transition for the main parameter correction to @xmath446 we obtain @xmath492 ^{\frac{1}{2 } } \label{119}\]]the obtained results for spinor particles are valid at the fulfillment of the condition @xmath493 . thus , the quantum picture of induced cherenkov interaction for charged spinor particles does not differ qualitatively from the case of scalar particles , i.e. , the spin interaction results only in quantitative corrections to the quantities describing the process . however , in the absence of charge interaction ( @xmath494 ) in the first order in the field , i.e. , for one - photon interaction , the first forbidden zone ( @xmath447 ) does not exist for scalar particles , but exists for spinor particles due to the spin interaction . coherent interaction of charged particles with a plane em wave of intensity smaller than the critical one in the induced cherenkov process leads to quantum modulation of the particle probability density and , consequently , current density after the interaction at the wave fundamental frequency and its harmonics . in contrast to classical modulation of particles current density proceeding in the free drift region after the interaction ( see , e.g. @xcite ) and conserving for short distances , the quantum modulation , being quantum feature of a single particle , is conserved after the interaction unlimitedly long @xcite . to reveal this quantum coherent effect it is necessary to take into account the quantum character of particle wave interaction entirely , i.e. the quantum recoil as well , in contrast to the eikonal approximation for particle wave function at the description of multiphoton interaction with strong em wave . the mathematical point of view requires taking into account in quantum equations of motion the second - order derivatives of the wave function as well , which are neglected in the eikonal approximation ( see , below the description of the diffraction effect ) . to describe the effect of particle quantum modulation with regard to the wave harmonics we will solve the klein gordon equation by perturbation theory in the field of monochromatic wave ( @xmath495 ) of intensity @xmath496 at which one can neglect again the constant term @xmath497 . then looking for the solution of klein gordon equation in the form which will be solved in the approximation of perturbation theory by wave function . thus , the amplitude of the particle wave function corresponding to @xmath16-photon induced radiation ( @xmath498 ) and absorption , respectively , reads @xmath499here the dimensionless parameter of one - photon interaction @xmath500is the small parameter of perturbation theory : @xmath501 @xmath502 and @xmath503are the dimensionless cherenkov resonance width and quantum recoil parameter , respectively . the current density of the particles after the interaction is given by the formula:@xmath504@xmath505 \cos s\omega \left ( t - nx / c\right)\]]@xmath506\]]@xmath507where @xmath508 is the current density of initially uniform particle beam . as is seen from eq . ( [ 3.109 ] ) as a result of direct and inverse induced cherenkov effect the current density of initially uniform particle beam is modulated at the wave fundamental frequency and its harmonics . this is a result of coherent superposition of particle states with various energy and momentum due to absorbed and emitted photons in the radiation field that remains after the interaction unlimitedly long ( for a monochromatic beam ) . the denominators in eq . ( [ 3.109 ] ) becomes zero at the fulfillment of exact quantum conservation law for multiphoton cherenkov process ( [ 75a ] ) . in this case perturbation theory is not applicable and the consideration in the scope of above - developed secular perturbation is required . however , in actual cases because of nonmonochromaticity of particle beams the width of cherenkov resonance is rather larger than quantum recoil ( @xmath509 ) and one can neglect the latter in eq . ( [ 3.109 ] ) . then the modulation depth ( @xmath510 ) at the wave fundamental frequency is expressed via critical intensity ( [ 9 ] ) : @xmath511 . the latter shows that the effect of quantum modulation at the stimulating wave harmonics proceeds at intensities smaller than the critical one when the induced cherenkov interaction of the particles with the periodic wave field ( photons ) occurs . in the opposite case the interaction proceeds with the potential barrier , and as was shown above the quantum modulation of the particles due to reflection phenomenon occurs because of interference of the incident and reflected electron s matter waves . for this reason the modulation frequency ( actually x - ray ) corresponds to particle s energy exchange in the result of interaction with the moving barrier . on the other hand , it is clear that the modulated by anyway particle beam is a coherent source of em radiation . consider now the induced compton and undulator processes in vacuum , at which the restriction on the wave intensity taking place in dielectric media for induced cherenkov process vanishes and one can use laser pulses of extremely large intensities to achieve laser acceleration of particles of superhigh energies by considering phenomena , as well as realization of many nonlinear qed phenomena from vacuum . at first let investigate the classical dynamics of a charged particle at the interaction with two bichromatic counterpropagating ( along the axis @xmath335 ) plane em waves ( with frequencies @xmath512 and @xmath513 ) in vacuum . relativistic equations of motion allow exact solution in case of monochromatic waves of circular polarization and if the particle initial momentum is directed along the axis of waves propagation ( initial transverse momentum @xmath514 ) . then for the particle energy in the field we have @xcite : as is seen from eq . ( [ 5.7 ] ) due to the effective interaction of the particle with the counterpropagating waves a slowed traveling wave in vacuum arises . the parameter @xmath517 denotes the refractive index of this interference wave and since @xmath521 ( see eq . ( [ 5.8 ] ) ) the phase velocity of the effective traveling wave @xmath522 . then the expression under the root in eq . ( [ 5.7 ] ) evidences the peculiarity in the interaction dynamics like the induced cherenkov one that causes the analogous threshold phenomena of particle reflection and capture by the interference wave in the induced compton process . hence , omitting the same procedure related to bypass of the multivalence and complexity of eq . ( [ 5.7 ] ) , which has been made in detail for the analogous expression in the cherenkov process , we will present the final results for particle reflection and capture by the effective interference wave in the induced compton process . the threshold value of the reflection phenomenon or the critical field for nonlinear compton resonance is @xmath523if one knows the longitudinal velocity @xmath166 of the particle in the field , then it is easy to see that @xmath524 is the value of the total intensity of counterpropagating waves at which @xmath525 becomes equal to the phase velocity of the effective interference wave : @xmath526 irrespective of the magnitude of particle initial velocity @xmath399 . the latter is the condition of coherency of induced compton process @xmath527under condition ( [ 5.10 ] ) the nonlinear resonance in the field of counterpropagating waves of different frequencies occurs and because of induced compton radiation / absorption the particle velocity becomes smaller or larger than the phase velocity of the interference wave and the particle leaves the slowed effective wave . in the rest frame of the latter the particle swoops on the motionless barrier ( if @xmath528 ) and the elastic reflection occurs . in the laboratory frame it corresponds to inelastic reflection and from eq . ( [ 5.7 ] ) for particle energy after the reflection ( @xmath529 adiabatically at @xmath530 ) we have @xmath531from this equation it follows that the energy of the particle with the initial velocity @xmath532 corresponding to the resonance value of the induced compton process does not change after the interaction ( @xmath52 ) . for such particle @xmath533 , i.e. , it can not enter the field : @xmath534 . the particle with the initial velocity @xmath535 after the reflection is decelerated , while at @xmath536 it is accelerated because of direct and inverse induced compton processes . at the acceleration the particle absorbs photons from the wave of frequency @xmath537 and coherently radiates into the wave of frequency @xmath513 if @xmath538 @xmath513 and at the deceleration the inverse process takes place . hence , at the particle acceleration the amplification of the wave of a smaller frequency holds , while at the deceleration the wave of a larger frequency is amplified . in the case of @xmath539 @xmath540 the refractive index of the interference wave @xmath541 and nonlinear interaction of the particle with the strong standing wave occurs . it is evident that in this case the process is elastic : @xmath542 ( see eq . ( [ 5.11 ] ) ) and for the longitudinal momentum of the particle in the field we have @xmath543from this equation it is seen that at @xmath544 @xmath545 the standing wave becomes a potential barrier for the particle and elastic reflection occurs : the root changes its sign and @xmath546 ( if @xmath547 @xmath548 we have @xmath549 ) . electrons `` reflection '' or capture phenomenon is also possible by a plane em wave propagating in electric or magnetic undulator / wiggler @xcite . as far as descriptions of electron dynamics in a plane monochromatic wave in the electric and magnetic undulators are coincide in many features , here we will consider only more important case of magnetic undulator / wiggler , which is currently the most perspective coherent tool with extremely large length of coherency , specifically due to which the x - ray free electron laser has been realized in the wiggler @xcite . consider the nonlinear dynamics of a charged particle at the interaction with a strong em wave in a magnetic undulator / wiggler . let a particle with an initial velocity @xmath550 enters into a magnetic undulator with circularly polarized field @xmath551(@xmath552 is the space period or step of an undulator ) along the axis of which propagates a plane monochromatic em wave of circular polarization . relativistic equations of motion in this case allow exact solution and for the particle energy we have @xcite : @xmath553@xmath554 \biggr]^{1/2}\biggr\ } \label{5.25}\]]where relativistic invariant dimensionless interaction parameter @xmath555(for large magnitudes of undulator field strength @xmath556 and space period @xmath552 when @xmath557 such undulator is called a wiggler ) . from eq . ( [ 5.25 ] ) it follows that at the particle wave nonlinear resonance interaction in the undulator an effective slowed traveling wave is formed as in the induced compton process . the parameter @xmath558is the refractive index of this slowed wave , which causes the analogous threshold phenomenon of particle reflection in the induced undulator process . the effective critical field at which the nonlinear resonance and then the particle reflection take place in the undulator , is @xmath559at this value of the resulting field the longitudinal velocity of the particle @xmath166 reaches the resonant value in the field at which the condition of coherency in the undulator @xmath560is satisfied . the latter has a simple physical explanation in the intrinsic frame of the particle . in this frame of reference the static magnetic field ( [ 5.18 ] ) becomes a traveling em wave with the frequency @xmath561and phase velocity @xmath562 . for coherent interaction process this frequency must coincide with the doppler - shifted frequency of stimulated wave . the energy of the particle after the reflection ( in eq . ( [ 5.25 ] ) @xmath563 at the sign @xmath56 before the root ) is @xmath564 . \label{5.30}\]]from this equation it follows that the particle with the initial velocity @xmath565 after the reflection accelerates , while at @xmath566 it decelerates because of induced undulator radiation . the reflection phenomenon of charged particles from a plane em wave , as was shown in the induced cherenkov process , may be used for monochromatization of the particle beams . note that the considered vacuum versions of this phenomenon are more preferable for this goal taking into account the influence of negative effects of the multiple scattering and ionization losses in a medium . on the other hand , the refractive index of the effective slowed waves in vacuum @xmath517 or @xmath567 in corresponding induced compton and undulator processes may be varied choosing the appropriate frequencies of counterpropagating waves or wiggler step . in particular , for monochromatization of particle beams with moderate or low energies via the induced cherenkov process one needs a refractive index of a medium @xmath568 that corresponds to solid states . meanwhile , such values of effective refractive index may be reached in the induced compton process at the frequencies @xmath569 of the counterpropagating waves . however , we will not consider here the possibility of particle beam monochromatization on the basis of the vacuum versions of reflection phenomenon since the principle of conversion of energetic or angular spreads is the same . to study the subject in more detail we refer the reader to original papers comp._mon.,und._mon.,conv .. considered above quantum effects in the induced cherenkov process as in the above critical regime , as well as below the critical point also take place in described vacuum versions of particle reflection and capture phenomena at the induced compton and undulator / wiggler nonlinear interaction . however , because of similar physical picture and investigation methods we will not repeat here the consideration of these effects . as was mentioned in introduction , at the wave intensities below the threshold of particle reflection and capture phenomena , the inelastic diffraction scattering of electron matter wave de broglie wave on a traveling light phase - lattice in the induced cherenkov , compton , and undulator processes is possible . for efficient probabilities of multiphoton diffraction effect on such wave - gratings the intensities of slowed traveling wave should be close to the critical value : @xmath570 to be near the nonlinear resonance of particle - wave interaction in each induced process . we will proceed the solution of diffraction effect in general form for cherenkov and vacuum processes . as the coordinates @xmath571 are cyclic , then the corresponding components of generalized momentum @xmath572 are conserved . hence , according to floquet s theorem the solution of klein gordon equation may be sought in the form @xmath573where @xmath574and we will assume that @xmath575 are slowly varying functions : @xmath576 and @xmath577 ( this condition is always satisfied for optical photons ) . from the klein gordon equation with eq . ( [ psi1 ] ) for the coefficients @xmath578 we obtain the set of equations@xmath579where@xmath580is the resonance width . the wave function ( [ psi1 ] ) at the initial condition @xmath581 describes inelastic scattering of the electron on the slowed traveling wave . the electron energy and momentum after the scattering are:@xmath582@xmath583the probability of these processes are determined by the coefficients @xmath584 @xmath585 we will represent the solution of equations eqs . ( [ m2 ] ) in the two interaction regimes : diffraction at the classical conditions of coherency ( [ 5.10 ] ) , ( [ 5.29 ] ) and bragg resonance at the exact quantum conditions of coherency with the quantum recoil of electron due to photon absorption / radiation . [ [ diffraction - regime - of - electron - coherent - scattering - on - the - traveling - wave - phase - lattice ] ] diffraction regime of electron coherent scattering on the traveling wave phase - lattice ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ for the finite electron - wave interaction time the interaction energy is uncertain by the quantity @xmath586 . the diffraction regime of electron coherent scattering corresponds to the short interaction times and intense wave fields . thus , at the satisfaction of the conditions @xmath587one can neglect the term @xmath588 in eqs . ( [ m2 ] ) and the dynamics of electron in the wavefields will be described by the following equation @xcite : @xmath589for which at the initial condition @xmath581 one can obtain the following solution : @xmath590consequently , the probability of this process according to eq . ( [ prob ] ) will be done by the formula : @xmath591 . \label{3.98}\]]in the case of a monochromatic wave from eq . ( [ 3.98 ] ) we have @xmath592where @xmath593 is the time - duration of the particle motion in the wavefield . as is seen from eq . ( [ main ] ) in the diffraction regime symmetric diffraction ( @xmath594 ) into many momentum states is possible . the process dynamics is defined by the argument of the bessel function @xmath595 . for @xmath596 only few diffraction maxima are possible . for the values @xmath597 , the process is essentially multiphoton . the most probable number of absorbed / emitted cherenkov photons is @xmath598 . the width of the main diffraction maximums @xmath599 and since @xmath600 then @xmath601 . for the concreteness let us explicitly write probability for the cherenkov diffraction and kapitza dirac effects . in case of cherenkov diffraction of electron on a traveling wave in a dielectric medium , from eq . ( [ main ] ) we have @xmath602for the actual values of the parameters @xmath603 , that is , the process is essentially multiphoton . the most probable number of absorbed / emitted cherenkov photons is @xmath604the scattering angles of the @xmath16-photon cherenkov diffraction are determined by the formula : @xmath605from eq . ( [ 3.101 ] ) it follows that at the inelastic diffraction there is an asymmetry in the angular distribution of the scattered particle : @xmath606 , i.e. , the main diffraction maxima are situated at different angles with respect to the direction of particle initial motion . however , in accordance with the condition @xmath607 of the eikonal approximation this asymmetry is negligibly small and for the scattering angles of the main diffraction maxima from eq . ( [ 3.101 ] ) we have @xmath608 . hence , the main diffraction maxima will be situated at the angles @xmath609with respect to the direction of the particle initial motion . for the kapitza dirac effect , which is the particular case of induced compton effect at @xmath610 , @xmath611 , from eqs . ( [ psi1 ] ) and ( [ main ] ) we can write@xmath612hence , the probability of @xmath16-photons kapitza dirac diffraction effect on the strong standing em wave in vacuum is determined by the formula : @xmath613note that formula ( [ kp2 ] ) for the nonrelativistic case @xmath614 coincides with analogous formula ( 7 ) of the paper @xcite up to a factor of @xmath615 ( in the paper @xcite ) the factor @xmath615 has been missed perhaps at the transformation of recurrent relation of bessel functions @xmath616 ) . in general case of inelastic compton diffraction on counterpropagating waves of different frequencies and electron diffraction on a travelling em wave in an undulator / wiggler the probabilities of @xmath16-photon diffraction effect may be written by analogy of formulas ( [ 3.99 ] ) and ( [ kp2 ] ) with corresponding parameters . note that the general formula ( [ 3.99 ] ) for cherenkov diffraction that expresses the probabilities of multiphoton absorption / radiation processes by electron at the induced cherenkov effect has been applied in the paper ushir for explanation of the experiment on energetic widening of an electron beam at the induced cherenkov process in a gaseous medium , implemented in slac @xcite ( see , also the next experiment of this group @xcite , made in the same conditions ) . formula ( [ 3.99 ] ) for cherenkov diffraction in a dielectric medium , which had been received forty years ago @xcite and included in the monographs @xcite and @xcite , recently has been considered in the paper @xcite because of reproduction of electron diffraction effect on a travelling em wave in a dielectric medium by the authors of the paper mis , nevertheless represented it as a new phenomenon predicted in the paper @xcite . repeating the results of the paper @xcite with the main formula ( [ 3.99 ] ) in other way of derivation -within the helmholtz kirchhoff diffraction theory ( as was mentioned in introduction , by the method developed in the paper @xcite ) , authors report in the abstract that they showed on the possibility of electrons diffraction effect on a travelling em wave in a dielectric medium . apart from such misunderstanding resulted to serious confusion in scientific literature , it is interesting the claim of the authors even in the abstract of the paper @xcite ( and this is the only difference from the paper @xcite ) that explains diffraction effect by the `` group velocity of light '' ( and this is repeated also in the text -and for a monochromatic wave ! ) . this is interesting fact since repeating the results , at the same time authors of the paper @xcite quote to original papers @xcite without a comment . while , as it has been shown above , diffraction effect is thoroughly the result of the phase relations and is conditioned exceptionally by the phase velocity of light that must be smaller than @xmath28 just the physical reason which causes electron diffraction on a light phase lattice due to phase matching between the particle and wave phase velocity . beside this rough mistake , authors of the paper @xcite ignored the existence of critical field in this process and influence of considered phenomenon of particle `` reflection '' or capture on the diffraction effect . meanwhile , one of the authors @xcite is also a co - author of both inelastic diffraction effect @xcite and `` reflection '' phenomenon in the undulators @xcite . [ [ bragg - regime - of - electron - coherent - scattering - on - the - traveling - wave - phase - lattice ] ] bragg regime of electron coherent scattering on the traveling wave phase - lattice ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ the condition ( [ res_q ] ) has transparent physical interpretation in the intrinsic frame of reference of the slowed wave . in this frame , due to the conservation of particle energy and transverse momentum the real transitions in this strongly quantum regime occur from a @xmath395 state to the @xmath620 one and we reach the bragg diffraction effect on a slowed traveling wave at the fulfilment of the condition : @xmath621the latter expresses the condition of exact resonance between the particle de broglie wave and the wave motionless lattice . in particular , in this case when the above mentioned particle capture regime by the slowed traveling wave @xcite takes place , we have the quantum effect of zone structure of particle states like the particle states in a crystal lattice , and at the condition ( res_q ) the diffraction maxima take place see fig . 1 . here we just write the solution for the resonant case @xmath622at the condition@xmath623from the set of equations ( [ m2 ] ) one can keep only resonant ones for @xmath624 and @xmath625:@xmath626with the solution@xmath627for the concreteness let us explicitly write probability for kapitza dirac effect - @xmath610 , @xmath628 , @xmath629 ) . from eq . ( [ res_q ] ) at @xmath630 we obtain resonant initial momentum @xmath631 and wave function can be writen as @xmath632 . \label{w}\]]with the probabilities @xmath633for the nonrelativistic case @xmath634 eqs . ( [ bregp ] ) coinside with the results in ref .

many important problems in the theory of integrable systems and approximation theory can be recast as riemann - hilbert problems for a matrix - valued unknown . via the connection with approximation theory , and specifically the theory of orthogonal polynomials , one can also study problems from the theory of random matrix ensembles and combinatorics . roughly speaking , solving a riemann - hilbert problem amounts to reconstructing a sectionally meromorphic matrix from given homogeneous multiplicative `` jump conditions '' at the boundary contours of the domains of meromorphy , from `` principal part data '' given at the prescribed singularities , and from a normalization condition . so , many asymptotic questions in integrable systems ( _ e.g. _ long time behavior and singular perturbation theory ) and approximation theory ( _ e.g. _ behavior of orthogonal polynomials in the limit of large degree ) amount to determining asymptotic properties of the solution matrix of a riemann - hilbert problem from given asymptotics of the jump conditions and principal part data . in recent years a collection of techniques has emerged for studying certain asymptotic problems of this sort . these techniques are analogous to familiar asymptotic methods for expanding oscillatory integrals , and we often refer to them as `` steepest - descent '' methods . the basic method first appeared in the work of deift and zhou @xcite . the first applications were to riemann - hilbert problems without poles , in which the solution matrix is sectionally holomorphic . later , some problems were studied in which there were a number of poles a number held fixed in the limit of interest in the solution matrix ( see , for example , the paper @xcite on the long - time behavior of the toda lattice with rarefaction initial data ) . the previous methods were extended to these more complicated problems through the device of making a local change of variable near each pole in some small domain containing the pole . the change of variable is chosen so that it has the effect of removing the pole at the cost of introducing an explicit jump on the boundary of the domain around the pole in which the transformation is made . the result is a riemann - hilbert problem for a sectionally holomorphic matrix , which can be solved asymptotically by pre - existing `` steepest - descent '' methods . recovery of an approximation for the original sectionally meromorphic matrix unknown involves putting back the poles by reversing the explicit change of variables that was designed to get rid of them to begin with . yet another category of riemann - hilbert problems consists of those problems where the number of poles is not fixed , but becomes large in the limit of interest , with the poles accumulating on some closed set @xmath0 in the finite complex plane . a problem of this sort has been addressed @xcite by making an explicit transformation of the type described above in a single fixed domain @xmath1 that contains the locus of accumulation @xmath0 of all the poles . the transformation is chosen to get rid of all the poles at once . in order to specify it , discrete data related to the residues of the poles must be interpolated at the corresponding poles by a function that is analytic and nonvanishing in all of @xmath1 . once the poles have been removed in this way , the riemann - hilbert problem becomes one for a sectionally holomorphic matrix , with a jump at the boundary of @xmath1 given in terms of the explicit change of variables . in this way , the poles are `` swept out '' from @xmath0 to the boundary of @xmath1 resulting in an analytic jump . there is a strong analogy in this procedure with the concept of balayage ( meaning `` sweeping out '' ) from potential theory . in establishing asymptotic formulae for such riemann - hilbert problems , it is essential that one make judicious use of the freedom to place the boundary of the domain in which one removes the poles from the problem . placing this boundary contour in the correct position in the complex plane allows one to convert oscillations into exponential decay in such a way that the errors in the asymptotics can be rigorously controlled . if the poles accumulate with some smooth density on @xmath2 , the characterization of the correct location of the boundary of @xmath1 can be determined by first passing to a continuum limit of the pole distribution in the resulting jump matrix on the boundary of @xmath1 , and then applying analytic techniques or variational methods . the continuum limit is justified as long as the boundary of @xmath1 remains separated from @xmath0 . this idea leads to an interesting question . what happens if the boundary of @xmath1 as determined from passing to the continuum limit turns out to intersect @xmath0 ? far from being a hypothetical possibility , this situation is known to occur in at least three different problems : 1 . * the semiclassical limit of the focusing nonlinear schrdinger hierarchy with decaying initial data . * see @xcite . this is an inverse - scattering problem for the nonselfadjoint zakharov - shabat operator . on an _ ad hoc _ basis , one replaces the true spectral data for the given initial condition with a formal wkb approximation . there is no jump in the riemann - hilbert problem associated with inverse - scattering for the modified spectral data , but there are poles accumulating asymptotically with the wkb density of states on an interval @xmath0 of the imaginary axis in the complex plane of the eigenvalue . the methods described above turn out to yield rigorous asymptotics for this modified inverse - scattering problem as long as the independent time variable in the equation is not zero . for @xmath3 , the argument of passing to the continuum limit in the pole density leads one to choose the boundary of @xmath1 to _ coincide _ in part with the interval @xmath0 . strangely , if one sets @xmath3 in the problem from the beginning , an alternative method due to lax and levermore @xcite and extended to the nonselfadjoint zakharov - shabat operator with real potentials by ercolani , jin , levermore , and macevoy @xcite can be used to carry out the asymptotic analysis in this special case ; this alternative method is not based on matrix riemann - hilbert problems , and therefore when taken together with the methods described in @xcite does not result in a uniform treatment of the semiclassical limit for all @xmath4 and @xmath5 . at the same time , the lax - levermore method that applies when @xmath3 fails in this problem when @xmath6 . 2 . * the zero - dispersion limit of the korteweg - de vries equation with potential well initial data . * as pointed out above , the original treatment of this problem by lax and levermore @xcite was not based on asymptotic analysis for a matrix - valued riemann - hilbert problem . but it is possible to pose the inverse - scattering problem with modified ( wkb ) spectral data as a matrix - valued riemann - hilbert problem and ask whether the `` steepest descent '' techniques for such problems could be used to reproduce and/or strengthen the original asymptotic results of lax and levermore . in particular , we might point out that the lax - levermore method only gives weak limits of the conserved densities , and that a modification due to venakides @xcite is required to extract any pointwise asymptotics ( _ i.e. _ to reconstruct the microstruture of the modulated and rapidly oscillatory wavetrains giving rise to the leading - order weak asymptotics ) . on the other hand , `` steepest descent '' techniques for matrix - valued riemann - hilbert problems typically give pointwise asymptotics automatically . it would therefore be most useful if these techniques could be applied to provide a new and unified approach to this problem . + if one tries to enclose the locus of accumulation of poles ( wkb eigenvalues for the schrdinger operator with a potential well ) with a contour and determine the optimal location of this contour for zero - dispersion asymptotics , it turns out that the contour must contain the support of a certain weighted logarithmic equilibrium measure . it is a well - known consequence of the lax - levermore theory that the support of this measure is a subset of the interval of accumulation of wkb eigenvalues . consequently , the enclosing contour `` wants '' to lie right on top of the poles in this problem , and the approach fails . in a sense this failure of the `` steepest descent '' method is more serious than in the analogous problem for the focusing nonlinear schrdinger equation because the contour is in the wrong place for all values of @xmath4 and @xmath5 ( the independent variables of the problem ) , whereas in the focusing nonlinear schrdinger problem the method fails generically only for @xmath3 . * the large degree limit of certain systems of discrete orthogonal polynomials . * fokas , its , and kitaev @xcite have shown that the problem of reconstructing the orthogonal polynomials associated with a given continuous weight function can be expressed as a matrix - valued riemann - hilbert problem . it is not difficult to modify their construction to the case when the weight function is a sum of dirac masses . the corresponding matrix - valued riemann - hilbert problem has no jump , but has poles at the support nodes of the weight . the solution of this riemann - hilbert problem gives in this case the associated family of discrete orthogonal polynomials . if one takes the nodes of support of the discrete weight to be distributed asymptotically in some systematic way , then it is natural to ask whether `` steepest descent '' methods applied to the corresponding riemann - hilbert problem with poles could yield accurate asymptotic formulae for the discrete orthogonal polynomials in the limit of large degree . indeed , similar asymptotics were obtained in the continuous weight case @xcite using precisely these methods . + unfortunately , when the poles are encircled and the optimal contour is sought , it turns out again to be necessary that the contour contain the support of a certain weighted logarithmic equilibrium measure ( see @xcite for a description of this measure ) which is supported on a subset of the interval of accumulation of the nodes of orthogonalization ( _ i.e. _ , the poles ) . for this reason , the method based on matrix - valued riemann - hilbert problems would appear to fail . in this paper , we present a new technique in the theory of `` steepest descent '' asymptotic analysis for matrix riemann - hilbert problems that solves all three problems mentioned above in a general framework . we illustrate the method in detail for the first case described above : the inverse - scattering problem for the nonselfadjoint zakharov - shabat operator with modified ( wkb ) spectral data , which amounts to a treatment of the semiclassical limit for the focusing nonlinear schrdinger equation at the initial instant @xmath3 . this work thus fills in a gap in the arguments in @xcite connecting the rigorous asymptotic analysis carried out there with the initial - value problem for the focusing nonlinear schrdinger equation . application of the same techniques to the zero dispersion limit of the korteweg - de vries equation will be the topic of a future paper , and a study of asymptotics for discrete orthogonal polynomials using these methods is already in preparation @xcite . the initial - value problem for the focusing nonlinear schrdinger equation is @xmath7 subject to the initial condition @xmath8 . in @xcite , this problem is considered for cases when the initial data @xmath9 where @xmath10 is some positive real function @xmath11 $ ] . the function @xmath10 is taken to decay rapidly at infinity and to be even in @xmath4 with a single genuine maximum at @xmath12 . thus @xmath13 , @xmath14 , and @xmath15 . also , the function @xmath10 is taken to be real - analytic . with this given initial data , one has a unique solution of ( [ eq : nls ] ) for each @xmath16 . to study the semiclassical limit then means determining asymptotic properties of the family of solutions @xmath17 as @xmath18 . this problem is associated with the scattering and inverse - scattering theory for the nonselfadjoint zakharov - shabat eigenvalue problem @xcite : @xmath19 for auxiliary functions @xmath20 and @xmath21 . the complex number @xmath22 is a spectral parameter . under the conditions on @xmath10 described above , it is known only that for each @xmath16 the discrete spectrum of this problem is invariant under complex conjugation and reflection through the origin . however , a formal wkb method applied to ( [ eq : zs ] ) suggests for small @xmath23 a distribution of eigenvalues that are confined to the imaginary axis . the same method suggests that the reflection coefficient for scattering states obtained for real @xmath22 is small beyond all orders . it is therefore natural to propose a modification of the problem . rather than studying the inverse - scattering problem for the true spectral data ( which is not known ) , simply replace the true spectral data by its formal wkb approximation in which the eigenvalues are given by a quantization rule of bohr - sommerfeld type , and in which the reflection coefficient is neglected entirely . for each @xmath16 , this modified spectral data is the _ true _ spectral data for some other ( @xmath23-dependent ) initial condition @xmath24 . since there is no reflection coefficient in the modified problem , it turns out that for each @xmath23 the solution of ( [ eq : nls ] ) corresponding to the modified initial data @xmath24 is an exact @xmath25-soliton solution , with @xmath26 . we call such a family of @xmath25-soliton solutions , all obtained from the same function @xmath10 by a wkb procedure , a _ semiclassical soliton ensemble _ , or sse for short . we will be more precise about this idea in [ sec : sses ] . in @xcite , the asymptotic behavior of sses was studied for @xmath6 . although the results were rigorous , it was not possible to deduce anything about the _ true _ initial - value problem for ( [ eq : nls ] ) with @xmath27 because the asymptotic method failed for @xmath3 . in this paper , we will explain the following new result . let @xmath10 be real - analytic , even , and decaying with a single genuine maximum at @xmath12 . let @xmath28 be for each @xmath16 the exact initial value of the sse corresponding to @xmath10 ( see [ sec : sses ] ) . then , there exists a sequence of values of @xmath23 , @xmath29 for @xmath30 , such that @xmath31 and such that for all @xmath32 , there exists a constant @xmath33 such that @xmath34 for all @xmath35 . [ theorem : mainresult ] as @xmath28 is obtained by an inverse - scattering procedure applied to wkb spectral data , this theorem establishes in a sense the validity of the wkb approximation for the zakharov - shabat eigenvalue problem ( [ eq : zs ] ) . it says that the true spectral data and the formally approximate spectral data generate , via inverse - scattering , potentials in the zakharov - shabat problem that are pointwise close . the omission of @xmath12 is merely technical ; a procedure slightly different from that we will explain in this paper is needed to handle this special case . we will indicate as we proceed the modifications that are necessary to extend the result to the whole real line . the pointwise nature of the asymptotics is important ; the variational methods used in @xcite suggest convergence only in the @xmath36 sense . rigorous statements about the nature of the wkb approximation for the zakharov - shabat problem are especially significant because the operator in ( [ eq : zs ] ) is nonselfadjoint and the spectrum is not confined to any axis ; furthermore sturm - liouville oscillation theory does not apply . each @xmath25-soliton solution of the focusing nonlinear schrdinger equation ( [ eq : nls ] ) can be found as the solution of a meromorphic riemann - hilbert problem with no jumps ; that is , a problem whose solution matrix is a rational function of @xmath37 . the @xmath25-soliton solution depends on a set of discrete data . given @xmath25 complex numbers @xmath38 in the upper half - plane ( these turn out to be discrete eigenvalues of the spectral problem ( [ eq : zs ] ) ) , and @xmath25 nonzero constants @xmath39 ( which turn out to be related to auxiliary discrete spectrum for ( [ eq : zs ] ) ) , and an index @xmath40 , one considers the matrix @xmath41 solving the following problem : find a matrix @xmath41 with the following two properties : 1 . * rationality : * @xmath41 is a rational function of @xmath22 , with simple poles confined to the values @xmath42 and the complex conjugates . at the singularities : @xmath43\sigma_1^{\frac{1-j}{2}}\,,\\\\ \displaystyle \mathop{\rm res}_{\lambda=\lambda_k^ * } { \bf m}(\lambda)&=&\displaystyle\lim_{\lambda\rightarrow\lambda_k^*}{\bf m}(\lambda ) \sigma_1^{\frac{1-j}{2}}\left[\begin{array}{cc}0 & -c_k(x , t)^*\\0 & 0\end{array } \right]\sigma_1^{\frac{1-j}{2}}\ , , \end{array } \label{eq : residuerelation}\ ] ] for @xmath44 , with @xmath45 2 . * normalization : * @xmath46 [ rhp : meromorphic ] here , @xmath47 denotes one of the pauli matrices : @xmath48\ , , \hspace{0.2 in } \sigma_2:=\left[\begin{array}{cc } 0 & -i\\i & 0\end{array}\right]\ , , \hspace{0.2 in } \sigma_3:=\left[\begin{array}{cc } 1 & 0\\0 & -1\end{array}\right]\,.\ ] ] the function @xmath17 defined from @xmath41 by the limit @xmath49 is the @xmath25-soliton solution of the focusing nonlinear schrdinger equation ( [ eq : nls ] ) corresponding to the data @xmath42 and @xmath50 . the index @xmath51 will be present throughout this work , so it is worth explaining its role from the start . it turns out that if @xmath52 , then the solution @xmath41 of riemann - hilbert problem [ rhp : meromorphic ] has the property that for all fixed @xmath22 distinct from the poles , @xmath53 as @xmath54 . likewise , if @xmath55 , then @xmath56 as @xmath57 . so as far as scattering theory is concerned , the index @xmath51 indicates an arbitrary choice of whether we are performing scattering `` from the right '' or `` from the left '' . both versions of scattering theory yield the same function @xmath17 via the relation ( [ eq : potentialrecovery ] ) , and are in this sense equivalent . however , the inverse - scattering problem involves the independent variables @xmath4 and @xmath5 for ( [ eq : nls ] ) as parameters , and it may be the case that for different choices of @xmath4 and @xmath5 , different choices of the parameter @xmath51 may be more convenient for asymptotic analysis of the matrix @xmath41 solving riemann - hilbert problem [ rhp : meromorphic ] . that this is indeed the case was observed and documented in @xcite . so we need the freedom to choose the index @xmath51 , and therefore we need to carry it along in our calculations . a _ semiclassical soliton ensemble _ ( sse ) is a family of particular @xmath25-soliton solutions of ( [ eq : nls ] ) indexed by @xmath58 that are formally associated with given initial data @xmath59 via an _ ad hoc _ wkb approximation of the spectrum of ( [ eq : zs ] ) . note that the initial data @xmath59 may not exactly correspond to a pure @xmath25-soliton solution of ( [ eq : nls ] ) for any @xmath23 , and similarly that typically none of the @xmath25-soliton solutions making up the sse associated with @xmath59 will agree with this given initial data at @xmath3 . we will now describe the discrete data @xmath42 and @xmath50 that generate , via the solution of riemann - hilbert problem [ rhp : meromorphic ] and the subsequent use of the formula ( [ eq : potentialrecovery ] ) , the sse associated with a function @xmath59 . we suppose that @xmath10 is an even function of @xmath4 that has a single maximum at @xmath12 , and is therefore `` bell - shaped '' . we will need @xmath10 to be rapidly decreasing for large @xmath4 , and we will suppose that the maximum @xmath60 is genuine in that @xmath15 . most importantly in what follows , we will assume that @xmath10 is a real - analytic function of @xmath4 . the starting point is the definition of the wkb eigenvalue density function @xmath61 : @xmath62 defined for positive imaginary numbers @xmath63 in the interval @xmath64 , where @xmath65 and @xmath66 are the ( unique by our assumptions ) negative and positive values of @xmath4 for which @xmath67 . the wkb eigenvalues asymptotically fill out the interval @xmath64 , and @xmath61 is their asymptotic density . this function inherits analyticity properties in @xmath63 from those of @xmath10 via the functions @xmath68 . our assumption that @xmath10 is real - analytic makes @xmath61 an analytic function of @xmath63 in its imaginary interval of definition . also , our assumption that @xmath10 should be rapidly decreasing makes @xmath61 analytic at @xmath69 , and our assumption that @xmath10 have nonvanishing curvature at its maximum makes @xmath61 analytic at @xmath70 . from this function it is convenient to define a measure of the number of wkb eigenvalues between a point @xmath71 on the imaginary axis and @xmath72 : @xmath73 now , each @xmath25-soliton solution in the sse for @xmath10 will be associated with a particular value @xmath29 , namely @xmath74 where @xmath75 . in this sense we are taking the values of @xmath23 themselves to be `` quantized '' . clearly for any given @xmath10 , @xmath76 which goes to zero as @xmath25 becomes large . for each @xmath75 , we then define the wkb eigenvalues formally associated with @xmath10 according to the bohr - sommerfeld rule @xmath77 and the auxiliary scattering data by @xmath78 here , @xmath79 is an arbitrary integer . clearly the bohr - sommerfeld rule ( [ eq : bs ] ) implies that choosing different integer values of @xmath79 in ( [ eq : gammainterpolate ] ) will yield the same set of numbers @xmath50 . however , we take the point of view that the right - hand side of ( [ eq : gammainterpolate ] ) furnishes an analytic function that interpolates the @xmath50 at the @xmath42 ; for different @xmath80 these are different interpolating functions which is a freedom that we will exploit to our advantage . in fact , we will only need to consider @xmath81 or @xmath82 . for @xmath10 given as above , the sse is a sequence of exact solutions of ( [ eq : nls ] ) such that the @xmath25th element @xmath83 of the sse ( i ) solves ( [ eq : nls ] ) with @xmath29 as given by ( [ eq : quantization ] ) and ( ii ) is defined as the @xmath25-soliton solution corresponding to the eigenvalues @xmath42 given by ( [ eq : bs ] ) and the auxiliary spectrum @xmath50 given by ( [ eq : gammainterpolate ] ) via the solution of riemann - hilbert problem [ rhp : meromorphic ] with @xmath29 . for each @xmath25 , we restrict the sse to @xmath3 to obtain functions @xmath84 it is this sequence of functions that is the subject of theorem [ theorem : mainresult ] . in the following sections we will set up a new framework for the asymptotic analysis of sses in the limit @xmath85 , a problem closely related to the computation of asymptotics of solutions of ( [ eq : nls ] ) for fixed initial data @xmath59 in the semiclassical limit . the asymptotic method we will now develop for studying riemann - hilbert problem [ rhp : meromorphic ] for sses is especially well - adapted to studying the case of @xmath3 , where the method described in detail in @xcite fails . to illustrate the new method , we therefore set @xmath3 in the rest of this paper . also , we anticipate the utility of tying the value of the parameter @xmath86 to the remaining independent variable @xmath4 by setting @xmath87 in all subsequent formulae in which the index @xmath51 appears it should be assumed to be assigned a definite value according to ( [ eq : jdef ] ) . we now want to convert riemann - hilbert problem [ rhp : meromorphic ] into a new riemann - hilbert problem for a sectionally holomorphic matrix so that the `` steepest - descent '' methods can be applied . as mentioned in the introduction , in @xcite this transformation can be accomplished by encircling the locus of accumulation of the poles , here the imaginary interval @xmath64 , with a loop contour in the upper half - plane and making a specific change of variables based on the interpolation formula ( [ eq : gammainterpolate ] ) for some value of @xmath80 in the interior of the region enclosed by the loop and also in the complex - conjugate region . one then tries to choose the position of the loop contour in the complex plane that is best adapted to asymptotic analysis of the resulting holomorphic riemann - hilbert problem . the trouble with this approach is that it turns out that for @xmath3 the `` correct '' placement of the contour requires that part of it should lie on a subset of the imaginary interval @xmath64 , that is , right on top of the accumulating poles ! for such a choice of the loop contour , the boundary values taken by the transformed matrix on the outside of the loop would be singular and the `` steepest descent '' theory would not apply . so taking the point of view that making any particular choice of @xmath80 in ( [ eq : gammainterpolate ] ) leads to problems , we propose to simultaneously make use of _ two distinct values of @xmath79 _ in passing to a riemann - hilbert problem for a sectionally holomorphic matrix . consider the contours illustrated in figure [ fig : twohalves ] , arranged such that @xmath88 . for @xmath89 , set @xmath90\sigma_1^{\frac{1-j}{2}}\ , . \label{eq : lefttransform}\ ] ] for @xmath91 , set @xmath92\sigma_1^{\frac{1-j}{2}}\ , . \label{eq : righttransform}\ ] ] for @xmath93 set @xmath94 , and for all other complex @xmath22 set @xmath95 . so rather than enclosing the poles in a loop and making a single change of variables inside , we are splitting the region inside the loop in half , and we are using different interpolants ( [ eq : gammainterpolate ] ) of the @xmath50 at the @xmath42 in each half of the loop . some of the properties of the transformed matrix @xmath96 are the following . the matrix @xmath96 is analytic in @xmath97 where @xmath98 is the union of the contours @xmath99 , @xmath100 , and @xmath101 , and their complex conjugates . moreover , @xmath96 takes continuous boundary values on @xmath98 . [ prop : manalytic ] _ proof : _ the function @xmath102 is analytic in @xmath103 and @xmath104 if @xmath99 and @xmath100 are chosen close enough to the imaginary axis since @xmath61 is analytic there . by using the residue relation ( [ eq : residuerelation ] ) and the interpolation formula ( [ eq : gammainterpolate ] ) alternatively for @xmath81 and @xmath82 , one checks directly that the poles of @xmath41 are canceled by the explicit blaschke factors in ( [ eq : lefttransform ] ) and ( [ eq : righttransform ] ) . @xmath105 let @xmath106 denote the boundary values taken on the oriented contour @xmath98 , where the subscript `` @xmath107 '' ( respectively `` @xmath108 '' ) indicates the boundary value taken from the left ( respectively from the right ) . then for @xmath109 , @xmath110\sigma_1^{\frac{1-j}{2}}\ , . \label{eq : mjumpcl}\ ] ] for @xmath111 , @xmath112\sigma_1^{\frac{1-j}{2}}\ , . \label{eq : mjumpcr}\ ] ] for @xmath113 , @xmath114\sigma_1^{\frac{1-j}{2}}\ , . \label{eq : mjumpcm}\ ] ] on the contours in the lower half - plane the jump relations are determined by the symmetry @xmath115 . all jump matrices are analytic functions in the vicinity of their respective contours . [ prop : mjump ] _ proof : _ this is also a direct consequence of ( [ eq : lefttransform ] ) and ( [ eq : righttransform ] ) . the analyticity is clear on @xmath99 and @xmath100 since @xmath102 is analytic there , while on @xmath101 one observes that as a consequence of the bohr - sommerfeld quantization condition ( [ eq : bs ] ) , the cosine factor precisely cancels the poles on @xmath101 contributed by the product of blaschke factors . although we have specified the contour @xmath101 to coincide with a segment of the imaginary axis , the reader will see that the same statements concerning the analyticity of @xmath96 and the continuity of the boundary values on @xmath98 also hold when @xmath101 is taken to be absolutely any smooth contour in the upper half - plane connecting @xmath116 to @xmath117 . given a choice of @xmath101 , the contours @xmath99 and @xmath100 must be such that the topology of figure [ fig : twohalves ] is preserved . we also have specified that @xmath99 and @xmath100 should lie sufficiently close to @xmath101 ( a distance independent of @xmath118 ) so that @xmath102 is analytic in @xmath103 and @xmath104 . later we will also exploit the proximity of these two contours to @xmath101 to deduce decay properties of certain analytic functions on these contours from their oscillation properties on @xmath101 by the cauchy - riemann equations . taken together , proposition [ prop : manalytic ] and proposition [ prop : mjump ] indicate that the matrix @xmath119 satisfies a riemann - hilbert problem without poles , but instead having explicit homogeneous jump relations on @xmath98 given by the matrix functions on the right - hand sides of ( [ eq : mjumpcl ] ) , ( [ eq : mjumpcr ] ) , and ( [ eq : mjumpcm ] ) . the normalization of @xmath96 at infinity is the same as that of @xmath120 since no transformation has been made outside a compact set , so if @xmath96 can be recovered from its jump relations and normalization condition , then the sse itself can be obtained for @xmath3 from ( [ eq : potentialrecovery ] ) with @xmath41 replaced by @xmath96 . we now introduce a further change of dependent variable involving a scalar function that is meant to capture the dominant asymptotics for the problem . let @xmath121 be a complex - valued function that is independent of @xmath23 , analytic for @xmath122 taking continuous boundary values , satisfies @xmath123 , and @xmath124 . setting @xmath125 we find that for @xmath109 , @xmath126 \sigma_1^{\frac{1-j}{2}}\ , , \label{eq : cljumpn}\ ] ] where @xmath127 similarly , for @xmath111 , we find @xmath128 \sigma_1^{\frac{1-j}{2}}\ , , \label{eq : crjumpn}\ ] ] where @xmath129 finally , for @xmath113 , @xmath130 \sigma_1^{\frac{1-j}{2}}\ , , \label{eq : cmjumpn}\ ] ] where @xmath131 and @xmath132 this means that given a function @xmath121 with the properties described above , one finds that the matrix @xmath133 satisfies another holomorphic riemann - hilbert problem with jump conditions determined from ( [ eq : cljumpn ] ) , ( [ eq : crjumpn ] ) , and ( [ eq : cmjumpn ] ) . because @xmath124 and @xmath121 is analytic near infinity , it follows that the correct normalization condition for @xmath133 is again that @xmath134 as @xmath135 . these same conditions on @xmath121 show that if @xmath133 can be found from its jump conditions and normalization condition , then the sse can be found via ( [ eq : potentialrecovery ] ) with @xmath120 replaced by @xmath133 . the function @xmath121 is called a _ complex phase function_. the advantage of introducing it into the problem is that by choosing it correctly , the jump matrices ( [ eq : cljumpn ] ) , ( [ eq : crjumpn ] ) , and ( [ eq : cmjumpn ] ) can be cast into a form that is especially convenient for analysis in the semiclassical limit of @xmath136 . the idea of introducing the complex phase function to assist in finding the leading - order asymptotics and controlling the error in this way first appeared in the paper @xcite as a modification of the `` steepest - descent '' method proposed in @xcite . for our purposes , we would like to have each element of the jump matrix for @xmath133 of the form @xmath137 for some appropriate function @xmath138 that is independent of @xmath118 . while this is not true strictly speaking , it becomes a good approximation in the limit @xmath136 with @xmath22 held fixed ( the approximation is not uniform near @xmath116 or @xmath117 ) . in this section , we describe the pointwise asymptotics of the jump matrix for @xmath133 with the aim of writing all nonzero matrix elements asymptotically in the form @xmath137 with a small relative error whose magnitude we can estimate . roughly speaking , the intuition is that the product over @xmath139 of blaschke factors should be replaced with an exponential of a sum over @xmath139 of logarithms . the latter sum goes over to an integral that scales like @xmath140 in the semiclassical limit . on the contour @xmath101 , the cosine that cancels the poles must also be encorporated into the asymptotics . the branch of the logarithm that is convenient to use here is most conveniently viewed as a function of two complex variables : @xmath141 as a function of @xmath22 for fixed @xmath63 , it is a logarithm that is cut downwards in the negative imaginary direction from the logarithmic pole at @xmath142 . equivalently , @xmath143 can be viewed as the branch of the multivalued function @xmath144 for which @xmath145 . suppose @xmath146 . the boundary value of @xmath147 taken on @xmath101 as @xmath22 approaches from the left ( respectively right ) side is denoted by @xmath148 ( respectively @xmath149 ) . the average of these two boundary values is denoted by @xmath150 . all the results we need will come from studying the asymptotic behavior of two quotients : @xmath151 and @xmath152 the function @xmath153 is analytic and nonvanishing for @xmath154 . let us denote by @xmath155 the domain of analyticity of @xmath156 restricted to the upper half - plane , so that by our assumptions on @xmath10 , @xmath157 . then , due to the zeros of the cosine on the imaginary axis , which match the poles of the product below @xmath117 and are not cancelled above @xmath117 , @xmath158 is analytic and nonvanishing for @xmath159 , where @xmath160 is the vertical ray from @xmath117 to infinity along the positive imaginary axis . the domain of analyticity for @xmath158 is a subset of @xmath161 rather than of the whole upper half - plane due to the presence of the averages of the logarithms in the integrand of ( [ eq : t ] ) . whereas these are boundary values defined _ a priori _ only on @xmath101 , the integrals extend from @xmath101 to analytic functions in the domain @xmath162 via the introduction of the function @xmath102 ( _ cf . _ equation ( [ eq : logjump ] ) ) . for all @xmath22 in the upper half - plane with latexmath:[$\hbar_n\le sufficiently small , but fixed as @xmath136 , @xmath164 [ lemma : souter ] _ proof : _ let us define the function @xmath165 by @xmath166 this analytic function takes the imaginary interval @xmath167 $ ] to the real interval @xmath168 $ ] where @xmath169 since @xmath170 does not vanish on @xmath101 , we have the inverse function @xmath171 defined for @xmath172 near the real interval @xmath168 $ ] . using these tools , we get the following representation for @xmath153 : @xmath173 and @xmath174\,dm - \left[l^0_{-e(m_k)}(\lambda)-l^0_{e(m_k)}(\lambda)\right]\,,\ ] ] with @xmath175 . expanding the logarithms , we find that @xmath176\ , . \label{eq : integralformula}\ ] ] this quantity is clearly @xmath177 for @xmath22 fixed away from @xmath101 . now , when @xmath178 as @xmath179 , we can estimate the denominator in the integrand to obtain two different bounds : @xmath180 and @xmath181 the idea is to use the estimate ( [ eq : lambdadenominator ] ) when @xmath182 is close to @xmath183 and to use the estimate ( [ eq : xidenominator ] ) for the remaining terms . suppose first @xmath184 is bounded between @xmath185 and @xmath186 , _ i.e. _ there are small fixed positive numbers @xmath187 and @xmath188 so that @xmath189 , and let @xmath190 be a small positive scale tied to @xmath23 and satisfying @xmath191 , and let @xmath192 be chosen from @xmath193 so that @xmath194 is as close as possible to @xmath195 , and likewise let @xmath196 be chosen from @xmath193 so that @xmath197 is as close as possible to @xmath198 . using ( [ eq : lambdadenominator ] ) we then find that @xmath199 because the sum contains @xmath200 terms and the volume of the region of integration for each term is @xmath201 , and we must take into account the overall factor of @xmath202 . now in each of the remaining terms @xmath203 , we have @xmath204 so using ( [ eq : xidenominator ] ) and summing over @xmath139 we get both @xmath205 the total estimate of @xmath206 is then optimized by a dominant balance among the three partial sums . this balance requires taking @xmath207 , upon which we deduce that under our assumptions on @xmath22 , we indeed have @xmath208 when @xmath184 is bounded between @xmath185 and @xmath186 . when @xmath209 or @xmath210 , the estimate ( [ eq : lambdadenominator ] ) should be used only for those terms that correspond to @xmath172 near zero or @xmath172 near @xmath211 respectively . in both of these exceptional cases , the same estimate is found . when @xmath184 is bounded below by @xmath186 , there is no need to use the estimate ( [ eq : lambdadenominator ] ) at all , and the relative error is of order @xmath118 uniformly in @xmath212 . this completes the proof . we now use this information about @xmath153 to effectively replace the sums of logarithms by integrals , at least on some portions of the contour @xmath98 . suppose that the contour @xmath99 is independent of @xmath118 and that for some sufficiently small positive number @xmath213 , @xmath99 lies in the strip @xmath214 and meets the imaginary axis only at its endpoints and does so transversely . then @xmath215 as @xmath118 goes to zero through positive values , for all @xmath216 with @xmath217 and @xmath218 . [ proposition : clout ] suppose that the contour @xmath100 is independent of @xmath118 and that for some sufficiently small positive number @xmath213 , @xmath100 lies in the strip @xmath219 and meets the imaginary axis only at its endpoints and does so transversely . then @xmath220 as @xmath118 goes to zero through positive values , for all @xmath221 with @xmath217 and @xmath218 . [ proposition : crout ] _ proof of propositions [ proposition : clout ] and [ proposition : crout ] : _ these propositions follow directly from lemma [ lemma : souter ] upon using the transversality of the intersections with the imaginary axis to replace @xmath222 by @xmath223 . @xmath105 we notice that the first factor on the second line in ( [ eq : lasymp ] ) and the first factor on the second line in ( [ eq : rasymp ] ) are both exponentially small as @xmath118 goes to zero through positive values , as a consequence of the fact that @xmath224 is an analytic negative real measure on @xmath101 . this follows from the cauchy - riemann equations and the geometry of figure [ fig : twohalves ] . it will be a very useful fact for us shortly . now we turn our attention to the function @xmath158 . the result analogous to lemma [ lemma : souter ] is the following . for all @xmath22 in the upper half - plane with latexmath:[$\hbar_n\le sufficiently small , but fixed as @xmath136 , @xmath225 [ lemma : touter ] _ proof : _ we begin with the jump condition @xmath226 relating the boundary values of the logarithm @xmath143 on the imaginary axis . using this jump relation and the definition of @xmath150 as the average of the boundary values of @xmath227 and @xmath228 , we see that for @xmath229 , we have @xmath230 while for @xmath231 , we have @xmath232 now , using the fact that @xmath61 is an analytic function satisfying @xmath233 for @xmath234 , we see by the cauchy - riemann equations that in both cases , the exponential relative error term is of the order @xmath235 for some @xmath236 . since this is negligible compared with the relative error associated with the asymptotic approximation of @xmath153 given in lemma [ lemma : souter ] , the proof is complete . @xmath105 unfortunately , we need asymptotic information about @xmath158 right on the imaginary axis , which contains the contour @xmath101 , so we need to improve upon lemma [ lemma : touter ] . we begin to extract this additional information by noting that under some circumstances , it is easy to show that @xmath158 remains bounded in the vicinity of the imaginary axis . if either ( i ) @xmath22 is real or ( ii ) @xmath237 and @xmath238 , and if for some @xmath239 sufficiently small @xmath240 , then @xmath158 is uniformly bounded as @xmath136 . [ lemma : tbounded ] _ proof : _ it suffices to show that @xmath153 is bounded under the same assumptions , because from ( [ eq : tleft ] ) and ( [ eq : tright ] ) and the cauchy - riemann equations , we see easily that @xmath241 . using the function @xmath242 and its inverse @xmath243 , we have the following @xmath244 where @xmath245 when @xmath246 , we see immediately that @xmath247 , and therefore @xmath248 and hence @xmath249 . now consider @xmath250 with @xmath251 sufficiently small independent of @xmath118 . the idea is that of the terms on the right - hand side of ( [ eq : difference ] ) , the discrete sum is a riemann sum approximation to the integral . the riemann sum is constructed using the midpoints of @xmath25 equal subintervals as sample points . if @xmath252 is bounded uniformly , then this sort of riemann sum provides an approximation to the integral that is of order @xmath253 or equivalently @xmath254 . in this case , we deduce that @xmath255 and in particular this is bounded as @xmath118 tends to zero . but as @xmath22 approaches the imaginary axis , the accuracy of the approximation is lost . for @xmath250 , the function @xmath256 satisfies @xmath257 and takes its maximum when @xmath258 , with a maximum value @xmath259 therefore , as @xmath251 tends to zero , @xmath256 becomes unbounded , growing logarithmically in @xmath251 . as a consequence of this blowup the approximation of the integral by the riemann sum based on midpoints for @xmath237 fails to be second - order accurate uniformly in @xmath251 . however , because the maximum of @xmath256 always occurs at the right endpoint , it is easy to see that when the error becomes larger than @xmath177 in magnitude its sign is such that the riemann sum is always an _ underestimate _ of the value of the integral , and consequently the right - hand side of ( [ eq : difference ] ) is negative . this is concretely illustrated in figure [ fig : riemannsums ] where we have taken the example of the gaussian function @xmath260 in order to supply the function @xmath61 and therefore the function @xmath261 needed to build @xmath256 . in this case , @xmath262 and @xmath263 . the error of the riemann sum is worst when @xmath264 . in this case it is easy to see that the discrepancy contributed by only the subinterval adjacent to the logarithmic singularity of @xmath256 is @xmath265 , which clearly dominates the @xmath177 error contributed by the majority of the subintervals bounded away from @xmath258 . consequently , for those @xmath22 on the circle @xmath237 for which @xmath266 is not asymptotically small in @xmath118 , it is negative , and therefore @xmath153 is uniformly bounded for @xmath237 , as is @xmath158 . @xmath105 using this information , we can finally extract enough information about @xmath158 on the imaginary axis to approximate @xmath267 for @xmath113 . let @xmath101 be a fixed contour from @xmath116 to @xmath117 lying between @xmath99 and @xmath100 , possibly coinciding with the imaginary axis . then , for @xmath268 arbitrarily small , @xmath269 as @xmath118 goes to zero through positive values , for all @xmath113 with @xmath217 and @xmath218 . [ prop : a_m ] _ proof : _ let @xmath270 be the closed contour illustrated in figure [ fig : c ] . this counter - clockwise oriented contour consists of two vertical segments , one horizontal segment that lies on the real axis , and an arc of the circle of radius @xmath186 centered at the origin . the function @xmath158 is analytic on the interior of @xmath270 and is continuous on @xmath270 itself . in fact it is analytic on most of the boundary , failing to be analytic only at @xmath116 and @xmath117 . therefore for any @xmath22 in the interior , we may write @xmath271 if we let @xmath272 denote the part of @xmath270 with @xmath273 , and let @xmath274 denote the remaining portion of @xmath270 , then we get @xmath275 using the estimate guaranteed by lemma [ lemma : touter ] in the integral over @xmath274 , and the uniform boundedness of @xmath276 ( and therefore of @xmath277 ) guaranteed by lemma [ lemma : tbounded ] in the integral over @xmath272 , we find @xmath278 for some positive constants @xmath279 and @xmath280 . replacing the logarithm by a slightly cruder estimate of @xmath281 for arbitrarily small positive @xmath282 completes the proof . we have therefore succeeded in showing that , at least away from the self - intersection points of the contour @xmath98 , the jump matrices for @xmath133 as defined by ( [ eq : cljumpn ] ) for @xmath216 , ( [ eq : crjumpn ] ) for @xmath111 , and ( [ eq : cmjumpn ] ) for @xmath113 are well - approximated in the semiclassical limit @xmath136 by matrices in which all nonzero matrix elements are of the form @xmath137 with @xmath138 being independent of @xmath118 . the fact that this approximation is valid even when the `` active '' contour @xmath101 is taken to be right on top of the poles of the meromorphic riemann - hilbert problem for @xmath120 is an advantage over the approach taken in @xcite . using these approximations , we can introduce an _ ad hoc _ approximation of the matrix @xmath133 . first , define @xmath283 and for @xmath109 or @xmath100 , define @xmath284 then we pose the following problem . given a complex phase function @xmath121 find a matrix @xmath285 satisfying : 1 . * analyticity : * @xmath286 is analytic for @xmath287 . * boundary behavior : * @xmath286 assumes continuous boundary values on @xmath98 . 3 . * jump conditions : * the boundary values taken on @xmath98 satisfy @xmath288\sigma_1^{\frac{1-j}{2}}\ ] ] for @xmath109 , @xmath289\sigma_1^{\frac{1-j}{2}}\ ] ] for @xmath111 , and @xmath290\sigma_1^{\frac{1-j}{2 } } \label{eq : ntildemiddle}\ ] ] for @xmath113 . for all other @xmath291 ( that is , in the lower half - plane ) , the jump is determined by the symmetry @xmath292 . * normalization : * @xmath286 is normalized at infinity : @xmath293 let @xmath294 be defined by the equation @xmath295 , the fact that @xmath294 is an analytic function for @xmath22 away from the imaginary interval @xmath296 $ ] , and the normalization that for large @xmath22 , @xmath297 . for @xmath298 , let @xmath299 it is easy to check directly that for all @xmath298 , we have @xmath300 . also , using the fact that @xmath301 is purely imaginary on the imaginary axis , and that @xmath302 is purely imaginary in the domain of integration , where it satisfies @xmath303 , we see that @xmath304 furthermore , it follows easily from ( [ eq : rhodefine ] ) that for all @xmath298 , we have @xmath305 with the lower constraint being achieved only at the endpoint being extended to all of @xmath101 by setting @xmath306 for @xmath22 above the endpoint @xmath307 . in this case one views the lower constraint as being active on the whole imaginary interval @xmath308 $ ] . ] of @xmath309 , @xmath310 , and the upper constraint being achieved only at the origin in accordance with ( [ eq : rhosequal ] ) . now , set @xmath311 this function satisfies all of the basic criteria set out earlier : it is analytic in @xmath312 and takes continuous boundary values , it satisfies @xmath123 , and it satisfies @xmath124 because @xmath313 note that @xmath121 is analytic across @xmath101 for @xmath22 above @xmath307 . consequently @xmath314 for all such @xmath22 . for @xmath113 below @xmath307 , @xmath315 becomes ( _ cf . _ equation ( [ eq : defthetaoriginal ] ) ) @xmath316 we now describe a number of important consequences of our choice of @xmath121 . for all @xmath317 $ ] , @xmath318 . [ prop : phitildezero ] to prove the proposition , we first point out that @xmath319 simply as a consequence of the fact that both @xmath61 and @xmath320 are purely imaginary on @xmath101 . next we point out that @xmath321 whenever @xmath322 $ ] . this follows from a direct calculation in which all integrals are evaluated by residues and the formula ( [ eq : rho0 ] ) is used . next we consider @xmath323 for @xmath324 $ ] , that is , above the endpoint of the support . clearly , @xmath325 is the boundary value on @xmath101 of an analytic function defined near @xmath101 in @xmath103 . since the boundary value taken below the endpoint is @xmath326 because @xmath327 there , and the boundary value taken above the endpoint is @xmath323 because @xmath328 there , we obtain the formula @xmath329 valid for @xmath113 above @xmath307 , where by @xmath330 for @xmath63 in the imaginary interval @xmath331 we mean the function @xmath320 defined by ( [ eq : rhodefine ] ) for @xmath63 in the imaginary interval @xmath332 , analytically continued from @xmath332 in the clockwise direction about the endpoint @xmath310 . in particular for such @xmath22 we have @xmath333 carrying out the analytic continuation , we find from ( [ eq : rhodefine ] ) that for @xmath334 , @xmath335 from this formula we see easily that for all @xmath22 strictly above the endpoint @xmath307 , @xmath336 is positive real . consequently , from ( [ eq : phiprime ] ) and since @xmath318 for @xmath310 , we get the following result . the function @xmath323 is negative real and decreasing in the positive imaginary direction for @xmath337 $ ] . [ prop : tildephineg ] now we consider the behavior of the function @xmath338 on @xmath99 and @xmath100 . from the definitions of the functions @xmath338 and @xmath323 , we see that for @xmath109 , @xmath339 and for @xmath111 , @xmath340 that is , the analytic function @xmath338 takes boundary values from the left on @xmath101 equal to @xmath341 and from the right on @xmath101 equal to @xmath342 . first consider the situation to the left or right of the imaginary interval @xmath343 $ ] . since @xmath327 in @xmath343 $ ] , the function @xmath338 on @xmath99 will be the analytic continuation of @xmath344 from @xmath101 and the function @xmath338 on @xmath100 will be the analytic continuation of @xmath345 from @xmath101 . from ( [ eq : upperlowerconstraints ] ) we see that for @xmath346 $ ] one has @xmath347 . therefore , it follows from the cauchy - riemann equations that for @xmath22 in portions of @xmath99 and @xmath100 close enough ( independently of @xmath118 ) to the interval @xmath343 $ ] one has @xmath348 for @xmath22 on _ both _ @xmath99 and @xmath100 . furthermore , it follows from the fact that @xmath233 that @xmath349 for @xmath109 and @xmath350 for @xmath111 . therefore @xmath351 for @xmath109 near the portion of @xmath101 below @xmath307 , and @xmath352 for @xmath22 in the analogous portion of @xmath100 . next consider the situation to the left or right of the portion of @xmath101 lying above the endpoint @xmath310 . since @xmath328 and @xmath353 for @xmath354 $ ] we see that for @xmath99 and @xmath100 close enough ( again independently of @xmath118 ) to this part of @xmath101 we again find that we have ( [ eq : decayleft ] ) on @xmath99 and ( [ eq : decayright ] ) on @xmath100 . this shows that the jump matrix on both contours @xmath99 and @xmath100 is an exponentially small perturbation of the identity for small positive @xmath118 , pointwise in @xmath22 bounded away from the origin and @xmath72 . for @xmath355 $ ] , the jump matrix for @xmath286 factors ( recall @xmath327 here ) : @xmath356 = \left[\begin{array}{cc } 1 & \displaystyle -i\exp\,\left(\frac{i\theta(\lambda)}{\hbar_n}\right)\\\\ 0 & 1\end{array}\right ] \left[\begin{array}{cc } 0 & i\\\\i&0\end{array}\right ] \left[\begin{array}{cc } 1 & \displaystyle -i\exp\,\left(-\frac{i\theta(\lambda)}{\hbar_n}\right)\\\\ 0 & 1\end{array}\right]\ , . \label{eq : factorization}\ ] ] let @xmath357 and @xmath358 be two boundaries of a lens surrounding @xmath167 $ ] . see figure [ fig : twohalveslens ] . using the factorization ( [ eq : factorization ] ) , we new define a new matrix function @xmath359 . in the region between @xmath357 and @xmath101 set @xmath360\sigma_1^{\frac{1-j}{2}}\ , . \label{eq : onleft}\ ] ] in the region between @xmath101 and @xmath358 , set @xmath361\sigma_1^{\frac{1-j}{2}}\ , . \label{eq : onright}\ ] ] elsewhere in the upper half - plane set @xmath362 . and in the lower half - plane define @xmath359 by symmetry : @xmath363 . these transformations imply jump conditions satisfied by @xmath364 on the contours in figure [ fig : twohalveslens ] since the jump conditions for @xmath286 are given . for @xmath365 we have @xmath366\sigma_1^{\frac{1-j}{2}}\ ] ] which is an exponentially small perturbation of the identity except near the endpoints . and for @xmath367 we have @xmath368\sigma_1^{\frac{1-j}{2}}\ ] ] which is also a jump that is exponentially close to the identity . for @xmath355 $ ] we get @xmath369\ ] ] as a consequence of the factorization ( [ eq : factorization ] ) . since @xmath370 for all @xmath22 in the upper half - plane outside the lens bounded by @xmath357 and @xmath358 , we see that @xmath359 satisfies the following jump condition on @xmath99 : @xmath371\sigma_1^{\frac{1-j}{2}}\ ] ] the following jump relation on @xmath100 : @xmath372\sigma_1^{\frac{1-j}{2}}\,,\ ] ] and the following jump relation on the imaginary interval @xmath308\subset c_m$ ] : @xmath373 \sigma_1^{\frac{1-j}{2}}\,.\ ] ] all three of these matrices are exponentially close to the identity matrix pointwise in @xmath22 for interior points of their respective contours . the matrix @xmath359 is related to @xmath286 by explicit transformations . however , taking the pointwise limit of the jump matrix for @xmath359 leads us to the following _ ad hoc _ model problem . find a matrix @xmath374 satisfying : 1 . * analyticity : * @xmath374 is analytic for @xmath375 , where @xmath309 is the imaginary interval @xmath376 $ ] . * boundary behavior : * @xmath374 assumes boundary values that are continuous except at @xmath377 , where at worst inverse fourth - root singularities are admitted . 3 . * jump condition : * for @xmath378 , @xmath379\,.\ ] ] 4 . * normalization : * @xmath374 is normalized at infinity : @xmath380 this model problem is easily solved explicitly . the unique solution of this riemann - hilbert problem is @xmath381\ , , \label{eq : otildesolution}\ ] ] where @xmath382 and @xmath383 with both functions @xmath294 and @xmath384 being analytic in @xmath385 , normalized according to @xmath386 and @xmath387 as @xmath135 . using this matrix , we define an `` outer '' model for the matrix @xmath388 as follows . the idea is to recall the relationship between the matrix @xmath286 and @xmath359 , and simply substitute @xmath374 for @xmath359 in these formulae . for @xmath22 in between @xmath357 and @xmath101 , we use ( [ eq : onleft ] ) to set @xmath389\sigma_1^{\frac{1-j}{2}}\,.\ ] ] for @xmath22 in between @xmath101 and @xmath358 , we use ( [ eq : onright ] ) to set @xmath390\sigma_1^{\frac{1-j}{2}}\,.\ ] ] for all other @xmath22 in the upper half - plane , set @xmath391 , and in the lower half - plane set @xmath392 . the important properties of this matrix are the following . the matrix @xmath393 is analytic for all complex @xmath22 except at the contours @xmath357 , @xmath358 , the imaginary interval @xmath343 $ ] , and their complex - conjugates . it satisfies the following jump conditions : @xmath394\sigma_1^{\frac{1-j}{2}}\ , , \hspace{0.2 in}\mbox{for}\hspace{0.2 in } \lambda\in l_l\,,\ ] ] @xmath395\sigma_1^{\frac{1-j}{2}}\ , , \hspace{0.2 in}\mbox{for}\hspace{0.2 in } \lambda\in l_r\,,\ ] ] @xmath396 \sigma_1^{\frac{1-j}{2}}\,,\hspace{0.2 in } \mbox{for}\hspace{0.2 in}\lambda\in [ 0,ia(x)]\,,\ ] ] with the jump matrices on the conjugate contours in the lower half - plane being obtained from these by the symmetry @xmath397 . in particular , note that for @xmath355 $ ] , we have @xmath398 . also , if @xmath399 is any given open set containing the endpoint @xmath310 , then @xmath393 is uniformly bounded for @xmath400 with a bound that depends only on @xmath399 and not on @xmath118 . [ prop : noutproperties ] in justifying formally the local model @xmath401 , we ignored the fact that the pointwise asymptotics for the jump matrices for @xmath359 that we used to obtain the matrix @xmath374 were not uniform near the origin or near the moving endpoint @xmath310 . we also neglected the breakdown of the asymptotics for @xmath402 , @xmath403 , and @xmath267 near the points @xmath116 and @xmath117 . consequently , we do not expect the outer model @xmath401 to be a good approximation to @xmath133 near @xmath116 , @xmath310 , or @xmath117 . in this section , we examine the neighborhoods of these three points in more detail , and we will obtain accurate local models for @xmath133 in the corresponding neighborhoods . . let @xmath404 and @xmath405 be small scales tied to @xmath118 such that @xmath406 as @xmath179 . let @xmath407 be defined as the unique integer for which exactly @xmath408 of the numbers @xmath38 lie strictly below @xmath409 on the positive imaginary axis . we want to compute uniform asymptotics for @xmath153 defined by ( [ eq : s ] ) for @xmath410 , and for @xmath158 defined by ( [ eq : t ] ) for @xmath113 when @xmath411 . [ lemma : originpeelfirst ] when @xmath412 and @xmath411 and with @xmath407 defined as above , @xmath413 _ proof : _ we recall the integral formula ( _ cf . _ equation ( [ eq : integralformula ] ) ) @xmath414 in which we expand the integrand in partial fractions : @xmath415 since @xmath412 , for @xmath416 and @xmath417 , we get @xmath418 for such @xmath419 we therefore have @xmath420 so summing over @xmath139 gives @xmath421 because @xmath422 , which proves the lemma . so only the fraction of terms @xmath203 with @xmath423 contribute significantly to the sum for @xmath206 . it is easy to check directly that @xmath424 is an analytic function for @xmath411 whenever @xmath425 , so it makes no difference in these terms whether it is @xmath143 or @xmath150 that appears in the definition of @xmath426 . therefore , the terms in @xmath153 and @xmath158 that can be significant for @xmath22 near the origin are thus @xmath427 and @xmath428 here we have written the integrals in the exponent using the change of variables @xmath429 . so lemma [ lemma : originpeelfirst ] simply says that @xmath430 and @xmath431 uniformly for @xmath432 . when @xmath22 is close to the origin along with the points @xmath433 contributing to @xmath158 , the ladder of discrete nodes appears to become equally spaced . the next lemma shows that this is indeed the case . [ lemma : originpeelsecond ] let @xmath434 for @xmath435 be the sequence of numbers defined by the relation : @xmath436 which results from expanding the bohr - sommerfeld relation ( [ eq : bs ] ) for @xmath437 small , and keeping only the dominant terms . define @xmath438 and @xmath439 then , for @xmath412 and @xmath411 , @xmath440 where we suppose that the scale @xmath404 is further constrained so that the relative error is asymptotically small . if @xmath22 is additionally bounded outside of some sector containing the positive imaginary axis , then @xmath441 _ proof : _ we begin by observing that for @xmath442 , the distance between @xmath433 and @xmath443 is much smaller than the distance between @xmath433 and @xmath444 , as long as @xmath445 . more precisely , we have @xmath446 decompose the quotients as follows : @xmath447 where @xmath448 @xmath449 @xmath450 -[\overline{l}^0_{-e(m)}(\lambda)-\overline{l}^0_{-e'(0)m}(\lambda)]\right)\,dm \right)\ , , \label{eq : ellbar}\ ] ] and @xmath451 -[l^0_{-e(m)}(\lambda)-l^0_{-e'(0)m}(\lambda)]\right)\,dm \right)\ , . \label{eq : ell}\ ] ] first we deal with @xmath452 and @xmath453 . since @xmath261 is differentiable and @xmath172 is small we have @xmath454 . also , the interval of integration is @xmath455 in length . although the integrands in ( [ eq : ellbar ] ) and ( [ eq : ell ] ) are not pointwise small , upon integration it follows that @xmath456 uniformly for all @xmath22 in the upper half - plane satisfying @xmath411 . here we are assuming that @xmath445 . for the moment , let s drop the conditions @xmath412 and @xmath411 and instead consider @xmath22 to lie on the sides of the square centered at the origin , one of whose sides is parallel to the real axis and intersects the positive imaginary axis halfway between the points @xmath457 and @xmath458 . note that the estimate ( [ eq : twinning ] ) implies that the sides of the square intersect the real and imaginary axes a distance from the origin that is approximately @xmath404 . therefore the square asymptotically contains the closed disk @xmath411 because @xmath422 . we will show that for @xmath22 on the four sides of the square , both @xmath459 and @xmath460 are very close to one . we write @xmath459 in the form @xmath461 first consider the top of the square : for @xmath462 , we easily see that @xmath463 for @xmath442 . combining this with ( [ eq : twinning ] ) , we get @xmath464 summing these estimates over @xmath139 ( it is convenient to approximate sums by integrals in doing so ) , we find that @xmath465 consequently , for @xmath22 on the top of the square , @xmath466 an estimate of the same form holds when @xmath22 is on the bottom of the square , where @xmath467 . when @xmath22 is on the left or right side of the square , so that @xmath468 , both @xmath469 and @xmath470 are @xmath471 . by the same arguments as above , we then have for such @xmath22 that @xmath472 now we look at @xmath460 on the same square . generally , for such @xmath22 which are of order @xmath404 in magnitude , we have @xmath473 when @xmath22 is on the top or bottom of the square , we have @xmath474 and when @xmath22 is on the left or right sides of the square , the same quantity is exponentially small . it follows easily that for @xmath22 on any of the sides of the square , @xmath475 so uniformly on the four sides of the square , we have @xmath476 but the product @xmath477 is analytic within the square , so by the maximum principle it follows that the same estimate holds for all @xmath22 on the interior of the square , and in particular for all @xmath22 in the upper half - plane with @xmath411 . this shows that @xmath478 holds for all such @xmath22 . now to control the relationship between @xmath479 and @xmath480 we consider @xmath22 to lie outside of some symmetrical sector about the positive imaginary axis , of arbitrarily small nonzero opening angle @xmath481 independent of @xmath118 . since @xmath412 , we get @xmath482 summing these estimates over @xmath139 one finds that @xmath472 combining this with the estimate ( [ eq : elelbar ] ) of @xmath483 , we find that @xmath484 for all @xmath22 in the upper half - plane with @xmath432 and bounded outside of the sector of opening angle @xmath481 about the positive imaginary axis . this completes the proof . @xmath105 without any approximation , @xmath480 can be rewritten in the form : @xmath485 and @xmath486 can be rewritten in the form : @xmath487 where @xmath488 and we are introducing a transformation @xmath489 to a local variable @xmath490 given by @xmath491 these formulae come from evaluating the logarithmic integrals exactly , which is possible because @xmath261 has been replaced by the linear function @xmath492 , taking advantage of the equal spacing of the @xmath443 to write the product explicitly in terms of gamma functions , and then using the reflection identity for the gamma function to eliminate the cosine from @xmath486 . now , the integer @xmath493 is large , approximately of size @xmath494 . but for @xmath411 , @xmath493 is asymptotically large compared to @xmath490 because @xmath422 . these observations allow us to apply stirling - type asymptotics to @xmath480 and @xmath486 . [ lemma : originpeelthird ] in addition to all prior hypotheses , suppose that @xmath495 . then , @xmath496 and @xmath497 _ proof : _ asymptotically expanding the gamma functions for large @xmath493 , we find that @xmath498 and @xmath499 where @xmath500 next , expanding @xmath501 , one gets worse error terms : @xmath502 combining these estimates and noting that @xmath503 completes the proof of the lemma . @xmath105 with these results in hand , we can easily establish the following . let @xmath22 be in the upper half - plane , with @xmath504 , where @xmath505 , and let @xmath22 be bounded outside of some fixed symmetrical sector containing the positive imaginary axis . then @xmath506 where @xmath507 . [ prop : sinner ] _ proof : _ according to lemmas [ lemma : originpeelfirst ] , [ lemma : originpeelsecond ] , and [ lemma : originpeelthird ] , the total relative error is a sum of three terms : @xmath508 note that since @xmath509 , the order @xmath510 term is always dominated asymptotically by the order @xmath511 term . the error is optimized by picking @xmath404 so that the two possibly dominant terms are in balance . this forces us to choose @xmath512 . the proposition follows upon taking @xmath513 . @xmath105 let @xmath22 be in the upper half - plane , with @xmath504 , where @xmath505 . then for all @xmath35 , however small , @xmath514 where @xmath507 . [ prop : tinner ] _ proof : _ in this case , according to lemmas [ lemma : originpeelfirst ] , [ lemma : originpeelsecond ] , and [ lemma : originpeelthird ] , the total relative error is a sum of three different terms : @xmath515 again , since @xmath509 , the order @xmath510 term is always dominated asymptotically by the order @xmath511 term . for any @xmath516 , we have @xmath517 so we can eliminate the logarithm at the expense of a slightly larger error . taking @xmath518 as in the statement of the proposition , and using the cruder estimate ( [ eq : cruder ] ) , the nearly optimal value of @xmath404 to minimize the total relative error is achieved by a dominant balance between the right - hand side of ( [ eq : cruder ] ) and the term of order @xmath511 . the balance gives @xmath519 , with @xmath520 with this choice of @xmath404 , the total relative error is of the order @xmath521 , with @xmath522 with the inequality following because @xmath516 and @xmath523 . the inequality fails in the limit @xmath524 . therefore , for each arbitrarily small @xmath35 , we can find a @xmath516 sufficiently small that @xmath525 . this gives us a slightly less optimal estimate of the relative error : simply @xmath526 , which completes the proof . @xmath105 to repair the flaw in our model @xmath393 for the matrix @xmath133 related to the nonuniformity of the approximation of the jump matrices near the origin , we need to provide a different approximation of @xmath133 that will be valid when @xmath504 for some @xmath527 . the local failure of the approximation is gauged by the deviation of the matrix quotient @xmath528 from the identity matrix near the origin . it turns out to be more convenient to study a conjugated form of this matrix ( which also deviates from the identity for @xmath22 near the origin ) . namely , for @xmath504 , set @xmath529 if @xmath229 and @xmath530 if @xmath231 . it is easy to check that as a consequence of the boundary conditions satisfied by the matrix @xmath374 on the imaginary axis near the origin , the conjugating factors are analytic throughout the disk @xmath504 . for later convenience , let us assume without loss of generality that the auxiliary contours @xmath99 , @xmath100 , @xmath357 and @xmath358 are straight rays in some @xmath118-independent neighborhood of the origin . it is easy to write down the jump conditions satisfied by @xmath531 on these four rays and also on the positive imaginary axis . we find : @xmath532\,,\hspace{0.2 in } \mbox{for}\hspace{0.2 in}\lambda\in c_l\,,\ ] ] @xmath533\,,\hspace{0.2 in } \mbox{for}\hspace{0.2 in}\lambda\in c_r\,,\ ] ] @xmath534\ , , \hspace{0.2 in } \mbox{for}\hspace{0.2 in}\lambda\in l_l\,,\ ] ] @xmath535\ , , \hspace{0.2 in } \mbox{for}\hspace{0.2 in}\lambda\in l_r\,,\ ] ] and for @xmath113 , @xmath536 & -ie^{i(\theta(\lambda)-\theta(0))/\hbar_n } [ ia_m(\lambda)+e^{\tilde{\phi}(\lambda)/\hbar_n}]\\\\ -ie^{-i(\theta(\lambda)-\theta(0))/\hbar_n } [ ia_m(\lambda)+e^{\tilde{\phi}(\lambda)/\hbar_n } ] & 1-[ia_m(\lambda)+e^{\tilde{\phi}(\lambda)/\hbar_n}]\end{array}\right]\ , . \label{eq : fjumpcm}\ ] ] the jump relations satisfied by @xmath531 on the complex conjugate contours in the lower half - plane follow from these by the symmetry @xmath537 . now , for @xmath109 with @xmath504 , @xmath538 where in the last line @xmath539 with the change of coordinate being given by ( [ eq : varphi0 ] ) . in these steps , we used the relation ( [ eq : tauleft ] ) , the fact that from proposition [ prop : phitildezero ] we get @xmath327 , and , according to ( [ eq : deftheta0 ] ) and the quantization condition ( [ eq : quantization ] ) on @xmath118 , @xmath540 . we have also used the fact that @xmath541 which follows directly from the definition ( [ eq : deftheta0 ] ) of @xmath102 , the definition ( [ eq : deftheta ] ) of @xmath315 , and the relation ( [ eq : rhosequal ] ) . in a similar way , for @xmath111 with @xmath504 , we get @xmath542 and for @xmath543 with @xmath504 , @xmath544 with @xmath539 . finally , when @xmath113 and @xmath504 we have for arbitrarily small @xmath35 , @xmath545\\\\ & = & \displaystyle 1-t(\lambda)\\\\ & = & \displaystyle 1-\frac{2\pi e^{2i\zeta}(-i\zeta)^{-2i\zeta } } { \gamma(1/2-i\zeta)^2 } \left(1+o\left(\hbar_n^{4\alpha/3 - 1-\nu}\right)\right ) \\\\ & = & \displaystyle 1-\frac{2\pi e^{2i\zeta}(-i\zeta)^{-2i\zeta}}{\gamma(1/2-i\zeta)^2 } + o\left(\hbar_n^{4\alpha/3 - 1-\nu}\right)\ , , \end{array } \label{eq : ammodel}\ ] ] again with @xmath539 . the last step follows because @xmath546 is uniformly bounded on @xmath101 . let @xmath547 , @xmath548 , @xmath549 , @xmath550 , and @xmath551 denote the straight rays that agree with the corresponding contours in a fixed neighborhood of the origin in the @xmath22-plane , but lying in the @xmath490-plane ( according to ( [ eq : varphi0 ] ) , @xmath490 is a simple rescaling of @xmath22 by a positive number ) . these rays are oriented contours , with the same orientation as the original contours : @xmath551 , @xmath549 , and @xmath550 are oriented outwards from the origin toward infinity , and @xmath547 and @xmath548 are oriented inwards from infinity toward the origin . let the union of these contours with their complex conjugates be denoted @xmath552 . consider the following riemann - hilbert problem . find a matrix @xmath553 with the following properties : 1 . * analyticity : * @xmath553 is analytic for @xmath554 . * boundary behavior : * @xmath553 assumes continuous boundary values on @xmath552 . 3 . * jump conditions : * the boundary values taken on @xmath552 satisfy @xmath555\,,\hspace{0.2 in}\zeta\in \vec{c_l}\,,\ ] ] @xmath556\,,\hspace{0.2 in}\zeta\in \vec{c_r}\,,\ ] ] @xmath557\,,\hspace{0.2 in}\zeta\in \vec{l_l}\,,\ ] ] @xmath558\,,\hspace{0.2 in}\zeta\in \vec{l_r}\,,\ ] ] and @xmath559 \\\\ \displaystyle ie^{\pi\zeta } \left[\frac{2\pi e^{2i\zeta}(-i\zeta)^{-2i\zeta } } { \gamma(1/2-i\zeta)^2}-1\right ] & \displaystyle \frac{2\pi e^{2i\zeta}(-i\zeta)^{-2i\zeta}}{\gamma(1/2-i\zeta)^2 } \end{array}\right]\,,\hspace{0.2 in}\zeta\in \vec{c_m}\ , . \label{eq : fhatjumpcm}\ ] ] on the contours in the lower half - plane , the jump conditions are implied by the symmetry @xmath560 . * normalization : * @xmath553 is normalized at infinity : @xmath561 [ rhp : localorigin ] unfortunately , we can not solve riemann - hilbert problem [ rhp : localorigin ] explicitly . luckily , we will not require an explicit solution . however , existence of a solution is an issue that must be resolved , and we need to obtain a decay estimate that quantifies the normalization condition ( [ eq : fhatnorm ] ) . these questions are addressed via the abstract theory of riemann - hilbert problems . the local model riemann - hilbert problem [ rhp : localorigin ] has a unique solution satisfying @xmath562 as @xmath563 , uniformly with respect to direction . also , @xmath564 . _ proof : _ each riemann - hilbert problem is equivalent to an inhomogeneous system of linear singular integral equations . it must be shown that the matrix singular integral operator associated with these equations is of fredholm type , with index zero . then it must be shown that there are no homogeneous solutions , at which point one has existence and uniqueness of a solution to the inhomogeneous system . finally , one maps the solution of the integral equations to the unique solution of the riemann - hilbert problem and it remains to verify the rate of decay to the identity matrix as @xmath563 . the theory we will use is the theory of matrix riemann - hilbert problems on self - intersecting contours , with boundary values taken in spaces of hlder continuous functions . this theory is summarized in a self - contained way in the appendix of @xcite . the first step is to establish that the operator of the associated system of singular integral equations is fredholm index zero on an appropriate space of functions . as described in @xcite , this follows from two facts . first , on each ray of the contour @xmath552 the jump matrix @xmath565 is uniformly lipschitz with respect to @xmath490 , and differs from the identity by a quantity that is @xmath566 for large @xmath490 . second , the limiting values of the jump matrix , taken as @xmath567 along each ray of @xmath552 , are consistent with a bounded solution @xmath568 near @xmath569 . this means the following . suppose that @xmath553 has a limiting value , say a matrix @xmath570 , as @xmath567 inside one of the sectors of @xmath571 . using the limiting value of the jump matrix @xmath572 at the origin along one of the rays of @xmath552 bounding that sector , one can compute the limiting value of @xmath553 at the origin in the neighboring sector . this procedure can be continued , moving from sector to sector of @xmath571 in the same direction , until one arrives once again in the original sector , with a matrix @xmath573 . the consistency condition is simply that @xmath574 , which upon elimination of @xmath575 can be viewed as a cyclic relation among the limiting values of the jump matrix @xmath576 taken along each ray of @xmath552 as @xmath567 . it is easily checked that this cyclic relation indeed holds for riemann - hilbert problem [ rhp : localorigin ] . the second step is to establish existence and uniqueness of the solution @xmath553 . the fact that the associated singular integral equations are fredholm index zero means that , in a certain precise sense , the fredholm alternative applies to our riemann - hilbert problem . the inhomogeneity is the normalization to the identity matrix at @xmath577 . the corresponding homogeneous riemann - hilbert problem has exactly the same form except that the normalization condition is replaced by the condition @xmath578 as @xmath563 . we will have a unique solution of riemann - hilbert problem [ rhp : localorigin ] if it can be shown that no such homogeneous solutions exist . for this purpose , it is sufficient that the jump matrix @xmath572 should have a certain symmetry with respect to schwartz reflection through the real axis in the @xmath490-plane . for the orientation of @xmath552 described above , the required relation is : @xmath579 for @xmath580 . it is easily checked , using the symmetry @xmath581 and structural details of @xmath582 in the upper half - plane , that this relation holds , and this means that riemann - hilbert problem [ rhp : localorigin ] has a unique solution @xmath553 . the third step is to establish that the unique solution @xmath568 decays to the identity for large @xmath490 like @xmath583 . the hlder theory that we have been using generally provides a solution @xmath553 under these circumstances that takes boundary values on @xmath552 that are hlder continuous with exponent @xmath282 and that differs from the identity matrix by @xmath584 as @xmath585 , for all @xmath282 strictly less than 1 . this fact can be traced to the compact embedding of each hlder space into all hlder spaces with strictly smaller exponents . the compactness is needed to establish the fredholm property of the riemann - hilbert problem . so to obtain the required decay , we need an additional argument . the condition that is required to obtain the @xmath566 decay is that a signed sum of the mean values of @xmath586 taken as @xmath563 along each ray of @xmath552 ( the signs are related to the orientation of the individual rays ) is zero @xcite . now along each ray of @xmath552 except for @xmath551 and its conjugate ( _ i.e. _ the imaginary axis in the @xmath490-plane ) , @xmath572 decays to the identity exponentially fast as @xmath563 . so these rays to do not contribute to the sum and it is only necessary to check the imaginary axis . when @xmath587 , @xmath588+o\left(\frac{1}{\zeta}\right)\,,\ ] ] as @xmath563 , and for @xmath490 on the negative imaginary axis oriented upwards , @xmath589+o\left(\frac{1}{\zeta}\right)\,,\ ] ] as @xmath563 . the limits of these quantities do not exist as @xmath563 due to the oscillations on the off - diagonal . but the mean values exist and are equal , and it turns out that they enter the sum with opposite signs due to the orientation of the contour rays . thus , the required sum of signed mean values indeed vanishes . this , along with the analyticity of the jump matrix @xmath572 along each ray of @xmath552 establishes that @xmath590 as @xmath563 . finally , we check that @xmath591 . taking determinants in the jump relations we see that on all rays of the contour , @xmath592 . since the boundary values taken by @xmath553 on @xmath552 are continuous , we discover that @xmath593 is an entire function . since this function tends to one at infinity , it follows from liouville s theorem that @xmath594 . this completes the proof of the proposition . @xmath105 from ( [ eq : fdefleft ] ) and ( [ eq : fdefright ] ) we can express @xmath388 in terms of @xmath531 for @xmath504 . for @xmath229 we have @xmath595 and for @xmath231 we have @xmath596 to obtain a local model for @xmath133 near the origin , we simply replace @xmath531 in these formulae by the approximation @xmath597 . with @xmath504 , we set for @xmath229 , @xmath598 and for @xmath231 we set @xmath599 the most important properties of this matrix function are easily seen to be the following . the matrix @xmath600 is a piecewise analytic function of @xmath22 in the disk @xmath601 , with jumps only on the locally straight - line contours @xmath99 , @xmath100 , @xmath357 , @xmath358 , and @xmath101 , and their conjugates in the lower half - disk . the jump relations satisfied by @xmath600 on these contours are the following : @xmath602 \sigma_1^{\frac{1-j}{2}}\,,\hspace{0.2 in}\lambda\in c_l\,,\ ] ] @xmath603 \sigma_1^{\frac{1-j}{2}}\,,\hspace{0.2 in}\lambda\in c_r\,,\ ] ] @xmath604\sigma_1^{\frac{1-j}{2}}\ , , \hspace{0.2 in } \lambda\in l_l\,,\ ] ] @xmath605\sigma_1^{\frac{1-j}{2}}\ , , \hspace{0.2 in } \lambda\in l_r\,,\ ] ] @xmath606 where @xmath607 with @xmath608 and where @xmath539 . the jumps on the corresponding contours in the lower half - plane are obtained from the symmetry @xmath609 . the matrix @xmath600 is uniformly bounded for @xmath601 , with a bound that is independent of @xmath118 . also , @xmath610 and when @xmath611 , @xmath612 [ prop : originmodelproperties ] as before , we suppose that @xmath404 and @xmath405 are small scales satisfying @xmath613 as @xmath118 tends to zero . we redefine the integer @xmath407 so that exactly the first @xmath407 of the numbers @xmath38 lie on the positive imaginary axis above @xmath614 . we will suppose that @xmath615 , and @xmath616 and we will deduce asymptotic formulae for @xmath158 given by ( [ eq : t ] ) valid for such @xmath22 , and for @xmath153 given by ( [ eq : s ] ) when @xmath22 is also bounded outside of some downward - opening sector with vertex at @xmath72 , in the semiclassical limit @xmath617 . first , we establish a result that is the analogue of lemma [ lemma : originpeelfirst ] . [ lemma : iapeelfirst ] when @xmath615 and @xmath616 and with @xmath407 defined as above , @xmath618 _ proof : _ we again estimate @xmath203 using the integral formula ( [ eq : integralformulag ] ) with integrand @xmath619 given by ( [ eq : glambdaxi ] ) . given our conditions on @xmath22 , for @xmath416 and @xmath423 we have @xmath620 for all such @xmath419 we therefore have the estimate @xmath621 summing over @xmath139 gives @xmath622 which is @xmath623 because @xmath422 , and the lemma is proved . @xmath105 as was the case when @xmath22 was near the origin , only certain terms are important when @xmath624 is small , as a direct consequence of lemma [ lemma : iapeelfirst ] and the fact that @xmath424 is analytic for such @xmath22 when @xmath423 . the important terms when @xmath616 are @xmath625 and @xmath626 lemma [ lemma : iapeelfirst ] states that @xmath627 and @xmath628 as @xmath136 for @xmath616 . the analogue of lemma [ lemma : originpeelsecond ] says that for @xmath624 small the sequence of numbers @xmath629 contributing to @xmath153 and @xmath158 can be replaced essentially by a `` straightened - out '' sequence with uniform density . [ lemma : iapeelsecond ] let @xmath443 for @xmath630 be the sequence of numbers defined by the relation : @xmath631 which results from expanding the bohr - sommerfeld relation ( [ eq : bs ] ) for @xmath433 near @xmath72 and keeping only the dominant terms . define @xmath632 and @xmath633 then , for @xmath615 and @xmath616 , @xmath634 where we suppose that the scale @xmath404 is further constrained so that the relative error is asymptotically small . if @xmath22 is additionally bounded outside of some downward opening sector with vertex at @xmath72 , then @xmath635 _ proof : _ the proof of this lemma follows that of lemma [ lemma : originpeelsecond ] almost exactly and will not be repeated here . the only difference is that the square in that proof should be replaced here by the rectangle whose top side is @xmath636 and @xmath637 and whose bottom is @xmath638 . @xmath105 without any approximation , @xmath639 and @xmath640 can be rewritten in a more transparent form by expressing the products in terms of gamma functions and evaluating the logarithmic integrals exactly . introduce a local variable @xmath490 in terms of a transformation @xmath641 given by the relation @xmath642 and let @xmath213 be the positive constant @xmath643 in terms of these quantities , one finds that @xmath639 and @xmath640 take a simple form : @xmath644 and @xmath645 where @xmath646 and @xmath647 now , @xmath648 because @xmath213 is proportional to @xmath140 and @xmath407 is of the order of @xmath649 while @xmath650 . so again we can use stirling s formula to extract the dominant asymptotic contributions to @xmath639 and @xmath640 as @xmath118 tends to zero . [ lemma : iapeelthird ] as @xmath118 tends to zero through positive values , @xmath651 and @xmath652 _ proof : _ using stirling s formula , one expands @xmath653 to find @xmath654 the error here is dominated by the fact @xmath655 is the smallest large number involved . now we expand the powers that remain in conjunction with those in @xmath656 to find @xmath657 since @xmath658 , the relative error is dominated by @xmath659 . using this expression for the product @xmath660 in ( [ eq : s2iarewrite ] ) and ( [ eq : t2iarewrite ] ) , the lemma is proved . @xmath105 . in exactly the same way as in our study of the local behavior near the origin , we may combine lemmas [ lemma : iapeelfirst ] , [ lemma : iapeelsecond ] , and [ lemma : iapeelthird ] and choose the `` internal '' scale @xmath404 in terms of @xmath405 and @xmath118 to obtain asymptotics for @xmath153 and @xmath158 with optimized relative error . let @xmath615 , with @xmath661 , where @xmath505 , and let @xmath22 be bounded outside some fixed symmetrical sector with vertex at @xmath72 and opening downward . then @xmath662 where @xmath663 . [ prop : iaasymps ] let @xmath615 , with @xmath661 , where @xmath505 . then for all @xmath35 , however small , @xmath664 where @xmath663 . [ prop : iaasympt ] in particular it follows from these considerations that both @xmath153 and @xmath158 are uniformly bounded functions on their respective contours in any fixed neighborhood @xmath665 of @xmath117 . we claim that the quotient of the jump matrices for @xmath133 and @xmath393 is uniformly close to the identity matrix in @xmath665 as @xmath136 . since @xmath666 is , by definition , analytic throughout @xmath665 , it suffices to show that the jump matrix @xmath667 is uniformly close to the identity in @xmath665 . this will be the case if @xmath402 , @xmath403 , and @xmath267 are uniformly small on their respective contours . now for @xmath113 , @xmath668 so , with @xmath323 being real and strictly negative for @xmath669 according to proposition [ prop : tildephineg ] and @xmath158 being bounded , we see that @xmath267 is in fact exponentially small as @xmath118 tends to zero through positive values . similarly , for @xmath109 , @xmath670 where we have used ( [ eq : tauleft ] ) and the fact that @xmath328 on @xmath101 above @xmath671 . since @xmath672 , it is possible to choose the neighborhood @xmath665 small enough ( independent of @xmath118 ) so that @xmath673 throughout @xmath665 . since @xmath153 is bounded , it then follows that for @xmath674 , @xmath402 is exponentially small as @xmath118 tends to zero through positive values . virtually the same argument using ( [ eq : tauright ] ) in place of ( [ eq : tauleft ] ) shows that @xmath403 is also exponentially small for @xmath675 ( it may be necessary to make @xmath665 slightly smaller to have @xmath676 throughout @xmath665 ) . it follows that @xmath677 is exponentially small uniformly for the contours within @xmath665 . for this reason we expect that the outer model @xmath401 will be a good approximation to @xmath133 near @xmath117 even though @xmath678 and @xmath679 are not small . we do not need to construct a special - purpose local model for @xmath133 in this case . let @xmath399 be a circular disk centered at @xmath117 of sufficiently small radius ( independent of @xmath118 ) that @xmath680 and @xmath681 , and that @xmath357 and @xmath358 each have exactly one intersection with @xmath682 ( of course @xmath101 will have two intersection points with @xmath682 ) . this situation is possible as long as @xmath32 . the case of @xmath12 is a degenerate case that we will not treat in detail here . since @xmath32 and therefore @xmath683 , it follows from the definition ( [ eq : rhodefine ] ) of @xmath320 that as @xmath22 tends to @xmath307 along @xmath309 , @xmath315 vanishes like @xmath684 , and not to higher order . since @xmath315 may be extended from @xmath309 to be an analytic function in @xmath399 except for a branch cut along the part of @xmath101 in @xmath399 lying above the center , and since @xmath685 extended to this cut domain from @xmath309 is nonzero , it is easy to see that the function @xmath686 defines an invertible conformal mapping of all of @xmath399 to its image . consider the local variable @xmath490 defined by the relation @xmath687 . the image of @xmath399 in the @xmath490-plane is a neighborhood of @xmath569 that scales with @xmath118 such that it contains the disk centered at @xmath569 with radius @xmath688 for some constant @xmath689 . the transformation ( [ eq : varphiiax ] ) maps @xmath690 to a ray segment of the positive real @xmath490-axis , and takes the portion of @xmath101 in @xmath399 lying above @xmath310 to a ray segment of the negative real @xmath490-axis . we suppose that the contours @xmath357 and @xmath358 have been chosen so that @xmath691 and @xmath692 are straight ray segments with angles @xmath693 and @xmath694 respectively . for @xmath695 , the matrix @xmath696 satisfies the following jump relations : @xmath697\ ] ] for @xmath698 , oriented from right to left , @xmath699\ ] ] on the part of the ray @xmath700 in @xmath701 , oriented toward the origin , @xmath702\ ] ] on the part of the ray @xmath703 in @xmath701 , oriented toward the origin , and finally @xmath704\ ] ] for @xmath705 , oriented from right to left . the jump relation on the negative real @xmath490-axis follows from the formula ( [ eq : ntildemiddle ] ) which applies because @xmath706 here . in ( [ eq : ntildemiddle ] ) one uses the fact that @xmath707 for @xmath113 above @xmath307 , and the relation ( [ eq : phifromtheta ] ) giving @xmath323 above @xmath307 in terms of the analytic continuation of @xmath315 from @xmath309 , which one writes in terms of the local coordinate @xmath490 . as @xmath563 on all of the rays except for @xmath708 , the jump matrix for @xmath709 decays exponentially to the identity matrix . these jump conditions were precisely the ones that were neglected in obtaining the outer model . that is , the matrix @xmath710 defined for @xmath711 is analytic except on the positive real @xmath490-axis , where it satisfies @xmath712\,.\ ] ] it follows that the matrix @xmath713 can be decomposed into a product of a holomorphic prefactor depending on @xmath118 and a universal ( _ i.e. _ independent of @xmath118 ) local factor that takes care of the jump . we therefore may write @xmath714 where @xmath715 is holomorphic in @xmath701 and where @xmath716=\frac{1}{\sqrt{2 } } \left[\begin{array}{cc } ( -\zeta)^{1/4 } & ( -\zeta)^{1/4}\\\\ -(-\zeta)^{-1/4 } & ( -\zeta)^{-1/4}\end{array}\right]\,.\ ] ] note that @xmath715 has determinant one . its matrix elements are of size @xmath717 for @xmath718 . it is easy to write down an explicit formula for @xmath715 because both @xmath713 and @xmath719 are known . we will now approximate @xmath709 by @xmath720 where @xmath721 is the solution of the following riemann - hilbert problem . let @xmath722 be the contour shown in figure [ fig : airycontour ] . find a matrix @xmath721 with the following properties : 1 . * analyticity : * @xmath721 is analytic for @xmath723 . * boundary behavior : * @xmath721 assumes continuous boundary values on @xmath722 . 3 . * jump conditions : * the boundary values taken on @xmath722 satisfy @xmath724\,,\hspace{0.2 in } \mbox{for}\hspace{0.2 in}\arg(\zeta)=0\,,\ ] ] @xmath725\ , , \hspace{0.2 in}\mbox{for}\hspace{0.2 in } \arg(\zeta)=-\pi/3\,,\ ] ] @xmath726\ , , \hspace{0.2 in}\mbox{for}\hspace{0.2 in } \arg(\zeta)=\pi/3\,,\hspace{0.2 in}\mbox{and}\,,\ ] ] @xmath727\ , , \hspace{0.2 in}\mbox{for}\hspace{0.2 in}\arg(\zeta)=\pi\,.\ ] ] 4 . * normalization : * @xmath721 is normalized at infinity so that @xmath728 [ rhp : airy ] we will describe how riemann - hilbert problem [ rhp : airy ] can be solved explicitly . first , we make an explicit change of variable to a new matrix @xmath729 by setting @xmath730\ , . \label{eq : slocfromt}\ ] ] if @xmath721 satisfies riemann - hilbert problem [ rhp : airy ] , then it follows that @xmath729 obeys the following jump relations : @xmath731\,,\hspace{0.2 in}\mbox{for}\hspace{0.2 in } \arg(\zeta)=0\,,\ ] ] @xmath732\,,\hspace{0.2 in}\mbox{for}\hspace{0.2 in } \arg(\zeta)=\pm\pi/3\,,\hspace{0.2 in}\mbox{and}\,,\ ] ] @xmath733\,,\hspace{0.2 in}\mbox{for}\hspace{0.2 in } \arg(\zeta)=\pi\,.\ ] ] it is a general fact that @xmath734 matrix functions that satisfy piecewise constant jump conditions , like the matrix @xmath729 does , can be expressed in terms of solutions of @xmath25th order linear differential equations with meromorphic ( and often rational , or even polynomial ) coefficients . in the @xmath735 case , classical special functions therefore play a key role . in this case , we see immediately from the fact that the boundary values taken by @xmath736 on @xmath722 are continuous that the matrix @xmath737{\bf s}^{\rm loc}(\zeta)^{-1}\ ] ] is analytic for @xmath738 . if we suppose that @xmath739 has a bounded derivative near the origin in each sector of @xmath740 a hypothesis that must be verified later then we see that in fact @xmath741 is analytic at the origin and is consequently an entire function of @xmath490 . to work out how @xmath741 behaves for large @xmath490 , we need to use the normalization condition ( [ eq : snorm ] ) for @xmath736 . we interpret ( [ eq : snorm ] ) to mean both that @xmath742 and also that @xmath743 both ( [ eq : snorm2 ] ) and ( [ eq : snorm2deriv ] ) are again hypotheses that must be verified once we obtain a solution for @xmath736 . they are not true _ a priori _ by virtue of ( [ eq : snorm ] ) alone ; for example the decay rate in ( [ eq : snorm ] ) might not be as fast as @xmath583 , and the error term might have rapid oscillations that would make its derivative larger than @xmath744 thus violating ( [ eq : snorm2deriv ] ) . it follows from our hypotheses that @xmath741 must be a polynomial ; in fact , @xmath745\,.\ ] ] under our hypotheses , it then follows that the matrix @xmath729 solves the linear differential equation @xmath746{\bf t}(\zeta)\,.\ ] ] upon introducing the new independent variable @xmath747 we see that the elements of the second row of @xmath748 satisfy airy s equation : @xmath749 and that the elements of the first row are given by @xmath750 so which solutions of airy s equation are the appropriate ones for our purposes ? the first observation is that we need to specify different solutions of airy s equation in each simply - connected region of the complex plane where @xmath751 and @xmath752 are analytic . the assignment of solutions in these regions must be consistent with the jump conditions and asymptotics for @xmath729 . from the jump conditions for @xmath729 , we can see that in fact @xmath751 is analytic in @xmath753 , while @xmath752 is analytic except when @xmath754 or @xmath755 . we now want to use the normalization condition ( [ eq : snorm2 ] ) to find sectors in which @xmath756 and @xmath757 are exponentially decaying . then we will be able to uniquely identify these functions with particular solutions of airy s equation that also decay . from the presumed asymptotic relation ( [ eq : snorm2 ] ) we have ( in terms of the variable @xmath419 ) , @xmath758 and @xmath759 as @xmath760 with @xmath761 . these show that @xmath751 is exponentially decaying for @xmath762 and also for @xmath763 , while @xmath752 is exponentially decaying for @xmath764 . significantly , both matrix elements are analytic throughout the sectors where they are exponentially decaying for large @xmath419 . as a basis of linearly independent solutions of airy s equation we take the functions @xmath765 and @xmath766 . these decay in different sectors , and have the asymptotic expansions @xmath767 as @xmath760 for @xmath761 , and @xmath768 as @xmath760 for @xmath762 . comparing the expansion of @xmath766 with that of @xmath751 in the sector @xmath762 , we find that here @xmath769 since @xmath756 is analytic in the lower half @xmath419-plane , this relation holds identically for @xmath770 . similarly , comparing the expansion of @xmath771 with that of @xmath752 in the sector @xmath764 , we find that here @xmath772 being as @xmath757 is analytic for @xmath773 , this identity holds throughout the sector of analyticity . restoring the original independent variable @xmath490 , we find that for @xmath774 , @xmath775 and therefore throughout the same domain , @xmath776 for @xmath490 with @xmath777 or @xmath778 , @xmath779 and therefore throughout the same domain , @xmath780 the sector of @xmath740 that is contained in both of these domains is @xmath781 . it is sufficient to have specified the matrix elements of @xmath729 in this sector , since it may be consistently obtained in the remaining sectors of @xmath740 by making use of the jump relations for @xmath729 . the procedure is consistent because the cyclic product of these jump matrices is the identity : @xmath782 \left[\begin{array}{cc } 1 & 1 \\\\ 0 & 1\end{array}\right]^{-1 } \left[\begin{array}{cc } 0 & -1 \\\\ 1 & 0\end{array}\right]^{-1 } \left[\begin{array}{cc } 1 & 1 \\\\ 0 & 1\end{array}\right]^{-1 } = { \mathbb i}\,.\ ] ] once @xmath729 is known for @xmath783 , the original unknown matrix @xmath736 solving riemann - hilbert problem [ rhp : airy ] is obtained directly by the transformation ( [ eq : slocfromt ] ) . it suffices to give a formula that holds for @xmath781 . we find @xmath784 we will have found a solution to riemann - hilbert problem [ rhp : airy ] if we can verify the two hypotheses ( _ cf . _ equations ( [ eq : snorm2 ] ) and ( [ eq : snorm2deriv ] ) ) we made regarding the interpretation of the normalization condition ( [ eq : snorm ] ) and the differentiability of @xmath736 at @xmath569 . one verifies these directly , using the explicit formulae given here . to build a better model for @xmath133 in the disk @xmath399 than @xmath393 , we begin by recalling the exact relationship between the matrix @xmath359 and the matrix @xmath785 : @xmath786 for all @xmath787 . the matrix @xmath286 is also explicitly related to @xmath359 . for @xmath787 in the part of the lens between the contours @xmath357 and @xmath101 , @xmath788\sigma_1^{\frac{1-j}{2}}\,,\ ] ] for @xmath787 in the part of the lens between the contours @xmath101 and @xmath358 , @xmath789\sigma_1^{\frac{1-j}{2}}\,,\ ] ] and for all other @xmath787 , we simply have @xmath790 . as we do not expect the difference between @xmath133 and the formal continuum limit approximation @xmath286 to be important in the disk @xmath399 since it is isolated from the points @xmath116 and @xmath117 , we can obtain a guess for an approximation for @xmath133 that should be valid in @xmath399 simply by substituting @xmath791 for @xmath792 in these formulae . putting these steps together , the model for @xmath133 for @xmath787 that we will use is defined as follows . for @xmath787 in the lens between @xmath357 and @xmath101 , set @xmath793\sigma_1^{\frac{1-j}{2}}\,,\ ] ] for @xmath787 in the lens between @xmath101 and @xmath358 , set @xmath794\sigma_1^{\frac{1-j}{2}}\,,\ ] ] and for all other @xmath787 , set @xmath795 where @xmath796 is defined by ( [ eq : shatdef ] ) . the most important properties of the matrix @xmath797 in the disk @xmath399 are the following . the matrix @xmath798 is piecewise analytic in the left and right half - disks of @xmath399 . on the imaginary axis ( which bisects @xmath399 ) oriented in the positive imaginary direction , @xmath799 that is , the local model has exactly the same jump as @xmath285 . for @xmath787 , the matrix function @xmath800 is bounded by a constant of order @xmath801 . also , for @xmath802 , @xmath803 [ prop : endpointproperties ] _ proof : _ computing the jump matrix for @xmath797 is straightforward . to show the bound on @xmath798 , we recall that @xmath804 is bounded by a quantity of order @xmath805 , and note that the other factor of the same size comes from the factor @xmath806 via the normalization condition on this matrix and the fact that @xmath807 grows like @xmath808 for @xmath787 . similar reasoning using ( [ eq : snorm2 ] ) establishes the error in matching onto @xmath666 on the boundary of @xmath399 . we are now in a position to put all of our models together to build a guess for a uniformly valid approximation of @xmath133 . such a guess is called a parametrix . to build the parametrix @xmath809 as a sectionally holomorphic matrix function , we simply combine the outer and local models . for all @xmath22 satisfying @xmath504 , where the parameter @xmath810 is to be determined later , set @xmath811 for @xmath787 , we set @xmath812 and by symmetry for all @xmath813 we set @xmath814 finally , for all remaining @xmath815 , set @xmath816 the parametrix @xmath809 is holomorphic for @xmath817 , where @xmath818 is the contour illustrated in figure [ fig : sigmahat ] . to determine the accuracy of the parametrix , we compare it directly with the original matrix @xmath133 . that is , we consider the error matrix defined by @xmath819 this matrix is sectionally analytic in the complex @xmath22-plane , with discontinuities on a contour @xmath820 that is illustrated in figure [ fig : sigmae ] . note that as a consequence of the symmetry of @xmath809 and @xmath133 under complex conjugation , we have @xmath821 . if the parametrix is indeed a good model for @xmath133 , then we must be able to show that the matrix @xmath822 is uniformly close to the identity matrix in the whole complex plane . while we do not know @xmath822 explicitly like we know @xmath809 , we know from the normalization condition of both factors that @xmath823 it turns out that we can also calculate explicitly the ratio of boundary values taken by @xmath822 from both sides on each arc of @xmath820 . that is , we know the jump matrix for @xmath824 , and can express it explicitly in terms of @xmath825 and the jump matrix for @xmath133 , both of which are known in terms of the matrix function @xmath826 for which we have an existence proof and a characterization , but not an explicit formula . ] . this means that the matrix @xmath822 itself is a solution of a particular riemann - hilbert problem for which we know the data . by solving this riemann - hilbert problem , we will show that indeed @xmath822 is uniformly close to the identity matrix . there are two kinds of arcs in the contour @xmath820 : `` matching '' arcs of the circles @xmath682 , @xmath827 , and @xmath611 where two different components of the parametrix have to match well onto each other , and the remaining arcs within the disks and outside the disks where the jump matrix for @xmath809 should be a good approximation to that of @xmath388 . consider one of the arcs of @xmath820 oriented in some convenient way , and as usual let the subscript `` @xmath107 '' ( respectively `` @xmath108 '' ) denote a boundary value taken on the arc from its left ( respectively right ) . we can easily see from the definition ( [ eq : error ] ) that for @xmath22 on this arc , @xmath828 where @xmath829 and @xmath830 denote the jump matrices on the arc for @xmath133 and the parametrix @xmath809 respectively . if the arc under consideration is a `` matching '' arc , then the discontinuity in @xmath824 is wholly due to the mismatch of components of the parametrix , and the jump matrix @xmath829 is therefore replaced with the identity matrix in ( [ eq : ejump ] ) . it then follows that an equivalent formula for @xmath831 on a `` matching '' arc is the following : @xmath832 in this case , the two boundary values represent different components of the parametrix , for example @xmath393 would play the role of @xmath833 and @xmath797 would play that of @xmath834 if the `` matching '' arc under consideration is an arc of @xmath682 , oriented clockwise . the key fact that we need now is the following . the optimal value of the radius parameter @xmath810 is @xmath835 . for this value of @xmath810 , and for all @xmath35 arbitrarily small , @xmath836 uniformly for all @xmath837 . [ prop : errorjumpbound ] _ proof : _ we begin by considering the `` matching '' arcs . we take the circle @xmath682 to be oriented in the clockwise direction . here we find @xmath838 with the error estimate coming from proposition [ prop : endpointproperties ] . an estimate of the same form necessarily holds on the `` matching '' arcs of @xmath827 according to the conjugation symmetry of @xmath822 . the remaining `` matching '' arcs lie on the circle @xmath611 , which again we take to be oriented in the clockwise direction . here we find @xmath839 with the error estimate coming from proposition [ prop : originmodelproperties ] . we continue by considering the arcs of @xmath820 with @xmath601 . using the fact recorded in proposition [ prop : originmodelproperties ] that @xmath840 has determinant one and is uniformly bounded , we see from ( [ eq : ejump ] ) that the important quantity to estimate is simply @xmath841 , the difference between the jump matrix ratio and the identity . first , consider the portion of the contour @xmath99 with @xmath842 . using proposition [ prop : originmodelproperties ] and ( [ eq : cljumpn ] ) , we find that here @xmath843 \sigma_1^{\frac{1-j}{2}}\ , , \label{eq : clquotient}\ ] ] where @xmath539 . using ( [ eq : almodel ] ) and the uniform boundedness of the leading - order term on the right - hand side of ( [ eq : almodel ] ) for @xmath601 , we see that the matrix quotient in ( [ eq : clquotient ] ) differs from the identity matrix by an order @xmath844 amount . virtually the same argument using ( [ eq : crjumpn ] ) and ( [ eq : armodel ] ) in conjunction with proposition [ prop : originmodelproperties ] establishes that on @xmath100 the matrix quotient @xmath845 differs from the identity by a quantity of order @xmath844 . next , consider the contour @xmath357 for @xmath601 . on this contour , there is no jump for @xmath133 , so @xmath846 in the formula ( [ eq : ejump ] ) for @xmath847 . but from the formula for @xmath848 for @xmath365 given in proposition [ prop : originmodelproperties ] , we see by ordinary taylor expansion that for @xmath601 , we have @xmath849 for @xmath365 . virtually the same argument yields the same estimate for @xmath850 on @xmath358 with @xmath601 . finally consider the contour @xmath101 ( the positive imaginary axis ) with @xmath601 . using ( [ eq : cmjumpn ] ) and the jump matrix @xmath830 for @xmath113 recorded in proposition [ prop : originmodelproperties ] , we find that here @xmath851\sigma_1^{\frac{1-j}{2}}\,,\ ] ] where @xmath852 is given in terms of @xmath539 by ( [ eq : delta ] ) and where @xmath853 note that @xmath854 and @xmath855 are both of order @xmath856 for @xmath601 by taylor expansion arguments . also , @xmath852 is uniformly bounded in the disk of radius @xmath857 , and @xmath858 both have modulus one for @xmath113 in this disk . also , from ( [ eq : ammodel ] ) we get @xmath859 for all @xmath35 since @xmath327 on this part of @xmath101 . this error dominates those arising from taylor approximation , and thus on @xmath101 with @xmath601 we find that @xmath860 . the corresponding estimates hold on the corresponding contours in the lower half - disk , by conjugation symmetry . putting this information together with the uniform boundedness of @xmath600 and its inverse , we find that for all @xmath861 with @xmath601 , @xmath862 for all @xmath35 . now we proceed to study the jump matrix @xmath831 inside the disk @xmath399 centered at the endpoint @xmath310 , assuming @xmath32 is fixed so that @xmath399 is fixed and bounded away from the origin and from @xmath117 . the only contour we need to consider is the imaginary axis . the jump matrix @xmath831 is given by ( [ eq : ejump ] ) . this time , the conjugating factors of @xmath863 and its inverse are not uniformly bounded in @xmath399 as @xmath118 tends to zero . according to proposition [ prop : endpointproperties ] each conjugating matrix contributes an amplifying factor of @xmath801 . also according to proposition [ prop : endpointproperties ] , we have that @xmath864 is the same as the jump matrix for @xmath286 . therefore , using ( [ eq : cmjumpn ] ) and ( [ eq : ntildemiddle ] ) , we find that @xmath865\sigma_1^{\frac{1-j}{2}}\,.\ ] ] using proposition [ prop : a_m ] and the fact that @xmath866 while @xmath867 has modulus one on @xmath101 within @xmath399 , we get that @xmath868 for all @xmath268 . combining this with the bounds on the conjugating factors , we find that within @xmath399 , @xmath869 for all @xmath268 . by conjugation symmetry , the same estimate holds for the jump matrix @xmath831 when @xmath813 . finally , we consider the parts of @xmath820 outside all of the disks , where we have set @xmath870 . on all of these parts of @xmath820 , proposition [ prop : noutproperties ] guarantees that the conjugating factors @xmath871 and @xmath872 are uniformly bounded as @xmath118 tends to zero . so it remains to determine the magnitude of the difference between the quotient @xmath873 and the identity . first consider the part of @xmath101 between the disk at the origin and the disk @xmath399 . using proposition [ prop : noutproperties ] to find @xmath848 and recalling ( [ eq : cmjumpn ] ) , we find @xmath874\sigma_1^{\frac{1-j}{2}}\,.\ ] ] using proposition [ prop : a_m ] and the fact that @xmath327 while @xmath867 has modulus one here then allows us to conclude that @xmath875 for all @xmath268 . the @xmath810 appears because we need the estimate down to the outside boundary of the shrinking disk at the origin . next we look at the contours @xmath357 and @xmath358 . on these contours there is no jump for @xmath133 , and we see directly from proposition [ prop : noutproperties ] that @xmath876 is exponentially small as @xmath118 tends to zero through positive values . the next contour we examine is the portion of @xmath101 lying above the disk @xmath399 . here we observe that @xmath401 has no jump , and because we have on this contour the strict inequality @xmath353 , the matrix @xmath877 is exponentially close to the identity matrix . to see this , note that here @xmath878\sigma_1^{\frac{1-j}{2}}\ ] ] because @xmath328 . while the relative error in replacing @xmath267 by @xmath879 is not small near @xmath117 , it is bounded . so the exponential decay afforded by the strict inequality on the real part of @xmath323 is maintained . virtually the same arguments show that on the contours @xmath99 and @xmath100 outside of the disk at the origin , the quotient @xmath873 is again an exponentially small perturbation of the identity matrix , since on these contours there is again no jump of the parametrix @xmath393 . therefore , for all @xmath861 outside all disks , we have @xmath880 for all @xmath268 . we come up with an overall estimate for @xmath881 for @xmath837 by combining the estimates ( [ eq : ve1 ] ) , ( [ eq : ve2 ] ) , ( [ eq : ve3 ] ) , ( [ eq : ve4 ] ) , and ( [ eq : ve5 ] ) , and optimizing the error by choosing the parameter @xmath810 . the optimal balance among all @xmath882 comes from taking @xmath835 , which gives an overall error estimate of @xmath883 uniformly for all @xmath861 , for all @xmath35 . this proves the proposition . @xmath105 the following is then a consequence of the @xmath36 theory of riemann - hilbert problems ( see the analogous discussion in @xcite ) . for @xmath118 sufficiently small , the riemann - hilbert problem for @xmath822 has a unique solution . let @xmath884 be sufficiently large so that @xmath820 is contained in the circle of radius @xmath885 centered at the origin . then uniformly for all @xmath22 outside of this circle , and for all @xmath268 however small , the matrix @xmath824 satisfies @xmath886 with the size of the matrix measured in any matrix norm . _ proof : _ only one thing must be verified in order to deduce existence and uniqueness from the general theory : the cauchy - kernel singular integral operators defined on the contour @xmath820 , which depends on @xmath118 because of the shrinking boundary of the circle at the origin , have @xmath887 norms that can be bounded uniformly in @xmath25 . but this fact follows in this case from the fact that for @xmath25 sufficiently large , the circle @xmath611 intersects only radial straight - line segments ( we chose the contours to all be exactly straight lines in some fixed neighborhood of the origin ) , so that the portion of @xmath820 near the origin simply scales with @xmath857 . the uniform bound we need can then be established using the fact that the cauchy operators commute with scaling . a similar result was established under more general conditions in @xcite . once existence and uniqueness have been established , the estimate of @xmath822 follows from an integral representation formula for this matrix : @xmath888 in which the matrix function @xmath41 for @xmath889 is an element of @xmath887 with a norm that is bounded independently of @xmath118 ( and in fact converges to the identity matrix in @xmath887 as @xmath118 tends to zero ) . @xmath105 we are now in a position to prove our main result , which we presented as theorem [ theorem : mainresult ] in the introduction . _ proof of theorem [ theorem : mainresult ] : _ we have been setting @xmath3 all along , so the function @xmath890 given by ( [ eq : psi0define ] ) can be found from the matrix @xmath133 by the relation ( _ cf . _ equation ( [ eq : potentialrecovery ] ) ) @xmath891 writing @xmath892 , we get @xmath893 now , the first term on the right - hand side of ( [ eq : controlpsi ] ) can be evaluated explicitly since near @xmath894 we have @xmath895 , and we have an explicit formula ( _ cf . _ equation ( [ eq : otildesolution ] ) ) for @xmath374 . we find @xmath896 which is the `` true '' initial data that we started with , before making any modifications based on the wkb approximation of the spectral data . when we consider the second term on the right - hand side of ( [ eq : controlpsi ] ) , we see that as a consequence of the normalization of the matrices @xmath822 and @xmath825 , @xmath897 as @xmath135 . therefore the second term on the right - hand side of ( [ eq : controlpsi ] ) vanishes identically . finally , for the third term on the right - hand side of ( [ eq : controlpsi ] ) we can again apply the normalization condition for @xmath809 to obtain @xmath898 putting these steps together , we have @xmath899 the proof of the theorem is finished upon using the integral formula ( [ eq : cauchyformula ] ) for @xmath822 and proposition [ prop : errorjumpbound ] . using the new technique of simultaneous interpolation of residues by two different analytic interpolating functions , combined with `` steepest - descent '' techniques for matrix - valued riemann - hilbert problems , we have established the validity of the formal wkb approximation of the spectrum in the nonselfadjoint zakharov - shabat eigenvalue problem ( [ eq : zs ] ) in the sense of pointwise convergence of the potentials . strictly speaking , our analysis applies to certain classes of potential functions whose most important property for our purposes is their real analyticity , and then we obtain convergence for all nonzero values of @xmath4 . in order to extend the result of theorem [ theorem : mainresult ] to @xmath12 , some different steps are required . since @xmath900 as @xmath901 , the local analysis that we carried out independently for @xmath902 ( _ cf . _ [ sec : local_ia ] ) and for @xmath903 ( _ cf . _ [ sec : local_ia(x ) ] ) will need to be combined . consequently , a different local model for @xmath133 will need to be constructed near @xmath904 . due to the presence of the gamma functions in the asymptotics established in [ sec : local_ia ] and given in proposition [ prop : iaasymps ] and proposition [ prop : iaasympt ] , it is likely that the construction of the local model will require knowledge of the solution of a new riemann - hilbert problem that , like that for the matrix @xmath568 in [ sec : origin ] , can not be solved explicitly . nonetheless , one expects that to establish the validity of theorem [ theorem : mainresult ] for @xmath12 will require only technical modifications of what we have done here . understanding the nature of the wkb approximation at the level of the potentials is one step in a larger ongoing program to obtain corresponding information at the level of the ( unknown ) spectrum itself . indeed , quantifying the difference between the true spectrum of a given potential @xmath10 and the wkb approximation of the spectrum , in terms of motion of eigenvalues , will be necessary before it can be proven that the rigorous asymptotic analysis of sses is relevant to the problem of semiclassical asymptotics for the initial - value problem for the focusing nonlinear schrdinger equation ( [ eq : nls ] ) . one imagines that a study of the semiclassical limit for ( [ eq : nls ] ) should proceed by first generating from the given initial data @xmath905 the corresponding well - defined sse , and then using the fact that by combining the results of @xcite with theorem [ theorem : mainresult ] from this paper , one has a complete picture of the limiting behavior of the sse for an open interval of time @xmath5 that is independent of @xmath23 and includes @xmath3 , and moreover that according to theorem [ theorem : mainresult ] the sse is pointwise close to the given initial data @xmath10 for @xmath3 . the problem here is that the focusing nonlinear schrdinger equation is known to have modulational instabilities whose exponential growth rates become arbitrarily large in the semiclassical limit . there is , therefore , the very real possibility that while the sse is close to the initial data @xmath10 at @xmath3 , it is not close to the corresponding solution of ( [ eq : nls ] ) for any positive @xmath5 . in order to control the difference for positive time , it is necessary to know in advance how much the sse spectral data differs from the true ( unknown ) spectral data , as it is the spectral data that is the starting point for analysis ( _ cf . _ riemann - hilbert problem [ rhp : meromorphic ] ) . one can imagine obtaining this sort of information from the pointwise estimate given in theorem [ theorem : mainresult ] by rayleigh - schrdinger perturbation theory . i would like to thank jinho baik , thomas kriecherbauer , and ken mclaughlin for useful feedback . this work was supported in part by the national science foundation under award number dms-0103909 . kmm00 j. baik , t. kriecherbauer , k. t .- mclaughlin , and p. d. miller , `` strong asymptotics for discrete orthogonal polynomials '' , in preparation . p. deift , s. kamvissis , t. kriecherbauer , and x. zhou , `` the toda rarefaction problem '' , _ comm . pure appl . _ , * 49 * , 3583 , 1996 . p. deift , t. kriecherbauer , k. t .- mclaughlin , s. venakides , and x. zhou , `` strong asymptotics of orthogonal polynomials with respect to exponential weights '' , _ comm . pure appl . math._,*52 * , 14911552 , 1999 . p. deift , s. venakides , and x. zhou , `` the collisionless shock region for the long - time behavior of solutions of the kdv equation '' , _ comm . pure appl . _ , * 47 * , 199206 , 1994 . p. deift and x. zhou , `` a steepest descent method for oscillatory riemann - hilbert problems : asymptotics for the mkdv equation '' , _ ann . _ , * 137 * , 295368 , 1993 . n. ercolani , s. jin , c. d. levermore , and w. macevoy , jr . , `` the zero dispersion limit of the nls / mkdv hierarchy for the nonselfadjoint zs operator '' , preprint , 1993 . a. s. fokas , a. r. its , and a. v. kitaev , `` the isomonodromy approach to matrix models in 2d quantum gravity '' , _ comm . _ , * 147 * , 395430 , 1992 . s. kamvissis , k. t .- mclaughlin , and p. d. miller , `` semiclassical soliton ensembles for the focusing nonlinear schrdinger equation '' , submitted to _ ann . online copy available at ` http://arxiv.org/abs/nlin.si/0012034 ` a. b. j. kuijlaars and e. a. rakhmanov , `` zero distributions for discrete orthogonal polynomials '' , _ j. comput . _ , * 99 * , 255274 , 1998 . corresponding corrigendum : _ j. comput . _ , * 104 * , 213 , 1999 . p. d. lax and c. d. levermore , `` the small dispersion limit of the korteweg - de vries equation . i , ii , iii '' , _ comm . pure appl . _ , * 36 * , 253290 , 571593 , 809830 , 1983 . s. venakides , `` the korteweg - de vries equation with small dispersion : higher order lax - levermore theory '' , _ comm . pure appl . _ , * 43 * , 335361 , 1990 . v. e. zakharov and a. b. shabat , `` exact theory of two - dimensional self - focusing and one - dimensional self - modulation of waves in nonlinear media '' , _ sov . jetp _ , * 34 * , 6269 , 1972 .

the standard thin accretion disk model @xcite has been widely applied to x - ray binaries and active galactic nuclei . due to the basic assumption of the energy balance between the viscous heating and the radiative cooling , such a model was known to be invalid for super - eddington accretion case , where the advective cooling is probably significant . instead , the slim disk model @xcite was introduced to describe super - eddington accretion disks . however , there exists some conflict between the theory and the observation . the theory predicts that the advection is negligible for @xmath1 ( e.g. , * ? ? ? * ; * ? ? ? * ) , where @xmath2 is the eddington luminosity , which indicates that the standard disk model should be valid up to @xmath2 . on the contrary , some works on the black hole spin measurement showed that the standard disk model is likely to be inaccurate for @xmath3 ( e.g. , * ? ? ? * ) . moreover , even the recent general model for optically thick disks ( e.g. , * ? ? ? * ; * ? ? ? * ) , which unifies the standard thin disk and the slim disk , could not help to obtain a self - consistent spin parameter for @xmath3 ( e.g. , * ? ? ? * ) . in our opinion , the above conflict may be resolved if the vertical structure is well incorporated . most previous works on accretion disks focused on the radial structure in cylindrical coordinates ( @xmath4 , @xmath5 , @xmath6 ) . for the vertical structure , however , a simple well - known relationship @xmath7 " or @xmath8 " was widely adopted , where @xmath9 is the half - height of the disk , @xmath10 is the sound speed , and @xmath11 is the keplerian angular velocity . such a relationship comes from the vertical hydrostatic equilibrium with two additional assumptions . one is the approximation of gravitational potential : @xmath12 , and the other is a one - zone approximation or a polytropic relation @xmath13 in the vertical direction ( e.g. , * ? ? ? * ) , where @xmath14 is the total ( gas plus radiation ) pressure and @xmath15 is the density . obviously , the above assumptions work well for geometrically thin disks , but may be inaccurate for the mass accretion rate @xmath16 approaching the eddington one @xmath17 , for which the disk is probably not thin . consequently , the relationship @xmath8 " may be invalid for @xmath18 . without the potential approximation , our two previous works investigated the geometrical thickness of accretion disks and the validity of the relationship @xmath8 " . @xcite adopted the explicit gravitational potential in cylindrical coordinates and found that the above relationship is inaccurate for @xmath18 , and therefore the disk can be geometrically thick . @xcite took spherical coordinates to avoid the approximation of gravitational potential , and found that an advection - dominated accretion disk is likely to be quite thick . in these two works , however , the polytropic relation is still adopted in the vertical direction , which takes the place of the energy balance per unit volume between the viscous heating and the advective cooling plus the radiative cooling . the validity of such a polytropic relation , however , remains questionable , in particular for large @xmath16 due to dominant radiation pressure . the purpose of this paper is to revisit the vertical structure of radiation pressure - supported disks by taking into account the local energy balance and to study the variation of energy advection with mass accretion rates . the paper is organized as follows . equations and boundary conditions are derived in section 2 . a global view of the solutions in the @xmath19-@xmath20 diagram is presented in section 3 . for a typical radius @xmath21 , the vertical structure and the energy advection are investigated in section 4 . the two - dimensional solutions and the convective stability are studied in section 5 . summary and discussion are made in section 6 . we consider a steady state axisymmetric accretion disk in spherical coordinates ( @xmath20 , @xmath22 , @xmath5 ) and use the newtonian potential , @xmath23 , where @xmath24 is the black hole mass . following @xcite , we assume @xmath25 for simplicity , which means a hydrostatic equilibrium in the @xmath22 direction . simulations , ( e.g. , * ? ? ? * figure 3 ) , however , revealed that @xmath26 will be significant for extremely high accretion rates such as @xmath27 . as shown in the following sections , our solutions mainly correspond to @xmath16 around @xmath17 . for such accretion rates , the validity of @xmath25 remains a question . the basic equations of continuity and momentum take the forms ( e.g. , * ? ? ? * ) : @xmath28 @xmath29 @xmath30 @xmath31 where @xmath32 and @xmath33 are respectively the radial and azimuthal velocity , @xmath34 and @xmath35 are respectively the radial and vertical radiation flux , @xmath36 is the gas pressure , @xmath37 is the opacity of electron scattering , and @xmath38 is the @xmath39 component of the viscous stress tensor , @xmath40 . following the spirit of @xmath41 stress prescription , we assume the kinematic viscosity coefficient @xmath42 , where @xmath10 is the sound speed defined below ( equation ( 7 ) ) , and @xmath43 is the keplerian velocity . we would stress that , even though the @xmath41 stress prescription has been widely adopted for theoretical studies , simulations of magnetorotational turbulence have shown that the stress does not well scale locally with the pressure . for instance , the simulations on thin disks by a shearing box showed that , the time and box - averaged results are likely to support that the stress is proportional to the thermal ( gas plus radiation ) pressure ( e.g. , * ? ? ? * figure 3 ) . however , figure 16 of @xcite shows that the maximal thermal pressure is located on the equatorial plane , whereas figure 11 shows that the maximal stress is obviously not at the same place . these two figures reveal that the stress is not proportional to the pressure locally . in the present study , for simplicity , we will keep the local @xmath41 stress prescription for numerical calculation , which is a weak point of this work . the energy equation including gas and radiation is written as ( e.g. , * ? ? ? * ) @xmath44 = -p { { \mbox{\boldmath $ \nabla$ } } } \cdot { { \mbox{\boldmath $ v$ } } } - { { \mbox{\boldmath $ \nabla$ } } } { { \mbox{\boldmath $ v$ } } } { { \mbox{\boldmath $ : $ } } } { { \mbox{\boldmath $ p$ } } } - { { \mbox{\boldmath $ \nabla$ } } } \cdot { { \mbox{\boldmath $ f$ } } } + \phi_{\rm vis } \ , \ ] ] where @xmath45 and @xmath46 are the internal energy density of the gas and the radiation , respectively . @xmath47 is the radiation pressure tensor , @xmath48 is the viscous dissipative function , and the radiation flux @xmath49 is expressed as @xmath50 in this work , we focus on the region inside the photosphere , so we can take the well - known eddington approximation , i.e. , @xmath51 and the eddington tensor @xmath52 . since we only study the radiation pressure - supported disks , the gas pressure @xmath36 and the gas internal energy density @xmath45 will be dropped in our calculation . in order to avoid directly solving the partial differential equations , some assumptions on the radial derivatives ( @xmath53 ) are required . following the spirit of self - similar assumptions ( e.g. , * ? ? ? * ; * ? ? ? * ) , we adopt the following radial derivatives for @xmath10 and @xmath46 : @xmath54 where the sound speed @xmath10 is defined as @xmath55 based on the above two radial derivatives , the following four derivatives can be inferred from equations ( 1)-(7 ) : @xmath56 with all the above derivatives , we can remove the @xmath53 " terms in equations ( 2 ) and ( 4)-(6 ) , and the following equations are then obtained from equations ( 2)-(6 ) : @xmath57 @xmath58 @xmath59 @xmath60 @xmath61 @xmath62 the seven equations , equations ( 7)-(13 ) , enable us to solve for the seven variables : @xmath32 , @xmath33 , @xmath10 , @xmath15 , @xmath46 , @xmath34 , and @xmath35 . there are two differential equations in this system . in addition , the position of the surface is unknown . thus , totally three boundary conditions are required to determine a unique solution . an obvious boundary condition on the equatorial plane is @xmath63 . however , this condition is not applicable for numerical calculation since it is automatically matched as indicated by equation ( 9 ) . combining equations ( 9 ) and ( 11 ) we can derive the following equation : @xmath64 an alternative boundary condition on the equatorial plane is then obtained from the above equation ( the left - hand side is zero thus the right - hand side should also be zero ) : @xmath65 the second boundary condition is a definition of the surface . we define the photosphere as the position above which the optical depth is around unity . the condition can be written as @xmath66 where @xmath67 ( @xmath68 ) is the polar angle of the photosphere . the third condition is related to the mass accretion rate : @xmath69 in our calculation we set @xmath70 , @xmath71 @xmath72g@xmath73 , and @xmath74 , where the value of @xmath41 is taken from recent simulations @xcite . the eddington accretion rate is expressed as @xmath75 , where @xmath76 is a radiative efficiency of the flow . we choose @xmath77 since it is comparable to the schwarzschild black hole efficiency of 0.057 . the dimensionless accretion rate is defined as @xmath78 . with the equations and boundary conditions in section 2 , we can numerically derive the @xmath22-direction distribution of physical quantities for a given @xmath19 at a certain radius @xmath20 . the radiation pressure - supported disk solutions in the @xmath19-@xmath20 diagram are shown in figure 1 , where @xmath79 is the schwarzschild radius . the parameter space is divided into three regions by two parallel solid lines , roughly with @xmath80 . the region above the upper solid line is denoted by outflow " , where we can not find solutions . no solution exists probably due to the assumption of @xmath81 in advance . in our view , a real flow located in this region may have @xmath82 and the inflow accretion rate may decrease inward . the physical understanding could be that , for high accretion rates and particularly for the inner radii , the viscous dissipation may be sufficiently large such that the radiation pressure is too strong to be balanced by the gravitational force . thus , outflows may be driven by the radiation pressure and the inflow @xmath19 drops inward . on the other hand , simulations of supercritical accretion flows ( e.g. , * ? ? ? * figure 6 ) showed that the inflow accretion rate roughly follows the @xmath80 relationship for @xmath83 at @xmath84 ( corresponding to @xmath85 due to the definition of @xmath17 with @xmath77 ) . the slope of the upper solid line in figure 1 , which may be regarded as maximal accretion rates due to our calculation , agrees well with the slope in the above simulations . the region under the lower solid line is denoted by gas pressure " , where no solution is found either . in our understanding , it is probably because the gas pressure can not be ignored in this region , which may be in conflict with the radiation pressure - supported assumption . we would point out that , the lower solid line in this diagram is higher than the well - known line which separates the inner and middle regions of standard thin disks @xcite . the reason is that , the gas and radiation pressure are comparable for the latter , whereas the radiation pressure - supported disk may require the accretion rate to be higher such that the radiation pressure sufficiently dominate over the gas pressure , and therefore the effect of gas pressure on the vertical structure can be completely ignored . the region between the two solid line , denoted by radiation pressure " , which means that the radiation pressure is completely dominated , corresponds to the solutions of our main interest in this work . in section 4 , we will focus on the vertical structure and the energy advection at a typical radius , @xmath86 , as indicated by the vertical dashed line in figure 1 . in section 5 , we will study the two - dimensional solutions for a typical accretion rate @xmath87 in the range @xmath88 and @xmath89 , as indicated by the horizontal dot - dashed line . in addition , we would point out that for inner radii such as @xmath90 , the two solid lines in figure 1 may deviate from a real black hole accretion system due to the newtonian potential used in this work . in this section , we will focus on the solutions at a typical radius @xmath91 . figure 2 shows the vertical structure of the disk with @xmath87 . in figure 2(a ) , the dot - dashed , dotted , solid , and dashed lines show the vertical distribution of the dimensionless density ( @xmath92 ) , radial velocity ( @xmath93 ) , azimuthal velocity ( @xmath94 ) , and sound speed ( @xmath95 ) , respectively , where @xmath96 is the density on the equatorial plane . it is seen that @xmath15 significantly decreases , whereas @xmath10 and @xmath97 increases , from the equatorial plane to the surface . in figure 2(b ) , the solid line shows the variation of @xmath98 ( defined in equation ( 15 ) ) , where the photosphere ( @xmath99 ) is located at @xmath100 , quite close to the polar axis . the disk seems to be extremely thick according to the position of the photosphere . however , the profile of @xmath15 implies that most of the accreted matter exists in a moderate range around the equatorial plane , such as @xmath101 , which is more clear in figures 4 and 5 ( discussed below ) . the dashed line shows the variation of @xmath102 . there exists @xmath103 for the whole solution , which indicates that the eddington approximation is valid and the solution is therefore self - consistent . with a more general viscosity , @xcite studied a geometrically thick , radiation pressure - supported model for supercritical accretion disks . they showed that there exists a narrow empty funnel along the rotation axis with a half - opening angle @xmath104 . as seen in our figure 2(a ) , the density drops sharply close to the photosphere , thus a nearly empty funnel also seems to exist in our model . the difference is that , the disk surface in @xcite is the position where some physical quantities such as @xmath32 diverges , whereas our surface is defined as the position where equation ( 15 ) is matched , and no divergence appears in our solutions . for a real disk with @xmath87 , the photosphere may exist between @xmath105 and the present result ( @xmath106 ) . our argument is as follows . there are two possible reasons that may cause the present photosphere quite close to the polar axis . one is that we have ignored the radiation force from one side ( e.g. , @xmath107 and @xmath108 ) to the other ( e.g. , @xmath107 and @xmath109 ) . the other reason is that we consider only the @xmath39 component of the stress tensor , which may cause inaccurate results for small @xmath22 , such as strong shearing of the angular velocity @xmath110 in the vertical direction , where @xmath111 . nevertheless , the @xmath112 component assumption may work well for moderate @xmath22 : @xmath101 . the fact that the surface condition could not be matched in the range @xmath101 indicates that the half - opening angle of the disk ( @xmath113 ) is likely to be larger than @xmath114 . the profile of @xmath10 in figure 2(a ) is quite different from that in the previous works with a vertical polytropic assumption ( e.g. , * ? ? ? * ; * ? ? ? . under the polytropic relation @xmath13 ( normally @xmath115 , and for radiation pressure - dominated case , @xmath14 can be replaced by @xmath116 ) , @xmath10 will decrease continuously from the equatorial plane to the surface . the reason for the opposite behavior of @xmath10 , as implied in figure 2(a ) , is that @xmath15 drops faster than @xmath46 from the equatorial plane to the surface . figure 3 shows the variation of the quantity @xmath117 with @xmath22 for @xmath118 ( dashed line ) , @xmath87 ( solid line ) , and @xmath119 ( dotted line ) . if the polytropic relation works well , @xmath117 should be a constant of @xmath120 . it is clearly shown in figure 3 that , however , @xmath117 varies significantly with @xmath22 rather than being a constant . more importantly , @xmath121 indicates that @xmath122 is negative thus unacceptable . we therefore argue that the polytropic relation should be unsuitable for describing the vertical structure of radiation pressure - supported disks . moreover , since the energy advection is relevant to @xmath10 , we may expect essentially different results on the strength of advection . figure 4 shows the variation of the vertically averaged advection factor @xmath123 with the mass accretion rate @xmath19 , where @xmath123 is defined as @xmath124 . the quantities @xmath125 and @xmath126 are expressed as follows : @xmath127 @xmath128 where @xmath129 and @xmath130 are respectively the advective cooling rate and the viscous heating rate per unit volume , as implied by the left - hand side of equation ( 11 ) . the solid line in figure 4 corresponds to the total accretion rate integrating from @xmath67 to @xmath131 , as shown by equation ( 16 ) , whereas the dashed line corresponds to the specific accretion rate integrating from @xmath132 to @xmath133 , i.e. , @xmath134 the reason why we calculate for @xmath135 is that the @xmath112 stress assumption may work well for @xmath101 . as shown by the horizontal range of the solid and dashed lines , most of the accreted matter exists in this specific range , e.g. , @xmath136 corresponding to @xmath87 . the figure also shows that @xmath123 rapidly increases with increasing @xmath19 in the range @xmath137 . more importantly , the value of @xmath123 ( @xmath138 ) indicates that the energy advection is significant even for sub - eddington accretion disks . such a result is quite different from the previous one , where advection was found to be significant only for super - eddington accretion case . @xcite introduced an elegant formula to describe the @xmath139 relationship based on their numerical solutions under the well - known paczyski - wiita potential @xcite . their equations ( 15)-(19 ) imply that , for the position @xmath91 , advection is negligible for @xmath140 ( @xmath141 ) . for the whole disk , advection is negligible for @xmath142 ( @xmath143 ) . such a critical @xmath16 for the whole disk was confirmed by some recent global solutions under the general relativity . figure 4.11 of @xcite shows that advection is negligible for @xmath1 for any spin parameter @xmath144 . for @xmath145 , i.e. , the schwarzschild black hole , the critical @xmath16 is just around @xmath17 . in our opinion , the different results on the advection between the above two works and ours are related to the different approach to describing the vertical structure . a significant difference is that @xcite and @xcite chose the cylindrical coordinates whereas we adopt the spherical coordinates . of course , the final results should not depend on the coordinates used . however , as pointed out by @xcite , there are some interesting differences between the equations written in cylindrical and spherical coordinates . there is no centrifugal force in the @xmath6 direction in cylindrical coordinates , whereas there is no gravitational force in the @xmath22 direction in spherical coordinates . @xcite claimed that it is exactly this property that makes the spherical coordinates much better adapted for describing the flow near the black hole horizon . here , we argue that the spherical coordinates should be more suitable for describing geometrically thick disks as follows . in cylindrical coordinates , in the @xmath6 direction , whether with a polytropic relation between the pressure and the density @xcite , or with the local energy balance @xcite , an approximation for the gravitational force , i.e. , @xmath146 , was adopted for describing the vertical structure . such an approximation will probably be invalid for @xmath147 . in particular for @xmath148 , the approximate force goes to infinity whereas the real force ought to vanish . thus , the cylindrical coordinates seem unsuitable for studying geometrically thick disks . in other words , a geometrically thick disk solution in cylindrical coordinates may not be self - consistent . @xcite limited their solutions by @xmath149 probably due to this reason . as shown by their figure 10 , the maximal value of @xmath150 for @xmath151 is @xmath152 . for higher @xmath16 , the value of @xmath150 will be even larger thus the solution based on the approximate force may be inaccurate . on the contrary , in spherical coordinates , there is no need to make approximation for the gravitational force . the centrifugal force in the @xmath22 direction , which takes the place of the @xmath6-direction gravitational force in cylindrical coordinates , is derived in this work by solving the vertical differential equations . thus , our approach to the vertical structure seems to be more reasonable . we would agree that the solutions in @xcite are likely to be self - consistent since their @xmath150 is significantly less than unity , in particular for the solutions with @xmath153 . then what are the reasons for the quantitative difference in the advection for @xmath153 between their solutions and ours ? in our understanding , there exist three possible reasons as follows . first , as mentioned in section 3 of @xcite for their numerical methods , the vertical structure is derived by a given advection factor @xmath123 in advance . the value of @xmath123 is probably obtained by solving the radial structure on the equatorial plane . moreover , their @xmath123 is assumed to be uniform in the @xmath6 direction . on the contrary , we obtain a varying @xmath123 by solving the vertical equations . as shown by our figure 8 , @xmath123 increases significantly with @xmath6 . we can therefore expect that the vertically averaged @xmath123 at a cylindrical radius will also be significantly larger than that at @xmath154 , which may explain why our @xmath123 is larger than that in @xcite . second , @xcite assumed a uniform @xmath155 and @xmath33 in the @xmath6 direction , and used the keplerian strain to calculate the viscous dissipation . in our method , however , we include varying @xmath32 and @xmath33 in the vertical direction , and the viscous dissipation is calculated based on @xmath33 instead of @xmath156 . third , @xcite assumed @xmath157 whereas we have @xmath25 . as stressed by @xcite , since the stationary accretion flows resemble quasi - spherical flows ( @xmath158 constant ) much more than quasi - horizontal flows ( @xmath159 constant ) , @xmath25 may be a more reasonable approximation than @xmath157 . moreover , the above three reasons may also be responsible for the different results in the convective stability , as will be discussed in section 5.3 . as mentioned in section 1 , some works on the black hole spin measurement showed that the standard thin disk model is likely to be inaccurate for @xmath3 ( e.g. , * ? ? ? * ; * ? ? ? one explanation is that the inner disk edge is still located at the innermost stable circular orbit ( isco ) , but its emission is shaded by the outer disk . thus , the inner disk radius obtained from the spectral fitting is not true . however , @xcite showed that the disks in black hole and neutron star x - ray binaries trace the same evolutionary pattern for @xmath3 . in addition , for the neutron star system xte j1701 - 462 , the boundary emission area maintains nearly constant despite the varying luminosity of the disk ( * ? ? ? * figure 17 ) , which indicates that the neutron star s surface is not shaded . @xcite therefore argued that the inner disk of the black hole system should not be shaded either due to the similar phenomenon . they suggested that the inner disk radius moves outward because of the increasing radiation pressure . in our opinion , from the energy advection , it is easy to understand that the standard disk model seems to be inaccurate above @xmath160 . as revealed by the lines in figure 4 , @xmath123 is likely to be non - negligible ( probably @xmath161 ) for @xmath162 at @xmath21 . we would point out that , compared with the paczyski - wiita potential , the newtonian potential in the present work may magnify the viscous heating rate at small radii such as @xmath163 , thus the real @xmath123 at @xmath163 may be smaller than the values showed in figure 4 . on the other hand , for the same @xmath19 , since the viscous heating rate at a smaller radius such as @xmath164 will probably be larger than that at @xmath91 , so does the advection factor . we can therefore expect that , even for the paczyski - wiita potential , the advection at the position close to the isco should be non - negligible for @xmath162 . consequently , the standard thin disk model , based on the energy balance between the viscous heating and the radiative cooling with the advective cooling being ignored , may be inaccurate . figure 2 shows that @xmath15 decreases significantly with decreasing @xmath22 , and figure 4 implies that most of the accreted matter exists in the range @xmath101 . in order to have a more clear view , we define an averaged dimensionless height as @xmath165 , where the surface density @xmath166 takes the form : @xmath167 figure 5 shows the variation of @xmath123 with @xmath168 ( solid line ) . even though the photosphere is close to the polar axis , the averaged height @xmath168 is geometrically slim with @xmath169 . furthermore , the figure shows that @xmath123 increases with increasing @xmath168 or @xmath19 , which agrees with the classic picture . for quantitative comparison , we plot the function @xmath170 ( dashed line ) in figure 5 due to the relationship @xmath171 introduced by @xcite , which is equivalent to @xmath172 here . it is seen that @xmath123 is not well proportional to @xmath173 . in the range @xmath174 or @xmath175 , however , we may regard the formula @xmath170 as a rough approximation . in section 4 , we focus on the solutions at a typical radius @xmath21 . in this section we will study the disk solutions for various radii . since the vertical solutions are based on the assumptions of partial derivatives in the radial direction ( presented in section 2.1 ) , it is necessary to derive vertical solutions for various radii to check whether these assumptions are self - consistent . following the example solution in figure 2 , we study the two - dimensional solutions for @xmath87 in the range @xmath176 and @xmath89 . figure 6 shows the radial variations of @xmath10 and @xmath46 ( solid lines ) for five polar angles , i.e. , @xmath177 , @xmath178 , @xmath179 , @xmath180 , and @xmath181 . for comparison , the radial profile of @xmath156 , which is proportional to @xmath182 , is shown in figure 6(a ) , and an example slope of @xmath183 is shown in figure 6(b ) . the figure shows that @xmath10 and @xmath46 behave roughly as @xmath184 and @xmath183 , respectively , which agrees with the original assumptions of radial derivatives . as mentioned in section 2.1 , once the radial derivatives of @xmath10 and @xmath46 are given , the other ones can be inferred from equations ( 1)-(7 ) . thus , we can expect that the radial derivatives of @xmath32 , @xmath33 , @xmath15 , and @xmath34 in the two - dimensional solutions should also be in agreement with the assumptions . our solutions are therefore likely to be self - consistent . in our calculation , the location of the photosphere does not vary much with the radius , i.e. , @xmath185 for various radii . as discussed in section 4.1 , the real position of the photosphere is likely to be located in the range @xmath186 . we will make some comparison with simulations for the photosphere in section 6 . as shown in figure 6 , the derivatives of @xmath10 and @xmath46 deviate a little for @xmath187 and @xmath177 . such a divergence may be well understood from figure 1 , which shows that the solution for @xmath87 and @xmath187 is quite close to the lower solid line , which indicates that the gas pressure may not be negligible . in particular for the equatorial plane , the mass density has the maximal value there , thus the gas pressure may be most significant at this position . in this section , we will study the convective stability of the radiation pressure - supported disks . the well - known solberg - hiland conditions in cylindrical coordinates ( @xmath4 , @xmath5 , @xmath6 ) take the forms ( e.g. , * ? ? ? * ) : @xmath188 @xmath189 where @xmath190 is the specific angular momentum per unit mass , @xmath191 is the total pressure , @xmath192 is the specific heat at constant pressure , and @xmath193 is the entropy expressed as @xmath194 where @xmath195 is the adiabatic index . the @xmath4 and @xmath6 components of the well - known brunt - visl frequency are written as @xmath196 @xmath197 and the epicyclic frequency takes the form : @xmath198 thus , the first solberg - hiland condition , equation ( 21 ) , can be simplified as @xmath199 where @xmath200 is defined as an effective frequency . for accretion disks , there usually exists @xmath201 ( as shown in figure 8) , so the second solberg - hiland condition , equation ( 22 ) , reduces to @xmath202 in numerical calculation , we adopt @xmath203 and @xmath204 according to the radiation pressure - supported assumption . based on the two - dimensional solutions for @xmath87 in section 5.1 , we can obtain the variations of physical quantities in cylindrical coordinates and therefore investigate the convective stability by equations ( 24)-(25 ) . we take the cylindrical radius @xmath205 as a typical position to study the convective stability . figure 7 shows the @xmath6-direction variations of @xmath206 , @xmath207 , @xmath208 , @xmath209 , and @xmath210 , where the former four quantities are normalized by @xmath211 , and the last one is normalized by @xmath212 . the positive values for both @xmath209 and @xmath210 indicate that the disk should be convectively stable . for the equatorial plane , the result of @xmath213 can be inferred by equation ( 15 ) of @xcite , which revealed that the disk will always be convectively stable for @xmath204 at @xmath154 . for further understanding the convectively stable results for @xmath214 , we plot figure 8 to show the @xmath6-direction variations of @xmath15 , @xmath46 , @xmath190 , @xmath215 , and the advection factor @xmath123 . it is seen that @xmath15 drops faster than @xmath46 with increasing @xmath6 , so the quantity @xmath215 increases with @xmath6 . from equation ( 23 ) we immediately have @xmath216 which is known as the schwarzschild criterion for a constant angular velocity at a cylindrical surface ( @xmath217 ) , corresponding to the so - called barytropic flows where the pressure depends only on the density . as the dashed line in figure 8 shows , the angular momentum @xmath190 ( or equivalently the angular velocity @xmath110 ) does not vary significantly with @xmath6 . thus , the convectively stable results are easy to understand from @xmath218 . in a similar study ( vertical structure based on the local energy balance ) of the general model for optically thick disks , however , the disk was found to be convectively unstable ( e.g. , * ? ? ? * ; * ? ? ? as interpreted by the three possible reasons in section 4.2 , the significant difference in the results between their works and ours is probably related to the different approach to describing the vertical structure . in addition , here we would mention two more details which may help to understand the difference in convective stability . first , the profiles of @xmath15 and @xmath46 in figure 8 reveal that the absolute value of radial velocity @xmath97 increases with @xmath6 ( @xmath219 inferred from equations ( 7 ) and ( 10 ) ) . compared with the uniform @xmath155 in @xcite , our increasing @xmath97 with @xmath6 may result in faster drop of @xmath15 in the @xmath6 direction ( steeper slope of @xmath15 ) if we simply assume the mass supply to be comparable . second , compared with @xcite , the advection in our solutions is significantly stronger , which means that for the same @xmath19 thus comparable viscous heating rate , the vertical radiation flux @xmath220 will be less in our results . thus , @xmath46 may decrease slower in the @xmath6 direction ( flatter slope of @xmath46 ) due to the less @xmath220 and lower @xmath15 ( inferred from equation ( 6 ) ) . as indicated by equation ( 26 ) , the flatter slope of @xmath46 and the steeper slope of @xmath15 will both make contribution to @xmath218 , and the disk is therefore likely to be convectively stable . we would stress that , our solutions are limited by the radiation pressure - supported case . thus , the present results can not directly show the convective stability of disks for either the gas pressure - supported case or the case of comparable gas and radiation pressure . actually , we have made some additional calculation for thin disks in cylindrical coordinates to check the convective stability , following the method of @xcite but without considering advection . we found that the disk is convectively stable for the gas pressure - supported case , whereas the disk is convectively unstable for the case of radiation pressure being significant . thus , we would agree with @xcite on the convectively unstable disks for moderate accretion rates such as @xmath221 , corresponding to significant radiation pressure and non - negligible gas pressure . moreover , for @xmath222 , the disk will be geometrically thin , thus there is no difference between the assumptions @xmath25 and @xmath157 , and @xmath123 is probably negligible . as a consequence , the solutions of @xcite ought to be accurate . furthermore , as revealed by the profiles of @xmath46 and @xmath15 in figure 8 , @xmath117 is less than unity in the @xmath6 direction . following the argument in section 4.1 , the polytropic relation seems unsuitable either in the @xmath6 direction . in addition , as shown in figures 7 and 8 , our example solution for @xmath87 at @xmath205 is terminated at @xmath223 . the reason is that the solutions in spherical coordinates is limited by @xmath224 ( as shown by the horizontal dot - dashed line in figure 1 ) , which corresponds to @xmath223 at @xmath205 in cylindrical coordinates . for the region close to the equatorial plane , we can make some analysis of the convective stability by the taylor expansion method . obviously , we have @xmath226 at @xmath154 from symmetric conditions . thus , for @xmath225 , the value of @xmath227 can be estimated by the second - order derivative at @xmath228 : @xmath229 based on equations ( 7)-(12 ) , we can eliminate @xmath32 , @xmath46 , and @xmath35 and therefore obtain a set of three equations for the three quantities @xmath33 , @xmath10 , and @xmath15 . by using the taylor expansion method , together with the boundary condition of equation ( 14 ) , we derive the following three relationships for the second - order derivatives of @xmath33 , @xmath10 , and @xmath15 ( the ram - pressure term in equation ( 8) is ignored ) : @xmath230 @xmath231 @xmath232 where @xmath233 is defined as @xmath234 , which is a small value in the analysis . in our solutions we have @xmath235 at @xmath154 , e.g. , @xmath236 for @xmath87 at @xmath237 . for the simple case with @xmath238 , the above three relationships provide @xmath239 the second - order derivative of entropy at @xmath240 can therefore be derived by the following coordinate transformation : @xmath241 thus , equations ( 27)-(28 ) indicate @xmath218 for the region close to the equatorial plane . the disk in this region is therefore likely to be convectively stable . in this paper , we have studied the vertical structure , energy advection , and convective stability of radiation pressure - supported disks in spherical coordinates . in the @xmath22 direction , we replaced the pressure - density polytropic relation by the local energy balance per unit volume between the viscous heating and the advective cooling plus the radiative cooling , and obtained the distribution of physical quantities such as @xmath15 , @xmath32 , @xmath33 , @xmath10 , @xmath46 , and @xmath35 . the photosphere was found close to the polar axis and therefore the disk seems to be extremely thick . however , most of the accreted matter exists in a moderate range around the equatorial plane such as @xmath242 . we showed that the polytropic relation is unsuitable for describing the vertical structure of radiation pressure - supported disks . more importantly , we found that the energy advection is significant even for slightly sub - eddington accretion disks , which is quite different from the previous result that the advection is of importance only for super - eddington accretion disks . we argued that , the non - negligible advection may help to understand why the standard thin disk model is likely to be inaccurate for @xmath3 . in addition , we studied the two - dimensional solutions to check our basic assumptions of radial derivatives , which indicates our solutions to be self - consistent . furthermore , we investigated the convective stability of the disks in cylindrical coordinates by the two - dimensional solutions derived in spherical coordinates . the disk solutions satisfy the solberg - hiland conditions , which reveals that the disk ought to be convectively stable . to our knowledge , there are mainly two series of simulations on optically thick accretion flows . one is three - dimensional , radiation magnetohydrodynamic ( rmhd ) simulations by a shearing box , which focused on thin disks ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the other is two - dimensional , either radiation hydrodynamic ( rhd ) or rmhd simulations for global flows , which included thin disks and super - eddington accretion flows ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? we would like to compare our numerical results with the latter owing to the global , radiation pressure - supported , and geometrically not thin case . @xcite studied rhd simulations in spherical coordinates with the @xmath41 stress assumption for only the @xmath112 component , for extremely high accretion rates @xmath243 , @xmath244 , and @xmath245 . moreover , @xcite studied rmhd simulations in cylindrical coordinates for @xmath246 . in our results , the position of the photosphere is quite close to the polar axis with @xmath185 . as discussed in section 4.1 , the real position may exist in the range @xmath186 . in the simulations , @xcite showed that the photosphere is around @xmath247 corresponding to @xmath248 . @xcite did not mention the position of photosphere , but their figure 4 for the density distribution reveals that the photosphere should exist at a certain @xmath67 significantly less than @xmath180 . thus , we would express that , for accretion rates around and beyond @xmath17 , the photosphere may exist far from the equatorial plane with @xmath249 . in addition , we discuss the possible link between our disk model and ultraluminous x - ray sources . as shown by the upper solid line in figure 1 , there exists a maximal accretion rate @xmath250 varying with the radius . we argue that the possible upper limit of the accretion rate may help to understand why most ultraluminous x - ray sources ( ulxs ) are not in thermal dominant state . as mentioned in the review paper of @xcite , there may exist three classes of black holes in ulxs : normal stellar mass black holes ( @xmath251 ) , massive stellar black holes ( @xmath252 ) , and intermediate mass black holes ( @xmath253 ) . a massive stellar black hole with a moderate super - eddington accretion rate seems to account for most sources up to luminosities @xmath254 a few @xmath255 erg s@xmath73 . the slim disk model , which is the classic model for super - eddington accretion disks , predicts a dominant thermal radiation from the disk . however , observations have shown that ulxs are not in the thermal dominant state except only a few sources such as m82 x-1 @xcite and hlx-1 @xcite . in our understanding , the radiation of ulxs may be interpreted by an optically thick disk with @xmath256 plus strong outflows . the disk will provide a thermal radiation , which is normally not dominant because of the moderate @xmath250 . on the other hand , the outflows may make contribution to the non - thermal radiation by the bulk motion comptonization @xcite or through the jet of the radiation - pressure driven and magnetically collimated outflow @xcite . the author is particularly grateful to ramesh narayan for constructive suggestions and beneficial discussions . the author also thanks ivan hubeny , da - bin lin , aleksander sadowski , lijun gou , yucong zhu , and shan - shan weng for helpful discussions , and the referee for providing useful comments to improve the paper . this work was supported by the national natural science foundation of china under grants 11073015 and 10833002 , the national basic research program ( 973 program ) of china under grant 2009cb824800 , and the scholarship from china scholarship council under grant 2009835057 . abramowicz , m. a. , chen , x. , kato , s. , lasota , j .- p . , & regev , o. 1995 , , 438 , l37 abramowicz , m. a. , czerny , b. , lasota , j .- p . , & szuszkiewicz , e. 1988 , , 332 , 646 abramowicz , m. a. , lanza , a. , & percival , m. j. 1997 , , 479 , 179 begelman , m. c. , & meier , d. l. 1982 , , 253 , 873 davis , s. w. , narayan , r. , zhu , y. , et al . 2011 , , 734 , 111 feng , h. , & kaaret , p. 2010 , , 712 , l169 feng , h. , & soria , r. 2011 , new astron . rev . , 55 , 166 gu , w .- m . , & lu , j .- f . 2007 , , 660 , 541 gu , w .- m . , xue , l. , liu , t. , & lu , j .- f . 2009 , , 61 , 1313 hirose , s. , blaes , o. , & krolik , j. h. 2009a , , 704 , 781 hirose , s. , krolik , j. h. , & blaes , o. 2009b , , 691 , 16 hirose , s. , krolik , j. h. , & stone , j. m. 2006 , , 640 , 901 hshi , r. 1977 , prog . phys . , 58 , 1191 kato , s. , fukue , j. , & mineshige , s. 2008 , black - hole accretion disks : towards a new paradigm ( kyoto : kyoto univ . press ) krolik , j. h. , hirose , s. , & blaes , o. 2007 , , 664 , 1045 lin , d. , remillard , r. a. , & homan , j. 2009 , , 696 , 1257 mcclintock , j. f. , shafee , r. , narayan , r. , et al . 2006 , , 652 , 518 narayan , r. , & yi , i. 1994 , , 428 , l13 narayan , r. , & yi , i. 1995 , , 444 , 231 ohsuga , k. , & mineshige , s. 2011 , , 736 , 2 ohsuga , k. , mineshige , s. , mori , m. , & kato , y. 2009 , , 61 , l7 ohsuga , k. , mori , m. , nakamoto , t. , & mineshige , s. 2005 , , 628 , 368 paczyski , b. , & wiita , p. j. 1980 , , 88 , 23 sadowski , a. 2011 , phd thesis , arxiv:1108.0396 sadowski , a. , abramowicz , m. , bursa , m. , et al . 2011 , , 527 , a17 sadowski , a. , abramowicz , m. , bursa , m. , et al . 2009 , , 502 , 7 shakura , n. i. , & sunyaev , r. a. 1973 , , 24 , 337 straub , o. , bursa , m. , sadowski , a. , et al . 2011 , , 533 , a67 tassoul , j .- l . 2000 , stellar rotation ( cambridge : cambridge univ . press ) titarchuk , l. , & zannias , t. 1998 , , 493 , 863 watarai , k .- y . , fukue , j. , takeuchi , m. , & mineshige , s. 2000 , , 52 , 133 weng , s .- s . , & zhang , s .- n . 2011 , , 739 , 42

the understanding of confinement and , more generally , of the principles that govern the low energy dynamics in qcd is , perhaps , the most interesting unanswered question in the theory of strong interactions . the strongly coupled dynamics of the four dimensional non abelian gauge theories has so far proven to be forbiddingly complicated . a fruitful approach to cope with such situation can be to turn to simpler , more tractable models , that nevertheless capture at least some of the important features of the real theory . from this point of view , three dimensional non abelian gauge theories are of particular interest since they are , indeed , confining , and have a strong coupling dynamics . on the other hand they can be continuously deformed into the weak coupling region without encountering any phase transition . the prototypical case is the @xmath1 gauge theory with an adjoint higgs field - the georgi - glashow model . these weakly coupled theories are confining , just like their strongly coupled counterparts . here , however , confinement can be studied using semiclassical methods , as first done by polyakov for the @xmath1 case @xcite , and later on generalized to @xmath2 theories @xcite . a related feature of these models is that their spectra has a large gap between the light ( light ` photons ' ) and heavy ( heavy massive gauge bosons ) states . this makes the identification of the relevant low energy degrees of freedom relatively easy , and also allows one to construct the low energy effective lagrangian explicitly . it turns out that this effective lagrangian exhibits the confinement phenomenon in a very simple and straightforward way _ on the classical level_. in a nutshell , the natural degrees of freedom turn out to be magnetic vortex fields first introduced by thooft @xcite , and the confinement mechanism is realized as confinement of topological solitons for those vortex fields . the structure of the effective lagrangian is robustly determined by the spontaneously broken magnetic @xmath0 symmetry @xcite . this approach has been motivated by earlier work @xcite , it was advanced and formulated in @xcite , and further clarified in @xcite . it has also been argued in @xcite , that the same type of effective lagrangian remains valid in the strongly coupled limit of pure gluodynamics . in this situation , the components of the vortex field are naturally identified with the scalar and pseudoscalar glueballs , which according to lattice results are indeed the two lowest states in the spectrum of @xmath3 dimensional gluodynamics @xcite . the confinement mechanism for the strongly coupled regime remains , in this picture , essentially the same as for the weakly coupled phase . all previous work dealt exclusively with theories having no matter fields in fundamental representation . the purpose of this note is to address the question of what effect dynamical fundamental ` quarks ' have on this low energy structure . that that effect exists and must be nontrivial can be seen from one basic observation : in the presence of fundamental charges , the vortex field ( which is the effective low energy degree of freedom ) is no longer a local field . the effective lagrangian can not , therefore , be a simple local scalar theory as in @xcite . we will show below that , in the presence of fundamental matter , the magnetic @xmath0 in the effective lagrangian is _ gauged _ , the value of the gauge coupling being inversely proportional to the mass of the ( lightest ) dynamical quark . we will construct such an effective lagrangian , which moreover in the limit of the infinitely heavy quarks reduces to the effective theory of @xcite . the presence of fundamental quarks implies the existence of a conserved baryon number charge . we will see how this charge is represented in the low energy theory , and will study the structure of finite energy baryon states . quite surprisingly , we find that these states naturally have a bag - like structure . the value of the vortex field in the region of space which contains the quarks , is much lower than its value in the vacuum . the potential energy associated with this difference in the expectation values can also be naturally identified with the bag constant . the plan of this paper is the following . in section [ glue ] , we give a lightning review of the results of @xcite . we discuss the magnetic @xmath0 symmetry , its order parameter ( the vortex operator ) , and the structure of the effective lagrangian in pure gluodynamics . in section [ fund ] we explain the basic physics of the changes in this structure in the presence of the fundamental charges . in section [ low ] we show how to define the relevant low energy degrees of freedom , and how to construct the effective lagrangian in this case . in section [ baryon ] , we study the structure of the baryon states . finally we conclude in section [ discus ] with a short discussion . consider the pure su(n ) gauge theory in @xmath3 dimensions @xmath4 as was argued by thooft in @xcite , this theory possesses a global @xmath0 symmetry which is spontaneously broken in the vacuum . the explicit realization of this symmetry was found in @xcite . the generator of the @xmath0 group is ( up to a multiplicative constant factor ) the fundamental wilson loop along the spatial boundary of the system @xmath5 in @xcite , the conservation of @xmath6 was proved taking the pure gluodynamics limit of the georgi - glashow model . this , however , can also be shown directly by calculating the commutator of @xmath6 with the yang - mills hamiltonian @xmath7 : @xmath8=\lim_{c\to\infty}\oint_c dx_i { \rm tr } \left [ p e_i(x)\exp\{i\oint_{c(x , x ) } dy_ia^i(y)\ } \right ] \to_{c\to\infty}0 \;. \label{commutator}\ ] ] here , the integral in the exponential on the right hand side starts and ends at the point of insertion of the electric field . the vanishing of the commutator follows from the fact that it only involves electric fields at spatial infinity , and in a theory with a finite mass gap those fields should vanish at infinity . to dispel any doubt about the vanishing of this commutator we note that the situation is completely analogous to the commutator of any ` conserved charge ' which is defined as an integral of a local charge density @xmath9 the commutator of such a charge with a hamiltonian contains also a surface term , since the charge density @xmath10 never commutes with the hamiltonian density but rather gives a total derivative in the commutator . for a conserved charge , due to the continuity equation , this surface term is equal to the circulation of the spatial component of the current @xmath11=i\oint_c dx^ij_i\;.\ ] ] the vanishing of this term again is the consequence of the vanishing of the physical fields at infinity in a theory with a mass gap . when the charge is not conserved , the commutator contains also a bulk term in addition to the surface contribution . it is the absence of the bulk terms that is the distinctive property of a conserved charge . the same conclusion is reached if , rather than considering the generator of the algebra , one considers the commutator of the group element for either continuous or discreet symmetry groups . we see , therefore , that the commutator in ( [ commutator ] ) indeed tells us that @xmath6 is a conserved operator . next , a simple argument establishes that the magnetic @xmath0 symmetry is spontaneously broken in the vacuum if the wilson loop @xmath12 has an area law behaviour @xcite . the argument goes like this . the vev of @xmath12 is the overlap of the vacuum state @xmath13 with the state @xmath14 which is obtained from the vacuum by acting on it with @xmath12 . when acting on the vacuum state , @xmath12 performs the @xmath0 transformation at all points within the area @xmath15 bounded by the loop . if the vacuum wavefunction depends on the configuration of the @xmath0 non invariant degrees of freedom ( the state in question is not @xmath0 invariant ) the action of @xmath12 affects the state everywhere inside the loop . in the local theory with finite correlation length the overlap between the two states approximately factorizes into the product of the overlaps taken over the region of space of linear dimension of order of the correlation length @xmath16 @xmath17 where the label @xmath18 is the coordinate of the point in the center of a given small region of space . for @xmath18 outside the area @xmath15 the two states @xmath19 and @xmath20 are identical and therefore the overlap is unity . however for @xmath18 inside @xmath15 the states are different and the overlap is therefore some number @xmath21 smaller than unity . the number of such regions inside the area is obviously of order @xmath22 and thus @xmath23 the vev of @xmath12 then falls off as an area . conversely , if the vacuum is @xmath0 invariant , the wavefunction does not depend on the configuration of the non invariant degrees of freedom . the action of @xmath12 then alters the state only along the perimeter and @xmath12 has the perimeter law behaviour in the unbroken phase . thus the area law behaviour of the fundamental wilson loop is tantamount to the breaking of the magnetic @xmath0 symmetry in the vacuum of pure gluodynamics . the only requirement that this argument presupposes is the existence of * local * degrees of freedom which are non invariant under @xmath0 , or in other words the existence of a * local * order parameter . such an operator can indeed be constructed explicitly . this is the magnetic vortex creation operator @xmath24 @xcite,@xcite,@xcite . the defining property of @xmath24 is that it satisfies the following commutation relation with the spatial fundamental wilson loop @xmath25 @xmath26 where @xmath27 is the linking number of the curve @xmath28 and the point @xmath18 , or in other words , @xmath29 if @xmath30 , and @xmath31 if @xmath32 , where @xmath15 is the region of the plane bounded by @xmath28 . an explicit representation for @xmath33 is given by writing it as the operator that performs a ` singular gauge transformation ' @xmath34 here , the hypercharge generator @xmath35 is defined as @xmath36 the electric field is written in the matrix notation @xmath37 , with @xmath38 denoting the @xmath2 generator matrices in the fundamental representation , and @xmath39 is the hypercharge density due to gluons @xmath40 $ ] . the function @xmath41 is the planar angle of the vector @xmath18 . using gauss law @xmath42 this operator ( on the states that satisfy gauss law ) may be written in an alternative representation @xmath43 the integration here is along the branch cut in the definition of the planar angle @xmath44 , which is an infinite line that starts at the point @xmath18 and goes to infinity . there are two crucial properties of the definition eq.([v2 ] ) that insure that the operator @xmath24 is a local physical field : 1 . gauge invariance . the definition is not explicitly gauge invariant . nevertheless , one can show @xcite,@xcite that @xmath24 transforms a physical state into another physical state . that is to say that the matrix elements of @xmath24 between a physical and an unphysical state vanish . by physical state we mean a state which satisfies the non abelian gauss law . by virtue of this property the definition eq.([v2 ] ) can be used as long as the matrix elements of @xmath24 ( or any power of @xmath24 with insertions of an arbitrary number of gauge invariant operators ) are calculated between physical states . 2 . locality . again , the definition is not explicitly local , since it contains the integral of the electric field along an infinite line . however in a theory that does not contain fundamental charges , @xmath24 in fact does not depend on the curve @xmath28 , but only on its endpoint @xmath18 . physically , this is due to the fact that the bohm - aharonov phase between @xmath24 and any charged state in the theory vanishes . more formally , consider changing the position of the curve @xmath28 to @xmath45 . this adds to the phase in the definition eq.([v2 ] ) a contribution @xmath46 , where @xmath15 is the area bounded by @xmath47 . in our normalization , the hypercharge of gluons is @xmath48 or @xmath49 and no particles with smaller value of the hypercharge are present in the theory . therefore the hypercharge within any closed area is an integer multiple of the gauge coupling @xmath50 , and the extra phase factor is always unity . one can also show directly @xcite , using canonical commutation relations , that @xmath33 commutes with all the local gauge invariant operators @xmath51 , unless @xmath52 , which formally establishes the locality of @xmath24 . the universal properties of the low energy dynamics of the pure yang mills theory are determined by the spontaneous breaking of the magnetic @xmath0 symmetry . the effective low energy lagrangian is constructed in terms of the vortex field @xmath24 in principle we have to include in this lagrangian higher order potential terms which stabilize the potential energy at large values of @xmath24 . although we will not indicate them explicitly , it is assumed in the following that they are indeed present whenever necessary . ] @xmath53 this lagrangian can be derived in a controllable way in the weakly coupled phase@xcite,@xcite . the coupling constants in eq.([ldualgg ] ) are determined in the weakly coupled region from perturbation theory and dilute monopole gas approximation . it was argued in @xcite that the same low energy lagrangian is also valid in the strongly coupled regime ( pure gluodynamics ) . several features of this lagrangian are worth noting : * first , @xmath54 , and therefore the @xmath0 symmetry is spontaneously broken . * second , the field components of @xmath24 can be reinterpreted in terms of the physical glueball fields . using a polar decomposition for @xmath24 @xmath55 we find that the fluctuation field @xmath10 is the charge conjugation even scalar , while the phase @xmath56 is the conjugation odd pseudoscalar . indeed , according to lattice results @xcite , the two lightest glueballs in the spectrum of pure gluodynamics are the scalar @xmath57 and pseudoscalar @xmath58 states . the lagrangian ( [ ldualgg ] ) , is thus the natural effective lagrangian that one would write to describe the dynamics of the lightest particles . in this respect the identification of the glueball fields with the modulus and the phase of @xmath24 constrains the possible terms in the effective lagrangian by imposing the @xmath0 symmetry on possible interaction terms . * finally , this effective lagrangian exhibits in a simple way confining properties of the theory . the colour charged ` gluon ' states ( or massive vector bosons in the weakly coupled regime ) are represented in the low energy theory by topological solitons carrying unit winding of the field @xmath24 . if not for the @xmath59 symmetry breaking term in the potential in ( [ ldualgg ] ) , the lowest lying state in the topologically nontrivial sector would have a rotationally invariant hedgehog - like structure . its energy would be logarithmically divergent in the infrared , which corresponds to the logarithmic coulomb potential in 2 + 1 dimensions . however since the symmetry of the potential is @xmath0 rather than @xmath59 , the energy of the hedgehog configuration diverges quadratically with the volume . this is so since the number of vacuum states is discreet and in the hedgehog configuration the phase of the field @xmath24 is far from its vacuum value everywhere in space . the configuration that actually minimizes the energy in the one soliton sector breaks rotational invariance by confining the winding of the phase of @xmath24 to some quasi one - dimensional strip , like in figure 1 . it is clear that , dynamically , the width of this strip is of the order of the inverse glueball mass and the strip itself is nothing but the confining electric flux string . the energy of such an isolated soliton now diverges _ linearly _ in the infrared , due to the finite energy density per unit length of the string . [ fig1 ] in the state of unit charge in the presence of the symmetry breaking terms in the effective lagrangian , for the @xmath60 model.,title="fig : " ] the soliton we are discussing is a dynamical particle in the effective theory - it thus represents a particle with an adjoint color charge , and the string is the adjoint string . this string is of course not absolutely stable since it can break into a soliton - antisoliton pair when it is too long . the fundamental string also appears naturally in the low energy description . it is the stable domain wall which separates the regions of space with different expectation values of the vortex field @xmath24 @xcite , see figure 2 . in this section we wish to discuss the modifications introduced by the presence of fundamental quarks in the structure just described . for simplicity we will consider scalar quarks . the lagrangian of the theory we are interested in is @xmath61 where the scalar field @xmath62 transforms according to the fundamental representation of the @xmath2 color group . first thing to note is that the fundamental wilson loop still commutes with the hamiltonian . this is obvious , since the extra terms in the hamiltonian in the presence of the quarks do not involve electric field operator , but only vector potential . the wilson loop commutes with the vector potential , and thus with the additional terms in the hamiltonian . we thus conclude that the theory still has the magnetic @xmath0 symmetry . this may seem somewhat surprising at first sight . we have seen in the previous section that spontaneous breaking of the magnetic @xmath0 implies the area law for the wilson loop , and conversely the perimeter law of @xmath25 implies unbroken @xmath0 . in the theory with fundamental charges the wilson loop is known to have perimeter law due to breaking of the confining string at any finite value of the fundamental mass @xmath63 . we might then conclude that the magnetic @xmath0 is restored at any , arbitrarily large value of @xmath63 but is broken at @xmath64 . the common lore is that the @xmath0 breaking phase transition for @xmath65 is first order , and this then implies a discontinuous behaviour of the theory in the infinite mass limit . this of course is completely counterintuitive and in fact plain wrong . the caveat in this line of reasoning is the following . the relation between the behaviour of the wilson loop and the mode of the realization of the magnetic symmetry hinges crucially on the existence of a local order parameter of the magnetic @xmath0 . in the absence of such an order parameter it is not true that the wilson loop locally changes the quantum state inside the loop and this invalidates the whole argument . in particular in the absence of a local order parameter , the @xmath0 symmetry can be spontaneously broken but the wilson loop can have a perimeter law . in fact it is easy to see that for any finite @xmath63 the magnetic @xmath0 does not have a local order parameter . the only candidate for such an order parameter is the vortex operator @xmath33 defined in eq.([v2 ] ) , since it has to be local also relative to the purely gluonic operators by any explicitly local gauge invariant operator . such modifications are however irrelevant as far as the locality properties of @xmath24 are concerned , and we will not consider them in the following . ] . however in the presence of fundamental quarks the operator @xmath33 is not local anymore . to see this consider the dependence of @xmath66 on the curve @xmath28 which enters its definition . as before the operators @xmath67 and @xmath68 are related by @xmath69 where @xmath15 is the area bounded by @xmath47 . as before , due to the gauss law the integral in the exponential is equal to the total hypercharge in the area @xmath15 . however the hypercharge of fundamental quarks has eigenvalues @xmath70 . the extra phase factor is therefore not unity anymore but can rather take values @xmath71 depending on the state and the choice of the contour @xmath28 . the status of the magnetic @xmath0 is thus quite different in a theory with fundamental quarks - it does not have a local order parameter . nevertheless it is clear that at least as long as the mass @xmath63 is large @xmath72 , the relevant degrees of freedom for the effective infrared dynamics should still be the vortices @xmath24 and the main factor which determines their dynamics should still be the magnetic @xmath0 . at large @xmath63 the dependence of @xmath24 on the curve @xmath28 is weak , since the probability of the vacuum fluctuations which involve fundamental charges is small . the probability of appearance of a virtual @xmath73 pair separated by a distance @xmath16 is suppressed by the exponential factor @xmath74 . the typical distance scale for the `` glueball '' physics is @xmath75 . thus at these distances such fluctuations are unimportant and should not affect much the dynamics . things are different however if one is interested also in the baryonic sector of the theory . the baryons are necessarily heavy and in order to be able to discuss their structure we must understand the main dynamical effects also at shorter distances . therefore our aim now is to understand what is the main effect of the non locality discussed above on the dynamics of magnetic vortices . the situation we have just described - a symmetry without a local order parameter - is not exceptional in quantum field theory . this is precisely the property of the global part of any abelian gauge group . consider for example quantum electrodynamics . the global electric charge is of course a physical gauge invariant charge with the corresponding gauge invariant local charge density . @xmath76 nevertheless there is no local operator that carries this charge . this is a direct consequence of the gauss law @xmath77 any physical , gauge invariant operator that carries @xmath78 must also carry the long range electric field , which can not fall off faster than a power of the distance . the gauge invariant qed lagrangian is written in terms of `` local '' charged fields @xmath79 . but appearances are deceptive : these fields are not gauge invariant , and therefore not physical . a gauge invariant charged field can be constructed from @xmath79 by multiplying it by a phase factor @xmath80 with the @xmath81-number field @xmath82 satisfying @xmath83 for any @xmath82 satisfying this condition the operator @xmath84 is gauge invariant , and therefore physical . it is however necessarily nonlocal . different choices of @xmath82 define different gauge invariant operators and correspond to different gauge fixings . thus for @xmath85 the field @xmath84 is the field @xmath79 in the coulomb gauge , while @xmath86 corresponds to the axial gauge @xmath87 , and so on . different definitions of @xmath84 differ from each other by a phase factor , which is precisely the gauge ambiguity of the original field @xmath79 . the @xmath59 gauge group is the most natural abelian gauge symmetry to consider in continuum field theory . one can however also consider discrete groups like @xmath0@xcite . in this case again no local operator that carries the global @xmath0 charge exists . the various gauge invariant charged operators are nonlocal and differ from each other by a local @xmath0 - valued phase . these different operators again correspond to different gauge fixings of the local @xmath0 group . this is precisely the structure that emerged in our discussion in the earlier part of this section . we have \1 . global magnetic @xmath0 symmetry generated by the fundamental wilson loop . the set of nonlocal vortex operators @xmath66 , which all carry the @xmath0 charge and differ from each other by a @xmath0 valued phase factors . it is very suggestive therefore to think about @xmath66 as of different gauge fixed versions of a field charged under local @xmath0 . this leads us to expect that the low energy theory we are after should be a @xmath0 gauge theory of the magnetic vortex field @xmath24 . in fact coming back to the discussion in the beginning of this section , we see that from this vantage point it is obvious why the wilson loop has a perimeter law , even if the global @xmath0 is broken spontaneously . the action of a wilson loop of a finite size inside the contour is a gauge transformation . thus in physical terms locally inside the contour the new state is _ the same _ as the old one and so the overlap between the two locally is unity . the only nontrivial contributions to the overlap come from the region close to the contour , thus giving the perimeter law . in the next section we shall construct the effective gauge theory and discuss its relation to the original qcd lagrangian . although it is possible to give a more ` analytic ' derivation of the effective lagrangian , we prefer to write it down directly guided by the previous discussion . we will then explain the physical meaning of the various fields that appear in it . the easiest way to construct a @xmath0 gauge theory in the continuum is to consider a @xmath59 gauge theory with the higgs field of charge @xmath88 which has a large expectation value @xcite . consider therefore the following lagrangian : @xmath89 here @xmath90 . we take the parameters such that , @xmath91 and @xmath92 . the ` higgs ' field @xmath93 has a large expectation value and breaks the @xmath59 gauge symmetry down to its @xmath0 subgroup @xmath94 . it should be noted that this gauge subgroup does not correspond to ` singular ' gauge transformations , since partial derivatives commute on the gauge parameter , despite the fact that there are @xmath95-like singularities . below the scale determined by the expectation value @xmath96 , the field @xmath93 is practically frozen and its fluctuations are unimportant . in this regime the model indeed describes the locally @xmath0 invariant theory . the global part of the gauge group is our @xmath0 magnetic symmetry generated by the wilson loop . the larger @xmath59 gauge structure at this point is just an auxiliary trick which enables us to write down a discrete gauge theory in continuous notations . we will however see later that it does in fact has a real physical meaning of its own and arises naturally in the effective theory . let us first discuss how this lagrangian reduces to the effective lagrangian of the pure yang - mills theory eq.([ldualgg ] ) in the limit of the large quark mass . in this limit not only the field @xmath93 must decouple , but also the gauge interactions of the field @xmath24 must vanish . there is another consistency requirement . in the limit of zero gauge coupling @xmath97 the gauge @xmath59 symmetry becomes global @xmath59 and is broken due to non vanishing expectation value of @xmath93 . the spectrum therefore contains a massless goldstone boson . this goldstone boson is of course the longitudinal component of the gauge field @xmath98 . there is however no such massless particle in the pure gluodynamics nor in the effective lagrangian eq.([ldualgg ] ) . this means that the couplings in eq.([efflfund ] ) should depend on the quark mass in such a way that the vector particles remains heavy for any finite @xmath63 and its mass goes to zero very sharply only in the limit when it is completely decoupled . in terms of the goldstone boson couplings it means that @xmath99 must be larger than any scale relevant to the dynamics of the vortex field @xmath24 . all these conditions can be met by choosing for example @xmath100 with this choice the mass of the vector boson @xmath101 stays finite as @xmath102 and can be arbitrarily large . the goldstone boson in the decoupling limit has an infinite @xmath99 and is completely decoupled just as the `` invisible axion '' . we now want to understand how the basic fields present in the effective lagrangian arise in the fundamental theory eq.([lqcd ] ) . we start with considering the symmetries . the lagrangian eq.([efflfund ] ) has two global @xmath59 symmetries . one is the global part of the local @xmath59 with the conserved current @xmath103 . note that due to the direct coupling between the @xmath24 and @xmath93 fields the @xmath59 rotations of @xmath24 and @xmath93 separately are not symmetries . the other current , conserved by virtue of the homogeneous maxwell equation is @xmath104 . the two charges have quite different nature . the second one , the dual magnetic flux @xmath105 has a local order parameter . it can be constructed in a way similar to the vortex field in qed @xcite,@xcite . the global @xmath59 gauge charge , on the other hand does not have a local order parameter , as discussed earlier . both these conserved currents should also exist in the fundamental theory . it is fairly straightforward to identify them . the qcd lagrangian has one obvious global @xmath59 charge - the baryon number . this charge has local order parameters - gauge invariant baryon fields of qcd , and is therefore identified with the dual magnetic flux @xmath106 the second charge can be expressed in terms of the spatial current components of the first one @xmath107 \label{qt}\ ] ] it is thus the vorticity associated with the baryon number current . the vortex operator @xmath24 which appears in eq.([efflfund ] ) is not gauge invariant and is only physical after complete gauge fixing of the @xmath59 gauge group . after such a gauge fixing ( which amounts to multiplying @xmath24 by an operator valued phase ) the gauss law requires that on the physical states the physical operator @xmath24 carry the charge @xmath108 . due to the identification eqs.([f],[qt ] ) this leads to a surprising conclusion that any physical operator @xmath24 in the effective theory creates a vortex of the original baryon number current . this somewhat unexpected conclusion is in fact quite natural for an eigenoperator of the magnetic @xmath0 . consider the vortex operator @xmath24 defined by eq.([v2 ] ) . in qcd just like in pure gluodynamics , it has an alternative representation of an operator of the singular gauge transformation of the form eq.([v1 ] ) @xmath109 the only difference is that now @xmath110 is the hypercharge operator due to both , gluons and fundamental quarks @xmath111+y_{\alpha\beta}(\phi^*_\alpha\pi_\beta- \phi_\alpha\pi^*_\beta)\big]\ ] ] consider the action of this operator on a quark field @xmath112 . the transformed quark field @xmath113 is a gauge transform of @xmath112 everywhere except along the branch cut of the function @xmath114 . across this cut the phase of @xmath115 is discontinuous - it jumps by @xmath116 for all color components of @xmath115 . due to this discontinuity the baryon current - the global @xmath59 current of @xmath112 - does not vanish at points along the cut . calculating explicitly the action of @xmath24 on the baryon number current we find @xmath117 where @xmath118 is a unit vector normal to the branch cut @xmath28 at the point @xmath18 . it is natural to define the local vorticity associated with the baryon number as the vorticity defined in this way is precisely the circulation of the phase of @xmath112 . ] @xmath119 \label{vort}\ ] ] the vortex operator @xmath24 therefore creates a vortex of baryon number current with fractional vorticity @xmath116 @xmath120 the operator @xmath93 due to the gauss law also carries baryon vorticity . since its gauge coupling is @xmath88 times the coupling of @xmath24 , it creates one unit of vorticity . in fact this simple exercise also helps us to identify the value of the gauge coupling constant @xmath121 in the effective theory . comparing eq.([vort ] ) with eq.([qt ] ) we find @xmath122 the same relation is obtained by comparing the current algebra in the fundamental and the effective theories . the commutator of the baryon charge density with the baryon current density in the fundamental theory is @xmath123=i\phi^*\phi\partial_i\delta^2(x - y)\ ] ] in the effective theory using the canonical commutators that follow from eq.([efflfund ] ) and the identification eq.([f ] ) we find @xmath123=i{e^2\over 4\pi^2}\partial_i\delta^2(x - y)\ ] ] again we deduce eq.([e2 ] ) . the operator on the right hand side of this equation in the effective theory is indeed a constant . recall , that our effective theory should be valid at long distances . in this regime in the leading order in the derivative expansion the operator @xmath124 should be approximated by its expectation value . taking into account fluctuations of @xmath124 is tantamount to including higher derivative terms in the effective lagrangian eq.([efflfund ] ) . at this , higher order in derivative expansion the gauge coupling constant would become a dynamical field . while this is perfectly legitimate , it is certainly beyond our present framework . when taking the expectation value of @xmath124 we should remember that it has to be calculated with the cutoff @xmath125 appropriate for the effective theory . this cutoff must be above the characteristic scale of the pure gluodynamics ( determined by the string tension ) but below the heavy quark mass . with this in mind we get @xmath126 so that finally @xmath127 which is consistent with the expected scaling in the large mass limit eq.([largem ] ) . having understood the origin of the fields and the symmetries in the effective theory , we would like to see how it encodes the qualitative features of the low energy qcd physics . since we are considering a heavy quark theory , below the fundamental mass scale the spectrum should be the same as in pure gluodynamics . we have already seen that , in the infinite mass limit , the effective theory reduces to that of ( [ ldualgg ] ) . indeed , even at finite but large @xmath63 , this is the case at low energies . formally , this can be seen as follows . since the vev of the field @xmath93 is large , we can impose the unitary gauge condition on it . in this unitary gauge the phase of @xmath93 disappears . the modulus of @xmath93 is very heavy , and so is the vector field @xmath98 . thus at low energies we recover the effective theory of pure glue sector . there is however a set of configurations , on which the unitary gauge can not be imposed . those are configurations in which @xmath93 vanishes at some points in space . indeed it is these configurations that are important for the baryonic sector . recall that @xmath93 is a vortex of baryon number current . thus one expects that the baryon charge is associated with the vortex configuration of the field @xmath93 . in the core of the vortex the field @xmath93 of course has to vanish , and so the unitary gauge is not admissible . thus in the baryon sector we can not think of @xmath93 as frozen at its expectation value and instead have to treat it as a dynamical field . that the baryon does indeed carry vorticity of the field @xmath93 , can be seen by the following simple argument . the baryon number is represented in the effective theory by the dual magnetic flux . the baryonic state must therefore be the dual magnetic vortex . such a vortex in a nonsingular gauge has a vector potential of the form @xmath128 . to have a finite energy it must be accompanied by the winding of the phases of both @xmath24 and @xmath93 . in fact since @xmath93 carries @xmath88 times the charge of @xmath24 , the only states that are allowed energetically are those that carry @xmath88 vortices of @xmath93 . this is natural in view of the ` dual ' relation between the effective and the fundamental theories . the field @xmath93 is the ` vortex ' dual to the fundamental quark . we thus expect that its elementary vortex would represent the fundamental quark itself , and so finite energy states must contain @xmath88 such elementary vortices . a single vortex must be confined . in the same sense , @xmath24 is ` dual ' to the adjoint gluon field . the elementary vortex of @xmath24 is then , in a sense , the ` constituent ' gluon . in fact , such single gluon should not exist as a finite energy state either , and we expect it also to be confined . the single vortex of @xmath24 should therefore bind either with the anti - vortex of the same type , or with @xmath88 vortices of @xmath93 . the former type of state is a glueball , and exists in the pure glue theory @xcite , while the latter type is a baryon . interestingly enough , this line of reasoning leads us to expect that the baryon must have a bag - like structure . namely , the quarks are bound to the @xmath24 field vortex . inside this vortex , the value of @xmath24 is small - in fact it exactly vanishes in the middle , and than rises quite slowly ( relative to the scale of @xmath129 ) towards the edges . recall that @xmath130 corresponds to a non confining state @xcite . the quarks are therefore sitting in the ` perturbative ' region of space - where there are no confining forces . only when they separate far from each other - into the region with non vanishing @xmath24 , the linear potential pulls them inside again . let us look at this more carefully . consider for simplicity the @xmath60 symmetric case @xmath131 . the baryon is the dual vortex with dual magnetic flux @xmath132 . far from the vortex core , the field configuration is pure gauge , with the phases of @xmath24 and @xmath93 following the vector potential : @xmath133 the parameters of the model are such that the field @xmath24 is much lighter than both @xmath93 and @xmath134 . thus the size of the vortex core of @xmath24 is large - of the order of the inverse glueball mass . the two @xmath93-vortices which have a very small size core , sit inside this core . since the length associated with the dual magnetic field is much smaller than the core size of the @xmath24 - vortex , the dual flux is concentrated on the @xmath93 - vortices . from the low energy point of view , the picture is that two point - like magnetic vortices sit inside a soft core of a @xmath24 field vortex . the field configuration looks roughly as depicted on figure 3 . the magnetic flux is concentrated in the points @xmath135 and @xmath136 . the phase of the field @xmath93 follows the variation of the vector potential very closely . the phase of @xmath24 is also trying to do that , but it can not quite follow it all the way , since on the line between the two vortices it would have to be discontinuous . the most important energy contribution ( apart form the core energy of small vortices ) comes therefore from the vicinity of this line . the phase of @xmath24 obviously has to interpolate across this line between the values @xmath48 and @xmath137 . since the modulus of @xmath24 is not extremely rigid , it will be smaller along this line than in the immediate neighborhood . both , the variation of the modulus and the phase of @xmath24 along the line connecting the small vortices contribute to the energy which is clearly linear in the distance @xmath138 . [ fig3 ] . arrows represent the direction ( phase ) of @xmath24 , and the big dots at @xmath135 and @xmath136 correspond to the positions of the @xmath93 field vortices.,title="fig : " ] to study the structure of the baryon more quantitatively let us fix the gauge such that @xmath139\ ] ] @xmath140 this is a valid gauge choice for the configuration with two vortices of unit winding at the positions @xmath141 and @xmath142 . due to the conditions on the parameters of our model , the vector field @xmath98 follows the phase of @xmath93 in the whole space . thus @xmath143\ ] ] where : @xmath144 , @xmath145 , and @xmath16 , @xmath146 are the positions of the left and right vortices , respectively : @xmath147 , @xmath148 . there are two interesting limiting situations . the first is when the distance @xmath149 is larger than the dynamical distance scales @xmath150 and @xmath151 which determine the masses of the glueballs . the second interesting situation is the reverse , that is when the two vortices are sitting well inside the glueball correlation length . let us look at them in turn . when the distance between the vortices is large we expect the potential between them to be linear with the string tension calculated in the theory without the dynamical @xmath93 field . we will study the interquark potential in our low energy theory in the semiclassical approximation . to find the minimal energy configuration , we have to solve the classical equations of motion for the field @xmath24 at a fixed configuration of @xmath93 and @xmath134 given by eqs.([eq : uconf],[eq : bconf ] ) . let us concentrate on the points which are close to the @xmath18 - axis , with @xmath152 . the main contribution to the energy comes from this region of space . in this region the vector potential @xmath134 vanishes , and the phase of the field @xmath93 is zero . thus the equations of motion for the field @xmath24 are the same as in the pure gluodynamics . also as long as we stick to this region , the configuration of @xmath24 is translationally invariant in the @xmath18 direction . what determines the energy then are the boundary conditions on the field @xmath24 . in this configuration clearly the phase of the field @xmath24 is @xmath153 far above the @xmath18 axis and @xmath154 far below the axis , fig.3 . thus both , the equations and the boundary conditions are precisely the same as for the domain wall separating the two degenerate vacua in the effective theory of pure gluodynamics ( @xmath155 ) . there is an extra contribution to the energy that comes from the region of space close to the points @xmath135 and @xmath136 . but this energy does not depend on @xmath149 and is subleading for large @xmath149 . the rest of the space does not contribute to the energy , since the field configuration there is pure gauge . thus as expected the energy in this regime is @xmath156 , where @xmath157 is the domain wall tension ( fundamental string tension ) calculated in pure gluodynamics @xcite . it is perhaps more interesting to consider the opposite situation , that is when the distance between the quarks is smaller than the glueball correlation length . this is the regime in which we do not expect to see any stringy structure . instead we can ask whether the lowest energy configuration has any resemblance to a bag . to study this question we take the limit @xmath158 . in this case clearly the phase of the field @xmath24 will follow the phase of @xmath93 in the whole space @xmath159 given this condition , only the variation of the radial component of @xmath24 has to be determined . since the problem has rotational symmetry , the equation of motion for the modulus @xmath10 becomes @xmath160 with @xmath161 , which is assumed to be positive throughout this paper . the relevant boundary condition in this case is that at infinity @xmath10 approaches its vacuum value @xmath162 while in the vortex core it vanishes @xmath163 . with these boundary conditions , eq . ( [ eq : eqrho1 ] ) has the familiar form of the @xmath164 static kink equation in @xmath165 dimensions with the solution @xmath166 the energy of this solution is @xmath167_{a=0 } \;=\ ; \lambda \int_0^\infty \ , 2 \pi dr r \ , ( v^4 - \rho^4 ) \;=\ ; \frac{\pi^3}{12 } \ , \frac{\eta}{\lambda } \;=\ ; \frac{\pi^3}{12 } \ , ( \mu^2 - \frac{\xi u}{\lambda } ) \;.\ ] ] the picture is thus just as described in the previous subsection . since the two quark state is accompanied by the vortex of the field @xmath24 , the quarks effectively ` dig a hole ' in the vacuum . in their immediate vicinity the modulus @xmath10 vanishes , and therefore there is a ` bag ' of the nonconfining state . the radius of this bag is given by the mass of the scalar glueball @xmath168 . it is interesting to note that , although for large separation @xmath149 the energy of the string gets contributions from both , the scalar and the pseudoscalar glueballs ( the modulus and the phase of @xmath24 ) , the ` bag constant ' is determined solely by the scalar glueball . for small @xmath149 the phase of @xmath24 is not excited and only @xmath10 deviates from the vacuum state inside the ` bag ' . this is consistent with the common lore that the inside of the bag is distinguished from the outside by the value of the @xmath169 condensate . in fact since @xmath10 has vacuum quantum numbers and interpolates in our effective theory the scalar glueball , it is naturally associated with the operator @xmath169 which has a large overlap with the scalar glueball in qcd . in this note we have considered the effect of heavy fundamental quarks on the effective low energy theory description of nonabelian gauge theories in 2 + 1 dimensions . we found that the status of the magnetic @xmath0 symmetry changes . it becomes a local symmetry with the ` gauge coupling ' inversely proportional to the quark mass . thus the low energy theory becomes a discrete gauge theory . we have also studied the structure of the baryon in the framework of the effective theory . the picture that emerges is very reminiscent to the ` bag ' . the quarks in the baryon sit in the middle of the @xmath24-vortex , where the energy density differs from the vacuum energy density . the ` bag constant ' is equal to the difference of these two energy densities , and parametrically in the effective theory is given by @xmath170 , where @xmath171 is the mass of the scalar glueball and @xmath172 is the expectation value of the vortex radial field @xmath10 . an important thing to note is that the bag we are talking about here arises in a very different situation than in the usual bag model @xcite . there the bag describes the structure of the baryon containing _ light _ quarks . the radius of the bag in this situation is determined by the balance of the vacuum pressure and the pressure due to the free motion of the light quarks inside , and in fact depends on the quark wave function . in our case the inside of the bag contains heavy quarks . their kinetic energy is small , and we have treated them here as static . the radius of the bag thus is determined purely by the dynamics of the scalar glueball field and is not sensitive to the state of the heavy quarks . this is true for low lying excitations for which the radius of the quark state is smaller than the inverse glueball mass . when these two scales are comparable presumably the quark pressure will also be important and will play a role in the energy balance . thus in this intermediate regime we expect the @xmath24-vortex to be similar to the bag in the usual bag model . for states of even larger size the potential between quarks is linear with the fundamental string tension . the bag picture should therefore go smoothly into the string picture . it would be interesting to study these questions further in the framework of the effective theory discussed in this note . lastly we note that although we have so far considered scalar quarks , extending this discussion to spin @xmath173 quarks is fairly straightforward . it is well known @xcite that in the abelian higgs model the spin of the vortex is controlled by the coefficient of the chern - simons term of the vector field . thus adding the chern simons term for the vector field @xmath98 @xmath174 with @xmath175 , will endow the vortex of the @xmath93-field with a half integer spin - vortex and not the @xmath24-vortex , since in the dynamics of our effective model the vector potential @xmath98 follows the phase of the field @xmath93 . ] . our effective model with this extra term is relevant to the description of qcd with one flavour of heavy spinor quarks . just like the fermion mass term in 2 + 1 dimensions , the chern simons term breaks the parity symmetry in the baryonic sector of the theory . the extra term does not affect the classical analysis we have preformed here , but is clearly relevant for determination of the spin of the baryon as well as the spectrum of the excitations . this work was part of a collaboration supported by a fundacin antorchas - the british council grant . c. d. f. is a member of conicet , and is partially supported by anpcyt ( pict 97/1040 ) and fundacin antorchas grants . a. k. is supported by pparc . 99 a. m. polyakov , _ nucl . _ _ b120 _ , 429 ( 1977 ) . s. r. das and s. r. wadia , _ phys . * b106 * , 386 ( 1981 ) , erratum - ibid . * b108 * 435 ( 1982 ) ; n. j. snyderman _ nucl . _ * b218 * , 381 ( 1983 ) . g. thooft , _ nucl . phys . _ * b138 * , 1 ( 1978 ) . s. samuel , _ nucl . phys . _ * b154 * , 62,(1979 ) ; a. kovner , b. rosenstein and d. eliezer , _ mod . lett . _ * a5 * , 2661 ( 1990 ) ; _ nucl . phys . _ * b350 * , 325 ( 1991 ) ; a. kovner and b. rosenstein , _ phys . lett . _ * b266 * , 443 ( 1991 ) . a. kovner and b. rosenstein , _ int . j. mod . phys . _ * a7 * , 7419 ( 1992 ) . a. kovner and b. rosenstein , _ j. high energy phys . _ * 003 * 9809 ( 1998 ) . c. p. korthals - altes and a. kovner , ` magnetic z(n ) symmetry in hot qcd and the spatial wilson loop , ' hep - ph/0004052 . m. j. teper , _ phys . _ * d59 * 014512 ( 1999 ) . l. krauss and f. wilczek _ phys . lett . _ * 62 * , 1221 ( 1989 ) ; j. preskill and l. krauss _ nucl . phys . _ * b341 * 50 ( 1990 ) . a. chodos , r. l. jaffe , k. johnson and c. b. thorn , _ phys . _ * d10 * ( 1974 ) 2599 ; a. chodos , r. l. jaffe , k. johnson , c. b. thorn and v. f. weisskopf , phys . rev . * d9 * ( 1974 ) 3471 . see , for example : a. lerda , _ anyons _ , lecture notes in physics , springer - verlag , berlin , 1992 .

the minimum - cost flow ( mcf ) problem is a well - studied problem with many applications , for example , modeling transportation and communication networks @xcite . over the last half century many algorithms have been developed to solve it . the first algorithms proposed in the 1960s were all pseudo - polynomial . these include the out - of - kilter algorithm by minty @xcite and by fulkerson @xcite , the cycle canceling algorithm by klein @xcite , the network simplex ( ns ) algorithm by dantzig @xcite , and the successive shortest path ( ssp ) algorithm by jewell @xcite , iri @xcite , and busacker and gowen @xcite . in 1972 edmonds and karp @xcite proposed the capacity scaling algorithm , which was the first polynomial mcf algorithm . in the 1980s the first strongly polynomial algorithms were developed by tardos @xcite and by orlin @xcite . later , several more strongly polynomial algorithms were proposed such as the minimum - mean cycle canceling ( mmcc ) algorithm by goldberg and tarjan @xcite and the enhanced capacity scaling algorithm by orlin @xcite , which currently has the best worst - case running time . for a more complete overview of the history of mcf algorithms we refer to ahuja et al . @xcite . when we compare the performance of several mcf algorithms in theory and in practice , we see that the algorithms that have good worst - case bounds on their running time are not always the ones that perform best in practice . zadeh @xcite showed that there exist instances for which the network simplex ( ns ) algorithm has exponential running time , while the minimum - mean cycle canceling ( mmcc ) algorithm runs in strongly polynomial time , as shown by goldberg and tarjan @xcite . in practice however , the relative performance of these algorithms is completely different . kovcs @xcite showed in an experimental study that the ns algorithm is much faster than the mmcc algorithm on practical instances . in fact , the ns algorithm is even the fastest mcf algorithm of all . an explanation for the fact that the ns algorithm performs much better in practice than indicated by its worst - case running time is that the instances for which it needs exponential time are very contrived and unlikely to occur in practice . to better understand the differences between worst - case and practical performance for the ns algorithm and the mmcc algorithm , we analyze these algorithms in the framework of smoothed analysis . smoothed analysis was introduced by spielman and teng @xcite to explain why the simplex algorithm usually needs only a polynomial number of iterations in practice , while in the worst case it needs an exponential number of iterations . in the framework of smoothed analysis , an adversary can specify any instance and this instance is then slightly perturbed before it is used as input for the algorithm . this perturbation can model , for example , measurement errors or numerical imprecision . in addition , it can model noise on the input that can not be quantified exactly , but for which there is no reason to assume that it is adversarial . algorithms that have a good smoothed running time often perform well in practice . we refer to two surveys @xcite for a summary of results that have been obtained using smoothed analysis . we consider a slightly more general model of smoothed analysis , introduced by beier and vcking @xcite . in this model the adversary can not only specify the mean of the noisy parameter , but also the type of noise . we use the following smoothed input model for the mcf problem . an adversary can specify the structure of the flow network including all nodes and edges , and also the exact edge capacities and budgets of the nodes . however , the adversary can not specify the edge costs exactly . for each edge @xmath8 the adversary can specify a probability density @xmath9\rightarrow [ 0,\phi]$ ] according to which the cost of @xmath8 is drawn at random . the parameter @xmath3 determines the maximum density of the density function and can therefore be interpreted as the power of the adversary . if @xmath3 is large , the adversary can very accurately specify each edge cost and we approach worst - case analysis . if @xmath10 , the adversary has no choice but to specify the uniform density on the interval @xmath11 $ ] and we have average - case analysis . brunsch et al . @xcite were the first to show smoothed bounds on the running time of an mcf algorithm . they showed that the ssp algorithm needs @xmath12 iterations in expectation and has smoothed running time @xmath13 , since each iteration consists of finding a shortest path . they also provide a lower bound of @xmath14 for the number of iterations that the ssp algorithm needs , which is tight for @xmath15 . these bounds show that the ssp algorithm needs only a polynomial number of iterations in the smoothed setting , in contrast to the exponential number it needs in the worst case , and explains why the ssp algorithm performs quite well in practice . in order to fairly compare the ssp algorithm with other mcf algorithms in the smoothed setting , we need smoothed bounds on the running times of these other algorithms . brunsch et al . @xcite asked particularly for smoothed running time bounds for the mmcc algorithm , since the mmcc algorithm has a much better worst - case running time than the ssp algorithm , but performs worse in practice . it is also interesting to have smoothed bounds for the ns algorithm , since the ns algorithm is the fastest mcf algorithm in practice . however , until now no smoothed bounds were known for other mcf algorithms . in this paper we provide smoothed lower and upper bounds for the mmcc algorithm , and a smoothed lower bound for the ns algorithm . for the mmcc algorithm we provide an upper bound ( section [ ubmmcc ] ) for the expected number of iterations that the mmcc algorithm needs of @xmath0 . for dense graphs , this is an improvement over the @xmath16 iterations that the mmcc algorithm needs in the worst case , if we consider @xmath3 a constant ( which is reasonable if it models , for example , numerical imprecision or measurement errors ) . we also provide a lower bound ( section [ lbconstantphi ] ) on the number of iterations that the mmcc algorithm needs . for every @xmath1 , every @xmath17 , and every @xmath18 , we provide an instance with @xmath19 nodes and @xmath20 edges for which the mmcc algorithm requires @xmath4 iterations . for @xmath21 we can improve our lower bound ( section [ lbbigphi ] ) . we show that for every @xmath22 and every @xmath17 , there exists an instance with @xmath19 nodes and @xmath20 edges , and @xmath6 , for which the mmcc algorithm requires @xmath5 iterations . this is indeed a stronger lower bound than the bound for general @xmath3 , since we have @xmath23 for @xmath6 . for the ns algorithm we provide a lower bound ( section [ lbns ] ) on the number of non - degenerate iterations that it requires . in particular , we show that for every @xmath1 , every @xmath24 , and every @xmath25 there exists a flow network with @xmath19 nodes and @xmath20 edges , and an initial spanning tree structure for which the ns algorithm needs @xmath26 non - degenerate iterations with probability @xmath27 . the existence of an upper bound is our main open problem . note that our bound is the same as the lower bound that brunsch et al . @xcite found for the smoothed number of iterations of the ssp algorithm . this is no coincidence , since we use essentially the same instance ( with some minor changes ) to show our lower bound . we show that with the proper choice of the initial spanning tree structure for the ns algorithm , we can ensure that the ns algorithm performs the same flow augmentations as the ssp algorithm and therefore needs the same number of iterations ( plus some degenerate ones ) . in the rest of our introduction we introduce the mcf problem , the mmcc algorithm and the ns algorithm in more detail . in the rest of our paper , all logarithms are base @xmath28 . a _ flow network _ is a simple directed graph @xmath29 together with a nonnegative capacity function @xmath30 defined on the edges . for convenience , we assume that @xmath31 is connected and that @xmath32 does not contain a pair @xmath33 and @xmath34 of reverse edges . for the mcf problem , we also have a cost function @xmath35 $ ] on the edges and a budget function @xmath36 on the nodes . nodes with negative budget require a resource , while nodes with positive budget offer it . @xmath37 is a nonnegative function on the edges that satisfies the capacity constraints , @xmath38 ( for all @xmath39 ) , and flow conservation constraints @xmath40 ( for all @xmath41 ) . the cost @xmath42 of a flow @xmath43 is defined as the sum of the flow on each edge times the cost of that edge , that is , @xmath44 . the objective of the minimum - cost flow problem is to find a flow of minimum cost or conclude that no feasible flow exists . in our analysis we often use the concept of a _ residual network _ , which we define here . for an edge @xmath45 we denote the reverse edge @xmath34 by @xmath46 . for flow network @xmath31 and flow @xmath43 , the residual network @xmath47 is defined as the graph @xmath48 . here @xmath49 is the set of _ forward edges _ with capacity @xmath50 and cost @xmath51 . @xmath52 is the set of _ backward edges _ with capacity @xmath53 and cost @xmath54 . here @xmath55 is also called the _ residual capacity _ of edge @xmath8 for flow @xmath43 . the mmcc algorithm works as follows : * first we find a feasible flow using any maximum - flow algorithm . * next , as long as the residual network contains cycles of negative total cost , we find a cycle of minimum - mean cost and maximally augment flow along this cycle . * we stop when the residual network does not contain any cycles of negative total cost . for a more elaborate description of the mmcc algorithm , we refer to korte and vygen @xcite . in the following , we denote the mean cost of a cycle @xmath56 by @xmath57 . also , for any flow @xmath43 , we denote the mean cost of the cycle of minimum - mean cost in the residual network @xmath47 by @xmath58 . goldberg and tarjan @xcite proved in 1989 that the minimum - mean - cycle canceling algorithm runs in strongly polynomial time . five years later radzik and goldberg @xcite slightly improved this bound on the running time and showed that it is tight . in the following we will focus on the number of iterations the mmcc algorithm needs , that is , the number of cycles that have to be canceled . a bound on the number of iterations can easily be extended to a bound on the running time , by noting that a minimum - mean cycle can be found in @xmath59 time , as shown by karp @xcite . the tight bound on the number of iterations that the mmcc algorithm needs is as follows . [ strponbound ] the number of iterations needed by the mmcc algorithm is bounded by @xmath60 and this bound is tight . to prove our smoothed bounds in the next sections , we use another result by korte and vygen ( * ? ? ? * corollary 9.9 ) which states that the absolute value of the mean cost of the cycle that is canceled by the mmcc algorithm , @xmath61 , decreases by at least a factor @xmath62 every @xmath63 iterations . [ logbound ] every @xmath63 iterations of the mmcc algorithm , @xmath61 decreases by at least a factor @xmath62 . the network simplex ( ns ) algorithm starts with an initial spanning tree structure @xmath64 and associated flow @xmath43 , where each edge in @xmath32 is assigned to exactly one of @xmath65 , @xmath66 , and @xmath67 , and it holds that * @xmath68 for all edges @xmath69 , * @xmath70 for all edges @xmath71 , * @xmath38 for all edges @xmath72 , * the edges of @xmath65 form a spanning tree of @xmath31 ( if we consider the undirected version of both the edges of @xmath65 and the graph @xmath31 ) . if the mcf problem has a feasible solution , such a structure can always be found by first finding any feasible flow and then augmenting flow along cycles consisting of only edges that have a positive amount of flow less than their capacity , until no such cycles remain . note that the structure @xmath64 uniquely determines the flow @xmath43 , since the edges in @xmath65 form a tree . in addition to the spanning tree structure , the ns algorithm also keeps track of a set of node potentials @xmath73 for all nodes @xmath41 . the node potentials are defined such that the potential of a specified root node is @xmath74 and that the potential for other nodes is such that the reduced cost @xmath75 of an edge @xmath33 equals @xmath74 for all edges @xmath76 . in each iteration , the ns algorithm tries to improve the current flow by adding an edge to @xmath65 that violates its optimality condition . an edge in @xmath66 violates its optimality condition if it has strictly negative reduced cost , while an edge in @xmath67 violates its optimality condition if it has strictly positive reduced cost . one of the edges @xmath8 that violates its optimality condition is added to @xmath65 , which creates a unique cycle @xmath56 in @xmath65 . flow is maximally augmented along @xmath56 , until the flow on one of the edges @xmath77 becomes @xmath74 or reaches its capacity . the edge @xmath78 leaves @xmath65 , after which @xmath65 is again a spanning tree of @xmath31 . next we update the sets @xmath65 , @xmath66 , and @xmath67 , the flow and the node potentials . this completes the iteration . if any edges violating their optimality condition remain , another iteration is performed . one iteration of the ns algorithm is also called a pivot . the edge @xmath8 that is added to @xmath65 is called the entering edge and the edge @xmath78 that leaves @xmath65 is called the leaving edge . note that in some cases the entering edge can be the same edge as the leaving edge . also , if one of the edges in the cycle @xmath56 already contains flow equal to its capacity , the flow is not changed in that iteration , but the spanning tree @xmath65 still changes . such an iteration we call degenerate . note that in each iteration , there can be multiple edges violating their optimality condition . there are multiple possible pivot rules that determine which edge enters @xmath65 in this case . in our analysis we use the ( widely used in practice ) pivot rule that selects as the entering edge , from all edges violating their optimality condition , the edge for which the absolute value of its reduced cost @xmath79 is maximum . in case multiple edges in @xmath56 are candidates to be the leaving edge , we choose the one that is most convenient for our analysis . if a strongly feasible spanning tree structure @xcite is used , it can be shown that the number of iterations that the ns algorithm needs is finite . however , zadeh @xcite showed that there exist instances for which the ns algorithm ( with the pivot rule stated above ) needs an exponential number of iterations . orlin @xcite developed a strongly polynomial version of the ns algorithm , which uses cost - scaling . however , this algorithm is rarely used in practice and we will not consider it in the rest of our paper . for a more elaborate discussion of the ns algorithm we refer to ahuja et al . in this section we show an upper bound of @xmath0 for the expected number of iterations that the mmcc algorithm needs starting from the initial residual network @xmath80 for the feasible starting flow @xmath81 for flow network @xmath29 . note that we assumed in section [ intromcf ] that @xmath31 is simple and that @xmath32 does not contain a pair @xmath33 and @xmath34 of reverse edges . this implies that for each pair of nodes @xmath82 , there is always at most one edge from @xmath83 to @xmath84 and at most one edge from @xmath84 to @xmath83 in any residual network @xmath47 . we first show that the number of cycles that appears in at least one residual network @xmath47 for a feasible flow @xmath43 on @xmath31 , is bounded by @xmath85 , where @xmath86 . [ numcycles ] the total number of cycles that appears in any residual network @xmath47 for a feasible flow @xmath43 on @xmath31 , is bounded by @xmath85 . first we show that the number of directed cycles of length @xmath87 ( @xmath88 ) is bounded by @xmath89 . we identify each cycle @xmath56 of length @xmath87 with a path @xmath90 of length @xmath87 starting and ending at the same arbitrarily chosen node of @xmath56 and then following the edges of @xmath56 in the direction of their orientation . every such path @xmath90 can be identified with a unique cycle . the number @xmath91 of possible paths of length @xmath87 is bounded by @xmath92 the first inequality of equation follows since there are at most @xmath1 possible choices for the first node of the path , at most @xmath93 choices for the second node , etc . the lemma follows by observing that the number of possible lengths of cycles in residual networks @xmath47 is bounded by @xmath94 . we next show that the probability that any particular cycle has negative mean cost close to @xmath74 can be bounded . in the rest of this section , @xmath95 . [ probsmallcost ] the probability that an arbitrary cycle @xmath56 has mean cost @xmath96 can be bounded by @xmath97 . we can only have @xmath96 if @xmath98 , since @xmath56 consists of at most @xmath1 edges . we now draw the costs for all edges in @xmath56 except for the cost of one edge @xmath8 . the cost of cycle @xmath56 depends linearly on the cost of edge @xmath8 , with coefficient @xmath27 if @xmath8 is a forward edge in @xmath56 and coefficient @xmath99 if @xmath8 is a reverse edge in @xmath56 . therefore , the width of the interval from which the cost of @xmath8 must be drawn such that @xmath98 is @xmath100 . since the density function according to which the cost of @xmath8 is drawn has maximum density @xmath3 , the probability that the cost of @xmath8 is drawn from this interval is at most @xmath97 . [ probsmallcostallcycles ] the probability that there exists a cycle @xmath56 with @xmath96 is at most @xmath101 . the corollary follows directly from lemma [ numcycles ] and lemma [ probsmallcost ] . [ linknumitmeancyclecosts ] if none of the residual networks @xmath47 for feasible flows @xmath43 on @xmath31 contain a cycle @xmath56 with @xmath96 , then the mmcc algorithm needs at most @xmath102 iterations . assume to the contrary that none of the residual networks @xmath47 for feasible flows @xmath43 on @xmath31 contain a cycle @xmath56 with @xmath96 , but the mmcc algorithm needs more than @xmath102 iterations . let @xmath81 denote the starting flow found using any maximum - flow algorithm . since all edge costs are drawn from the interval @xmath11 $ ] , we have that @xmath103 . according to theorem [ logbound ] , after @xmath102 iterations we have that for the current flow @xmath104 holds that @xmath105 . now either @xmath106 contradicting the assumption that the mmcc algorithm needs more than @xmath102 iterations , or @xmath107 contradicting the assumption that none of the residual networks @xmath47 for feasible flows @xmath43 on @xmath31 contain a cycle @xmath56 with @xmath96 . the expected number of iterations that the mmcc algorithm needs is at most @xmath0 . let @xmath65 be the expected number of iterations that the mmcc algorithm needs . we have @xmath108 here equation follows from lemma [ linknumitmeancyclecosts ] . equation follows by bounding the probability for the first @xmath109 terms of the summation by @xmath27 and the probability for the other terms using corollary . finally , equation follows from the inequality @xmath110 , which holds for @xmath111 . in this section we describe a construction that , for every @xmath1 , every @xmath17 , and every @xmath18 , provides an instance with @xmath19 nodes and @xmath20 edges for which the mmcc algorithm requires @xmath4 iterations . for simplicity we describe the initial residual network @xmath31 , which occurs after a flow satisfying all the budgets has been found , but before the first minimum - mean cycle has been canceled . for completeness , we will explain at the end of the description of @xmath31 how to choose the initial network , budgets , and starting flow such that @xmath31 is the first residual network . we now describe how to construct @xmath31 given @xmath1 , @xmath2 , and @xmath3 . in the following , we assume @xmath112 . if @xmath3 is smaller than @xmath113 , the lower bound on the number of iterations reduces to @xmath114 and a trivial instance with @xmath19 nodes and @xmath20 edges will require @xmath114 iterations . we define @xmath115 and @xmath116 . note that this implies that @xmath117 or @xmath118 . for the edge costs we define intervals from which the edge costs are drawn uniformly at random . we define @xmath119 as follows ( see figure [ lbconstruction ] ) . * @xmath120 , where @xmath121 , @xmath122 , @xmath123 , and @xmath124 . * @xmath125 . * @xmath126 is an arbitrary subset of @xmath127 of cardinality @xmath2 . each edge @xmath128 has capacity @xmath27 and cost interval @xmath129 $ ] . * @xmath130 contains the edges @xmath131 , @xmath132 contains the edges @xmath133 , @xmath134 contains the edges @xmath135 , and @xmath136 contains the edges @xmath137 ( @xmath138 ) . all these edges have infinite capacity and cost interval @xmath129 $ ] . * @xmath139 contains the edges @xmath140 and @xmath141 ( @xmath142 . an edge @xmath140 has capacity @xmath2 and cost interval @xmath129 $ ] . an edge @xmath141 has capacity @xmath2 and cost interval @xmath143 $ ] . * @xmath144 contains the edges @xmath145 and @xmath146 ( @xmath147 . an edge @xmath145 has capacity @xmath2 and cost interval @xmath129 $ ] . an edge @xmath146 has capacity @xmath2 and cost interval @xmath148 $ ] . note that all cost intervals have width @xmath149 and therefore correspond to valid probability densities for the edge costs , since the costs are drawn uniformly at random from these intervals . the edges of the types @xmath141 and @xmath146 have a cost interval that corresponds to negative edge costs . the residual network with these negative edge costs can be obtained by having the following original instance ( before computing a flow satisfying the budget requirements ) : all nodes , edges , costs and capacities are the same as in @xmath31 , except that instead of the edges of type @xmath141 we have edges @xmath150 with capacity @xmath2 and cost interval @xmath151 $ ] and instead of the edges of type @xmath146 we have edges @xmath152 with capacity @xmath2 and cost interval @xmath153 $ ] . in addition , node @xmath154 has budget @xmath155 , node @xmath156 has budget @xmath157 , the nodes of the types @xmath158 and @xmath159 have budget @xmath160 and all other nodes have budget @xmath74 . if we now choose as the initial feasible flow the flow that sends @xmath2 units from @xmath154 to each node of type @xmath158 and from @xmath156 to each node of type @xmath159 then we obtain the initial residual network @xmath31 . we now show that the mmcc algorithm needs @xmath4 iterations for the initial residual network @xmath31 . first we make some basic observations . the minimum - mean cycle @xmath56 never contains the path @xmath161 if the path @xmath162 has positive residual capacity for some @xmath163 , since the mean cost of @xmath56 can be improved by substituting @xmath164 by @xmath165 in @xmath56 . analogously , @xmath56 never contains the path @xmath166 if the path @xmath167 has positive residual capacity for some @xmath163 . also , since all cycles considered have mean cost strictly less than @xmath149 , cycles will never include more edges with cost at least @xmath168 than necessary . in addition , since the edges of type @xmath141 and @xmath146 are saturated in the order cheapest to most expensive , none of these edges will ever be included in reverse direction in the minimum - mean cycle . the above observations lead to three candidate types for the minimum - mean cycle : cycles of type @xmath169 , of type @xmath170 , and of type @xmath171 . here @xmath83 and @xmath84 are arbitrary nodes in @xmath67 and @xmath172 , respectively . in the following series of lemmas we compare the mean costs of these cycle types . here @xmath83 and @xmath84 are again arbitrary nodes in @xmath67 and @xmath172 , possibly different for the cycles that are compared . in our computations we always assume worst - case realization of the edge costs , that is , if we want to show that a cycle @xmath173 has lower mean cost than a cycle @xmath174 , we assume that all edges in @xmath173 take the highest cost in their cost interval , while all edges in @xmath174 take the lowest cost in their cost interval ( an edge that appears in both @xmath173 and @xmath174 can even take its highest cost in @xmath173 and its lowest cost in @xmath174 in the analysis ) . [ adbc ] the cycle @xmath175 has lower mean cost than the cycle @xmath176 . since the cycles have equal length , we can compare their total costs instead of their mean costs . we have @xmath177 here the second inequality holds since @xmath178 . [ bcadiplus1 ] the cycle @xmath179 has lower mean cost than the cycle @xmath180 . since the cycles have equal length , we can compare their total costs instead of their mean costs . we have @xmath181 here the second inequality holds since @xmath182 . [ adcycle ] the cycle @xmath183 has lower mean cost than the cycle @xmath184 . we have @xmath185 here the second inequality holds since @xmath178 . [ bccycle ] the cycle @xmath179 has lower mean cost than the cycle @xmath186 . we have @xmath187 here the second inequality holds since @xmath182 . the above observations and lemmas allow us to determine the number of iterations that the mmcc algorithm needs for residual network @xmath31 . [ lowerboundg ] the mmcc algorithm needs @xmath188 iterations for residual network @xmath31 , independent of the realization of the edge costs . for the first iteration only cycles of the types @xmath169 and @xmath189 are available . according to lemma [ adcycle ] , all cycles of type @xmath169 have lower mean costs than all cycles of type @xmath189 and therefore the first iteration will augment along a cycle of type @xmath169 . after the first iteration , an edge from @xmath172 to @xmath67 will become available , and therefore a cycle of the type @xmath170 . according to lemma [ adbc ] this cycle has higher mean cost than cycles of the type @xmath169 however , and therefore the first @xmath2 iterations will be of the type @xmath169 . after the first @xmath2 iterations , the edge @xmath190 , the edge @xmath191 , and all edges in @xmath126 will be saturated . the available cycle types are now @xmath170 and @xmath192 . according to lemma [ bccycle ] , all cycles of type @xmath170 have lower mean cost than all cycles of type @xmath192 . the next iteration will therefore augment along a cycle of type @xmath170 . after this iteration , an edge from @xmath67 to @xmath172 becomes available and therefore a cycle of type @xmath193 , but according to lemma [ bcadiplus1 ] all cycles of type @xmath170 have lower mean cost than cycles of type @xmath193 and therefore in iterations @xmath194 the mmcc algorithm augments along cycles of type @xmath170 . after @xmath195 iterations , we obtain @xmath31 , except that now edges @xmath190 , @xmath191 , @xmath196 , and @xmath197 are saturated and that there is some flow on the infinite capacity edges of the types @xmath131 , @xmath133 , @xmath135 , and @xmath137 . the mmcc algorithm will keep augmenting among @xmath2 cycles of type @xmath169 followed by @xmath2 cycles of type @xmath170 until all edges of types @xmath141 and @xmath146 are saturated and no negative cost cycles remain . the total number of iterations the mmcc algorithm needs is therefore @xmath188 . the instance @xmath31 and theorem [ lowerboundg ] allow us to state a lower bound on the number of iterations that the mmcc algorithm needs in the smoothed setting . for every @xmath1 , every @xmath17 , and every @xmath18 , there exists an instance with @xmath19 nodes and @xmath20 edges for which the mmcc algorithm requires @xmath4 iterations , independent of the realization of the edge costs . follows directly from the instance @xmath31 , theorem [ lowerboundg ] and the definition of @xmath198 and @xmath199 . for which the mmcc algorithm needs @xmath200 iterations for @xmath201 , @xmath202 , and @xmath203 . next to the edges are the approximate edge costs.,scaledwidth=70.0% ] in section [ lbconstantphi ] we considered the setting where @xmath3 does not depend on @xmath1 . in this setting we showed that the mmcc algorithm needs @xmath4 iterations . we can improve the lower bound if @xmath3 is much larger than @xmath1 . in this section we consider the case where @xmath21 . in particular , we show that for every @xmath22 and every @xmath204 there exists an instance with @xmath19 nodes , @xmath20 edges , and @xmath6 for which the mmcc algorithm needs @xmath5 iterations . the initial residual network @xmath205 that we use to show our bound is very similar to the initial residual network @xmath31 that was used to show the bound in section [ lbconstantphi ] . below we describe the differences ( see figure [ lbconstructionbigphi ] for an illustration ) . we set @xmath206 . the constant of @xmath207 is large , but for the sake of readability and ease of calculations we did not try to optimize it . * the node set @xmath208 now consists of @xmath1 nodes @xmath209 and the node set @xmath91 now consists of @xmath1 nodes @xmath210 . * node @xmath154 is split into two nodes @xmath211 and @xmath212 . from node @xmath211 to @xmath212 there is a directed path consisting of @xmath1 edges , all with infinite capacity and cost interval @xmath129 $ ] . edges @xmath131 are replaced by edges @xmath213 with infinite capacity and cost interval @xmath129 $ ] . edges @xmath141 are replaced by edges @xmath214 with capacity @xmath2 and cost interval @xmath215 $ ] . * node @xmath156 is split into two nodes @xmath216 and @xmath217 . from node @xmath216 to @xmath217 there is a directed path consisting of @xmath1 edges , all with infinite capacity and cost interval @xmath129 $ ] . edges @xmath135 are replaced by edges @xmath218 with infinite capacity and cost interval @xmath129 $ ] . edges @xmath146 are replaced by edges @xmath219 with capacity @xmath2 and cost interval @xmath220 $ ] . note that this is a valid choice of cost intervals for the edges @xmath214 and @xmath219 and that they all have negative costs , since @xmath221 is the most expensive of them and we have @xmath222 as in section [ lbconstantphi ] , there are three candidate types for the minimum - mean cost cycle : cycles of type @xmath223 , cycles of type @xmath224 , and cycles of type @xmath225 . again we assume worst - case realizations of the edge costs and compare the mean costs of cycles of the different types in a series of lemmas . [ adbcbigphi ] the cycle @xmath175 has lower mean cost than the cycle @xmath176 . since the cycles have equal length , we can compare their total costs instead of their mean costs . we have @xmath226 here the second inequality holds since @xmath227 and @xmath228 for @xmath22 . [ bcadiplus1bigphi ] the cycle @xmath179 has lower mean cost than the cycle @xmath180 . since the cycles have equal length , we can compare their total costs instead of their mean costs . we have @xmath229 here the second inequality holds since @xmath230 and @xmath228 for @xmath22 . [ adcyclebigphi ] the cycle @xmath183 has lower mean cost than the cycle @xmath184 . we have @xmath231 here the second inequality holds since @xmath227 and @xmath228 for @xmath22 . [ bccyclebigphi ] the cycle @xmath179 has lower mean cost than the cycle @xmath186 . we have @xmath232 here the second inequality holds since @xmath230 and @xmath228 for @xmath22 . the above lemmas allow us to determine the number of iterations that the mmcc algorithm needs for initial residual network @xmath205 . [ lowerboundh ] the mmcc algorithm needs @xmath233 iterations for initial residual network @xmath205 , independent of the realization of the edge costs . the proof is similar to the proof of theorem [ lowerboundg ] . since cycles of type @xmath169 have lower mean cost than both cycles of type @xmath170 and of type @xmath189 according to lemma [ adbcbigphi ] and lemma [ adcyclebigphi ] , the first @xmath2 iterations will augment flow along cycles of type @xmath169 . after these @xmath2 iterations , edges @xmath190 and @xmath234 are saturated and the edges in @xmath126 have positive residual capacity only in the direction from @xmath172 to @xmath67 . since cycles of type @xmath170 have lower mean cost than both cycles of type @xmath193 and cycles of type @xmath192 according to lemma [ bcadiplus1bigphi ] and lemma [ bccyclebigphi ] , the next @xmath2 iterations will augment flow along cycles of type @xmath170 . after the first @xmath195 iterations , the residual network is the same as @xmath205 , except that edges @xmath190 , @xmath234 , @xmath196 , and @xmath235 are saturated and there is some flow on several edges of infinite capacity . the mmcc algorithm will keep augmenting along @xmath2 cycles of type @xmath169 followed by @xmath2 cycles of type @xmath170 , until all edges of type @xmath214 and type @xmath219 are saturated . at this point no negative cycles remain in the residual network and the mmcc algorithm terminates after @xmath233 iterations . initial residual network @xmath205 and theorem [ lowerboundh ] allow us to state a lower bound for the number of iterations that the mmcc algorithm needs in the smoothed setting for large @xmath3 . for every @xmath22 and every @xmath17 , there exists an instance with @xmath19 nodes and @xmath20 edges , and @xmath6 , for which the mmcc algorithm requires @xmath5 iterations , independent of the realization of the edge costs . follows directly from the instance @xmath205 and theorem [ lowerboundh ] . for which the mmcc algorithm needs @xmath5 iterations for @xmath201 , @xmath202 , and @xmath206 . next to the edges are the approximate edge costs.,scaledwidth=70.0% ] in this section we provide a lower bound of @xmath26 for the number of iterations that the network simplex ( ns ) algorithm requires in the setting of smoothed analysis . the instance of the minimum - cost flow problem that we use to show this lower bound is very similar to the instance used by brunsch et al . @xcite to show a lower bound on the number of iterations that the successive shortest path algorithm needs in the smoothed setting . the differences are that they scaled their edge costs by a factor of @xmath3 , which we do not , that we add an extra path from node @xmath236 to node @xmath237 , and that the budgets of the nodes are defined slightly differently . for completeness , we describe the construction of the ( slightly adapted ) instance by brunsch et al . below . for a more elaborate description of the instance , we refer to the original paper . in the following we assume that all paths from @xmath236 to @xmath237 have pairwise different costs , which holds with probability 1 , since the edge costs are drawn from continuous probability distributions . for given positive integers @xmath1 , @xmath24 , and @xmath25 let @xmath238 and @xmath239 . like brunsch et al . we assume that @xmath112 , which implies @xmath240 . we construct a flow network with @xmath19 nodes , @xmath20 edges , and smoothing parameter @xmath3 for which the ns algorithm needs @xmath241 iterations . the construction of the flow network @xmath31 that we use to show our lower bound consists of three steps . first we define a flow network @xmath242 . next we describe how to obtain flow network @xmath243 from flow network @xmath244 . finally , we construct @xmath31 using @xmath245 . as before , we give an interval of width at least @xmath149 for each edge from which its costs are drawn uniformly at random . [ [ construction - ofg_1 . ] ] construction of @xmath242 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for the first step , consider two sets @xmath246 and @xmath247 of @xmath1 nodes and an arbitrary set @xmath248 containing exactly @xmath249 edges . the initial flow network @xmath242 is defined as @xmath250 for @xmath251 and @xmath252 the edges @xmath8 from @xmath253 have capacity @xmath27 and costs from the interval @xmath254 $ ] . the edges @xmath255 have capacity equal to the out - degree of @xmath256 , the edges @xmath257 have capacity equal to the in - degree of @xmath258 and both have costs from the interval @xmath259 $ ] ( see figure [ g1 ] ) . with @xmath260 and @xmath261 with capacities different from 1 shown next to the edges and the cost intervals shown below each edge set . dashed edges are in the initial spanning tree for the network simplex algorithm , while solid edges are not.,scaledwidth=70.0% ] [ [ construction - ofg_i1-fromg_i . ] ] construction of @xmath243 from @xmath262 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + now we describe the second step of our construction . given a flow network @xmath263 , we define @xmath264 , where @xmath265 and @xmath266 let @xmath267 be the value of the maximum @xmath268-@xmath269 flow in the graph @xmath262 . the new edges @xmath270 have capacity @xmath271 and costs from the interval @xmath259 $ ] . the new edges @xmath272 also have capacity @xmath271 , but costs from the interval @xmath273 $ ] ( see figure [ giplus1 ] ) . with @xmath262 as sub - graph with edge costs next to the edges . dashed edges are in the initial spanning tree for the network simplex algorithm , while solid edges are not.,scaledwidth=70.0% ] [ [ construction - ofg - fromg_k . ] ] construction of @xmath31 from @xmath245 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath274 be the value of a maximum @xmath275-@xmath276 flow in @xmath245 . we will now use @xmath245 to define @xmath29 as follows ( see also figure [ gnetworksimplex ] ) . * @xmath277 , with @xmath278 , @xmath279 , @xmath280 , @xmath281 , and @xmath282 . * @xmath283 . * @xmath130 contains the edges @xmath284 , @xmath285 , with cost interval @xmath286 $ ] and infinite capacity ; @xmath287 , @xmath288 , with cost interval @xmath289 $ ] and capacity @xmath274 ; and @xmath290 with cost interval @xmath291 $ ] and infinite capacity . * @xmath132 contains the edges @xmath292 , @xmath285 , with cost interval @xmath286 $ ] and infinite capacity ; @xmath293 , @xmath288 , with cost interval @xmath289 $ ] and capacity @xmath274 ; and @xmath294 with cost interval @xmath286 $ ] and infinite capacity . * @xmath134 contains the edges @xmath295 , @xmath285 , with cost interval @xmath286 $ ] and infinite capacity ; @xmath296 , @xmath288 , with cost interval @xmath289 $ ] and capacity @xmath274 ; and @xmath297 with cost interval @xmath286 $ ] and infinite capacity . * @xmath136 contains the edges @xmath298 , @xmath285 , with cost interval @xmath286 $ ] and infinite capacity ; @xmath299 , @xmath288 , with cost interval @xmath289 $ ] and capacity @xmath274 ; and @xmath300 with cost interval @xmath291 $ ] and infinite capacity . * @xmath301 contains the edges @xmath302 , @xmath303 , with cost interval @xmath286 $ ] and infinite capacity ; @xmath304 , with cost interval @xmath286 $ ] and infinite capacity ; and @xmath305 with cost interval @xmath286 $ ] and infinite capacity . with @xmath245 as sub - graph with approximate edge costs next to the edges . dashed edges are in the initial spanning tree for the network simplex algorithm , while solid edges are not.,scaledwidth=100.0% ] the budgets of all nodes are @xmath74 , except for node @xmath236 and @xmath237 , which have budgets @xmath306 and @xmath307 . we choose as the initial spanning tree @xmath65 for the ns algorithm the edges * @xmath308 ( @xmath138 ) and @xmath309 ( @xmath138 ) , * @xmath310 ( @xmath311 ) and @xmath312 ( @xmath311 ) , * @xmath313 , @xmath284 ( @xmath314 ) , and @xmath290 , * @xmath297 and @xmath315 ( @xmath314 ) , * @xmath292 ( @xmath314 ) and @xmath294 , * @xmath316 , @xmath300 , and @xmath317 ( @xmath314 ) , * @xmath304 , @xmath318 ( @xmath319 ) , and @xmath305 . ( see figure [ g1 ] , figure [ giplus1 ] , and figure [ gnetworksimplex ] ) . in addition , @xmath320 and @xmath321 . spanning tree structure @xmath322 corresponds to the flow that sends @xmath323 units of flow on the path @xmath324 and does not send any flow on other edges . to prove our lower bound on the number of iterations that the ns algorithm needs for flow network @xmath31 , we link the iterations of the ns algorithm to the iterations of the successive shortest path ( ssp ) algorithm for flow network @xmath205 , where @xmath205 is obtained from @xmath31 by removing all nodes @xmath325 ( @xmath326 ) and all their incident edges . as shown by brunsch et al . * theorem 23 ) , the ssp algorithm needs @xmath26 iterations for @xmath205 . we show that each non - degenerate iteration of the ns algorithm on @xmath31 corresponds with an iteration of the ssp algorithm on @xmath205 and that therefore the ns algorithm needs @xmath26 iterations for flow network @xmath31 as well . in our analysis we use many results on the costs of paths in @xmath205 by brunsch et al . we will not prove these results again , but refer to the original paper . we now first observe that path @xmath324 is the most expensive path from @xmath236 to @xmath237 in @xmath31 . since paths from @xmath275 to @xmath276 have cost less than @xmath327 ( * ? ? ? * lemma 21 ) , the most expensive @xmath328 paths that do not use the nodes @xmath325 are @xmath329 and @xmath330 . both those paths have the same distribution for their costs . if we arbitrarily choose path @xmath329 and compare its cost with the cost of path @xmath324 , assuming worst case edge cost realizations , we have @xmath331 by the definition of @xmath87 and @xmath332 and @xmath112 , which shows that path @xmath324 is the most expensive path from @xmath236 to @xmath237 in @xmath31 . since @xmath324 is the most expensive @xmath328 path , we can obtain a negative cycle @xmath56 by combining the reverse edges of @xmath324 with the edges of another @xmath328 path @xmath90 . clearly , the cheaper @xmath90 is , the cheaper is @xmath56 . in addition , if all edges of @xmath56 except for one edge @xmath333 are in the current spanning tree @xmath334 , then edge @xmath333 has reduced cost equal to the cost of @xmath56 , since all edges in @xmath334 have reduced cost @xmath74 . using these two observations , we can conclude that as long as all negative cycles that can be formed using edges in the current spanning tree @xmath334 plus one edge outside of @xmath334 that has positive residual capacity in the current residual network consist of path @xmath335 plus an @xmath328 path @xmath90 , then the ns algorithm with the most negative edge pivot rule will pivot on the edge that together with the edges in @xmath334 forms the cheapest @xmath328 path . brunsch et al . @xcite showed that the ssp algorithm encounters @xmath323 paths on @xmath205 . let @xmath336 be the paths encountered on @xmath205 by the ssp algorithm , ordered from cheapest to most expensive . we refer the reader to the paper of brunsch et al . for a description of these paths . in the following we show that in the @xmath337 non - degenerate iteration of the ns algorithm on @xmath31 , flow is sent along cycle @xmath338 . we will not provide all the calculations needed to compare the reduced cost of the candidate edges for addition to the spanning tree in each iteration , but it can be checked by tedious computation that the claimed edge is indeed the one with lowest reduced cost . if flow is sent over one of the edges in @xmath253 , that edge is always a candidate leaving edge , since edges in @xmath253 have capacity 1 and all other capacities are integral . for simplicity we assume that in cases where multiple edges in the cycle become saturated simultaneously by the flow augmentation , it is always the edge from @xmath253 that leaves the spanning tree . in the first iteration of the ns algorithm on @xmath31 , all edges in @xmath253 have negative reduced cost and positive residual capacity , and the edge @xmath339 that together with the starting spanning tree @xmath65 contains path @xmath340 is added to @xmath65 , flow is augmented along cycle @xmath341 , and @xmath339 becomes saturated and leaves @xmath65 again . for the second iteration , edge @xmath339 is saturated and therefore the edge that together with @xmath65 contains @xmath342 will be added to @xmath65 . this will continue for the first @xmath2 iterations . at this point , the cheapest @xmath328 path using edges in @xmath65 plus one edge with positive residual capacity is the path that is obtained by using either edge @xmath343 or edge @xmath344 ( depending on the realization of the edge costs ) . the next two iterations will therefore be degenerate . in one of these iterations @xmath343 is added to the spanning tree , but edge @xmath345 is saturated , prevents any flow being sent along the cycle , and is therefore removed from the spanning tree . in the other iteration @xmath344 is added to the spanning tree and @xmath346 is removed . after these two iterations the edges in @xmath253 become eligible for augmenting flow in the reverse direction and the next @xmath2 iterations augment flow along the cycles @xmath347 , @xmath348 , @xmath349 . analogously to the above , every time an edge @xmath350 gets saturated , two degenerate iterations take place in which edges @xmath351 and @xmath352 are added to the spanning tree . this allows flow to be sent through @xmath262 in reverse direction , that is , from @xmath269 to @xmath268 . similarly , every time an edge @xmath353 ( that is , the reverse edge of an original edge @xmath351 ) in the residual network gets saturated , two degenerate iterations take place in which edges @xmath354 and @xmath310 are added to the spanning tree . after @xmath274 iterations , there are no paths with positive residual capacity from @xmath275 to @xmath276 . at this point another two degenerate iterations take place . in one of them edge @xmath355 is added to the spanning tree , but no flow is sent since @xmath313 has zero residual capacity and therefore @xmath313 leaves the spanning tree . in the other iteration @xmath356 is added to the spanning tree and @xmath299 leaves . now the edges in @xmath253 can be added to the spanning tree again and flow is augmented along them in reverse direction during the next @xmath2 iterations . in particular , in the next iteration flow is augmented along cycle @xmath357 . analogously to the above , every time an edge @xmath287 gets saturated , two degenerate iterations take place . in one of them @xmath296 enters the spanning tree and @xmath287 leaves . in the other @xmath293 enters the spanning tree and @xmath316 leaves . also , every time an edge @xmath293 gets saturated , two degenerate iterations take place . in one of them @xmath358 enters the spanning tree and @xmath293 leaves . in the other @xmath359 enters the spanning tree and @xmath296 leaves . finally , after @xmath323 iterations none of the edges not in the spanning tree has both negative reduced cost and positive residual capacity and therefore the ns algorithm terminates . from the above discussion we can conclude that the ns algorithm on @xmath31 needs @xmath323 non - degenerate iterations plus several degenerate ones . [ networksimplexg ] for flow network @xmath31 and initial spanning tree structure @xmath322 , the ns algorithm needs @xmath323 non - degenerate iterations with probability @xmath27 . follows immediately from the discussion above . theorem [ networksimplexg ] provides a lower bound for the number of iterations that the ns algorithm needs in the smoothed setting . for every @xmath1 , every @xmath24 , and every @xmath25 there exists a flow network with @xmath19 nodes and @xmath20 edges , and an initial spanning tree structure for which the network simplex algorithm needs @xmath26 non - degenerate iterations with probability @xmath27 . follows directly from theorem [ networksimplexg ] and the definition of @xmath332 and @xmath274 . in section [ lbns ] we showed a smoothed lower bound of @xmath26 for the number of iterations that the ns algorithm needs . this bound is the same as the smoothed lower bound that brunsch et al . @xcite showed for the ssp algorithm . for the ssp algorithm this lower bound is even tight in case @xmath360 . still , the ns algorithm is usually much faster in practice than the ssp algorithm . we believe that the reason for this difference is that the time needed per iteration is much less for the ns algorithm than for the ssp algorithm . in practical implementations , the entering edge is usually picked from a small subset ( for example of size @xmath361 ) of the edges , which removes the necessity of scanning all edges for the edge which maximally violates its optimality conditions . also , the spanning tree structure allows for fast updating of the flow and node potentials , in particular when the flow changes on only a small fraction of the edges . for the ssp algorithm an iteration consists of finding a shortest path , which takes @xmath362 time . the experimental results of kovcs @xcite seem to support this claim , since on all test instances the ssp algorithm is slower than the ns algorithm , but never more than a factor @xmath2 . to allow a better comparison of the ssp algorithm and the ns algorithm in the smoothed setting , it would be useful to have a smoothed upper bound on the running time of the ns algorithm . finding such an upper bound is our main open problem . there is a gap between our smoothed lower bound of @xmath4 ( section [ lbconstantphi ] ) for the number of iterations that the mmcc algorithm requires and our smoothed upper bound of @xmath0 . since our lower bound for the mmcc algorithm is weaker than the lower bound for the ssp algorithm , while the mmcc algorithm performs worse on practical instances than the ssp algorithm , we believe that our lower bound for the mmcc algorithm can be strengthened . our stronger lower bound of @xmath5 in case @xmath21 ( section [ lbbigphi ] ) is another indication that this is likely possible . tobias brunsch , kamiel cornelissen , bodo manthey , heiko rglin , and clemens rsner . smoothed analysis of the successive shortest path algorithm . computing research repository 1501.05493 [ cs.ds ] , arxiv , 2015 . preliminary version at soda 2013 . james b. orlin . genuinely polynomial simplex and non - simplex algorithms for the minimum cost flow problem . technical report , sloan school of management , mit , cambridge , ma , 1984 . technical report no . 1615 - 84 .

as molecular dynamics ( md ) is limited to microscopic systems and time scales , most chemical or biological reactions can not be simulated using straightforward md . one can literally wait ages before detecting a single event in a typical computer simulation . in the early 1930s , wigner and eyring made the first attempts to overcome this problem by introducing the concept of the transition state ( ts ) and the so - called ts theory ( tst ) approximation @xcite . later on , keck @xcite demonstrated how to calculate the dynamical correction , the transmission coefficient . this work has later been extended by bennett @xcite , chandler @xcite and others @xcite , resulting in a two - step approach . first the free energy as function of a reaction coordinate ( rc ) is determined . this can be done by e.g. umbrella sampling ( us ) @xcite or thermodynamic integration ( ti ) @xcite . then , the maximum of this free energy profile defines the approximate ts dividing surface and the transmission coefficient can be calculated by releasing dynamical trajectories from the top . this approach is , in principle , exact and independent of the choice of rc . however , the method becomes inefficient when the transmission coefficient is small . a proper choice of the rc can maximize the transmission coefficient and is hence crucial for the efficiency of the method . there exist different formalisms for the transmission coefficient formula which differ in the way trajectories are counted . we discuss the standard bennett - chandler ( bc ) @xcite , the history dependent bc ( bc2 ) @xcite , and the effective positive flux ( epf ) @xcite formalism . we show that the latter should always be preferred due to a lower average pathlength and a faster convergence . however , whenever a lot of correlated recrossings occur , the transmission coefficient will be very low and all these methods become inefficient . in high dimensional complex systems it can be a very difficult task to find a proper rc . moreover , whenever the dynamics is diffusive , even an optimal rc can result in a very low transmission and hence a poor efficiency . a new approach came with transition path sampling ( tps ) @xcite that is not based on the free energy barrier as starting point . tps is rather an importance sampling of dynamical trajectories . hence , it is a monte carlo ( mc ) sampling in path space rather than phase space . the tps method has been advocated as a method that does not need a rc and is akin to throwing ropes in the dark @xcite . this might be true if one wants to sample a set of reactive trajectories , but it is not for the calculation of reaction rates . in fact , the original approach to calculate reaction rates within the framework of tps required the definition of an order parameter and the calculation of the reversible work when the endpoint of the path is dragged along this parameter . for the sampling of reactive pathways , the order parameter needs only to distinguish between the two stable states . however , in the algorithmic procedure to calculate reaction rates with tps , the order parameter becomes very similar to a rc . still , it has been speculated that this approach is less sensitive to the problems related to an improper rc ( or order parameter ) . indeed , in this article we prove for the first time that this is true using the approach of transition interface sampling ( tis ) @xcite . tis increases the efficiency of the original tps rate calculation considerably by allowing the pathlength to vary and by counting only positive effective crossings . the overall reaction rate in tis is obtained from an importance sampling technique that uses a discrete set of interfaces between the stable states . hence , tis could be considered a dynamical analogue of us in path space . for diffusive systems the partial path tis ( pptis ) was invented that uses the assumption of memory loss @xcite . in this article we discuss the case of sharp barriers . here recrossings occur mainly due to the wrong choice of rc . in a follow - up article we will treat the diffusive case . up to now , it is not clear how these methods compare in efficiency and the need for benchmark systems has been put forward several times . it is not always easy to perform comparative calculations since it is not simple to know if each method is equally optimized for a specific system . therefore , in this paper we analyze a system for which the efficiency of the methods can be calculated analytically with only a few approximations . this does not only give a transparent comparison of the efficiency of the different methods , but also allows to obtain scaling laws for how this efficiency changes as function of the barrier height and width . moreover , we give some rules for how the methods can be optimized , for instance , by choosing the proper width of the us windows and the position of interfaces in tis . these rules are important for the simulation community , as they can be used as a rule of thumb in daily practice , when any method needs to be optimized . the principal component to measure the efficiency of the methods will be the _ cpu efficiency time _ @xmath0 which is the lowest computational cost needed to obtain an overall statistical error equal to one . we give a detailed analysis of how @xmath0 can be calculated for some very general cases in the appendix sections [ sgn ] and [ secect ] . it also gives the important result of how one should divide a total fixed simulation time over a set of different simulations to obtain the best overall efficiency . in sec . [ secmet1 ] , we outline the first class of methods , the reactive flux ( rf ) methods , and present their principal formulas . in sec . [ secmet2 ] , we do the same for the second class of methods , the path sampling methods . [ secsys ] introduces the 2d benchmark system where the angle @xmath1 indicates how far the chosen rc is deviated from the optimal one . [ secres ] is the main section of our paper in which we apply the different methods of sec . [ secmet1 ] and [ secmet2 ] to the analytical benchmark system of sec . [ secsys ] . finally , we discuss the important point of hysteresis for the two types of methods and show that this is less likely to occur for the path sampling methods . we summarize the results in sec . [ seccon ] . moreover , to support the readability of this paper we have added a list of symbols and abbreviations in app . [ alos ] and [ aloa ] . in all combined free energy and transmission coefficient based methods , the rate equation follows from @xmath2 where the reaction rate @xmath3 is expressed as a quasi - plateau value at a time @xmath4 of a time dependent reaction rate function @xmath5 . this function is given by the corrected flux through an hypersurface @xmath6 , that is is a collection of phase points @xmath7 , defined by the reaction coordinate @xmath8 and transition state ( ts ) value @xmath9 . the ts value @xmath9 is standardly taken as the maximum in the free energy profile along @xmath10 . both tst , bc , bc2 and epf can be expressed as @xmath11 { \rangle}}{{\langle}\theta(\lambda^*-\lambda(x_0 ) ) { \rangle } } , \label{kgen}\end{aligned}\ ] ] where @xmath12 is the dirac delta function , @xmath13 the heaviside step function , @xmath14 is the phase point at a time @xmath15 , and @xmath16 is a trajectory that includes @xmath17 . ensemble averages @xmath18 in phase and path space are defined in app . [ subsensem ] . ( [ kgen ] ) measures the flux contributed by pathways leaving the surface @xmath9 at @xmath19 under the influence of a correction functional @xmath20 $ ] . the functional @xmath21 has different forms for tst , bc , bc2 , and epf . the rate equation ( [ kgen ] ) can be rewritten as a product of two factors : the probability to be on top of the barrier times the transmission function @xmath22 : @xmath23 with @xmath24 and @xmath25 { \rangle}_{\delta(\lambda(x_0)-\lambda^*)}. \label{defr}\end{aligned}\ ] ] here , @xmath26 implies that the ensemble @xmath27 is constrained at the surface @xmath28 . substitution of eqs . ( [ defr],[defrp ] ) in eq . ( [ kpk ] ) using eq . ( [ wens ] ) gives back eq . ( [ kgen ] ) . eqs . ( [ kpk]-[defrp ] ) show the two - step procedure . the probability @xmath29 and the time dependent transmission function @xmath22 are calculated in two separate simulations . as for the rate @xmath3 , the unnormalized transmission coefficient @xmath30 follows from a plateau in this time dependent function : @xmath31 . this factor corrects for the correlated recrossings . in [ subsfe ] , we discuss the methods to compute @xmath29 or , equivalently , the free energy barrier @xmath32 . then , in [ sectrans ] we discuss the methods to determine the transmission coefficient @xmath30 . define the following @xmath33 box functions : @xmath34 \theta[\lambda_r+i\gamma+\gamma-\lambda(x ) ] , \nonumber \\ w_m(x)&=\theta[\lambda(x)+d\lambda-\lambda^ * ] \theta[\lambda^*-\lambda(x ) ] , \nonumber \\ w_0(x)&=\theta[\lambda^*-\lambda(x ) ] , \label{windows1 } \\ w_j(x)&= \theta[\lambda(x)-\lambda_r-(j-1 ) \gamma ] \theta[\lambda_r+j \gamma+\gamma - \lambda(x ) ] , \nonumber \end{aligned}\ ] ] with @xmath35 and @xmath36 . here @xmath9 is the ts , @xmath37 is a value in the reactant well and @xmath38 is a small length scale . @xmath39 and @xmath40 represent the dimensions of the us windows ; @xmath41 is the width of the window and @xmath39 is the overlap such that @xmath42 ( see fig . [ umbrel ] ) . neglecting higher orders terms in @xmath38 , we can write @xmath43 to calculate eq . ( [ freeensam ] ) , we can simply run an md simulation and count the number of times that the transition state region interval is visited . the weight function @xmath44 in the ensemble acts like an infinite wall at @xmath45 and prevents the unnecessary exploration of the product region . however , as @xmath46 is vanishingly small for high barriers , this straight - forward method will usually fail . using eq . ( [ wens ] ) and the relations @xmath47 , @xmath48 for all @xmath49 we can rewrite eq . ( [ freeensam ] ) as @xmath50 the final property is now calculated from a series of simulations in which each pair @xmath51 so that they can be determined accurately . the implementation of us using rectangular windows via mc is straightforward . the standard mc sampling is performed starting from a point inside the window . as soon as the mc procedure generates a point outside this window , this point is automatically rejected and the old point is kept . if the procedure is performed by means of md , the window boundaries simply act as infinitely hard walls . however , due to practical problems related to a discontinuous force profile , md simulations are usually performed with parabolic windows instead of rectangular ones . instead of performing several simulations using rectangular windows , one can also use a single biasing function @xmath52 : @xmath53 again , the equivalence between eq . ( [ freeensam ] ) and eq . ( [ ussb ] ) is easy to prove via eq . ( [ wens ] ) . the advantage of eq . ( [ ussb ] ) is that the error does not propagate as in the case of several windows . on the other hand , one needs to have already a good idea about the shape of the barrier to construct a good bias @xmath54 . the series of rectangular windows is a more robust way to explore the barrier region when no knowledge is available . the algorithmic procedure to sample points in the biased ensemble @xmath55 is explained in sec . [ subsensem ] . many other methods exist for the calculation of free energy surfaces . ti is of equally importance and is based on the constrained sampling at surfaces from which the free energy s derivative can be obtained at a given value of @xmath10 . integration of these derivatives results in the requested free energy profile . in addition , many variations of us have been devised , such as wang - landau sampling @xcite , meta dynamics @xcite , and flooding @xcite , where the optimal biasing potential is created on the fly . in tst , all pathways leading towards the product site @xmath56 are assumed to stay in @xmath56 for a very long time . the tst approximation uses eq . ( [ defr ] ) with @xmath57= \theta(\dot{\lambda}(x_0 ) ) . \label{chitst}\end{aligned}\ ] ] if the barrier is sharp and a proper rc is taken , tst is a very good approximation or can even be exact . the calculation of @xmath30 , in tst , requires a simple numerical or analytical calculation . for instance , suppose that @xmath10 is a simple cartesian coordinate with mass @xmath58 , then @xmath59 and @xmath60 therefore , the free energy profile is sufficient to obtain @xmath3 whenever tst is valid . the bc equation is obtained using @xmath61 = \theta[\lambda(x_t)-\lambda^ * ] \label{chibc}\end{aligned}\ ] ] in eq . ( [ defr ] ) . the evaluation of @xmath22 in eq . ( [ defr ] ) consists of releasing many trajectories that start from the top of the barrier . these trajectories only make a contribution different from zero if they end up at the product side of the barrier . it is important to note that a trajectory with @xmath62 that leaves the surface heading to the reactant state @xmath63 but finally ends up at the product state @xmath56 at a time @xmath15 , gives a negative contribution . the final value , which has to be positive per definition , results from a cancellation of positive and negative terms . the bc2 equation uses @xmath64 & = \theta[\lambda(x_t)-\lambda^ * ] \theta[\lambda^*-\lambda(x_{-t } ) ] . \label{chibc2}\end{aligned}\ ] ] here , besides ending in the product state , the trajectories integrated backward in time also have to end in the reactant state to give a nonzero contribution to eq . ( [ defr ] ) . however , here as well , the contribution of some trajectories are negative . this happens when the systems starts with a negative @xmath65 , but the forward and backward trajectories end up in the product @xmath56 and reactant state @xmath63 , respectively . these so - called s - curves ( trajectories that cross the ts surface more than 2 times within a short time ) are less likely to occur for sharp barriers . the epf equation arises when we take @xmath20=\chi^{\rm epf}[x , t]$ ] in eq . ( [ defr ] ) with @xmath66 & = \begin{cases } 0 , & \textrm{if } \dot{\lambda}(x_0 ) < 0,\\ & \textrm{or if } \lambda_{t ' } > \lambda^ * \textrm { for any } t ' \in [ -t:0],\\ & \textrm{or if } \lambda(x_t ) < \lambda^ * , \\ 1 , & \textrm{otherwise . } \end{cases } \label{chiepf}\end{aligned}\ ] ] despite the apparent mathematical more complicated form of eq . ( [ chiepf ] ) compared to eqs . ( [ chibc ] ) and ( [ chibc2 ] ) , the relation is quite is natural . the approach counts only the first crossings and only when they are in the positive direction ( see fig . [ figepf ] ) . in the effective flux formalism ( [ chiepf ] ) , all contributions are either zero or positive . in ref . [ ] , we presented a similar transmission coefficient formula . in this approach the pathway was stopped whenever the stable state regions @xmath63 and @xmath56 were reached . for a formal mathematical proof of the equivalence between bc- and epf - type equations see ref . [ ] . some other relations for the transmission different from eq . ( [ chibc]-[chiepf ] ) have been proposed . for instance , hummer @xcite proposed a relation that uses all the crossing velocities , in case of correlated recrossings , instead of just @xmath65 . ruiz - montero _ et al . _ @xcite devised a transition zone method for diffusive barrier crossings . this will be treated in more detail in the follow up article that will discuss the diffusive barrier case . suppose @xmath67 and @xmath68 are such that any @xmath7 with @xmath69 is at the reactant side @xmath63 and any @xmath7 with @xmath70 is at the product side @xmath56 of the unknown optimal ts dividing surface . it is important to note that @xmath8 does not have to be a proper rc to fulfill this criterion , but only has to distinguish between the two stable states . the tis expression is now given by @xmath71 here @xmath72 gives the flux through interface @xmath67 and @xmath73 is a history dependent function describing whether the system was more recently in @xmath63 or in @xmath56 : @xmath74 , for a phase point @xmath7 and its corresponding trajectory , if the system was more recently in @xmath63 than in @xmath56 and @xmath75 otherwise . in practice , the whole factor @xmath76 is calculated by counting the number of crossings in a straight - forward md simulation , divided by the number of cycles with @xmath74 , divided by the time step @xmath77 . the calculation of this factor is very cheap as interface @xmath67 is very close to the basin of attraction of state @xmath63 . on the other hand , the crossing probability @xmath78 is a very small number and can not be accurately determined by straightforward md . this is the probability that whenever the system crosses interface @xmath67 , it will cross interface @xmath79 before crossing interface @xmath67 again . as interface @xmath79 is an interface at the other side of the barrier , this probability is very small . the tis method overcomes this by defining @xmath80 interfaces between @xmath81 and @xmath82 . then , the total crossing probability can be expressed in a formula that contains conditional short - distance crossing probabilities @xcite @xmath83 here , @xmath84 is a generalization of the previously described overall crossing probability and denotes the conditional probability that , whenever the system leaves state @xmath63 by crossing @xmath67 and crosses @xmath85 in turn , it will also cross @xmath86 before returning to @xmath63 . if the distances @xmath87 are sufficiently small , the probabilities @xmath84 will be large enough so that they can be computed using a shooting algorithm @xcite . the shooting algorithm takes a random time slice from the old existing path and makes a slight randomized modification of this phase point ( usually only the momenta are changed ) . then , this new phase point is used to propagate forward and backward in time yielding a new trajectory . in tis , this propagation of the trajectory is stopped whenever the system enters @xmath63 or @xmath56 or , equivalently , whenever @xmath88 or @xmath89 are crossed . the pathway is accepted only if the backward trajectory ends in @xmath63 _ and _ the total trajectory has at least one crossing with @xmath85 . the fraction of these paths that cross @xmath86 as well , determines @xmath84 . although the form of eq . ( [ tiseq ] ) deceptively resembles a markovian factorization , no assumption has been made . as @xmath84 are not simple hopping probabilities , but incorporate the full history - dependence from @xmath67 to @xmath85 , the relation is actually non - markovian and exact @xcite . in analogy with us , instead of using a discrete set of interfaces , we could also bias the trajectory in a continuous way using a bias on @xmath90 @xcite . first we can write @xmath91-\lambda_b ) { \rangle}. \label{plm}\end{aligned}\ ] ] here @xmath92 where the path @xmath16 is terminated when it leaves @xmath63 and enters region @xmath63 or @xmath56 . then , as in eq . ( [ ussb ] ) we can write @xmath93-\lambda_b ) \omega^{-1}(\lambda_{\max } ) { \rangle}_{\omega } } { { \langle}\omega^{-1}(\lambda_{\max } ) { \rangle}_{\omega } } .\end{aligned}\ ] ] the algorithmic procedure would be the same as tis without the interface crossing constraint . instead , the acceptance - rejection criterion is adjusted as explained in sec . [ subsensem ] . a continuous bias has been applied within the original tps scheme . in ref . [ ] a bias on the end point of the path was used . alternatively , one can also bias the direction of the momenta change at the shooting point as was done in ref . the original tps rate calculation algorithm was introduced in refs . it first creates ensemble of reactive trajectories of a fixed length . these trajectories should constitute a time interval that is longer than @xmath4 where @xmath5 reach its plateau . then , a second series of simulations is required that combines the mc of pathways with a us technique . also these simulations use a fixed pathlength , but these can be shorter using a rescaling approach @xcite . in the end , the final rate constant can be constructed from the results of these two types of simulations . moroni showed that this algorithm is always less efficient than the tis technique and that the computational advancement of tis is at least a factor two @xcite . the tis improvement is partly due to the flexible pathlength so that each individual path can be limited to its strictly necessary minimum . moreover , tis has abandoned the shifting moves and has a stronger convergence ( no cancellation positive and negative terms ) . depending on the system and the required accuracy , the tis approach can easily become more than an order of magnitude faster than the original approach . the pptis method reduces the average pathlength even further as compared to tis using the assumption of memory - loss @xcite . in this method , the trajectories do not have to start at @xmath67 . instead , an ensemble of trajectories is generated that start and end at either @xmath86 or @xmath94 , but must have at least one crossing with the middle interface @xmath85 . from this , four crossing probabilities can be constructed that still inhibit some history dependence . once these are known for each @xmath49 , the final overall crossing probability can be constructed via a recursive relation @xcite . the milestoning method @xcite is very similar to pptis , but relies on a stronger markovian assumption that the system remains in an equilibrium distribution at each interface . on the other hand , milestoning uses time - dependent hopping probabilities which supplies a very general way to coarse - grain a dynamical system . the two aspects of pptis and milestoning could be combined as was suggested in ref . [ ] . finally , we mention forward flux sampling ( ffs ) @xcite . ffs uses the same rate equation ( [ tiseq ] ) as tis , but the sampling is different . in ffs @xcite , the crossing points at the next interface of the successful pathways are stored . the next simulation uses this set of points to initiate new pathways . the advantage is that ffs does not require any knowledge on the distribution @xmath95 . this allows to treat non - equilibrium systems as well . moreover , ffs creates effectively more pathways than tis with the same amount of md steps and does not rely on a markovian assumption as in pptis . however , the correlation between the different pathways is much stronger than in tis or pptis . therefore , ffs can only be applied when the process is sufficiently stochastic . we consider the following 2d potential : @xmath96 with @xmath97 and @xmath98 here , @xmath99 is the height of our barrier , @xmath100 is the barrier width and @xmath101 are the dimensions of the reactant region ( see fig . [ f2d ] ) . ) . @xmath99 and @xmath100 are the barrier height and width . @xmath102 and @xmath103 are the dimensions of the reactant ( and product ) well . @xmath104 is the assumed rc , while @xmath105 is another important degree of freedom . @xmath106 and @xmath107 are the maxima of the free energy functions for these coordinates . @xmath1 is the angle between the chosen reaction coordinate @xmath104 and the optimal one @xmath108 of eq . ( [ lamortho ] ) . [ f2d],width=302 ] the chosen rc is @xmath104 , while @xmath105 represents another important degrees of freedom . @xmath108 is the unknown optimal rc as its direction is orthogonal to the barrier ridge , the exact ts dividing surface . the angle @xmath1 is , hence , a measure of the quality of the chosen rc . to facilitate the analytics we assume a simplified dynamics : the particles propagate without dissipation and move only along the @xmath104 direction . once they collide with the walls , they obtain a new random @xmath105 coordinate @xmath109 $ ] and velocities @xmath110 taken from a maxwellian distribution . this type of dynamics satisfies ergodicity and ensures that the reaction rate is well defined for high barriers . although the low dimensionality and this dynamics might seem artificial , this minimalistic model already illustrates clearly the problems that occur in complex systems when no adequate rc can be found as we will see in the forth - coming sections . moreover , the potential can be viewed as a projection of a high - dimensional complex system . in that case , eq . ( [ v2d ] ) represents a free energy in which @xmath104 and ( the unknown but important coordinate ) @xmath105 can be any non - linear function of all cartesian coordinates . for instance , @xmath104 could be the simple distance between two atoms to describe the formation or breaking of a bond , while @xmath105 represents a complex solvent rearrangement . the surface potential energy at the barrier @xmath111 equals @xmath112 and , hence , @xmath113 here , we assumed that @xmath114 . the transmission coefficient can be obtained using eq . ( [ defr ] ) and @xmath115 of eqs . ( [ chibc2],[chiepf ] ) . the two equations are identical since there are no trajectories that can cross the surface @xmath116 more than two times before colliding with the outer walls . eqs . ( [ chibc2],[chiepf ] ) have only non - zero contributions whenever the backward and forward trajectory end , respectively , at the left and right side of the barrier after a short time @xmath4 , which is deterministically determined by @xmath117 . hence , @xmath118=\chi(\dot{\lambda}_x , \lambda_y)|_{t=0}= \theta[\dot{\lambda}_x-\sqrt{2(h - v(\lambda_x^*,\lambda_y))/m}]$ ] , which gives @xmath119 with @xmath120 hence , @xmath121 which yields by eq . ( [ kpk ] ) @xmath122 the reaction rate is , thus , independent of the angle @xmath1 . to quantify the efficiency of the different methods , we will calculate the cpu efficiency time , @xmath0 , that is defined as the minimal computational cost to obtain an overall relative error equal to one . efficiency _ is sometimes defined as the inverse @xmath123 of this quantity @xcite . in app . [ secect ] , we show how this quantity can be computed for some very generic cases . as the force calculations are the predominant steps in any ab initio or large scale classical simulations , all cpu values will be expressed as an integer representing the number of required md steps . in this way , we obtain a measure that is independent of the computational resources . in the following , we make two assumptions : * the correlation number is assumed to be the same for each simulation @xmath49 in the simulation series : @xmath124 * the mean cycle length @xmath125 of the path - simulations and transmission coefficient calculations are proportional to the average pathlength @xmath126 of the accepted paths : @xmath127 where @xmath128 are given in app . [ sgn ] and [ secect ] and @xmath129 is the average number of md steps of the accepted trajectories . hence , we neglect the fact that @xmath130 and @xmath131 can differ for each simulation @xmath49 in a simulation series . in fact , the equations that we will derive for tis are true even if a softer assumption is valid , i.e. that @xmath132 is constant for each @xmath49 . in general , @xmath133 is smaller than 1 as rejected pathways are usually shorter than the accepted ones . some rejections are even immediate @xcite and do not require any force calculations . the transmission coefficient calculation has @xmath134 as each point on the ts , obtained from the first free energy calculation , with randomized maxwellian distributed velocities is automatically accepted . still , the pathlengths @xmath129 can differ . as explained in sec . [ secmet1 ] , the rf methods consist of two independent types of calculations : the free energy and the transmission coefficient calculation . moreover , there exist several techniques to determine these two quantities . as the tst approximation ( [ rtst ] ) is exact for the @xmath135 case ( which basically reduces the problem to one - dimension ) , the only computational cost follows from the free energy calculation . contrary , when @xmath1 is significantly larger than zero we can expect that the transmission coefficient calculation is the dominant computational factor even though the free energy calculation becomes problematic as well ( see sec . [ susechys ] ) . focusing on the most dominant contributions we will therefore assume @xmath135 for the free energy and @xmath136 for the transmission coefficient calculations . first , to compare the enhanced efficiency of us techniques we need to know the cpu efficiency time @xmath0 of straightforward md . examination of eq . ( [ freeensam ] ) reveals that it simply corresponds to the calculation of the ensemble average of a binary function as in eq . ( [ taubin ] ) with @xmath137 . henceforth , using our assumptions ( [ assum],[asstaucyc ] ) we have @xmath138 . it is clear that the exponentially dependence on @xmath139 make straightforward md prohibitive for any high barrier or low temperature system . to obtain the overall efficiency for us using rectangular windows as expressed in eq . ( [ usrw ] ) , we first need the efficiency time @xmath140 for a single window . the principal result of this simulation @xmath49 equals @xmath141 . the general approach to calculate the efficiency time for a composite of two averages that are obtained simultaneously within the same simulation is explained in sec . [ subscq ] . in the app . [ uscpu ] , eqs . ( [ covij]-[partcov ] ) , we derive that the efficiency time for this single window is given by @xmath142 where @xmath143 . here , @xmath144 is a system - specific parameter , @xmath145 is an intrinsic of the simulation , and @xmath39 and @xmath40 have to be optimized . if we assume that @xmath146 is very large , we can neglect the difference of the first @xmath44 and last @xmath147 windows . following eq . ( [ effconst ] ) , the overall efficiency time is given by @xmath148 and as @xmath149 is not dependent on @xmath39 , we can minimize @xmath140 as function of @xmath39 to obtain also the lowest overall efficiency time . this optimum is achieved for half infinite windows @xmath150 where @xmath151 for all @xmath49 . as a result , the overall efficiency time equals @xmath152 eq . ( [ tefftot ] ) has a minimum for @xmath153 which gives @xmath154 the efficiency time is quadratically dependent on the barrier height @xmath99 and the inverse temperature @xmath155 . compared to straightforward md this is an enormous improvement . the optimal window boundaries imply that the fraction of phase points that is sampled in the right part of the window is given by @xmath156 . hence , @xmath40 should be adjusted to have approximately one fifth of the sampled points in the most right up - hill part of the window . in appendix [ uscpu ] , eqs . ( [ sigmacovw],[teffbcw ] ) , we derive the efficiency time for an ensemble average @xmath157 when it is obtained via a different ensemble using a weight function @xmath54 : @xmath158 . then , @xmath0 is given by : @xmath159 \nonumber \\ & + \big({\langle}\omega^{-1}{\rangle}{\langle}\omega { \rangle}-1 \big ) \big [ 1 + 2 n_c^{cc } \big ] \nonumber \\ & - 2 \big ( \frac{{\langle}\hat{a } \omega^{-1}{\rangle}{\langle}\omega { \rangle}}{a } -1 \big ) \big [ 1 + n_c^{bc } + n_c^{cb } \big ] \big\ } \tau_{\rm cyc } \label{teffbias}\end{aligned}\ ] ] with @xmath160 and @xmath161 . note that the correlation functions @xmath162 ( and @xmath163 for path sampling ) can depend in principle on @xmath54 as well . as eq . ( [ teffbias ] ) does not change whenever @xmath54 is multiplied by a single factor , we apply the normalization @xmath164 . assuming that eqs . ( [ assum],[asstaucyc ] ) are also valid in the biased ensemble , we write @xmath165 if we assume that @xmath145 has only a weak dependency on @xmath54 , we can minimize eq . ( [ teffbias2 ] ) by taking the functional derivative @xmath166 which gives ( see eqs . ( [ taueffbiasl]-[w2 ] ) ) @xmath167 as optimal weight function . coming back to eq . ( [ ussb ] ) using @xmath168 , the optimal bias follows directly from eq . ( [ optw ] ) and is given by @xmath169 where we used eq . ( [ freeensam ] ) and @xmath170 . substitution of eq . ( [ bias ] ) and @xmath171 in eq . ( [ teffbias2 ] ) and using that @xmath172 ^ 2 $ ] and @xmath173 gives @xmath174 hence , a scaling behavior independent of the barrier height or system size can be achieved . we have assumed here that @xmath145 is independent of the biasing function @xmath54 which is only true for certain types of mc such as those in which each cycle can be generated really independently ( hence @xmath175 ) . in general , mc sampling is a diffusive type of motion and the bias ( [ optw],[bias ] ) should also aid the exploration from the top to the reactant well and back . therefore optimal bias function should result in a more or less uniform sampling distribution , which is achieved when @xmath176 . taking this into account , the efficiency time is a bit larger @xmath177 , but still independent of @xmath139 . for the reasons explained in sec . [ eff1d ] , here we will study the case @xmath178 and , in particular , we assume that @xmath179 . substituting @xmath180=\chi[x]$ ] in eq . ( [ taueff ] ) and using eqs . ( [ std],[defr],[r2d],[assum],[asstaucyc ] ) , we can write a general formula for the cpu efficiency time @xmath0 for the rf method : @xmath181 where we used @xmath182 . we remind you that @xmath134 for the transmission algorithm as all generated phase points on the surface @xmath106 are accepted and used to generate trajectories . in [ bc2d ] and [ bc22d ] , we will use formula ( [ teffr2d ] ) to calculate the efficiency of the bc , bc2 and epf method . [ [ bc2d ] ] bc ^^ in the bc algorithm the pathways are propagated only forward in time . say @xmath183 is the longest time that the system takes to leave the barrier when released somewhere at the surface @xmath106 . hence , to have a guaranteed plateau in the @xmath184 and @xmath5 functions we simply have to integrate @xmath185 timesteps so that @xmath186 . the second unknown factor in eq . ( [ teffr2d ] ) is @xmath187 . for the simplified dynamics we are considering eq . ( [ chibc ] ) can be reduced to @xmath188 with @xmath189 and @xmath190 . following the same lines as in eqs . ( [ r2d0]-[r2d ] ) gives @xmath191 with @xmath192}{\int { { \mathrm d}}v e^{-\beta \frac{1}{2 } m v^2}}. \label{fbc}\end{aligned}\ ] ] in eqs . ( [ v2chibc],[fbc ] ) we made use of the mirror symmetry along @xmath107 . as the gaussian integral ( [ fbc ] ) has a symmetry as well along the @xmath193 axis , we can rewrite eq . ( [ fbc ] ) as follows @xmath194 substitution of eq . ( [ fbc2 ] ) in eq . ( [ v2chibc ] ) reduces eq . ( [ teffr2d ] ) to @xmath195 for large barriers , the exponential dependence on @xmath196 makes the method already prohibitive for relatively small angles @xmath1 . in bc2 the trajectories have to be followed until both forward and backward trajectories are no longer on the barrier . in epf , the generated point on the surface heading toward reactant state are accepted , but assigned zero without further integration . points on the surface with a positive velocity are first integrated backward and , then , integrated forward in time . this gives the advantage that whenever the backward trajectory recrosses the surface @xmath106 within a short period , this trajectory s contribution is assigned zero as well and the forward trajectory can be omitted . hence , we have @xmath197 and @xmath198 moreover , as s - curves are absent in this system the two path - functionals are identical : @xmath199=\chi^{\rm epf}[x]$ ] for all @xmath16 . a non - zero contribution of @xmath21 occurs only when both the backward trajectory ends in the reactant state and the forward trajectory ends in the product state . this implies that the velocity @xmath200 should be positive with kinetic energy higher than @xmath201 . hence , @xmath202 therefore , we can write the same equations as ( [ v2chibc],[fbc ] ) with @xmath203 instead of @xmath204 in eq . ( [ fbc ] ) , which gives @xmath205\end{aligned}\ ] ] we can neglect the @xmath206-term by invoking eq . ( [ erfc ] ) and omitting the terms of order @xmath207^{-1/2})$ ] . substitution of this in eq . ( [ v2chibc ] ) gives @xmath208 and , hence , eq . ( [ teffr2d ] ) yields @xmath209 although , the efficiency of eq . ( [ tbc2epf ] ) has been quadratically improved compared to eq . ( [ tbc2d ] ) , the exponential dependence of @xmath210 makes this method prohibitive as well when @xmath1 is significantly different from zero . as for the rf methods , the tis methods consist also of two types of simulations : the initial flux through @xmath88 and the crossing probability . however , contrary to the rf methods , in tis we might expect that the crossing probability is always the computational bottleneck . as @xmath88 is placed in the low potential energy region at the foot of the barrier , this flux is easy to compute for all values of @xmath1 . we will henceforth concentrate on the crossing probability for the two cases @xmath135 and @xmath211 . say @xmath212 , @xmath213 , and @xmath214 . for the pathlength we write @xmath215 where the constants @xmath216 and @xmath217 will be determined later on . as @xmath218 is basically an average of a binary function , that is 1 for the successful trajectories and 0 otherwise , we can invoke eq . ( [ taubin ] ) and use eqs . ( [ assum],[asstaucyc ] ) : @xmath219 @xmath84 is the flux through @xmath85 that reaches @xmath86 divided by the total flux through @xmath85 @xcite for trajectories that come from @xmath88 . hence , @xmath220 where @xmath221 equals 1 only if the backward ( forward ) trajectory crosses @xmath222 before @xmath223 @xcite . otherwise it is 0 . for the case @xmath224 , we can write @xmath225 and @xmath226 with @xmath227 the difference in potential energy of the two surfaces . hence , @xmath228 we take an equidistant interface separation such that @xmath229 and @xmath230 . moreover , @xmath231^m={\mathcal p}_a(\lambda_{m}|\lambda_0)=e^{- \beta h}$ ] or , equivalently , @xmath232 . this allows to rewrite eq . ( [ tefftisas ] ) as @xmath233 since we know the efficiency times for each simulation @xmath49 , we can calculate the total efficiency time of the whole simulation series . it is , however , important to note that @xmath140 is not a constant as in eq . ( [ taueffas],[tinf ] ) , but depends on @xmath49 . this raises an additional question of how one should divide a given total simulation time among the different simulations . a logical choice would be to use the same simulation time for each @xmath49 or to adjust the simulation times to obtain the same relative error in each part . we denote the total efficiency times using these two strategies @xmath234 and @xmath235 . surprisingly , for this case the two approaches are equally efficient and given by @xmath236 where we used eqs . ( [ effprod12],[effprod ] ) and @xmath237 . as we show in app . [ subsser ] , these two approaches do not yield the optimum efficiency , this would be attained when we assign each part @xmath49 a simulation time proportional to @xmath238 which yields @xmath239 minimizing eqs . ( [ efftistot ] ) with respect to @xmath218 shows that @xmath0 reaches a minimum for @xmath240 . hence , the tis procedure is optimized when approximately one out of five trajectories are successful . see the correspondence with us sampling [ usrec ] where one fifth was also the optimum for fraction of sampling point in the right uphill part of the window . using @xmath241 in eq . ( [ efftistot ] ) gives @xmath242 and @xmath243 . in practice , we have found a linear dependence ( @xmath244 ) of the pathlength on a steep barrier@xcite and quadratically dependence ( @xmath245 ) on a diffusive barrier @xcite . hence , @xmath0 is about 12 % and 33 % lower than @xmath234 and @xmath235 . note that eq . ( [ efftis ] ) has the same quadratically dependence on @xmath139 as us ( [ tauusrw ] ) . one should realize that normally @xmath246 and that @xmath247 usually has a strong @xmath100 dependence except for exceptional cases where mc cycles can be generated really independently . hence , the efficiency scaling of tis is comparable with that of us sampling using rectangular windows . as us has more flexibility to reduce @xmath145 than tis to reduce @xmath129 , us is probably a bit more efficient than tis by a single prefactor . in this specific system , the @xmath99 dependence of the tis efficiency is even a bit favorable than @xmath244 . in the app . [ app_efftis ] , eqs . ( [ tispath]-[erfc ] ) we derive that @xmath248 yielding @xmath249 due to a decreasing average pathlength , tis seems to have a slightly better scaling as function of @xmath99 than us using rectangular windows ( [ tauusrw ] ) . assuming a lower prefactor for us , this will imply that the tis efficiency approaches the us efficiency eq . ( [ tauusrw ] ) at increasing barrier heights . however , eqs . ( [ ggtis],[tefftisgg ] ) break down for very high barriers as the average tis pathlength @xmath129 can , of course , in practice never decrease below one md step . we can exactly follow the same lines as eqs . ( [ teffbias],[effussw ] ) which gives for the ideal biasing function [ biasmax ] @xmath250 and the overall efficiency @xmath251 hence , the ideal bias function ( [ tisbias ] ) allows to obtain a scaling independent of @xmath139 as in eq . ( [ effussw ] ) . note once more that generally @xmath252 . as for us , if we take into account the diffusive behavior of the sampling , @xmath145 is likely very large when the bias ( [ tisbias ] ) is used . this bias favors only trajectories that reach state @xmath56 , but does not aid the system to climb the barrier in successive cycles . therefore , a bias that generates a more uniform distribution is more advantageous than eq . ( [ tisbias ] ) . however , this does not affect the scaling dependency on @xmath139 . the estimation of the optimal interface separation on a straight slope is a bit difficult for path sampling methods like pptis , milestoning , and ffs ( note that the pptis memory - loss assumption is satisfied on the strictly increasing barrier even if the system is not diffusive @xcite ) . an efficiency analysis of ffs @xcite using the approximation ( [ assum_unc ] ) revealed that @xmath0 is constantly decreasing as function of the number of interfaces . the same result would be valid for pptis . however , as correctly stated in ref . [ ] , the apparent conclusion , that the computational cost can be made vanishingly small using an infinite set of infinitesimal spaced interfaces , is purely artifical . if one takes into account that there is a minimum path length ( of at least 1 md step ) , also the pptis and ffs show a quadratic dependence on @xmath139 . considering the lower path length , the efficiency is likely to be very close to us using rectangular windows . we take @xmath253 and @xmath254 ( see fig . [ top2d ] ) . , and @xmath89 . , width=302 ] with these outer - interface @xmath255 ensures that the system is at the flat left side of the barrier and @xmath256 ensures that the system has crossed the barrier ridge . as in eq . ( [ r2d ] ) we can write @xmath257 with @xmath258 the crossing probability along the line @xmath105 . now , the optimization of the full tis algorithm would yield an extended and tedious calculation . therefore , we derive an upper bound of the cpu efficiency time instead which is relatively easy . as in eq . ( [ patis ] ) , we can write for @xmath258 : @xmath259}].\end{aligned}\ ] ] the additional @xmath260-term compared to eq . ( [ patis ] ) is because the potential energy can decrease from @xmath85 to @xmath86 along some coordinate @xmath105 . using equidistant interfaces with spacing @xmath261 , we have @xmath262 . hence , @xmath263 and @xmath264 we remind you that the higher @xmath265 the lower @xmath140 due to eq . ( [ taubin ] ) . the equal sign in eq . ( [ pg ] ) for all @xmath49 , would correspond to the @xmath135 case with barrier height @xmath266 and barrier width @xmath267 . therefore , we can simply invoke eq . ( [ efftis ] ) and write @xmath268 which has only a quadratic dependence on @xmath269 . this is far superior to exponential scaling of eqs . ( [ tbc2d],[tbc2epf ] ) . as a result , for high barriers and non - vanishing angles @xmath1 , the tis method becomes orders of magnitude more efficient than the reactive flux methods . of course , one might object that the reactive flux methods for this 2d system improve dramatically if we would chose @xmath270 instead of @xmath104 as rc . the reactive flux efficiency is then again identical to the @xmath135 case . however , this is exactly the crucial point . it is quite simple to find a proper rc in a low dimensional system , but in high dimensional complex systems this is certainly not the case . some methods have been devised that systematically search for rcs , but they have their limitations . the optimal hyperplane method @xcite and the string method @xcite rely on harmonic approximations and on the assumptions that these hyperplanes can be described as a linear combination of cartesian coordinates . complex reaction mechanism , notably chemical reactions in solution , require a highly nonlinear function as rc . the inclusion of important solvent degrees of freedom is not an easy task . some success has been made using the coordination number as rc @xcite . however , the influence of the solvent can be more subtle . in ref . [ ] , we found that electric contributions due to nearby spontaneous formations of tetrahedral ordered water molecules can be crucial to give a last push over the potential energy barrier . to incorporate such an effect in a one - dimensional rc would be an enormous task and can not be made without a priori insight in the mechanism . automated multidimensional us sampling approaches @xcite allow to explore the free energy surface efficiently in a predetermined set of order parameters . from the reduced free energy potential the most optimal one - dimensional rc could be estimated @xcite . however , as the predetermined order - parameter space is still limited to e.g. 6 coordinates , it is still likely that important coordinates such as solvent degrees of freedom can be missed . up to now , we have given expressions for the efficiency times while treating the effective correlation @xmath145 as a simple constant factor . the calculation of this factor is difficult as it depend on the intrinsic diffusive behavior of the mc / md sampling itself . fact is that @xmath145 can be significant larger 1 and usually has a scaling dependence on some system parameters ( like @xmath101 , @xmath100 , and @xmath1 ) . especially , the @xmath1-dependence can be severe and will be discussed in this section . us sampling and ti constrain the system in a small window or on a surface that drags the system over the barrier . we have assumed that the time consuming step in the rate calculation for the 2d barrier is the transmission coefficient calculation . in fact , the calculation of the free energy barrier can also be very hard due to an improper rc . this problem is known as the hysteresis problem which basically results in a diverging @xmath145 . evidently , one could ask whether this problem occurs in tis as well . if this would be the case , the tis efficiency would be much less advantageous as suggested by eq . ( [ efftis2d ] ) . we will show that sampling of paths instead of phase points alleviates or even eliminates the hysteresis problem entirely . the hysteresis problem does not occur in the 2d system we are considering , but can still persist , although still less than in free energy methods , for systems that have multiple reaction channels . consider the potential defined by eq . ( [ v2d ] ) and the hypersurfaces @xmath271 and @xmath89 as depicted in fig . [ top2d ] . suppose that we apply ti or us using small windows located at these surfaces . the distribution at these surfaces along the @xmath105 direction is then given by @xmath272 where @xmath85 can be either @xmath271 and @xmath89 . in tis , we could look at the distribution of first crossing points with the @xmath85 interface . this distribution is given by ( @xmath189 ) @xmath273 the additional @xmath200-term in the nominator and denominator of eq . ( [ disttis ] ) compared to eq . ( [ distrf ] ) is due to the fact that tis looks at crossing points while the free energy distribution looks at points on the surface . a crossing point is a point that can cross the surface in a single timestep and because @xmath274 @xcite the additional @xmath200-term appears . the other term @xmath275 in eq . ( [ disttis ] ) that is missing in eq . ( [ distrf ] ) ensures that not all velocities are considered . starting from @xmath7 going backward in time , @xmath88 should be hit before @xmath85 . this implies that we can write @xmath276 if @xmath277 and @xmath278 for @xmath279 . substituting this in eq . ( [ disttis ] ) yields @xmath280 with @xmath281 . [ disttitis ] shows the distributions of eqs . ( [ titis ] ) for three interfaces for the case @xmath282 , @xmath283 and @xmath284 . coordinate for three surfaces and the ti / us and tis methods . , width=226 ] at the first interface in fig . [ disttitis ] , the two sampling distributions of eqs . ( [ titis ] ) are exactly the same . the distribution is maximized at the left side of @xmath107 . however , at the interface @xmath111 the two distributions are very different . the ti / us distribution has now two maxima at either side of @xmath107 . as md and most mc schemes generate a new phase points by making small displacements from the previous point , the low probability region in the middle needs to be crossed to sample the distribution at either side . however , as crossing this low probability region is a rare event on itself , the system might not cross this region during whole simulation period . on the other hand , once the ti surface or us window is dragged over the barrier , the distribution is peaked at the right side of @xmath107 as can been seen from the shape of the distribution at @xmath89 . if we move the surface back from @xmath89 to @xmath106 , the system would be stuck again , but now at the right side of @xmath107 , which illustrates the hysteresis problem . only when the sampling is done extensively long , both maxima at the surface @xmath106 can be sampled in a single simulation . however , all measurements between two crossing events are correlated and basically contribute as a single uncorrelated measurement . hence , @xmath285 becomes exceedingly large . tis does not have this problem . the distribution at @xmath106 has still only one maxima . the distribution becomes uniform at @xmath89 . the divergence of @xmath286 is , hence , not to be expected . only when there are distinctive different reaction channels , ergodic sampling becomes difficult as for any method . still the sampling of multiple reaction channels is likely more effective using path sampling than ti or us due to the non - locality of the shooting move @xcite . these findings and the results of [ tis2d ] actually prove the relative insensitivity of tis to the rc as compared to the rf methods . this quality has been assigned to path sampling methods before , but to our knowledge , this is the first time that this is rigorously proven for reaction rate calculation methods . this points out a significant advantage of tis for systems for which a proper rc can not easily be derived . it is important to note that other path sampling approaches such as pptis ( or milestoning ) and ffs do not have this advantage . the pptis approximation fails for @xmath211 except when the interfaces are very far apart . for instance , consider the interfaces @xmath287 with @xmath288 . the pptis memory - loss assumption reveals that the trajectories that start at @xmath94 and cross @xmath289 have on average the same probability to reach @xmath86 as the trajectories that start at @xmath290 and cross @xmath289 . this can only be valid if @xmath291 ( see fig . [ top2d ] ) . if @xmath94 is set closer to @xmath85 , many trajectories that cross @xmath94 along a lines @xmath292 will not come from @xmath88 if they are followed backward in time and the pptis approximation results in an overestimation of the reaction rate . in turn , the large interface distances results that some of the pptis crossing probabilities are very small and can only be determined efficiently using tis . ffs , although in principle exact , also has a problem for non - zero angles @xmath1 . as ffs only propagates forward in time using the successful paths from the previous simulation , all the simulations become correlated . this implies that whenever the simulation at the first interface is in error , all other simulation results will be erroneous as well even if these simulations are performed infinitely long . this is a direct effect of the non - validity of eq . ( [ assum_unc ] ) for ffs . coming back to fig . [ disttitis ] , the flat distribution at @xmath89 can only be reproduced using ffs when at the previous interfaces ( as @xmath106 and @xmath293 ) a significant number of pathways is sampled in low - populated right tail of the path distribution which requires the sampling of an extremely large number of pathways . we have derived analytical expressions to determine the efficiency of different computational methods for the calculation of reaction rates . the efficiency has been expressed as the computational cost to obtain an overall relative error of 1 when all algorithmic parameters are optimized . we have called this property the cpu efficiency time @xmath0 . in app . [ secect ] , we have derived the cpu efficiency for very general cases . this also reveals an important generic result of how one should divide a fixed total simulation time among a set of independent simulations to get the lowest overall error . after a first round of simulations , reasonable estimates of @xmath140 can be obtained . then , the minimal additional simulation time , needed in each simulation , to obtain the overall best performance can be calculated and used for a second round of simulations . this approach can be very profitable and it is not restricted to rate calculations . we have applied these rules to determine the efficiency of a simple 2d benchmark system that allows an analytical approach . this offers a way to compare the efficiency of the different methods in a very transparent way . the two classes of methods that we compared are the rf methods and the path sampling methods . the rf methods consists of the calculation of the free energy barrier and the calculation of the transmission coefficient . both for the free energy as for the transmission coefficient calculation different methods exist . for the free energy calculation we compared two approaches of us . one using a series of rectangular windows and one using a single optimal biasing potential . for the path sampling methods , we have concentrated on the tis technique which is an improvement upon the original tps algorithm to calculate rates . the pptis approach reduces the computational cost even more but relies on the approximation of memory loss. as for us , we suggest that a single bias potential based on @xmath294 $ ] could replace the discrete set of interfaces . the @xmath135 case corresponds to the situation where the chosen rc is optimal . the tst approximation is then exact so that the free energy calculation is sufficient for the rf methods . we found an efficiency scaling equal to @xmath295 for us using rectangular windows . we obtained the same scaling rules for standard tis . using a single continuous bias potential , the us and tis efficiency can , in the optimal case , become independent of the barrier height and temperature . however , knowing the optimal bias basically implies that one already knows the answer . us using several windows and standard tis are more robust approaches if no a - priori knowledge of the system is available . this shows that tis and us compare very well in efficiency for all barrier heights and temperatures and that the difference can only rely in a single prefactor . it is likely that us is a bit better than tis that has to be faithful to the natural dynamics of the system . us has more flexibility to optimize the method such as choosing the best mc moves that minimizes the number of correlations . when @xmath211 , the chosen rc , @xmath104 , is no longer optimal . in contrast to the @xmath135 case , the principal computational effort of the rf method lies in the calculation of the dynamical correction . we showed that the bc method scales as @xmath296 , while the bc2 and epf methods are quadratically more efficient . the epf method is , however , a bit more than a factor 2 more faster than the bc2 method . the exponential dependence on @xmath297 indicates that a small deviation from the optimal rc can have dramatic consequences for the efficiency of the rf methods if they are applied to high barrier or low temperature systems . in contrast , the tis efficiency scaling is only @xmath298 . we also discuss the potential problem of hysteresis in us and ti when a non - optimal rc is chosen and why this problem does not occur for tis for the 2d barrier . this gives another evidence that the tis method is less sensitive the choice of rc . the advantage that path sampling does not require a rc has been advocated many times . however , although this statement is quite evident if the main object is to sample a representative set of reactive trajectories , it is not so evident for the calculation of reaction rates . the calculation of the reaction rate still requires a rc ( the only exception we know of is the method proposed in ref . [ ] using the pathlength as transition parameter ) . however , we are now the first to prove that a path sampling - based reaction rate calculation method , using tis , is potentially orders of magnitude faster than the rf - based methods whenever the right rc can not be determined . the reason is because tis uses an importance sampling technique to calculate the dynamical factor . the importance sampling of the rf methods only concentrates on the free energy . therefore , whenever the dynamical factor is low ( e.g. due to a wrong choice of rc ) , these methods automatically run into problems . the main question remains whether it is more profitable to search for a good rc and use the rf methods , or take a simple order parameter ( non - optimal rc ) and use the tis method . this question has not an easy answer and depends on the complexity of the system . i would like to thank daniele moroni for fruitful discussions and carefully reading this paper . i am also grateful to rosalind allen who made useful suggestions to improve the first version of this paper . this work was support by a marie curie intra - european fellowships ( meif - ct-2003 - 501976 ) within the 6th european community framework programme . we denote with @xmath7 a phase space point @xmath299 where @xmath300 are the cartesian coordinates and @xmath301 the momenta . the examples presented here consider only 1 particle in a two - dimensional potential , but @xmath302 are in general multidimensional vectors . the distribution of @xmath7 is given by the probability density @xmath95 . in case of boltzmann statistics @xmath303 with @xmath304 the total energy at phase point @xmath7 and @xmath305 with @xmath306 the temperature and @xmath307 the boltzmann constant . suppose @xmath308 is the value of a quantity we want to compute . in many cases such a quantity equals the expectation value of a certain observable . we denote this observable with @xmath309 which is a function of @xmath7 . then the expectation value or population mean @xmath310 is given by @xmath311 and @xmath312 for this specific case . path sampling simulations require a more extensive formulation especially when stochastic dynamics is applied . the discrete representation of a path is the most convenient , where a path @xmath16 is defined as a set of @xmath313 successive phase points that determine the system at intervals of @xmath77 along a certain trajectory : @xmath314 . in tis , the backward @xmath315 and forward @xmath316 time indices are not fixed , but depend on when a certain interface is crossed . the weight of the path is given by the initial distribution at time @xmath19 and the probability densities corresponding to the history and future of the path : @xmath317=\rho(x_0 ) & \prod_{i=1}^{\tau^b } p_{\rm n}(x_{(1-i)\delta t } \leftarrow x_{-i \delta t})\nonumber \\ \times & \prod_{i=1}^{\tau^f } p_{\rm n}(x_{(i-1)\delta t}\rightarrow x_{i \delta t } ) \label{pathweight1 } \end{aligned}\ ] ] where @xmath318 is the probability that the natural dynamics of the system brings you from @xmath7 to @xmath319 given you are in @xmath7 , and @xmath320 is the probability that if you are in @xmath7 you came from @xmath319 in the past . as shown in ref . [ ] , whenever the system is in a steady state , eq . ( [ pathweight1 ] ) is identical to @xmath317=\rho(x_{-\tau^b \delta t})\prod_{i=1}^{\tau^b+\tau^f } & p_{\rm n}(x_{(i-\tau^b-1)\delta t } \rightarrow x_{(i-\tau^b)\delta t } ) \label{pathweight2 } \end{aligned}\ ] ] which corresponds to the path weight as expressed in the original tps papers @xcite with the only difference that the starting index is @xmath321 instead of 0 . now , if @xmath180 $ ] is a function defined in path space , the population mean is given by @xmath322 { \rangle}=\frac{\int { \mathcal d}x \ , \hat{a}[x ] p[x ] } { \int { \mathcal d}x \ , p[x ] } , \label{pathav}\end{aligned}\ ] ] with @xmath323 and @xmath324 $ ] given by eq.([pathweight1 ] ) . in the following , we use @xmath325 as a point that is defined in either phase or path space , i.e. @xmath325 can either represent @xmath7 or @xmath16 . besides the simple ensemble averages ( [ phaseav ] ) and ( [ pathav ] ) , we can also define the biased ensemble average @xmath326 using a weight function @xmath327 . this biased ensemble is defined as @xmath328 in practice , sampling the biased ensemble @xmath329 by means of md simply requires the addition of a term @xmath330 to the potential of the system . in mc , this is achieved by a change of the acceptance - rejection criterion from @xmath331 $ ] to @xmath332 $ ] with @xmath333 and @xmath334 the new and old generated mc points @xcite . ensemble averages in path space @xmath335 { \rangle}$ ] can depend on the type of dynamics ( hence , on the hopping probabilities @xmath336 ) . to specify this dependency , explicitly , we use @xmath337 , which allows to generalize these path ensemble averages to arbitrary dynamics . the hopping probabilities of a simulation method @xmath338 can be of any type , for instance monte carlo , and do not need to be related to the natural dynamics of the system . therefore , these ensemble averages are annotated by @xmath339 . when @xmath340 is not specified , we assume that the natural dynamics is applied or that the property is independent of the hopping probability @xmath301 . hence , in our notation @xmath335 { \rangle}={\langle}a[x ] { \rangle}_1={\langle}a[x ] { \rangle}_{;p_{\rm n}}= { \langle}a[x ] { \rangle}_{1;p_{\rm n}}$ ] . the standard deviation of @xmath341 is defined as @xmath342 related to the standard deviation is the variance of @xmath341 @xmath343 and the covariance of two functions @xmath341 and @xmath344 @xmath345 with @xmath346 . a simulation @xmath49 , which can either be mc , md , or tps / tis generates a set of points @xmath347 in either phase or path space . each point @xmath348 is simulated with a certain probability @xmath349 and the chance that after @xmath348 another point @xmath350 is sampled is given by @xmath351 . we denote @xmath352 . now , the sample mean @xmath353 for a simulation of length @xmath354 is given by @xmath355 eq . ( [ sammean ] ) converges to the exact value , @xmath356 , in the limit @xmath357 . the standard deviation in the mean @xmath358 is defined as the standard deviation in the set of points @xmath359 that is obtained after performing a large number of independent simulations , @xmath360 , each of length @xmath354 and with result @xmath361 . hence , we can write @xmath362 using time translation invariance , we can show that @xmath363 and @xmath364 . now , we assume an exponentially decay in correlation for large @xmath365 which yields for large @xmath354 : @xmath366 \big\ } \label{stdm2}\end{aligned}\ ] ] which gives @xmath367 in eq . ( [ stdm ] ) , @xmath368 is the correlation number or the integrated auto - correlation function , which is a special case of @xmath369 , defined as @xmath370 with @xmath371 if two functions @xmath341 and @xmath344 are uncorrelated , @xmath372 . hence , if the simulation data are not correlated @xmath373 and @xmath374 and @xmath375 . let @xmath376 be the absolute error of a quantity @xmath308 and let @xmath377 be the relative error . for a sequence of measurements @xmath378 , the absolute error in the average @xmath353 is defined as the standard deviation in the mean : @xmath379 using eq . ( [ stdm ] ) , the absolute and relative errors are @xmath380 this shows that it is of eminent importance for the efficiency of the method that the trajectories generated by the computer algorithm minimize the correlation number @xmath162 as much as possible . in mc , this gives rise to conflicting strategies . in general mc algorithms select a new phase point by making a random displacement from a previous point , which is then accepted or rejected . when the displacement is small , the new point resembles strongly to the old one which can be a source of correlations . on the other hand , if the randomized displacement is too large , it is likely rejected after which the old phase point is counted again . therefore , a high rejection probability also introduces correlations . the optimum maximum displacement is a compromise between these two effects . the creation of a single step in path sampling is much more expensive than standard mc or md , but , on the other hand , the correlation number @xmath162 is usually much lower . suppose a quantity @xmath308 is not equal to the expectation value of a single operator , but , for instance , depends on two other quantities @xmath381 and @xmath382 via a function @xmath383 : @xmath384 . the error in @xmath308 can then be determined using the error propagation rules . say @xmath385 and @xmath386 are the approximations of @xmath381 and @xmath382 after @xmath58 and @xmath354 simulation cycles respectively . the resulting error @xmath387 can then be derived as follows . as @xmath388 and @xmath389 are small for large values of @xmath58 and @xmath354 , we can invoke following taylor expansion : @xmath390 as in eq . ( [ deferr ] ) the error @xmath387 equals the standard deviation of the mean @xmath391 . hence , @xmath392 substitution of eq . ( [ defcov ] ) , ( [ deferr ] ) and ( [ stdm ] ) yields @xmath393 then , from @xmath394 , @xmath395 , @xmath396 we get the propagation rule for the relative error @xmath397 ^ 2 \delta b^2(m)+ \big[c \frac{\partial f}{\partial c}/f \big]^2 \delta c^2(n ) \nonumber \\ & + 2 \big [ \frac{\partial f}{\partial b}\frac{\partial f}{\partial c}/f^2 \big ] { \rm cov}(\bar{b}(m),\bar{c}(n ) ) . \label{da2}\end{aligned}\ ] ] eqs . ( [ damn ] ) and ( [ da2 ] ) are the most general rules for the propagation of errors when @xmath308 depends on two quantities @xmath381 and @xmath382 . it can straightforwardly be generalized to more arguments such that @xmath398 . let us define @xmath163 as the average cpu time to perform a single mc , md or path sampling cycle . moreover , we denote @xmath399 as the total cpu simulation time after @xmath354 cycles . we introduce now the cpu efficiency time @xmath400 to obtain a relative error @xmath401 equal to one . putting @xmath402 in eq . ( [ errs ] ) and using @xmath403 gives @xmath404 this is the general cpu efficiency time to obtain a relative error of 1 in a single simulation average . this cpu efficiency time is a characteristic property of the quantity @xmath308 ( actually of the whole function @xmath341 ) and the simulation method ( via the @xmath405 hopping probabilities and @xmath163 ) . whenever @xmath400 is known the absolute and relative errors after a simulation period @xmath406 are directly given by : @xmath407 if @xmath341 is a binary operator such that @xmath408 is either 1 or 0 , then @xmath409 in eq . ( [ std ] ) and eq . ( [ taueff ] ) equals @xmath410 therefore , the @xmath400 becomes very large for small values of @xmath308 and , hence , an accurate evaluation becomes problematic . the cpu efficiency time for a composite value @xmath384 , where @xmath411 and @xmath412 are determined simultaneously in a single simulation , can be derived following the the same lines as in eq . ( [ stdm2 ] ) for the covariance @xmath413 \big\ } \nonumber \\ & \approx \frac{1}{n } \big\ { { \langle}(\hat{b}-b)(\hat{c}-c ) { \rangle}\times \big [ 1 + \nonumber \\ & \sum_{l=1}^\infty \frac { { \langle}(b_0-b)(c_l - c ) { \rangle}_{;p_{\rm s } } + { \langle}(c_0-c)(b_l - b ) { \rangle}_{;p_{\rm s } } } { { \langle}(\hat{b}-b)(\hat{c}-c ) { \rangle } } \big ] \big\ } \nonumber \\ & = \frac{1}{n } { \rm cov}(\hat{b},\hat{c } ) \big [ 1 + n_c^{bc}+n_c^{cb } \big ] \label{cov2}.\end{aligned}\ ] ] here , we inserted eq . ( [ defcov ] ) and ( [ corfun ] ) in the last line . substitution of eqs.([erreff ] ) and ( [ cov2 ] ) into eq . ( [ da2 ] ) and using @xmath414 gives @xmath415 ^ 2 \frac{\tau_{\rm eff}(b)}{\tau_{\rm sim}}+ \big[c \frac{\partial f}{\partial c}/f \big]^2 \frac{\tau_{\rm eff}(c)}{\tau_{\rm sim } } \label{da3 } \nonumber \\ & + 2 \big [ \frac{\partial f}{\partial b}\frac{\partial f}{\partial c}/f^2 \big ] { \rm cov}(\hat{b},\hat{c } ) \big\{1 + n_c^{bc}+n_c^{cb } \big\ } \frac{\tau_{\rm cyc}}{\tau_{\rm sim}}.\end{aligned}\ ] ] taking @xmath416 and @xmath417 directly results in @xmath418 ^ 2 \tau_{\rm eff}(b)+ \big[c \frac{\partial f}{\partial c}/f \big]^2 \tau_{\rm eff}(c ) \label{cpucomp } \\ & + 2 \big [ \frac{\partial f}{\partial b}\frac{\partial f}{\partial c}/f^2 \big ] { \rm cov}(\hat{b},\hat{c } ) \big\ { 1 + n_c^{bc}+n_c^{cb } \big\ } \tau_{\rm cyc } \nonumber \end{aligned}\ ] ] where @xmath419 and @xmath420 are given by eq . ( [ taueff ] ) . for example , if @xmath421 , the efficiency time of @xmath308 equals @xmath422 suppose @xmath384 and @xmath385 and @xmath386 are obtained via two different simulations . hence , @xmath354 and @xmath58 are not necessarily the same . in the following we assume that the different simulations are uncorrelated . thus , we assume that for two different simulations ( 1 ) and ( 2 ) the following holds @xmath423 here , the subscript @xmath424 indicates that the ensemble average can depend on how the two simulations are connected . assumption ( [ assum_unc ] ) is , for instance , true for us when two independent simulations are performed using two overlapping windows . it also holds for tis where the outcome of an interface sampling simulation at a certain interface is independent of the result of the previous interface simulation results . ( [ assum_unc ] ) does not hold for the most common implementation of the reactive flux method . in this approach , the importance sampling to determine the free energy barrier is simultaneously used to obtain a representative set of configuration points at the ts dividing surface . these configurations with randomized gaussian distributed velocities initiate the dynamical trajectories that determine the transmission coefficient @xcite . moreover , at variance with tis , eq . ( [ assum_unc ] ) is not true for the forward flux sampling ( ffs ) method , that was devised by allen _ et al . _ @xcite . here , the result of the interface sampling at one interface depends on the results of all the previous interfaces . the importance of eq . ( [ assum_unc ] ) is further discussed in sec . [ susechys ] . if the total simulation time @xmath425 is fixed , we still have some freedom in choosing @xmath354 and @xmath58 or , equivalently , choosing @xmath426 and @xmath427 . first , by substitution of eqs . ( [ erreff ] ) and ( [ assum_unc ] ) into eq . ( [ da2 ] ) we obtain @xmath428 ^ 2 \frac{\tau_{\rm eff}(b)}{\tau_{\rm sim}(b ) } + \big [ c \frac{\partial f}{\partial c}/f \big]^2 \frac{\tau_{\rm eff}(c)}{\tau_{\rm sim}(c)}. \label{d2a}\end{aligned}\ ] ] a logical approach , although not the optimum , is to give each simulation the same simulation time . we will denote the efficiency time that results from this strategy @xmath429 . taking @xmath430 for @xmath431 and @xmath416 gives @xmath432 ^ 2 \tau_{\rm eff}(b ) + \big [ c \frac{\partial f}{\partial c}/f \big]^2 \tau_{\rm eff}(c ) \big ] . \label{teffa1}\end{aligned}\ ] ] alternatively , we could try to obtain the same error in each simulation . the corresponding efficiency time will be annotated as @xmath235 . then , we need to use simulation times proportional to @xmath433 . hence , we take @xmath434 $ ] and @xmath435 $ ] in eq . ( [ d2a ] ) which results in @xmath436 ^ 2 + \big [ c \frac{\partial f}{\partial c}/f \big]^2 \big ] . \label{teffa2}\end{aligned}\ ] ] in order to determine the lowest @xmath0 , we add lagrange - multipliers constrains to eq . ( [ d2a ] ) in order to fix the total simulation time @xmath437 @xmath438 \equiv \big [ b \frac{\partial f}{\partial b}/f \big]^2 \frac{\tau_{\rm eff}(b)}{\tau_{\rm sim}(b ) } \nonumber \\ & + \big [ c \frac{\partial f}{\partial c}/f \big]^2 \frac{\tau_{\rm eff}(c)}{\tau_{\rm sim}(c ) } -\eta^2 \big\ { \tau_{\rm sim}(a ) - \tau_{\rm sim}(b)-\tau_{\rm sim}(c ) \big\ } \label{lagrange}\end{aligned}\ ] ] and minimize eq . ( [ lagrange ] ) with respect to all its arguments . taking the derivative to @xmath426 gives @xmath439 ^ 2 \frac{\tau_{\rm eff}(b)}{\big ( \tau_{\rm sim}(b ) \big)^2 } + \eta^2 = 0.\end{aligned}\ ] ] therefore , @xmath440 \frac{\sqrt { \tau_{\rm eff}(b ) } } { \eta } \big| . \label{tsimb}\end{aligned}\ ] ] we have put the absolute signs @xmath441 as the simulation times needs to be positive . the same relation holds for @xmath427 . @xmath442 \frac{\sqrt { \tau_{\rm eff}(c ) } } { \eta } \big| . \label{tsimc}\end{aligned}\ ] ] we can sum up eqs . ( [ tsimb ] ) and ( [ tsimc ] ) and use @xmath443 which gives the solution for @xmath444 @xmath445 \sqrt { \tau_{\rm eff}(b ) } \big| + \big| \big [ c \frac{\partial f}{\partial c}/f \big ] \sqrt { \tau_{\rm eff}(c ) } \big| \big ) . \label{relq}\end{aligned}\ ] ] then , substitution of eq . ( [ relq ] ) in eqs . ( [ tsimb],[tsimc ] ) completes the equations for @xmath426 and @xmath427 . substitution of these two equations in eq . ( [ d2a ] ) results in @xmath446 \big| \sqrt { \tau_{\rm eff}(b ) } + \big| \big [ c \frac{\partial f}{\partial c}/f \big ] \big| \sqrt { \tau_{\rm eff}(c ) } \big)^2 . \label{ect2sim}\end{aligned}\ ] ] the efficiency time @xmath400 of eq ( [ ect2sim ] ) is always strictly less than or equal to @xmath447 and @xmath448 of eqs . ( [ teffa1],[teffa2 ] ) . these cpu efficiency times are straightforwardly generalized to simulation series of any number . suppose that the final desired value @xmath308 is obtained by @xmath449 , where @xmath218 refers to the exact value that should be produced by simulation @xmath49 and @xmath146 is the total number of independent simulations that are needed to determine @xmath308 . then , the cpu efficiency times eqs ( [ teffa1]-[ect2sim ] ) yield @xmath450 ^ 2 \tau_{\rm eff}^{(s ) } \big ] , \nonumber \\ \tau_{\rm eff}''(a ) & = \big ( \sum_{s=1}^m \tau_{\rm eff}^{(s ) } \big ) \big [ \sum_{s=1}^m \big [ a^{(s ) } \frac{\partial f}{\partial a^{(s)}}/f \big]^2 \big ] , \nonumber \\ \tau_{\rm eff}(a)&=\big ( \sum_{s=1}^m \big| \big [ a^{(s ) } \frac{\partial f}{\partial a^{(s)}}/f \big ] \big| \sqrt { \tau_{\rm eff}^{(s ) } } \big)^2 , \label{teffa12m}\end{aligned}\ ] ] where @xmath451 is the efficiency time of simulation @xmath49 . for example , in many methods , the final value @xmath308 is given by a product of simulation results : @xmath452 . then , @xmath453|=1 $ ] for any @xmath49 after which eqs . ( [ teffa12 m ] ) become @xmath454 and @xmath455 the same eqs . ( [ effprod12],[effprod ] ) are valid for the case @xmath456 with any @xmath457 $ ] . whenever @xmath140 is the same for all @xmath49 , eq . ( [ effprod12 ] ) and ( [ effprod ] ) become identical and equal to @xmath458 we will derive the efficiency time @xmath140 of a single window in us as depicted in fig . [ umbrel ] . the factor that needs to be computed in simulation @xmath49 is @xmath459 . here , @xmath460 , @xmath461 , @xmath462 , and @xmath463 . then , we can use eq . ( [ taucomij ] ) with @xmath464 and @xmath465 @xmath466 and applying the assumptions eqs . ( [ assum],[asstaucyc ] ) gives @xmath467 via eq . ( [ defcov ] ) , eq . ( [ taubin ] ) , and eqs . ( [ assum],[asstaucyc ] ) we get @xmath468 then using eq . ( [ windows1 ] ) and ( [ v2d ] ) and writing @xmath469 and @xmath470 we arrive at @xmath471 moreover , @xmath472 is only nonzero in case @xmath473 . hence , @xmath474 substitution of eqs ( [ bcus],[partcov ] ) in eqs . ( [ teffcov ] ) and , after that , substitution of eqs . ( [ teffcov ] ) in eq . ( [ covij2 ] ) yields eq . ( [ taueffas ] ) . + in order to derive the efficiency time for an ensemble average @xmath157 when it is obtained by a weighted ensemble via @xmath158 , we write again @xmath475 with @xmath476 , @xmath160 , @xmath477 , and @xmath161 . applying eq . ( [ wens ] ) gives @xmath478 and @xmath479 . then applying eqs . ( [ std ] ) , ( [ defcov ] ) using the ensemble @xmath329 gives : @xmath480 substitution of eqs . ( [ sigmacovw ] ) in eq . ( [ taueff ] ) yields @xmath481 \tau_{\rm cyc } , \nonumber \\ \tau_{\rm eff}(c)&=\big ( { \langle}\omega^{-1 } { \rangle}{\langle}\omega { \rangle}-1 \big ) \big[1 + 2 n_c^{cc}\big ] \tau_{\rm cyc}. \label{teffbcw}\end{aligned}\ ] ] substitution of eqs . ( [ sigmacovw],[teffbcw ] ) in eq . ( [ covij ] ) yields eq . ( [ teffbias ] ) . to obtain the optimal biasing function , we add a lagrange multiplier to eq . ( [ teffbias2 ] ) to fulfill the normalization constraint @xmath482 the functional derivative of @xmath483 to @xmath54 for any operator @xmath341 and function @xmath383 is given by @xmath484 with @xmath485 . hence taking @xmath486 in eq . ( [ taueffbiasl ] ) results in @xmath487 { \mathcal n_c}+\eta^2 \big\}p(x)=0\end{aligned}\ ] ] which has as solution @xmath488=\frac{{\mathcal n_c}}{\eta^2 } \big ( 1 - \frac{\hat{a}(x)}{a } \big)^2 . \label{w2}\end{aligned}\ ] ] @xmath489 can be obtained , if necessary , via the normalization requirement @xmath164 . moreover , from eq . ( [ w2 ] ) follows directly eq . ( [ optw ] ) as @xmath54 should also obey @xmath490 for each @xmath7 . the tis pathlength ( [ ggtis ] ) can be obtained as follows . say @xmath200 is the velocity at the foot of the barrier ( @xmath88 ) at a time @xmath19 . the returning time is obtained by solving following equation @xmath491 , which has a solution for @xmath492 . within the path ensemble @xmath49 all trajectories should cross @xmath85 . hence , at the foot of the barrier the kinetic energy @xmath493 must be larger than @xmath494 . therefore , for the average pathlength we can write @xmath495 with @xmath77 the md timestep that is required to express @xmath129 as an integer representing the average number of discrete timesteps . in eq . ( [ tispath ] ) , we have first used the approximation @xcite @xmath496 and , then , neglected the @xmath497 terms . using @xmath498 and @xmath499 results in eq . ( [ ggtis ] ) . [ cols= " < , < " , ]

in a remarkable series of experiments in the nineties , sophisticated cooling techniques were used to expose the quantum nature of ultracold dilute gases . for bosons , the realization of bose - einstein condensation ( bec ) in 1995 @xcite revealed the importance of the underlying aggregating statistics for the existence of a broken symmetry phase in weakly interacting systems . for fermions , the statistics has just the opposite effect . due to the pauli exclusion principle , low energy collisions of fermionic particles are strongly suppressed , worsening the thermalization between the atoms and , therefore , making the cooling techniques more involved . nevertheless , a few years later another milestone experiment was realized in which a quantum degenerate fermi gas was observed @xcite . since then , research with both types of degenerate quantum gases has experienced a strong growth , and different trends have been pursued , each showing interesting and relevant connections to other fields of physics such as atomic , molecular and condensed matter physics . rotating quantum gases constitute one of these fast growing areas . in the case of bosons in a harmonic trap , intensive investigations have been carried out , both theoretically @xcite and experimentally @xcite . if the rotation frequency @xmath1 is of the order of ( but smaller than ) the radial trapping frequency @xmath2 , i.e. @xmath3 , the rotation is called slow . in this regime , intrinsic properties of rotating systems like the temperature dependence of the moment of inertia @xcite , rotation induced excitations @xcite , and the formation of vortices @xcite have been considered . if one rotates faster , vortex arrays are created @xcite where small - amplitude oscillations , so called tkachenko modes , have been observed @xcite and theoretically investigated @xcite . in addition , fast rotating ( @xmath4 ) becs have been linked to fascinating many - body phenomena such as the integer quantum hall effect @xcite or its fractional version @xcite , and are suspected to display quantum phase transitions to non superfluid strongly correlated states @xcite . in the fast rotation regime , experiments in harmonic traps are limited by the loss of confinement due to centrifugal forces when the rotation frequency approaches the axial harmonic trapping frequency . to overcome this problem , a. l. fetter suggested adding a quartic term to the trap potential @xcite . the successful experimental implementation of such a quartic term was carried out in the group of j. dalibard in paris @xcite . in their set - up a laser is used to stir the quartically trapped condensates , making possible the exploration of many properties of the system such as , for instance , the vortex distribution , the optical density @xcite , and the shape oscillations @xcite . the experimental realization triggered further theoretical research focusing on bose - einstein condensates ( bec ) . for instance , their collective oscillations @xcite , thermodynamical @xcite and dynamical @xcite properties have been considered in the presence of the quartic term . thus , the question arises , what new features come about through the interplay between rotation and fermionic statistics . in a harmonic trap , some properties have already been investigated . for instance , the moment of inertia and its relation to quadrupole oscillations , in a framework valid for bosons and fermions @xcite , and the collective excitations @xcite have been studied . in the superfluid phase , formation of vortices @xcite has been observed and a rotation - induced phase separation has been proposed @xcite . in addition , the influence of the landau levels on the density profiles of two - dimensional polarized fermions in the context of the integer quantum hall effect @xcite and the experimental feasibility of fractional quantum hall states @xcite have also been theoretically considered . in the case of an anharmonic trap , apart from studying the breathing mode in the bardeen - cooper - schrieffer ( bcs)-bec crossover regime @xcite , fermions have not been considered and most of the interesting aspects of the physics of rotating systems remain restricted to pure harmonic confinement . at low temperatures , polarized atomic fermi gases are essentially non - interacting due to pauli blocking and the fact that p - wave interactions are negligible in such systems . however , due to the fermi pressure , particles tend to avoid the center of the trap , so that , effectively , a statistical interaction arises even in the case of an ideal gas , which bears some analogies with repulsively interacting bosons @xcite . in this work we are concerned with rotating ideal fermi gases of spin polarized particles rotating with frequency @xmath1 . we consider the same trap configuration as in the paris experiment , which will be described in detail in section [ sec:2 ] . in section [ sec:3 ] we outline a general approach to formally justify the semiclassical approximation for calculating the density of states in anharmonic traps . furthermore , we apply this technique to the paris trap and see that a treatment within the semiclassical approximation is justified . in section [ sec:4 ] we consider ground - state properties of such a system in a similar semiclassical approach . we start by evaluating the fermi energy for arbitrary rotation , then we obtain the particle density , the momentum distribution , and free expansion time - of - flight absorption pictures . in section [ sec:5 ] temperature effects are considered and we derive expressions for the grand - canonical thermodynamic properties in terms of incomplete fermi functions . in addition we analyze analytic low- and high - temperature expansions . for the latter , a virial theorem approach is also presented . section [ sec:6 ] discusses the main results of this work emphasizing the effects of rotation and anharmonicity in comparison to a non - rotating harmonic trap and presents an outlook for the field . in the following we consider a trap potential which is anharmonic and consists of harmonic axially symmetric magnetic trapping supplemented by an additional gaussian trapping term in the @xmath5-plane . the gaussian laser beam is approximated by its fourth order taylor series @xmath6 where @xmath7 and @xmath8 are the laser intensity and width , respectively . the radial distance to the origin is denoted by @xmath9 . the first term in is just a shift in the energy scale and will be neglected in the following . the second order term will be absorbed into the harmonic trapping term . the last term is the one responsible for the quartic confinement and is characterized by the anharmonicity constant @xmath10 . it is important to notice that the harmonic trapping in the @xmath11-direction is not affected by the gaussian laser potential and is characterized by its harmonic oscillator frequency @xmath12 . an issue in the field of rotating quantum gases is that the rotation frequency , with which the condensate effectively rotates , might be different from the one of the stirring laser , due to changes in the shape dependent rotation properties of the gas such as the moment of inertia @xcite . we neglect these effects and assume that a stirring laser can adiabatically add angular momentum to the particles , so that the gas rotates with frequency @xmath1 , yielding an additional centrifugal harmonic term to the trap potential in the co - rotating frame @xcite . under these considerations , the rotating anharmonic potential takes the form @xcite @xmath13 where @xmath14 and @xmath15 are the axial harmonic energy and length scales , respectively , @xmath16 denotes the anisotropy of the trap , @xmath17 represents the anharmonicity of the trap , and @xmath18 is the mass of the atomic species under consideration . the rotation is taken into account with the parameter @xmath19 . adopting the values given for the paris experiment @xcite , the resulting rotating trap is then described by the radial harmonic trapping frequency @xmath20 , the axial harmonic trapping frequency @xmath21 , the force constant @xmath22 of the quartic trapping term , the anisotropy @xmath23 , and a tunable rotation frequency @xmath1 induced by the stirring laser . in the paris experiment , a bec was obtained with the bosonic isotope @xmath24 , for which the dimensionless anharmonicity takes the value @xmath25 . since fermionic species like @xmath26 and @xmath27 have been sympathetic cooled by using the bosonic atoms @xmath24 @xcite and @xmath28 @xcite , respectively , we consider them likely candidates for experiments with ultracold degenerate fermi gases in the paris trap . for definiteness , we evaluate all results of this paper for the parameters of the trap of the paris experiment and the @xmath26 species , for which we have @xmath29 ( see table [ tab:1 ] ) . let us discuss the trap potential and some of its peculiarities concerning rotation . for the moment , we restrict ourselves to the @xmath5-plane . as depicted in fig . [ fig:1 ] , for slow rotations , i.e. @xmath30 , the minimum of the potential is located at the origin . rotating faster one eventually reaches the point where the harmonic confinement vanishes ( @xmath31 ) and the trap is purely quartic . hereafter this will be called critical rotation . if the rotation is made even faster , one achieves the supercritical regime , where the harmonic confinement becomes centrifugal and the trap minimum is moved to @xmath32 and , with respect to the origin , has a depth given by @xmath33 which naturally characterizes the anharmonic trap . the lowest available potential energy is , thus , @xmath34 , in the supercritical case . , for the parameters given at table [ tab:1 ] at both sub- and supercritical rotations . for the latter , a minimum at @xmath35 appears due to the centrifugal nature of the harmonic term.[fig:1],width=288 ] llllll @xmath36 ( hz ) & @xmath37 ( hz ) & @xmath38 & @xmath39 ( jm@xmath40 ) & @xmath41 & @xmath42 + @xmath43 & @xmath44 & @xmath45 & @xmath46 & @xmath47 & @xmath48 + in this section we briefly discuss the meaning and the implementation of the semiclassical approximation for a system of particles possessing one - body hamiltonian @xmath49 with the eigenvalue equation @xmath50 where @xmath51 and @xmath52 denote the @xmath53-th single - particle eigenstate and eigenvalue , respectively . in view of studying the various thermodynamic properties of the system , one is usually interested in evaluating sums of the form @xmath54 where @xmath55 is some function of the single - particle energy levels . the semiclassical approximation consists in considering the energy levels so close to one another that the sum over them can be taken as an integral over a phase space density : @xmath56 to this end we define the density of states @xmath57 where @xmath58 is the dirac delta distribution . in the following we investigate for the parameters of the table [ tab:1 ] whether it is justified to approximate the density of states semiclassically according to @xmath59,\label{eq : g-1b}\ ] ] where @xmath60 denotes the single - particle classical hamiltonian . in systems where the exact energy spectrum is known , e.g. a harmonic trap , the validity of this approximation can be determined by comparing and . however , in the case that the energy spectrum is unknown , as for the present anharmonic potential , we must take a different approach to justify the semiclassical approximation . to this end , we determine not only the leading semiclassical approximation @xmath61 , but also its subleading correction @xmath62 . after setting the relevant energy scale of the system , the validity of the semiclassical approximation can be determined by the _ relative _ magnitude of the correction . we first treat the general case and , then , specialize to the anharmonic trap of the paris experiment . to obtain the semiclassical expansion of the density of states of the anharmonic potential , we follow the general procedure worked out in ref . @xcite , which derives the density of states in form of a gradient expansion @xmath63 the first term corresponds to the semiclassical approximation @xmath61 in , which yields after integration over the momenta @xmath64\dfrac{\left[e - v({\bf x})\right]^{\frac{d}{2}-1}}{\gamma(d/2)}.\label{eq : approx}\ ] ] the second term is the subleading correction to the density of states @xmath65^{\frac{d}{2}}\dfrac{1}{\gamma(d/2 - 2)}\nonumber\\ & & \times\int d^{d}x\,\theta\left[e - v({\bf x})\right]\nabla^{2}v({\bf x})\left[e - v({\bf x})\right]^{\frac{d}{2}-3}.\nonumber\\ \label{eq : g1}\end{aligned}\ ] ] we note that , for the physically relevant dimension @xmath66 , the spatial integral diverges due to the behavior of the term @xmath67^{\frac{d}{2}-3}$ ] at the classical turning points . this problem can be solved through the method of analytic continuation as demonstrated later by applying to the anharmonic potential . it remains to determine the energy scales for which the semiclassical approach is supposed to be valid . at high enough temperatures , the particles are classical . when the temperature is gradually lowered , or other parameters are changed , classical statistical mechanics is no longer exact . at first , a semiclassical approach may work , but for certain conditions it may completely fail . the temperature , particle number , and trap potential parameter scales in which the semiclassical approximation is valid depend on the underlying statistics . for bosons in the gas phase , the thermal distribution is characterized by the thermal energy @xmath68 . if this energy is sufficiently larger than the the energy of the single - particle ground - state , the leading semiclassical approximation is valid . at the bec critical temperature @xmath69 , however , the ground - state occupancy becomes macroscopic . this effect must be treated outside of semiclassics because it corresponds to a broken symmetry state associated to the divergence of the bose distribution @xmath70 for @xmath71 . we will see explicitly in the case of the anharmonic oscillator that the subleading correction is not negligible for energies near the ground - state energy . the thermal energy at the critical temperature is thus the relevant energy scale for determining the validity of this approach . for fermions the considerations are slightly different . due to the pauli exclusion principle , as @xmath72 the particles in the system do not all drop to the ground - state but rather fill every energy level up to the fermi energy @xmath73 , which is defined as the energy of the highest occupied state . as the density of states @xmath74 typically increases with a power of @xmath75 , the majority of particles in the system will have energies much larger than the ground - state value and their average energy is of order @xmath73 , even at @xmath76 , provided @xmath42 is large . therefore , in principle , a spin - polarized fermi gas can be well described at all temperatures in the semiclassical approximation provided the particle number is large enough . the relevant energy determining the validity of this assertion is simply the fermi energy @xmath73 . numerical calculations indicate that this is indeed the case @xcite . to consider the anharmonic potential , we must first determine explicitly the leading and subleading terms of the semiclassical density of states . we can , then , apply the leading term to calculate the fermi energy , and with this result return to verify the a priori assumption of the validity of the approximation . the form of the semiclassical density of states in the anharmonic potential can be found from . performing the spatial integral we obtain the result @xmath77 where it is meant in this and other equations , that the function takes on nonzero values for @xmath78 only for super - critical rotations . the density of states for the parameters of the anharmonic trap described in table [ tab:1 ] is shown at several rotation frequencies @xmath1 in fig . [ fig:2 ] . for subcritical rotations , the density of states vanishes for @xmath78 . for supercritical rotations , the density of states is nonzero also for @xmath79 , due to the dip in the potential . for the anisotropy and the anharmonicity of the trap . the solid lines are from bottom to top at rotations of @xmath80 to @xmath81 in steps of @xmath82 . the dashed line is the density of states at the critical rotation @xmath83.[fig:2 ] ] there are two important limiting cases which shed some light on the behavior of the system . the first one is the harmonic limit @xmath84 , for which we have from @xmath85 which corresponds to the known result @xcite . note that the harmonic limit is valid only for subcritical rotations ; for supercritical rotations the density of states becomes negative corresponding to an instability due to the fact that particles escape the trap . the second physically important case is the pure quartic limit in radial confinement . it is realized by either @xmath86 , i.e. a critical rotation , or @xmath87 , so that the density of states takes the form @xmath88 examining and , we see that the rotation parameter @xmath89 appears in the density of states only in ratios to the anharmonicity parameter @xmath90 . for supercritical rotations , we will find that there is a symmetry between the range of rotations , @xmath91 ( @xmath92 ) to @xmath86 ( critical rotation ) , and the range of anharmonicities from small @xmath90 to @xmath93 . the supercritical properties of the system can thus be seen as an interpolation between the harmonic limit and the pure quartic limit . we now calculate the subleading semiclassical approximation @xmath62 as given in for the particular case of the anharmonic potential . a direct integration in for @xmath66 yields a divergent result . however , we can rewrite the spatial integral in terms of the beta function @xmath94 for which in ref . @xcite provides the identity @xmath95 thus , the formally divergent integrals can be treated as finite based on the analytic continuation of the gamma function through the relationship @xmath96 . to evaluate the integral we need the laplacian of the anharmonic potential , which reads : @xmath97 with this and the full potential given in , we obtain from the first correction to the semiclassical density of states : @xmath98 where , again , the lower line is valid only for supercritical rotations . to determine the effects of the trap parameters on the relevance of the correction term , we examine the ratio @xmath99 . we first see that it is in general reduced in both the sub- and supercritical regimes as @xmath100 increases . likewise , the importance of the correction term decreases as the anharmonicity @xmath90 decreases . as the correction is largest for low energies , the most interesting case for supercritical rotations is when @xmath78 , where we have @xmath101^{2 } , \;\;\;\;-\delta e < e<0.\label{eq : correction - ratio}\ ] ] from we see that for a fixed @xmath102 , the relevance of the correction increases linearly with the rotation parameter @xmath89 . furthermore , we see that the correction diverges as @xmath103 . however , this divergence is only of the form @xmath104 , and so any interesting integral of the form over the density of states will still have a finite contribution from this correction , provided @xmath105 is not itself divergent . for bose systems , @xmath105 contains a factor of @xmath106^{-1}$ ] . the phase transition to bose - einstein condensation happens as @xmath107 , where @xmath108 is the lowest energy eigenvalue . thus , in the condensate phase , an integral of the form has an improper lower limit for @xmath105 blows up as @xmath109 , since it is the lowest available energy . this problem can be overcome by dividing the gas explicitly into a condensate plus thermal particles . we can now check the validity of the semiclassical approximation in the case of the anharmonic trap . both the leading and the subleading term are plotted in fig . [ fig:3 ] for the parameters of table [ tab:1 ] , where we see that even for supercritical rotations the subleading correction becomes negligible on energy scales of more than a few @xmath110 . in the following section we determine the fermi energy @xmath73 of the system , from which we will see that , for the chosen parameters , the difference between @xmath73 and the minimum potential is orders of magnitude larger than @xmath110 . with this , we justify the semiclassical approximation to the density of states . . [ fig:3],width=302 ] though not fully accessible from the experimental point of view , the ground - state and its properties are crucial for understanding the system . the fermi energy , e.g. , carries information about aspects like dimensionality , trap geometry , and particle number and , thus , providing information necessary to handle important physical issues . in this section , we consider some properties of the ground - state like the dependence of the fermi energy on the particle number , the ground - state energy , the spatial and momentum distribution and , lastly , the ballistic expansion of the system . the fermi distribution is given by @xmath111 where @xmath112 is the chemical potential and @xmath113 with @xmath114 the boltzmann constant and @xmath115 the temperature of the system . at @xmath76 , i.e. , @xmath116 , the fermi distribution becomes a step function at the fermi energy , which is defined as the chemical potential at @xmath76 : @xmath117 the ground - state particle number is therefore given by @xmath118 evaluating this integral with the semiclassical density of states from yields @xmath119 \label{eq : n-1a}\ ] ] for the first case , in which one has @xmath120 . for the second case , in which @xmath121 , the particle number is given by @xmath122 the inverse of ( [ eq : n-1a]-[eq : n-1b ] ) determines the fermi energy as a function of the trap parameters and the particle number . in the harmonic @xcite and pure quartic limits we have @xmath123 respectively . we note that the fermi temperature @xmath124 is of the same order of magnitude as the critical temperature for non - interacting bose - einstein condensates in the quartic regime @xcite @xmath125^{\frac{2}{5 } } , \label{eq : eflimiting_bec}\ ] ] with @xmath126 . it should be pointed out that negative fermi energies are allowed by ( [ eq : n-1a]-[eq : n-1b ] ) for supercritical rotations , in which case the gas is confined entirely to the dip in the potential , creating a hole in the center of the trap analogous to the one found in two - dimensional becs @xcite . in this interesting regime ( @xmath121 ) we see that the fermi energy also takes on a simple analytic form , @xmath127 where the superscript do stands for donut , in reference to the shape that the gas assumes when all the particles are in this energy interval , as discussed below . notice that in the ` donut ' regime , for a fixed particle number the fermi energy is set a distance above the trap minimum that is independent of the rotation frequency , but is directly related to the anharmonicity and to the number of particles . we now have three cases where the fermi energy takes a simple analytic form corresponding to the three interesting rotation regimes : the harmonic limit for subcritical rotations , the pure quartic limit for critical rotations , and the ` donut ' regime for supercritical rotations . although we can not obtain from ( [ eq : n-1a]-[eq : n-1b ] ) a general explicit expression for the fermi energy , we can determine its behavior from these special cases . in fig . [ fig:4 ] we see that for low @xmath73 , the general supercritical behavior approaches the harmonic oscillator behavior . as rotation increases to the critical value , the curves approach the pure quartic limit . as the rotation becomes supercritical , the curves are offset to more negative energies and the ` donut ' regime eventually sets up . in fig . [ fig:4 ] we also see that for any fermi energy possible , the quantum correction term to the semiclassical approximation is negligible , thus justifying our a priori use of the semiclassical approximation . in the following we use the fermi energy at zero rotation as a convenient energy scale . for our parameters it is@xmath128 the fermi temperature @xmath129 at zero rotation provides a corresponding temperature scale , @xmath130 the total internal energy of the system in the ground - state is given by @xmath131 evaluating this integral with the density of states from yields @xmath132\\ w(e_{f } ) \end{cases } \label{eq : u-1}\ ] ] with the remark that the upper line is valid for @xmath120 and the lower one for @xmath133 and the abbreviation @xmath134 in the harmonic and purely quartic limits , respectively , the internal energy it is related to the particle number and fermi energy by @xmath135 there is also a simple relation in the ` donut ' regime , @xmath136 we see in fig . [ fig:5 ] that the internal energy for subcritical rotations is bounded by the @xmath84 harmonic oscillator limit and the @xmath93 pure quartic limit . this is due to the symmetry between the anharmonicity and the rotation , which causes the latter to enhance the former . for fixed @xmath42 and the parameters of table [ tab:1 ] . the dashed horizontal line is the harmonic limit and the dot - dashed horizontal line is the quartic limit . the vertical lines indicate the critical rotation ( @xmath137 ) and the start of the ` donut ' regime , where @xmath138.[fig:5],width=302 ] to obtain the particle density and the momentum distribution in the semiclassical approximation , we start from the wigner function @xcite in the semiclassical approximation @xmath139\ } } , \label{wig_func}\ ] ] which delivers the particle density through @xmath140 and the momentum distribution through @xmath141 at @xmath76 the wigner function becomes a heaviside theta function and is given by @xmath142 . \label{wig_func_zerot}\ ] ] in this limit , both the particle density and the momentum distribution assume a well - defined shape . the particle density is given for a general velocity independent potential @xmath143 by @xmath144^{\frac{3}{2}}}{3\pi^{2}a_{z}^{3}\epsilon_{z}^{\frac{3}{2}}}\theta\left[e_{f}-v({\bf x})\right].\label{eq : config - dist-1}\ ] ] from , we see that a given rotation @xmath145 , at which the hole in the trap center appears , the system looks like a ` donut ' , so to speak . the value of @xmath145 can , therefore , be determined from by setting @xmath146 and yields @xmath147 in an effectively two - dimensional ( uniform over an axial length @xmath148 ) , interacting bec trapped by the radial components of the anharmonic potential , rotation can lead as well to a hole in the center of the trap @xcite , where , in the thomas - fermi approximation , a similar analytic expression can be written down : @xmath149 where @xmath150 denotes the s - wave scattering length . in the case of non - interacting bosons , @xmath151 coincides with @xmath152 , but this is not true for fermions , where an effective interaction arises from the fermi statistics and @xmath145 does not equal @xmath2 even in the absence of interactions . a discussion of this pauli `` pseudopotential '' in the case of harmonically trapped fermi gases is provided by ref . the value of @xmath145 could also be obtained from the length scales @xmath153 set up by the fermi energy , describing the extent of the gas in space , @xmath154\right|_{x_{i}=r_{i}^{\rm tf}}=0 , \label{eq : fermir-1}\ ] ] where @xmath155 range over @xmath156 . these length scales describe the maximum extent of the coordinate space distributions at @xmath76 and are called the thomas - fermi radii @xmath157 . for the anharmonic potential , they are given by @xmath158 } , \label{eq : r - tf}\end{aligned}\ ] ] where the positive root @xmath159 gives the outer radius of the distribution and the negative root has a physical meaning only in the ` donut ' regime , where it gives the inner radius . we note that for supercritical rotations ( @xmath138 ) , the density distribution has the shape of a shell , of which we then designate the outermost limit in the @xmath160-plane as the thomas - fermi radius @xmath159 and the largest extention in the @xmath11-direction as @xmath157 . in the harmonic limit , the thomas - fermi radii are @xmath161 in particular , we are interested in the optical density which can be directly measured by absorption images . performing an integration over the axial @xmath11-direction , we obtain @xmath162^{2}}{4\pi a_{z}^{2}\epsilon_{z}^{2}}.\label{eq : config - dist-2}\ ] ] the rotation dependence of the @xmath76 column density is shown in fig . [ fig:6 ] for the anharmonic potential . just above the critical rotation the distribution becomes nearly homogeneous , and at high supercritical rotations the gas eventually enters the ` donut ' regime , where the gas is entirely confined to the dip in the trap . in contrast , without the quartic trapping the gas spreads out to become an infinite pancake as the rotation approaches the critical rotation , as we can see from the divergence of the harmonic oscillator thomas - fermi radius at the critical rotation . spatial density distribution for the parameters of the paris experiment at increasing rotations . the anharmonic term prevents the formation of a ` pancake ' at near - critical rotations and causes the gas to take on a ` donut ' shape for large supercritical rotations.[fig:6],width=298 ] the fermi energy also sets a momentum scale through the fermi momentum @xmath163}$ ] . it represents the maximum extent of the momentum distribution at @xmath76 . for the anharmonic trap we have @xmath164/{\epsilon_{z}}}$ ] . a simple dependence on the particle number emerges again in the harmonic limit @xmath165 at @xmath76 the momentum space distribution for the anharmonic potential reads @xmath166+\sqrt{q } ; & q\geq0\\ ( q+1)\pi ; & q\leq0\end{cases},\nonumber\\ \label{eq : mom - distr}\ ] ] with the dimensionless variable @xmath167 . \label{eq : xdef}\ ] ] we see that , in contrast to bec , where the momentum distribution of the condensed atoms inherits the anisotropy of the trap , the fermionic momentum distribution of a non - interacting gas remains isotropic despite the trap anharmonicity . this feature is drastically affected by presence of the anisotropic dipole - dipole interaction , as then the fock term of the two - body mean - field energy induces a characteristic deformation of the momentum distribution @xcite . note that we have only considered the co - rotating frame ; as we will see , the rotation leads to an anisotropy in the momentum distribution in the lab frame . measurements in cold gases are usually carried out by turning off the trap and shining a resonant laser on the sample after some time of free expansion . these are called time - of - flight measurements ( tof ) . so far we have treated the problem in the corotating frame . in considering tof measurements , it is necessary to consider the laboratory frame , where the gas expands ballistically . the transformation of the momenta from the corotating frame to the lab frame yields @xmath168 where @xmath169 is the tangential unit vector . since the trap is radially symmetric , the coordinate transformation of the position does not affect the distribution . we have then @xmath170 thus although the momentum distribution does not inherit the anisotropy of the trap , the trap rotation leads nonetheless to an anisotropic momentum distribution . to study the expansion in detail we apply the ballistic substitution @xmath171 which yields a relation between the lab frame distribution and the corotating distribution : @xmath172,\ ] ] where @xmath173 and @xmath174 denote respectively the radius and tangential vector at the position @xmath175 , and can be expressed as @xmath176 with @xmath177 the unit vector in the axial direction . of particular experimental interest is the density distribution at a time @xmath178 after the trap is turned off , @xmath179.\label{eq : ballistic}\end{aligned}\ ] ] for the case of harmonic trapping and zero rotation , the kinetic term can be combined with the harmonic term and the integral over momentum space in takes nearly the same form as the integration over momentum space in . unfortunately , in the case of the anharmonic potential with rotation , the integration can not be cast into a simple form and does not provide an insightful analytic expression . a measure of the spatial density is given by the aspect ratio @xmath180 with the respective widths @xmath181^{\frac{1}{2}}.\label{eq : wi}\ ] ] for harmonically confined fermi gases with zero rotation , it is known that anisotropic samples become asymptotically isotropic for large expansion time @xcite , since their widths scale in time as @xmath182 and the aspect ratio as @xmath183 in the case of a rotating gas with additional quartic trap , it is interesting to see how the rotation affects the asymptotic value of the aspect ratio , which reflects the lab frame momentum distribution . in fig . [ fig:7 ] the aspect ratio has been evaluated numerically for the anharmonic potential . comparing to the harmonic trap at zero rotation , we see that the anisotropy of the momentum distribution increases with the rotation frequency for the anharmonic trap . it is worth mentioning that the value of @xmath184 for large asymptotically large @xmath178 depends on the rotation , so that the aspect ratio could be used to measure the actual rotation frequency in an experiment . from top to bottom.[fig:7],width=302 ] after having studied the effects of rotation on the ground state of an anharmonically trapped polarized fermi gas , we turn our attention to the thermodynamic properties of such a system . the standard thermodynamical recipe is specialized to the case of the anharmonic potential and the consequences for different rotation frequencies are explored in both low- and high - temperature regimes . in this section we describe the thermodynamical properties of the system within the framework of the grand - canonical ensemble , where the most fundamental quantity is the grand - canonical partition function . for particles obeying fermi - dirac statistics it is given by @xmath185 , \label{eq : zg}\ ] ] where @xmath52 , as before , is the eigenvalue corresponding to the eigenstate @xmath186 . the grand - canonical free energy is obtained from the grand - canonical partition function through @xmath187 inserting in and making use of the semiclassical approximation , i.e. , writing the discrete sum as an integral over the density of states , we obtain @xmath188de.\label{eq : fg-3}\end{aligned}\ ] ] performing a partial integration of , one sees that the integral for the grand - canonical free energy contains the fermi distribution : @xmath189 where @xmath190 denotes the number of single - particle states with energy less than @xmath75 @xmath191 in view of using the semiclassical approximation for the density of states in , we define the incomplete fermi integral @xmath192 which is called incomplete due to the fact that its integration range is smaller than the one of the original fermi integral , defined by @xmath193 for convenience we introduce the further shorthand @xmath194 with and , we write the grand - canonical free energy of a system contained in the anharmonic potential in terms of the following fermi integrals : @xmath195 } { 2\beta^{\frac{7}{2 } } \epsilon_{z}^{\frac{5}{2 } } \kappa^{\frac{1}{2 } } } + \dfrac{\lambda^{2}\eta\zeta_{3}^{+,{\rm inc } } } { 2\beta^{3 } \epsilon_{z}^{2}\kappa } \nonumber \\ & & - \dfrac{\lambda^{6 } \eta^{3}\zeta_{2}^{+,{\rm inc } } } { 24\beta^{2 } \epsilon_{z } \kappa^{2 } } + \dfrac{\lambda^{10 } \eta^{5}\zeta_{1}^{+,{\rm inc } } } { 320\beta\kappa^{3 } } , \label{eq : fg-5}\end{aligned}\ ] ] where @xmath196 is the sign of @xmath89 , defined as @xmath197 to obtain the other thermodynamical quantities , we express the grand - canonical free energy in terms of a legendre transformation @xmath198 with the differential @xmath199 therefore , the particle number @xmath42 and entropy @xmath200 can thus be determined from the grand - canonical free energy according to @xmath201 for the particle number we obtain @xmath202 } { 2\beta^{\frac{5}{2 } } \epsilon_{z}^{\frac{5}{2 } } \kappa^{\frac{1}{2}}}-\dfrac{\lambda^{2}\eta\zeta_{2}^{+,{\rm inc}}}{2\beta^{2}\epsilon_{z}^{2}\kappa}\nonumber\\ & & + \dfrac{\lambda^{6}\eta^{3}\zeta_{1}^{+,{\rm inc}}}{24\beta\epsilon_{z}\kappa^{2 } } , \label{eq : n}\end{aligned}\ ] ] and for the entropy we get @xmath203 the internal energy can then be obtained from yielding @xmath204 } { 4\beta^{\frac{7}{2 } } \epsilon_{z}^{\frac{5}{2}}\kappa^{\frac{1}{2}}}-\dfrac{\lambda^{2}\eta\zeta_{3}^{+,{\rm inc}}}{\beta^{3}\epsilon_{z}^{2}\kappa } \nonumber\\ & & + \dfrac{\lambda^{6}\eta^{3}\zeta_{2}^{+,{\rm inc}}}{6\beta^{2}\epsilon_{z}}-n\delta e. \label{eq : u}\end{aligned}\ ] ] notice that , in the pure quartic limit @xmath205 , a simple relation between the free and the internal energies arises @xmath206 the heat capacity of the system at constant particle number and trap parameters is given by @xmath207 to evaluate this derivative , we must , as usual , obtain the temperature derivative of the chemical potential from the with we then obtain @xmath208}{8\beta^{\frac{5}{2 } } \epsilon_{z}^{\frac{5}{2 } } \kappa^{\frac{1}{2 } } } - \dfrac{3\lambda^{2}\eta\zeta_{3}^{+,{\rm inc } } } { \beta^{2}\epsilon_{z}^{2 } \kappa } + \dfrac{\lambda^{6 } \eta^{3}\zeta_{2}^{+,{\rm inc } } } { 12\beta \epsilon_{z}\kappa^{2 } } \nonumber\\ & & -\dfrac{\left\{\dfrac{5\sqrt{\pi}\left [ 2\theta(-\eta ) \zeta_{\frac{5}{2}}^{+ } + \sigma(\eta ) \zeta_{\frac{5}{2}}^{+,{\rm inc } } \right]}{\beta^{\frac{5}{2}}\epsilon_{z}^{\frac{5}{2}}\kappa^{\frac{1}{2 } } } - \dfrac{4\lambda^{2}\eta\zeta_{2}^{+,{\rm inc } } } { \beta^{2 } \epsilon_{z}^{2}\kappa } + \dfrac{\lambda^{6 } \eta^{3}\zeta_{1}^{+,{\rm inc } } } { 6\beta \epsilon_{z } \kappa^{2 } } \right\}^{2 } } { \dfrac { 8\sqrt{\pi}\left[2\theta(-\eta ) \zeta_{\frac{3}{2}}^{+ } + \sigma(\eta)\zeta_{\frac{3}{2}}^{+,{\rm inc } } \right ] } { \beta^{\frac{5}{2 } } \epsilon_{z}^{\frac{5}{2 } } \kappa^{\frac{1}{2 } } } - \dfrac{8\lambda^{2 } \eta\zeta_{1}^{+,{\rm inc}}}{\beta^2\epsilon_{z}^{2}\kappa}}\nonumber\\ \label{eq : cv - sub}\end{aligned}\ ] ] since this analytic expression is not particularly illuminating , we analyze the behavior of for a range of temperatures from fig . [ fig:8 ] . at low temperatures the heat capacity is nearly linear in temperature , and at high temperatures the heat capacity approaches @xmath209 for a fixed rotation . at high temperatures a dip appears with a minimum occurring at a rotation that increases with the temperature . this interesting case and the general high - temperature behavior of the heat capacity will be discussed in detail below . , the fermi energy of the system at zero rotation ( @xmath210 . the temperature scale is the fermi temperature , @xmath211 . the angular frequency scale is the critical frequency @xmath2.[fig:8],width=302 ] in the low - temperature regime , we can understand temperature effects by using the sommerfeld expansion as sketched in ref . @xcite . to this end we consider the incomplete fermi function with @xmath212 @xmath213 where @xmath214 is the fermi distribution . a partial integration yields @xmath215,\label{eq : fermi - som-1a}\end{aligned}\ ] ] where the derivative of the fermi distribution is given by @xmath216\left[e^{-\beta(e - x ) } + 1 \right ] } .\label{eq : df}\ ] ] consider the case that @xmath217 . we see that the derivative becomes very small . if @xmath218 , then the lower limit of integration in can be extended to @xmath219 and @xmath220 . if @xmath221 then the integrand vanishes and @xmath222 . we now expand @xmath223 in a taylor series about @xmath224 , @xmath225\frac{(e - x ) ^{n}}{\gamma(n+1)}\nonumber\\ \label{eq : fermi - taylor}\end{aligned}\ ] ] and integrate with to obtain @xmath226 it is worth remarking that the sommerfeld expansion ( [ eq : fermi - inc - low ] ) is valide for @xmath227 and vanishes for @xmath228 . furthermore , the @xmath229 are constants given in terms of the riemann zeta function @xmath230 as @xmath231\zeta(2n).\ ] ] for the case of the complete fermi integral , eq . reduces to the usual sommerfeld expansion given by @xmath232 for @xmath227 . although the series in and terminate for @xmath233 , the expansion is not exact since the limits of integration were extended in . it is useful to obtain an explicit equation for the temperature dependence of the chemical potential using the sommerfeld expansion and the particle number equation . to this end , we substitute @xmath234 and keep terms linear in @xmath58 as well as quadratic in temperature to obtain up to third order contributions @xmath235 ( k_{b}t)^{2 } , \label{eq : mulow}\ ] ] where @xmath236 is the density of states evaluated at the fermi energy . we see that for low temperatures the chemical potential decreases with the square of the temperature . this result is compared to the numerical inversion of the particle number equation with both the exact evaluation of the fermi integrals and with the evaluation using terms quartic in temperature from the sommerfeld expansion in fig . [ fig:9 ] . .[fig:9],width=302 ] in the high - temperature limit we expect quantum statistics for both bosons and fermions to approach classical statistics . here we derive the high - temperature expansions of the thermodynamic properties of a fermi gas in the anharmonic trap of the paris experiment . we also consider in more detail the unique rotation dependence of the heat capacity in the high - temperature limit . a high - temperature expansion of the incomplete fermi integrals is obtained as follows . we rewrite as @xmath237 where @xmath238 . as seen in fig . [ fig:9 ] , if the temperature is high enough , @xmath239 and , thus , @xmath240 so that we can use the binomial expansion for the factor @xmath241 . then we obtain @xmath242 with the definition @xmath243 where @xmath244 is the gamma function . for @xmath245 in the high - temperature limit , it is sufficient to keep only the first term in the sum , yielding @xmath246 for positive integer @xmath53 , @xmath247 can be expanded exactly : @xmath248 the high - temperature expansion is compared to the exact evaluation of the fermi integrals in fig . [ fig:9 ] in the context of a numerical inversion of the particle number equation to solve for the chemical potential . the approximation is in general very accurate for temperatures larger than half the fermi temperature . from we can determine the high - temperature behavior of the general expressions for the thermodynamic properties of the system . for the particle number , we get @xmath249^{\frac{1}{2 } } \left[2 \theta(-\eta ) + \sigma(\eta)q\left(\frac{5}{2},\beta{\delta e}\right)\right]%\right\ } + \cdots . \label{eq : hight - n}\ ] ] it should be remarked that ( [ eq : hight - n ] ) is to be understood as giving the temperature dependence of the chemical potential in the high - temperature regime , so that @xmath112 is changed to maintain a constant value of @xmath42 on the left - hand side of the equation . note that the high - temperature chemical potential can be determined explicitly by taking the logarithm of . now we are in a position to write down the high - temperature versions of the internal energy @xmath250}+\cdots \label{eq : hight - u}\ ] ] and of the heat capacity @xmath251\dfrac{\sigma(\eta)(\beta\delta e)^{\frac{1}{2}}\pi^{-1/2 } e^{-\beta{\delta e}}}{\left[2\theta(-\eta)+\sigma(\eta)q\left(\frac{5}{2},\beta{\delta e}\right)\right]}\nonumber\\ & & -\dfrac{\beta{\delta e}e^{-2\beta{\delta e}}}{\pi\left[2\theta(-\eta)+\sigma(\eta)q\left(\frac{5}{2},\beta{\delta e}\right)\right]^{2 } } + \cdots\;. \label{eq : hight - cv}\end{aligned}\ ] ] the high - temperature behavior of the heat capacity is shown in fig . [ fig:10 ] . for subcritical rotations we have @xmath252 , for critical rotations @xmath253 , and for supercritical rotations @xmath254 . for a fixed rotation , the heat capacity approaches in all cases @xmath255 as @xmath256 . for a large constant temperature , the heat capacity also approaches @xmath255 in the high rotation limit , @xmath257 . the most interesting feature of the high - temperature heat capacity is the dip found in the supercritical region . differentiation of with respect to @xmath1 shows that the minimum heat capacity occurs at @xmath258 . interestingly , the value of the minimum heat capacity , @xmath259 , is independent of temperature and rotation . the relation between the temperature of the system and the rotation @xmath260 corresponding to the minimum heat capacity can be written more clearly as @xmath261 the rotation @xmath260 at which the minimum heat capacity is located increases monotonically with temperature and has the asymptotic form @xmath262 . thus , although the minimum is present at any temperature , it occurs at increasingly high rotations so that the heat capacity flattens to @xmath255 as @xmath256 for any finite rotation . furthermore , the similarity of to and , which give the critical rotation for the transition to the zero temperature donut regime , suggests that the minimum of the heat capacity is related to the transition of the gas into a configuration where the particles are mostly contained in the minimum of the supercritical potential . recalling that @xmath263 is the potential height at the trap mininum , one should expect that the `` donut regime '' is realized when the thermal energy of the particles is of the order of @xmath263 . thus , setting @xmath264 should provide an estimate of @xmath265 . indeed , this gives @xmath266 which is not bad at all . and @xmath267 with . from eq . ( [ eq : cv - min ] ) , one obtains that the minimum of the heat capacity at @xmath267 is located at @xmath268.[fig:10],width=302 ] at high enough temperatures , the quantum mechanical effects may be neglected . thus , the high - temperature properties of systems in an anharmonic trap can also be explained in the context of classical statistical mechanics . to do so , one has to average the desired quantities over an ensemble . for a function @xmath269 of the phase space points such an average @xmath270 is given explicitly through @xmath271 where @xmath272 is again the classical hamiltonian . in this context , the classical virial theorem states that for @xmath42 particles in thermal equilibrium at temperature @xmath115 in a potential @xmath273 , the classical ensemble averaged potential energy @xmath274 and translational kinetic energy @xmath275 of a particle in the system are related according to @xcite : @xmath276 where @xmath277 are the components of the position vector of a particle and @xmath278 . for a cylinder symmetrical potential of the form @xmath279 , one has @xmath280 the anharmonic potential , as given in , can be decomposed according to @xmath281 with the terms @xmath282,\quad%\label{eq : pot-3}\\ v_{z}(z ) = \dfrac{\epsilon_{z}z^{2}}{2a_{z}^{2}}.\label{eq : pot-3b}\end{aligned}\ ] ] using , we obtain for the @xmath11-degrees of freedom the familiar harmonic potential result @xmath283 for the radial degrees of freedom we find correspondingly @xmath284 the total internal energy @xmath285 of a system of @xmath42 non - interacting particles in the potential @xmath273 is then given by @xmath286 before we consider the general result , let us make some remarks about eq . . by using the radial symmetry of the problem , one can express the average square radius through @xmath287 with the momenta @xmath288}.\ ] ] with the help of ref . @xcite , the limiting cases can be analytically explored . in the case of a pure quartic trap ( @xmath86 ) , the average radius is given by @xmath289 showing , as expected , that @xmath90 sets up the important length scale . for the case of a pure harmonic trap , i. e. @xmath290 , one has @xmath291 from , we conclude that the average position of the particles diverges as @xmath89 approaches @xmath292 . nevertheless , the energy per particle , as given from , remains @xmath293 . in general we obtain @xmath294\sqrt{\pi}}\right\}.\end{aligned}\ ] ] substituting this into we recover . we , thus , see that the classical statistical treatment of the system is equivalent to the high - temperature expansion which we have developed by keeping only the leading term in . we have considered non - interacting spin - polarized fermi particles in an anharmonic trap in the presence of a centrifugal potential due to rotation of the system . applying a semiclassical treatment to the density of states , we have worked out the temperature dependence of several thermodynamical properties . at absolute zero we have studied the density of particles and the momentum distribution for an anharmonically trapped system . the analysis of the particle density shows that a well defined geometry characterizes the ground state and has revealed a tendency for the particles to move away from the center . an analytical expression was obtained for the value of the rotation frequency @xmath0 at which a hole is created at the center of the trap , giving rise to a regime where the gas is ` donut'-shaped . the frequency @xmath0 was shown to depart from @xmath2 , which characterizes the stability limit for harmonic confinements , due to the interplay between the fermionic statistical interaction and the anharmonicity , in close analogy to repulsively interacting bosons . at finite temperature , expressions for the chemical potential , the internal energy , and the heat capacity were given . these were followed by discussions of their dependence on both anharmonicity and rotation frequency . at high - temperature , a virial theorem approach was presented and compared to previous results as well as shown to be particularly useful in the intuitive limits of pure harmonic and quartic radial confinements . the most natural extension of this work would be the inclusion of interactions . in the low - temperature regime , short - range interactions are strongly suppressed for fully polarized fermions . to overcome this problem , one could include other zeeman states by using an optical trap @xcite , yielding a spinor fermi gas . issues as ( anti-)ferromagnetism and magnetization dependence of the different quantities @xcite could be addressed in the presence of the additional quartic confinement . as for long - range interactions , no further zeeman states are necessarily required . due to its anisotropy , the dipole - dipole interaction is particularly interesting . atoms with non - zero nuclear spin are possible constituents for systems with large magnetic dipole moment @xmath295 . degenerate stable , for @xmath296cr ( @xmath297 ) @xcite , and meta - stable states , for @xmath298yb ( @xmath299 ) @xcite , are believed to be achievable . to date , only studies in two - dimensional systems consisting of a few atoms have already been performed @xcite , so that a three - dimensional many - body analog could be thought of . a further , and even simpler extension of the present work could be achieved by investigating the two - dimensional limit ( @xmath300 ) of the present system . by an appropriate tuning of the quartic constant , a quantum hall regime could be realized making possible the observation of underlying landau - levels through a plateau structure in the particle density when @xmath301 @xcite . for higher rotation frequencies , one would expect this structure to disappear and a hole to be created at a frequency value analogous to @xmath0 . we acknowledge financial support from the german academic exchange service ( daad ) and from the german research foundation ( dfg ) within the collaborative research center sfb / tr12 symmetries and universalities in mesoscopic systems .

a _ unit disk graph _ @xmath7 is a graph whose @xmath0 vertices can be mapped to points in the plane and whose @xmath8 edges are defined by pairs of points within euclidean distance at most @xmath9 from one another . alternatively , one can regard the vertices of @xmath7 as mapped to coplanar disks of unit diameter , so that two vertices are adjacent whenever the corresponding disks intersect . a _ dominating set _ @xmath10 is a subset of the vertices of a graph such that every vertex not in @xmath10 is adjacent to some vertex in @xmath10 independent set _ is a subset of mutually non - adjacent vertices . an _ independent dominating set _ is a dominating set which is also an independent set . note that any maximal independent set is an independent dominating set . dominating sets in unit disk graphs are widely studied due to their application in wireless ad - hoc networks @xcite . since it is * np*-hard to compute a minimum dominating set of a unit disk graph @xcite , several approximation algorithms have been proposed @xcite . such algorithms are of two main types . _ graph - based algorithms _ receive as input the adjacency representation of the graph and assume no knowledge of the point coordinates , whereas _ geometric algorithms _ work in the real ram model of computation and receive solely the vertex coordinates as input . if the coordinates of the @xmath0 disk centers are known , the @xmath8 edges of the corresponding graph @xmath7 can be figured out easily . it can be done in @xmath2 time under the real ram model with floor function and constant - time hashing , and in @xmath11 time without those operations @xcite . thus , for the price of a conversion step , graph - based algorithms can be used when the input is a unit disk realization of @xmath7 . however , unless * * p**=*np * , no efficient algorithm exists to decide whether a given graph admits a unit disk realization @xcite , let alone exhibit one . as a consequence , geometric algorithms can not be efficiently transformed into graph - based algorithms . in this paper , we introduce approximation algorithms of both types , benefiting from the same approximation factor analysis . the proposed graph - based algorithm runs in @xmath2 time , and the geometric algorithm runs in @xmath4 time regardless of @xmath8 . [ [ previous - algorithms ] ] previous algorithms + + + + + + + + + + + + + + + + + + + a graph - based @xmath1-approximation algorithm that runs in @xmath2 time was presented in @xcite . the algorithm computes a maximal independent set , which turns out to be a @xmath1-approximation because unit disk graphs contain no @xmath12 as induced subgraphs , as shown in that same paper . consists of a vertex with @xmath13 pendant neighbors . ] polynomial - time approximation schemes ( ptas ) were first presented as geometric algorithms @xcite and later as graph - based algorithms @xcite . also , a graph - based ptas for the more general disk graphs was proposed in @xcite . unfortunately , the complexities of the existing ptass are high - degree polynomials . for example , the ptas presented in @xcite takes @xmath14 time to obtain a @xmath1-approximation ( the analysis from @xcite ) . although its analysis is not tight , the running time is too high even for moderately large graphs . the reason is that these ptass invoke a subroutine that verifies by brute force whether a graph admits a dominating set with @xmath15 vertices . the verification takes @xmath16 time , and it is unlikely that this can be improved ( unless * * fpt**=*w[1 ] * , as proved in @xcite ) . such a subroutine is applied to several subgraphs , and the value of @xmath15 grows as the approximation error decreases . a similar strategy was used in @xcite to obtain a ptas for the minimum independent dominating set . the lack of fast algorithms with approximation factor less than @xmath1 was recently noticed in @xcite , where geometric algorithms with approximation factors of @xmath17 and @xmath18 and running times respectively @xmath19 and @xmath20 were presented . while a significant step towards approximating large instances , those algorithms require the geometric representation of the graph , and their running times are still polynomials of rather high degrees . linear and near - linear - time approximation algorithms constitute an active topic of research , even for problems that can be solved exactly in polynomial time , such as maximum flow and maximum matching @xcite . it is useful to contrast the minimum dominating set problem with the maximum independent set problem . while a maximal independent set is a @xmath1-approximation to both problems , it is easy to obtain a geometric @xmath18-approximation to the maximum independent set problem in @xmath4 time @xcite . in the graph - based version , a related strategy takes roughly @xmath21 time , though . no similar results are known for the minimum dominating set problem . the existing ptass for the minimum dominating set problem in unit disk graphs are based on some packing constraints that apply to unit disk graphs . one of these constraints is the _ bounded growth property _ : the size of an independent set formed by vertices within distance @xmath22 of a given vertex , in a unit disk graph , is at most @xmath23 . note , however , that the bounded growth property is not tight . for example , for @xmath24 , it gives an upper bound of @xmath25 vertices where the actual maximum size is @xmath1 . since the bounded growth property is strongly connected to the problem of packing circles in a circle , obtaining exact values for all @xmath22 seems unlikely @xcite . [ [ our - contribution ] ] our contribution + + + + + + + + + + + + + + + + our main result consists of the two approximation algorithms given in section [ s : algorithm ] : a graph - based algorithm , which runs in linear @xmath2 time , and its geometric counterpart , which runs in @xmath26 time in the real ram model , regardless of the number of edges . the approximation factor of both algorithms is @xmath3 . the strategy in both cases is to construct a @xmath1-approximate solution using the algorithm from @xcite , and then perform local improvements to that initial dominating set . our main lemma ( lemma [ l : irreducible ] ) uses forbidden subgraphs to show that a solution that admits no local improvement is a @xmath3-approximation . since the dominating sets produced by our algorithms are independent sets , the same approximation factor holds for the independent dominating set problem . proving that a certain graph is _ not _ a unit disk graph ( and is therefore a forbidden induced subgraph ) is no easy feat . we make use of an assortment of results from discrete geometry in order to prove properties of unit disk graphs that are interesting _ per se_. for example , we use universal covers and disk packings to show that the neighborhood of a clique in a unit disk graph contains at most @xmath27 independent vertices . these properties , along with a tighter version of the bounded growth property , are collected in section [ s : forbidden ] , and allow us to show that certain graphs are not unit disk graphs . consequently , the analyses of our algorithms employ a broader set of forbidden subgraphs which include , but are not limited to , the @xmath12 . additionally , in section [ s : partial ] , we show that a possible , somewhat natural refinement to our graph - based algorithm leads to a tighter @xmath5-approximation , albeit for the price of an extra @xmath28 multiplying factor in the time complexity of the algorithm . in this section , we introduce some lemmas about the structure of unit disk graphs . these lemmas will be applied to prove our approximation factors in sections [ s : algorithm ] and [ s : partial ] . we start by stating three previous results from the area of discrete geometry . the first lemma comes from the study of universal covers ( for a recent survey see @xcite ) . [ l : universal_cover ] if a set of points @xmath29 has diameter @xmath9 , then @xmath29 can be enclosed by a circle of radius @xmath30 . packing congruent disks in a circle is a well - studied problem . exact bounds on the radius of the smallest circle packing @xmath15 unitary disks are known for some small values of @xmath15 , namely @xmath31 and @xmath32 @xcite . the bound for @xmath33 will be useful to us . [ l : pack13 ] the radius of the smallest circle enclosing @xmath34 points with mutual distances at least @xmath9 is @xmath35 . the _ density _ of a packing is the ratio between the covered area and the total area . the following general upper bound is useful when no exact bound is known . [ l : density ] every packing of two or more congruent disks in a convex region has density at most @xmath36 . given a graph @xmath37 and a vertex @xmath38 , let @xmath39 denote the _ open neighborhood _ of @xmath40 and let @xmath41 = n(v ) \cup \{v\}$ ] denote the _ closed neighborhood _ of @xmath40 . more generally , the _ open @xmath22-neighborhood _ of a vertex @xmath40 is the set of vertices @xmath42 such that the distance between @xmath40 and @xmath42 in @xmath7 is exactly @xmath22 , while the _ closed @xmath22-neighborhood _ of a vertex @xmath40 is the set of vertices @xmath42 such that the distance between @xmath40 and @xmath42 in @xmath7 is at most @xmath22 . for a set @xmath43 , we let @xmath44 and @xmath45 = n[v ] \cap s$ ] . finally , given a subgraph @xmath46 of @xmath7 , the closed neighborhood of @xmath46 , denoted @xmath47 $ ] , is the set of vertices that belong to the closed neighborhood of some vertex of @xmath46 . the following two lemmas concern neighborhoods in unit disk graphs . [ l : clique ] the closed neighborhood of a clique in a unit disk graph contains at most @xmath27 independent vertices . by lemma [ l : universal_cover ] , the points which define a clique in a unit disk graph can be enclosed by a circle of radius @xmath30 . therefore , the points corresponding to the closed neighborhood of such a clique are enclosed by a circle of radius @xmath48 . by lemma [ l : pack13 ] , a circle enclosing @xmath34 points with mutual distances at least @xmath9 has radius at least @xmath35 . since @xmath49 , the lemma follows . [ l:2neighborhood ] given an integer @xmath50 , the closed @xmath22-neighborhood of a vertex in a unit disk graph contains at most @xmath51 independent vertices . all @xmath0 disks of diameter @xmath9 corresponding to the closed @xmath22-neighborhood of a vertex @xmath40 must be enclosed by a circle @xmath52 of radius @xmath53 centered on @xmath40 . each disk of diameter @xmath9 has area @xmath54 and @xmath52 has area @xmath55 . using lemma [ l : density ] , we have @xmath56 , and the lemma follows . we say that a graph @xmath7 is _ @xmath57-pendant _ if there is a vertex @xmath40 in @xmath7 with @xmath15 vertices of degree @xmath9 in the open neighborhood of @xmath40 and @xmath58 vertices of degree @xmath9 in the open @xmath59-neighborhood of @xmath40 . we refer to @xmath40 as a _ generator _ of the @xmath57-pendant graph . the following lemma bounds the value of the parameter @xmath58 for @xmath60-pendant unit disk graphs . [ l:4l - pendant ] if @xmath7 is a @xmath60-pendant unit disk graph , then @xmath61 . let @xmath40 be a generator of @xmath7 . since @xmath12 is a forbidden induced subgraph @xcite and @xmath40 has @xmath17 neighbors of degree @xmath9 , we have that the remaining neighbors of @xmath40 together with @xmath40 itself form a clique . by lemma [ l : clique ] , it follows that @xmath62 . next , we consider the case of @xmath63-pendant unit disk graphs . [ l:3l - pendant ] if @xmath7 is a @xmath63-pendant unit disk graph , then @xmath64 . let @xmath40 be a generator of @xmath7 . since two vertices are adjacent if and only if their euclidean distance is at most @xmath9 , the closed neighborhood of @xmath40 lies inside a circle of radius @xmath9 centered at @xmath40 . let @xmath65 be a neighbor of @xmath40 with degree @xmath9 . we divide the aforementioned circle into six congruent sectors @xmath66 in such a way that @xmath65 sits on the boundary of two adjacent sectors @xmath67 . since the diameter of each sector is @xmath9 and @xmath65 has degree @xmath9 , we have that @xmath68 only contains vertex @xmath65 . by the same argument , the remaining two neighbors of @xmath40 that have degree @xmath9 are contained in two of the remaining four sectors , which we call @xmath69 ( note that the sectors do not necessarily appear in the order @xmath66 ) . therefore , only sectors @xmath70 may contain the neighbors of @xmath40 that have degree greater than @xmath9 . notice that the neighbors of the vertices in @xmath71 should be located within euclidean distance at most @xmath9 from @xmath71 . consequently , the disks of unit diameter corresponding to such vertices should be completely contained in the region defined as the minkowski sum of @xmath72 and a disk of radius @xmath73 , i.e. the region within distance at most @xmath73 from @xmath72 . we compute upper bounds to the area of such a region by considering two cases , depending on whether @xmath74 and @xmath75 are opposite sectors ( bounded by the same pair of straight lines ) . first , we consider the case when @xmath74 and @xmath75 are _ not _ opposite sectors . in this scenario , @xmath72 is contained in a semicircle of radius @xmath9 centered at the generator @xmath40 , as represented in figure [ f : sectors](a ) . we define a region @xmath76 as the locus of points within distance at most @xmath73 from the aforementioned semicircle of radius @xmath9 . the area @xmath77 of @xmath76 is therefore @xmath78 . by lemma [ l : density ] , the number of disks of unit diameter contained in @xmath76 is at most @xmath79 since @xmath80 of these disks are the degree-1 neighbors of @xmath40 , we have @xmath81 . second , we consider the case when @xmath74 and @xmath75 are opposite sectors . in this scenario , the region @xmath72 is not convex . in order to obtain a convex region @xmath82 , we define @xmath82 as the locus of points within distance at most @xmath73 from the convex hull of @xmath72 ( see figure [ f : sectors](b ) ) . the area @xmath83 of @xmath82 is therefore @xmath84 . by lemma [ l : density ] , the number of disks of unit diameter contained in @xmath82 is at most @xmath85 since @xmath80 of these disks are the degree-1 neighbors of @xmath40 , we have @xmath86 . finally , the following lemma holds for the general case . [ l : kl - pendant ] if @xmath7 is a @xmath57-pendant unit disk graph , then @xmath87 . immediately from lemma [ l:2neighborhood ] with @xmath88 . in this section , we present two 44/9-approximation algorithms . the key property to analyze the approximation factor is presented in lemma [ l : irreducible ] , while the running time analyses are presented in sections [ s : graph - alg ] and [ s : geo - alg ] . hereafter , let @xmath37 be a unit disk graph , and let @xmath89 be an independent dominating set of @xmath7 . if @xmath90 and @xmath91 , we say that @xmath40 _ dominates _ @xmath65 and , conversely , that @xmath65 _ is dominated _ by @xmath40 . as already mentioned , unit disk graphs are free of induced @xmath12 . therefore , at most @xmath1 vertices of @xmath10 may belong to the closed neighborhood of any given vertex @xmath38 . corona _ is a set @xmath92 consisting of exactly @xmath1 neighbors of some vertex @xmath93 . such a vertex @xmath94 , which is not necessarily unique , is called a _ core _ of the corona @xmath95 , whereas the @xmath1 vertices of the corona are referred to as the corona s _ petals_. notice that the subgraph induced by a corona @xmath95 and a corresponding core @xmath94 is a @xmath96 . given a dominating set @xmath10 , a corona @xmath92 is said to be _ reducible _ if there is a core @xmath94 of @xmath95 such that @xmath97 is a dominating set . we refer to the operation that converts @xmath10 into the smaller dominating set @xmath97 as a _ reduction _ of @xmath95 with respect to @xmath94 . if there is no core allowing for a reduction of @xmath95 , than @xmath95 is dubbed _ irreducible_. if @xmath95 is an irreducible corona , then , for every core @xmath94 of @xmath95 , there must exist a vertex @xmath98 , such that : a. @xmath42 is not adjacent to @xmath94 ; b. @xmath42 is only dominated , in @xmath10 , by vertices that belong to @xmath95 . we call @xmath42 a _ witness _ of @xmath94 , conveying the idea that the corona having @xmath94 as a core can not be reduced with respect to @xmath94 due to the existence of @xmath42 . [ l : irreducible ] let @xmath37 be a unit disk graph , @xmath10 an independent dominating set in @xmath7 , and @xmath99 a minimum dominating set of @xmath7 . if all coronas in @xmath10 are irreducible , then @xmath100 . we use a charging argument to bound the ratio between the cardinalities of @xmath10 and @xmath99 . consider that each vertex @xmath101 splits a _ unit charge _ evenly among the vertices in the closed neighborhood @xmath102 $ ] . the function @xmath103}$ ] below corresponds to the total charge assigned to each vertex @xmath104 , accumulating the ( fractional ) charges that @xmath105 receives from the vertices in @xmath106 $ ] : @xmath107 } \frac{1}{\left|n_{d^*}[u]\right|}.\end{aligned}\ ] ] note that , since @xmath10 and @xmath99 are dominating sets , neither @xmath102 $ ] nor @xmath108 $ ] are ever empty , and @xmath109|$ ] . such function @xmath110 allows us to write the cardinality of @xmath10 as @xmath111 since @xmath112 is precisely the average value of @xmath113 over the elements of @xmath99 , we obtain the desired bound @xmath114 by showing that the existence of vertices @xmath115 in @xmath99 with @xmath116 is counterbalanced by a sufficiently large number of vertices @xmath117 in @xmath99 with @xmath118 . before we continue , note that @xmath116 means that @xmath119 , because the summation in ( [ eq : fv ] ) has at most @xmath1 terms , all of which are of the form @xmath120 for some integer @xmath121 . thus , let @xmath115 be a vertex in @xmath99 with @xmath119 . clearly , @xmath122 , otherwise @xmath123| = 1 $ ] , because @xmath10 is an independent set . moreover , @xmath115 must have exactly @xmath1 neighbors in @xmath10 , since a larger number of neighbors in @xmath10 would imply the existence of an induced @xmath12 in @xmath7 , which is not possible , and a smaller number would imply @xmath123| \leq 4 $ ] , a contradiction . therefore , vertex @xmath115 is a core . now let @xmath92 be the corona of which @xmath115 is a core . because there are no reducible coronas in @xmath10 , the core @xmath115 must have a witness @xmath42 . note that , for all petals @xmath124 , the only vertex in @xmath102 $ ] is the core @xmath115 . otherwise , the contribution of some @xmath124 in the summation yielding @xmath125 given by ( [ eq : fv ] ) would be at most @xmath126 , and @xmath125 would be at most @xmath127 , a contradiction . in particular , the above implies that the witness @xmath42 , which is adjacent to at least one vertex in @xmath95 , can not belong to @xmath99 . but @xmath99 is a dominating set , so there must exist a vertex @xmath128 that is adjacent to @xmath42 , and @xmath129 because a witness @xmath42 is not adjacent to the corresponding core by definition . we call @xmath117 a _ reliever _ of @xmath115 . figure [ f : witness ] illustrates this situation . we now show that @xmath130| \leq 4 $ ] . for sake of contradiction , assume @xmath130| > 4 $ ] . because @xmath7 is free of induced @xmath12 , such number must be exactly @xmath1 , so that @xmath117 is the core of a corona @xmath131 . such a corona must be disjoint from corona @xmath95 , otherwise there would be a vertex in adjacent to more than one vertex in @xmath99 , namely @xmath115 and @xmath117 , contradicting the fact that the only neighbor in @xmath99 of any petal of @xmath95 is the core @xmath115 . since , by definition , the witness @xmath42 is only dominated in @xmath10 by vertices of @xmath95 , we have @xmath132 . hence , @xmath133 is an independent set of @xmath7 , constituting , along with the core @xmath117 , an induced @xmath12 in @xmath7 . this is a contradiction , because @xmath7 is a unit disk graph . thus , @xmath130| \leq 4 $ ] . since @xmath134|$ ] , we have @xmath118 . we have just shown that the existence of a vertex @xmath115 in @xmath99 with implies the existence of a vertex @xmath128 such that @xmath118 . were this correspondence one - to - one , we would be able to state that the average of @xmath113 over the elements of @xmath99 was no greater than @xmath135 . unfortunately , this correspondence is not necessarily one - to - one , as exemplified by the graph in figure [ f : badgraph ] , for which a disk model is given in figure [ f : badgraph_model ] with coordinates presented in table [ t : coordinates ] . still , the lemmas in section [ s : forbidden ] allow us to bound the ratio between the number of vertices @xmath115 with @xmath119 and the number of vertices @xmath117 for which the values of @xmath110 are significantly lower . let @xmath136 be a reliever . in order to obtain the claimed bound , we consider two cases depending on the size of @xmath137 $ ] : \(i ) @xmath138| \leq 3.}$ ] by lemma [ l:2neighborhood ] , the closed @xmath17-neighborhood of @xmath117 contains at most @xmath139 independent vertices . since each corona contains @xmath1 independent vertices , at most @xmath140|)/5 \rfloor = 14 $ ] cores may share a common reliever . to derive an upper bound , let @xmath141 denote such cores . if @xmath130| \leq 3 $ ] , then the average value of @xmath113 among @xmath142 is at most @xmath143 \(ii ) @xmath138| = 4.}$ ] by lemma [ l:4l - pendant ] , if @xmath130| = 4 $ ] , then at most @xmath144 cores @xmath145 may have @xmath117 as their common reliever , for otherwise we obtain a @xmath146-pendant graph , which can not be a unit disk graph . thus , the average value of @xmath113 among @xmath147 is at most @xmath148 the worst case is @xmath130| = 4 $ ] , and therefore @xmath114 , concluding the proof . @xmath149 & @xmath150 & + @xmath151 & @xmath152 & + @xmath153 & @xmath154 & + @xmath155 & @xmath156 & + + @xmath157 & @xmath158 & @xmath159 + @xmath160 & @xmath161 & @xmath162 + @xmath163 & @xmath164 & @xmath165 + @xmath166 & @xmath167 & @xmath168 + by lemma [ l : irreducible ] , an independent dominating set with no reducible coronas is a @xmath3-approximation to the minimum dominating set . we now describe how to obtain such a set in linear time given the adjacency list representation of the graph . we can easily compute a maximal independent set @xmath10 , which is a @xmath1-approximation to the minimum dominating set @xcite , in @xmath2 time . an independent dominating set with no reducible coronas can then be obtained by iteratively performing reductions . however , naively performing such reductions leads to a running time of @xmath169 , since ( i ) there are @xmath170 candidate cores for a reducible corona , ( ii ) detecting whether a vertex @xmath40 is in fact the core of a reducible corona by inspecting the @xmath18-neighborhood of @xmath40 takes @xmath171 time , and ( iii ) we may need to reduce a total of @xmath170 coronas . fortunately , the following algorithm modifies the set @xmath10 and returns an independent dominating set with no reducible coronas in @xmath2 time . 1 . for each vertex @xmath172 , compute @xmath173 . 2 . for each vertex @xmath172 , if @xmath174 , add @xmath173 to the list of coronas @xmath175 ( unless it is already there ) . 3 . let @xmath176 . for each corona @xmath177 , if there is a vertex @xmath94 such that @xmath97 is a dominating set , then add @xmath94 to the set @xmath178 . 4 . choose a maximal subset @xmath179 of @xmath178 such that the pairwise distance of the vertices in @xmath179 is at least @xmath1 . 5 . for each vertex @xmath180 , perform a reduction @xmath181 . repeat all the steps above until @xmath182 . the algorithm is correct since all changes made to @xmath10 along its execution preserve the property that @xmath10 is an independent dominating set . notice that , in step ( 4 ) , we only reduce coronas that are sufficiently far from each other , in order to guarantee that we do not reduce a corona that may have ceased to be reducible due to a previous reduction . moreover , the algorithm always terminates because the size of @xmath10 decreases at every iteration , except for the last one . next , we show that the running time is @xmath2 . step ( 1 ) can be easily implemented to run in @xmath2 time . to execute step ( 2 ) in @xmath2 time , we must determine in constant time whether a corona is already in the list @xmath175 . this can be achieved by indexing each corona @xmath95 by an arbitrary vertex @xmath183 ( say , the one with the lowest index ) , and by storing with @xmath40 a list of coronas that are in @xmath184 and whose index is @xmath40 . note that , because of the packing constraints inherent to unit disk graphs , the number of coronas that contain a given vertex is constant . step ( 3 ) can be implemented as follows ( for each corona @xmath177 ) : 1 . let @xmath185 be the union of the open neighborhoods of the @xmath1 petals of @xmath95 . 2 . let @xmath186 be the set of vertices dominated only be vertices of @xmath95 , i.e. , @xmath187 contains every vertex @xmath188 such that @xmath189 . 3 . let @xmath190 be the intersection of the closed neighborhoods @xmath41 $ ] of all @xmath191 . 4 . if @xmath192 , then add an arbitrary vertex of @xmath190 to the set @xmath178 . the steps above take @xmath2 total time when executed for all coronas @xmath177 , because the number of coronas that contain or are adjacent to a given vertex is also constant by packing constraints . it is easy to perform steps ( 4 ) and ( 5 ) in linear time . it remains to show that the whole process is only repeated for a constant number of iterations . let @xmath193 denote the set of reducible coronas at each iteration of the algorithm with @xmath194 . note that the reductions performed in step ( 5 ) never create a new reducible corona . therefore @xmath195 . let @xmath95 denote a corona that was reduced in the last iteration @xmath15 . if @xmath95 was not reduced during a previous iteration @xmath196 , then another corona within distance @xmath1 from @xmath95 was reduced at that very iteration @xmath197 . since , again by packing constraints , the maximum number of coronas within constant distance from @xmath95 is itself a constant , we have @xmath198 . the following theorem summarizes the result from section [ s : graph - alg ] . [ thm : graph - alg ] given the adjacency list representation of a unit disk graph with @xmath0 vertices and @xmath8 edges , it is possible to find a @xmath3-approximation to the minimum dominating set problem in @xmath199 time . in this section , we describe how to obtain an independent dominating set with no reducible corona in @xmath4 time given the geometric representation of the graph . the input is therefore a set @xmath29 of @xmath0 points . without loss of generality , we assume that the corresponding unit disk graph is connected ( otherwise , we can compute the connected components in @xmath4 time using a delaunay triangulation @xcite ) . we use terms related to vertices of the graph and to the corresponding points interchangeably . for example , we say a set of points is independent if all pairwise distances are greater than @xmath9 . we want the points of @xmath29 to be structured in a suitable fashion . thus , as a preliminary step , we sort the points by @xmath200-coordinates and by @xmath201-coordinates separately ( such orderings will also be useful later on ) , and we partition the points of @xmath29 according to an infinite grid with unitary square cells by performing two sweeps on the sorted points . without loss of generality , we assume that no point lies on the boundary of a grid cell . given @xmath202 , let @xmath203 denote the grid cell that contains @xmath204 . we refer to the set of at most @xmath144 non - empty grid cells surrounding a cell @xmath205 as the _ open vicinity _ of @xmath205 , denoted @xmath206 , and to the union of @xmath205 and its open vicinity as the _ closed vicinity _ of @xmath205 , denoted @xmath207 $ ] . note that a point @xmath204 can only be adjacent to points in the closed vicinity of @xmath203 , that is , @xmath208 \subset n[\sigma(p)]$ ] . each point @xmath202 stores a pointer to its containing cell @xmath203 . also , each cell stores the list of points it contains and pointers to the cells in its open vicinity . since the graph is connected , the diameter of the point set is at most @xmath209 , and thus this whole step can be done in @xmath4 time . we are now able to show how to compute a maximal independent set @xmath10 efficiently . we begin by making a copy @xmath210 of @xmath29 , and by letting @xmath211 . then we repeat the two following steps while set @xmath210 is non - empty . ( i ) choose an arbitrary point @xmath212 and add it to set @xmath10 . ( ii ) for each point @xmath213 in the closed vicinity of @xmath203 , remove @xmath213 from @xmath210 if @xmath214 . when @xmath210 becomes empty , @xmath10 is an independent dominating set . this process takes @xmath170 time due to the two following facts . first , a cell belongs to the closed vicinity of a constant number of cells . second , the maximum number of points inside a cell with pairwise distances greater than @xmath9 is also a constant . we now have that @xmath10 is a maximal independent set , and therefore a @xmath1-approximation to the minimum dominating set . next , we show how to modify @xmath10 in order to produce an independent dominating set with no reducible corona , therefore a @xmath3-approximation to the minimum dominating set . the algorithm mirrors the one in section [ s : graph - alg ] , but each step takes no more than @xmath4 time using the geometric representation of the graph . since @xmath10 is an independent set and a grid cell @xmath205 has side @xmath9 , a simple packing argument shows that @xmath215 . we store the set @xmath216 in the corresponding cell @xmath205 . in order to compute @xmath217 , it suffices to inspect the at most @xmath218 points in @xmath216 for @xmath219 $ ] . we can then build a list of coronas in @xmath170 time ( steps ( 1 ) and ( 2 ) of section [ s : graph - alg ] ) . to perform step ( 3 ) , we need to find out whether there is a core @xmath94 such that @xmath97 is a dominating set , for each corona @xmath220 . first , we make @xmath185 the union of @xmath221 for @xmath222 . then , we let @xmath186 be the subset of @xmath185 containing only the points @xmath204 with @xmath223 . these first two steps are similar to steps ( 3a ) and ( 3b ) in section [ s : graph - alg ] . the remaining sub - steps of step ( 3 ) are significantly different , though . we proceed by making @xmath224 . we need to determine whether there is a point @xmath225 that is adjacent to all points in @xmath190 . for each @xmath225 , let @xmath226 denote the disk of radius @xmath9 centered at @xmath204 . let @xmath76 denote the convex region defined by the intersection of @xmath226 for all @xmath225 . a point @xmath204 is adjacent to all points in @xmath190 if and only if @xmath227 . we can compute the region @xmath76 in @xmath228 time using divide - and - conquer in a manner analogous to half - plane intersection @xcite . we can then test whether each point @xmath225 belongs to the region @xmath76 in logarithmic time using binary search ( remember the points were previously sorted ) . if there is at least one point @xmath229 , then we add @xmath204 to the set @xmath178 . therefore , the whole step ( 3 ) takes @xmath4 time . in step ( 4 ) of the geometric algorithm , we choose an alternative set @xmath230 which can be computed in @xmath170 time as follows . for each @xmath231 , we add @xmath204 to @xmath179 and then remove from @xmath178 all points that are contained in the cells within euclidean distance at most @xmath17 of @xmath203 . since by packing constraints there are @xmath232 points in the intersection of @xmath10 and the closed vicinity of a cell , we can easily perform step ( 5 ) in @xmath170 time . finally , the number of repetitions triggered by step ( 6 ) is constant by an argument identical to the one given for the graph - based algorithm . the following theorem summarizes the result from section [ s : geo - alg ] . [ thm : geo - alg ] given a set of @xmath0 points representing a unit disk graph , it is possible to find a @xmath3-approximation to the minimum dominating set problem in @xmath4 time in the real ram model of computation . in the previous section , a 44/9-approximation was obtained by reducing coronas of a maximal independent set @xmath10 of graph @xmath7 , that is , by subsequently replacing @xmath1 petals with @xmath9 core in @xmath10 as long as that operation preserved dominance . a natural step to tighten the approximation factor is to allow for _ weak reductions _ , whereby the @xmath1 petals of a corona @xmath95 are removed from the independent dominating set @xmath10 , yet not only is a core @xmath94 of @xmath95 inserted into @xmath10 but also some mutually non - adjacent witnesses of @xmath94 , as long as their number is no greater than @xmath18 and they dominate all witnesses of @xmath94 . by doing so , the weak reduction of @xmath95 ( with respect to @xmath94 ) preserves dominance and still shaves off at least one unit from the size of @xmath10 . if such operation is possible on a corona @xmath95 , then @xmath95 is said to be _ weakly reducible_. a core @xmath94 which has @xmath17 ( or more ) mutually non - adjacent witnesses is said to be an _ overwhelmed _ core . is an overwhelmed core of a corona @xmath95 , then it might be the case that a weak reduction on @xmath95 with respect to @xmath94 is still possible . if the subgraph @xmath233 $ ] induced by the set @xmath234 of witnesses of @xmath94 admits an independent dominating set @xmath235 of size no greater than @xmath18 , then @xmath236 is still an independent dominating set of @xmath7 , and its cardinality is strictly less than @xmath237 . however , one can not decide in ( close to ) linear time whether such a dominating set exists , and that will not be required by our algorithm . ] we consider the graph - based algorithm presented in section [ s : graph - alg ] with some modifications to cope with weak reductions . the whole modified algorithm becomes : 1 . for each vertex @xmath172 , compute @xmath173 . 2 . for each vertex @xmath172 , if @xmath174 , add @xmath238 to the list of coronas @xmath175 ( unless it is already there ) , and add @xmath40 to the list @xmath239 containing the cores of @xmath95 . 3 . let @xmath178 be an initially empty mapping of cores onto sets of witnesses . for each corona @xmath177 , if there is a vertex @xmath240 and an independent set @xmath241 with at most @xmath18 witnesses of @xmath94 such that @xmath242 is a dominating set , then add @xmath243 to @xmath178 . 4 . choose a maximal subset @xmath179 of the cores in @xmath178 such that the pairwise distance of the vertices in @xmath179 is at least @xmath1 . 5 . for each vertex @xmath180 , perform a ( weak ) reduction @xmath244 . 6 . repeat all the steps above until @xmath182 . the new step ( 3 ) can be implemented as follows ( for each corona @xmath177 ) : 1 . let @xmath185 be the union of the open neighborhoods of the @xmath1 petals of @xmath95 . 2 . let @xmath186 be the set of all vertices @xmath188 with @xmath189 . 3 . for each core @xmath240 , greedily obtain a maximal independent set @xmath241 of @xmath245 $ ] . if @xmath246 , add @xmath243 to @xmath178 and break . because each core is evaluated separately as to whether it allows for a weak reduction ( whereas in the algorithm of section [ s : graph - alg ] a constant number of set operations per corona was executed ) , step ( 3c ) dominates the complexity of the whole sequence of steps ( 1 ) to ( 5 ) , with a total @xmath247 time for running on all coronas @xmath177 . another important difference , as far as time complexity goes , is that in this modified algorithm the number of iterations of the main loop steps ( 1 ) to ( 5 ) is no longer @xmath232 , due to the fact that new ( weakly ) reducible coronas can be created , as illustrated in figure [ f : newcoronas ] . however , the number of iterations is certainly @xmath170 , because the size of @xmath10 decreases by at least @xmath9 in each iteration . hence the overall time complexity of the modified algorithm is @xmath6 . next , we establish the approximation factor of the modified algorithm . note that step ( 3c ) asserts that the coronas that are not ( weakly ) reduced by the algorithm present only overwhelmed cores . [ l : irreducible_partial ] let @xmath37 be a unit disk graph , @xmath10 an independent dominating set in @xmath7 , and @xmath99 a minimum dominating set of @xmath7 . if all coronas in @xmath10 have only overwhelmed cores , then @xmath248 . we follow a strategy similar to that in the proof of lemma [ l : irreducible ] ( also using the concept of reliever defined therein ) : by employing the same function @xmath103}$ ] defined in ( [ eq : fv ] ) , we prove there is an appropriate balance among vertices with high ( @xmath249 ) and low ( @xmath250 ) images under @xmath110 , thus yielding an average value for @xmath113 that is no greater than @xmath5 the claimed approximation factor . let @xmath115 be a vertex in @xmath99 with @xmath251 . because @xmath252 , and all ( at most @xmath1 ) terms in the summation that yields @xmath110 are either @xmath9 or no greater than @xmath253 , we have that @xmath251 implies @xmath119 . so @xmath115 is a core . by hypothesis , @xmath115 is overwhelmed . hence , @xmath115 possesses at least @xmath17 mutually non - adjacent witnesses @xmath254 , each one implying the existence of a reliever @xmath128 with @xmath118 . surely , such relievers need not be all distinct , and different cores may share a common reliever . still , geometric properties of unit disk graphs allow us to derive upper bounds for the core - to - reliever ratio . we can thus obtain an upper bound to the average value of @xmath113 in @xmath99 , and consequently to the approximation factor of the algorithm . suppose there are @xmath255 cores @xmath256 such that @xmath257 . for @xmath258 , we let @xmath259 be the corona having @xmath260 as a core , @xmath261 be a set of ( at least 4 ) mutually non - adjacent witnesses of @xmath260 , and @xmath262 be the set of relievers of @xmath260 . now , we construct a bipartite multigraph @xmath263 as follows . the parts of @xmath264 are @xmath265 and @xmath266 . the multiset @xmath267 contains , between each core @xmath268 and reliever @xmath269 , a number @xmath270 of parallel edges that is equal to the number of petals of @xmath259 adjacent to witnesses @xmath42 ( of @xmath260 ) such that @xmath42 is a neighbor of @xmath271 . thus , the total number of edges incident to a core @xmath268 is @xmath272 analogously , the total number of edges incident to a reliever @xmath269 is @xmath273 we now obtain an upper bound @xmath274 for the average value of @xmath113 over @xmath275 . observe that @xmath276 contains all vertices @xmath115 of @xmath99 such that @xmath251 , hence the average value @xmath277 of @xmath113 over the whole set @xmath278 can not be any greater . of course the average value we are interested in depends on the core - to - reliever ratio @xmath279 in @xmath264 : the more cores ( respectively , the fewer relievers ) in @xmath276 ( in @xmath82 ) , the greater the average . therefore , in order to obtain the desired upper bound @xmath280 , we must consider the case in which the elements of @xmath276 ( respectively , of @xmath82 ) have degrees in @xmath264 that are as low ( as high ) as possible . first , notice that if @xmath115 is a core of corona @xmath95 and @xmath119 , then it is not possible that more than two non - adjacent witnesses of @xmath115 sharing a common reliever @xmath117 are adjacent to the same petal @xmath281 . otherwise , let @xmath282 and @xmath283 be such witnesses . since @xmath204 and @xmath117 are non - adjacent ( due to the image of @xmath115 under @xmath110 being @xmath1 ) , the subgraph of @xmath7 induced by @xmath284 is a @xmath285 . this is a contradiction , because the graph @xmath285 is not a unit disk graph @xcite . thus , since @xmath115 has at least four witnesses , the number of petals of @xmath95 adjacent to witnesses of @xmath115 and therefore the degree of each core in @xmath276 is at least @xmath286 . the above lower bound can be improved for cores @xmath287 having a with exactly four neighbors in @xmath10 . if a reliever @xmath117 has four neighbors in ( the independent set ) @xmath10 , then the remaining neighbors of @xmath117 form a clique . since there is a witness @xmath42 of @xmath115 among these remaining neighbors of @xmath117 , then the other ( at least ) three mutually non - adjacent witnesses of @xmath115 must have relievers distinct from @xmath117 . moreover , by an argument analogous to the one used in the previous paragraph , those witnesses must be adjacent to at least @xmath288 petals of the corona having @xmath115 as a core . this means there are at least two edges connecting @xmath115 to its relievers in @xmath289 , plus one edge connecting @xmath115 to @xmath117 . hence , the degree of @xmath115 in @xmath264 is at least @xmath18 . the maximum degrees of vertices @xmath290 depend on the number @xmath291 of neighbors of @xmath117 in @xmath10 , and now we employ some of the geometric lemmas of section [ s : forbidden ] to infer suitable upper bounds . recall that any set of petals is an independent set . * if @xmath292 ( implying @xmath118 ) , then , by lemma [ l:4l - pendant ] , there are at most @xmath144 petals in the @xmath59-neighborhood of @xmath117 . hence , @xmath293 . * if @xmath294 ( implying @xmath295 ) , then , by lemma [ l:3l - pendant ] , there are at most @xmath296 petals in the @xmath59-neighborhood of @xmath117 . hence , @xmath297 . * if @xmath298 ( implying @xmath299 ) , then , by lemma [ l : kl - pendant ] , there are at most @xmath300 petals in the @xmath59-neighborhood of @xmath117 . hence , @xmath301 . * if @xmath302 ( implying @xmath303 ) , then , by lemma [ l : kl - pendant ] , there are at most @xmath304 petals in the @xmath59-neighborhood of @xmath117 . hence , @xmath305 . for @xmath306 , let @xmath307 denote the number of relievers in @xmath7 containing exactly @xmath15 neighbors in @xmath10 , so that @xmath308 . let also @xmath309 be the set of cores having at least one reliever with exactly @xmath17 neighbors in @xmath10,and @xmath310 be the set of cores whose relievers have at most three neighbors in @xmath10 . finally , we let @xmath311 and @xmath312 , so that @xmath313 . since @xmath264 is bipartite , the number of edges incident to @xmath276 and the number of edges incident to @xmath82 are the same . consequently , @xmath314 and @xmath315 the same reasoning holds for the subgraph of @xmath264 induced by the cores in @xmath316 and their relievers in @xmath82 , so @xmath317 by dividing both sides of ( [ eq:3 ] ) by @xmath59 and adding it to ( [ eq:2 ] ) , we obtain @xmath318 as for the average @xmath280 of @xmath113 over the elements of @xmath264 , we can write @xmath319 now , substituting @xmath320 in the expression above by the upper bound obtained from ( [ eq:4 ] ) , we have @xmath321 thus , @xmath280 is bounded by a multivariate rational function whose maximum is @xmath322 , achieved when @xmath323 . the following theorem summarizes the result from section [ s : partial ] . [ thm : graph - alg - partial ] given the adjacency list representation of a unit disk graph with @xmath0 vertices and @xmath8 edges , it is possible to find a @xmath5-approximation to the minimum dominating set problem in @xmath6 time . we introduced novel efficient algorithms for approximating the minimum dominating set and minimum independent dominating set in unit disk graphs . on one hand , a linear - time algorithm was devised attaining a sub-5 approximation factor , namely @xmath324 nevertheless , the best lower bound we know for the proposed algorithm is @xmath325 , which corresponds to the unit disk graph given in figure [ f : badgraph ] . closing this gap would likely require the development of new tools to prove that certain graphs are not unit disk graphs , for which computer generated proofs may be useful . on the other hand , an enhanced approximation factor of @xmath327 was obtained by allowing for more local replacements , yet a lower bound of @xmath328 , corresponding to the unit disk graph given in figure [ f : lowerbound425 ] ( with coordinates in table [ t : coordinates2 ] ) , is the best we are aware of . notwithstanding the @xmath28 factor increase on its time complexity , such a modified algorithm is still much faster than , say , the state - of - the - art 4-approximation algorithm from @xcite , which requires a geometric model as input and runs in @xmath329 time . moreover , since the number of ( weak ) reductions that are performed remains linear , it may be possible to conceive either a refined analysis or a smarter implementation . p. christiano , j.a . kelner , a. madry , d.a . spielman , s .- h . teng , electrical flows , laplacian systems , and faster approximation of maximum flow in undirected graphs , in : proc . 43rd annual acm symp . on theory of computing ( stoc ) , 2011 , pp . 273282 . t. erlebach , m. mihalk , a ( @xmath330)-approximation for the minimum - weight dominating set problem in unit disk graphs , in : proc . 7th workshop on approximation and online algorithms , vol . 5893 of lecture notes in computer science , 2010 , pp . 135146 . m. gibson , i. pirwani , algorithms for dominating set in disk graphs : breaking the @xmath331 barrier , in : proc . 18th annual european symposium on algorithms ( esa ) , vol . 6346 of lecture notes in computer science , 2010 , pp . 243254 . hunt iii , m.v . marathe , v. radhakrishnan , s. ravi , d.j . rosenkrantz , r.e . stearns , nc - approximation schemes for np- and pspace - hard problems for geometric graphs , journal of algorithms 26 ( 1998 ) 238274 . d. marx , parameterized complexity of independence and domination on geometric graphs , in : proc . 2nd international workshop on parameterized and exact computation ( iwpec ) , vol . 4169 of lecture notes in computer science , 2006 , pp . 154165 . f. zou , y. wang , x .- h . xu , x. li , h. du , p. wan , w. wu , new approximations for minimum - weighted dominating sets and minimum - weighted connected dominating sets on unit disk graphs , theoretical computer science 412 ( 3 ) ( 2011 ) 198208 .

the minimal supersymmetric standard model ( mssm ) is one of the most important candidates for the physics beyond the standard model . the strongly interacting superpartners , gluinos and squarks with mass lighter than 2.5 tev can be discovered at the large hadron collider at cern ( lhc ) @xcite . the physics run of lhc will start in 2008 . among the supersymmetric ( susy ) particles , the lightest susy particle ( lsp ) plays a key role . in cosmology , it is a candidate for the cold dark matter in the universe . at the lhc , the signals of the supersymmetric particles depends on the nature of the lsp . it may be the lightest neutralino @xmath4 , which escapes from detection and leads to missing @xmath5 in the event . another possibility for the lsp is the gravitino @xmath1 , the superpartner of the graviton . the gravitino coupling to other particles in the mssm sector is extremely small , namely proportional to @xmath6 . the next lightest susy particle ( nlsp ) , whose decay products necessarily include the gravitino , is therefore long - lived . we discuss the scenario where the lsp is the gravitino and the long - lived nlsp carries charge . a natural candidate for a charged nlsp ( cnlsp ) is the lightest scalar tau , @xmath7 , which can be significantly lighter than the other sleptons due to the left - right mixing induced by the off diagonal matrix element @xmath8 . the charged particle leaves tracks in the central detectors ( atlas and cms ) , giving additional information for the susy particle reconstructions . if the nlsp decays in the main detectors , a displaced vertex may be observed as well . the expected lifetime of the cnlsp @xmath9 is unconstrained , because it is proportional to @xmath10 the yet unknown gravitino mass squared . on the other hand , the gravitino mass is proportional to the total susy breaking scale in the hidden sector , therefore the determination of the lifetime is an important physics goal . the lifetime measurement gives direct information on the hidden sector . a particle decays efficiently in the main detectors ( cms and atlas ) if the decay length is sufficiently short , @xmath11 ( 10m)@xmath12 , where @xmath13 is the number of produced susy particles . on the other hand , for typical susy production cross sections , a direct observation of the decay is very difficult for @xmath14 sec . however , it has been pointed out that the cnlsp stopped by the ionization loss in the material may be studied in detail @xcite . an idea for a stopper based on a water tank is presented in ref.@xcite , where the water can be transported away from the detector site , for concentration and further study . in ref . @xcite , a detector consisting of a tracker and heavy stopping material is proposed , which can measure the arrival time and the location where the cnlsps are stopped , in addition to the energy of the decay products . another possibility , which requires minimal experimental modification , is to study the cnlsp which are stopped in the main detector or surrounding rock @xcite . as pointed out in ref . @xcite , the study of the cnlsp decays can probe the underlying supergravity in nature . with the gravitino mass inferred from the kinematics , the additional cnlsp lifetime measurement will test an unequivocal prediction of supergravity . moreover , the study of a rare 3body decay @xmath15 can reveal the peculiar couplings of the gravitino and the gravitino spin @xmath16 . in this paper , we consider the physics of the cnlsp @xmath9 decays that can be done with the heavy material stopper detector @xcite , because only a detector of this type can cover a wide range of lifetime @xmath17 @xmath18 years ) . we find that the mass of the gravitino can be measured if it is sufficiently heavy ( roughly @xmath19 ) . in that case , one can check if the lifetime is consistent with the supergravity interpretation . as we shall see , the lsp mass resolution is however poor if @xmath20 . in that case , it is very hard to prove that the decay @xmath21 plus an invisible particle @xmath22 is indeed caused by the supergravity interaction involving @xmath23 . we should note that , because of the extremely weak coupling of the gravitino , if there is any unknown supersymmetric particle @xmath22 which couples rather weakly to the mssm particles , the lightest mssm particle may decay into @xmath22 instead of @xmath1 even if the @xmath1 is the lsp . for example , the superpartner of the axion , the axino @xmath2 , can be such a particle . however , we found that it may be possible to discriminate the case of @xmath24 and the case of @xmath25 , by investigating the three body decay @xmath26 as suggested in @xcite , with enough statistics . this paper is organized as follows . in section [ sec : detector ] , we discuss a possible detector setup . we found an @xmath27 kton detector ( up to 8 kton ) may be placed next to the cms detector without serious modification of the cms experiment itself , but with non - negligible modifications to the cms cavern side walls . in section [ sec : modelpoints ] , we select several model points and estimate the expected number of the stopped particles for @xmath28 fb@xmath29 and 3000 fb@xmath29 . in section [ sec : twobody ] , we discuss the measurement of the two body decay @xmath30 . the three body decay of the cnlsp is studied in section [ sec : three - body ] . section [ sec : discussion ] contains the discussion and comments . 44mcms & 15 m & 12.5kt & 21 m [ atlascms ] in this section , we discuss the possibility to install massive stoppers next to the lhc detectors , cms and atlas . it turns out that the cms cavern may allow for an easier installation and more room for a massive stopper , compared to the atlas cavern . the parameters of the two detectors are listed in table . [ atlascms ] . the diameter of the cms detector is smaller , therefore the massive stopper can be placed closer to the interaction point at cms . the weight of the cms detector is about twice as large as that of atlas . because of the large weight , the cavern of cms is designed so that it can sustain a massive object safely , which includes a reinforced floor to spread the detector pressure more equally . a potential massive stopper with a weight of around a few kton can be placed on both sides of the cms detector , but it will need a reorganization of the scaffolding and gallery paths on the cavern walls , to make room for such an additional detector . the assembly and construction of the two detectors is also very different . most of the cms detector components are assembled on the ground , and about 15 large detector units will be lowered in the cavern for final assembly works . hence the installation is less integrated with the cavern , leaving relatively more freedom and thus changes needed in the cavern to install massive stoppers are somewhat simpler @xcite . on the other hand atlas detector is assembled mostly in its cavern . the huge magnets toroids and outside muon system fill up the cavern . the cryogenic system in the atlas cavern is also taking space outside the detector @xcite . is the collision point . right : two stopper detectors and a circle about the size of cms detector are superimposed on the cross section of cms cavern uxc 55 , drawing taken from ref . @xcite . , title="fig:",width=264 ] is the collision point . right : two stopper detectors and a circle about the size of cms detector are superimposed on the cross section of cms cavern uxc 55 , drawing taken from ref . , title="fig:",width=264 ] we assume two stoppers with the size 3.5 m @xmath31 15 m @xmath31 15 m and the average density @xmath32g/@xmath33 , hence , the total weight of the detector is 8 kton . this is maximum possible rectangular parallelepiped volume that can be placed in the cavern , with the given space to the cavern wall and with its long edge being the same as that of the cms barrel part ( see fig . [ fig : stopper ] ) . the stoppers are thus placed 8.5 m away from the interaction point . we also assume that stopping power of the cms detector is equivalent to @xmath34 g/@xmath35 iron , where @xmath36 is angle between the cnlsp direction and the beam direction . the number comes from the average density of cms detector , 3.37g/@xmath33 , which leads to the weight per @xmath35 for the radial direction of @xmath37 . as discussed in the previous paper @xcite , the stopper can be a hadronic and electromagnetic calorimeter simultaneously , if the detector consists of layers of dense stopper and tracking devices . the measurement of the energy of the decay product of the cnlsp is the key ingredient to explore the cnlsp interactions to the @xmath22 particle . in this paper we assume that the cnlsp is the scalar tau lepton @xmath0 , which decays mostly as @xmath38 where @xmath39 or @xmath2 . the @xmath40 decays into @xmath41 , or into @xmath42 and @xmath43 s . we do not consider the decays into @xmath44 , because the muon energy can not be measured unless the stopper contains a magnetic field . the energy of the leptons are much softer than the parent @xmath40 energy anyway , so that they are less useful for the study of the decay kinematics . a large volume detector is advantageous to measure the energy of the @xmath40 decay products , because the detector must contain most of the energy of the showers from the @xmath40 decay products . to fully absorb the hadronic cascade one needs sufficient thickness of the calorimeter . the required thickness for an iron calorimeter ( density is 7.87g/@xmath33 ) is listed in ref . @xcite , and is about 170 ( 120 ) cm for 100 gev single hadron energy for 99% ( 95% ) containment respectively , equivalent to 1337.9g/@xmath35 ( 944.4g/@xmath35 ) . to measure the energy with sufficient accuracy by a stopper with @xmath45 5g/@xmath33 , the distance between the decay position and the end of the detector along the shower path must be at least 190 cm . a simplified monte carlo simulation shows that 63% of the @xmath9 decays satisfy this condition . high energy photons and electrons initiate electromagnetic cascades , which are much easier to contain in a detector . the energy deposition along the axis of the cascade is well described by a gamma distribution , where the maximum occurs at @xmath46 , where @xmath47 is the radiation length . the energy deposition terminates around @xmath48 . ( see fig.27.17 of @xcite . ) for the case of iron ( @xmath49 ) , 200g/@xmath35 thickness of the material is needed to stop the electromagnetic cascade . the difference in the absorption length is used to discriminate isolated photons from hadron , which would be useful to study @xmath50 decay . in a previous paper @xcite , we discussed the possibility to re - use of the existing 1 kton detector , such as soudan ii @xcite as the cnlsp stopper . the soudan ii consists of the layers of @xmath51(m ) long drift tubes and thin iron plates . the physics goal of this detector is the search for proton decays . to be sensitive to the low energies involved in these decay processes , the average density of soudan ii is low , less than 2g/@xmath33 . the size of the detector is probably enough to stop a certain amount of cnlsps and measure the decay rate , and therefore it may well be appropriate for the a first stage of the cnlsp study . on the other hand , it is certainly not enough for a detailed energy measurement of the @xmath0 decay product . most of the hadronic decay cascades are not fully contained for a detector with the geometry allowed by the space in the cms cavern . a high density detector consisting of the layers of drift tubes / scintillators or rpcs to measure the charged particles between iron plates thicker than soudan ii will be more optimized for the cnlsp study . in this paper we assume a conservative energy resolution for hadronic showers , which is around @xmath52 . the value is not unrealistic for a simple massive and affordable detector , if a shower is sufficiently contained in the stopper . an additional complexity may occur for showers which develop parallel to the layers of the tracking devices . if the particles pass mostly through tracking devices , they feel a much lower average density , while the particles going mostly through the iron plates do not give detectable signature efficiently . we assume that the measured energy will be corrected depending on the shower directions . a detector uniform to all directions would be better to measure the energy of decay products of the stopped cnlsp . the calorimetery technologies studied for the ilc may satisfy such conditions ; see @xcite for the cnlsp study at ilc . in this section , we briefly describe supersymmetric models with a long - lived charged next lightest susy particle . we also select some model points , and estimate at each model point the number of cnlsp that can be stopped at the stopper detector . in the minimal supergravity ( msugra ) models , the scalar masses and gaugino masses , trilinear couplings are universal at the gut scale @xmath53 which are denoted by @xmath54 , @xmath55 , and @xmath56 respectively . the resulting mass spectrum is obtained by solving renormalization group equations ( rge s ) . when the rge is integrated up to @xmath57 , slepton soft masses and gaugino masses are approximately given by the following convenient formulas @xmath58 where @xmath59 is the gauge couplings at the unification scale.(30% ) if the mass scale is @xmath51(1tev ) . ] in this model , the mass of gravitino @xmath60 is order of @xmath54 or @xmath55 , because both of them are proportional to the @xmath61 with @xmath27 coefficient , where @xmath62 is the fundamental susy breaking scale . depending on the @xmath27 coefficients , the gravitino can be the lightest superparticle in general . also in the gaugino mediation models @xcite , the gravitino can be the lsp in a large domain of the parameter space @xcite . in this case the scalar masses are very small at the boundary , i.e. , @xmath63 , and the stau naturally becomes the nlsp . in the gauge mediation ( gm ) model @xcite the supersymmetry is broken at lower energy scale , and the susy breaking in the hidden sector is mediated to the mssm sector by gauge interactions . the simplest gm model is described by 6 parameters : @xmath64 , @xmath65 , @xmath66 , @xmath67 , sgn@xmath68 and gravitino mass . here @xmath66 is an effective number of vector - like heavy quarks and leptons in su(5 ) representations , @xmath69 and @xmath70 . ( @xmath71 for @xmath72 quarks and leptons , and @xmath73 for @xmath74 . ) a messenger field @xmath75 couples to the vector - like fields as @xmath76 and develops a vacuum expectation value @xmath77 . the gravitino mass is given by @xmath78 where @xmath79 is the total susy breaking scale of the theory . we take @xmath60 as a free parameter in this model . the masses of mssm particles at @xmath65 are obtained by relatively simple formula . gaugino masses satisfy the gut relation , and @xmath80 squark and slepton masses are given by @xmath81 n_5 \lambda^2 , \cr m^2_{\tilde{u}_r } & = & \left [ { 8\over 3 } \left({\alpha_s\over 4\pi}\right)^2 + { 8\over 15 } \left({\alpha_1\over 4\pi}\right)^2 \right ] n_5 \lambda^2 , \cr m^2_{\tilde{d}_r } & = & \left [ { 8\over 3 } \left({\alpha_s\over 4\pi}\right)^2 + { 2\over 15 } \left({\alpha_1\over 4\pi}\right)^2 \right ] n_5 \lambda^2 , \cr m^2_{\tilde{\ell}_l } & = & \left [ { 3\over 2 } \left({\alpha_2\over 4\pi}\right)^2 + { 3\over 10 } \left({\alpha_1\over 4\pi}\right)^2 \right ] n_5 \lambda^2 , \cr m^2_{\tilde{e}_r } & = & \left [ { 6\over 5 } \left({\alpha_1\over 4\pi}\right)^2 \right ] n_5 \lambda^2 . \label{gmeq}\end{aligned}\ ] ] all of the above models predict a large mass hierarchy between strongly interacting superpartners and weakly interacting superpartners . heavy gluino and squarks are copiously produced at the lhc , and they decay into the light weakly interacting susy particles . the lightest susy particle in the mssm sector is either the lighter stau @xmath7 or the lightest neutralino @xmath4 . the mass of the @xmath7 is the smaller eigenvalue of the mass matrix , @xmath82 because of the off - diagonal elements of @xmath0 mass matrix , @xmath7 could be significantly lighter than the other sleptons . if the @xmath7 is the nlsp , the stopper detector is useful to stop it and to study its decay . we therefore consider the phenomenology when @xmath83 in this paper . in the following , we omit the subscript of @xmath7 and denote the nlsp stau as @xmath0 . in most part of this paper , we will discuss the cnlsp physics as model independent as possible . however , to give some numerical reference , we choose a few model points . for msugra models , we take the points proposed in @xcite . the parameters are listed in table . [ masssugra ] and mass spectrum is given by the isajet ver . 7.69 . in @xcite , the gravitino mass is taken to be @xmath84 in those model points . @xmath85 @xmath55 & 440 & 1000 & 1000 @xmath54 & 20 & 100 & 20 @xmath67 & 15 & 21.5 & 23.7@xmath86[gev ] & 1025 & 2191 & 2190 @xmath87[gev ] & 175 & 417 & 416@xmath88[gev ] & 154.2 & 343.5 & 324.3 @xmath89(susy)[pb ] & @xmath90 & @xmath91 & @xmath92 stopped in the stopper detector per @xmath93 events & 255 & 250 & 254 stopped for 300 fb@xmath29 & 4636 & 34 & 36 @xmath86[tev ] & 0.93 & 1.13 & 1.34 & 1.54 & 1.74 @xmath87[gev ] & 161.7 & 205.3 & 248.7 & 292.1 & 335.4@xmath88[gev ] & 120.5 & 150.1 & 179.9 & 209.8 & 239.9@xmath89(susy)[pb ] & 5.24 & 1.68 & 0.64 & 0.28 & 0.13 stopped in the stopper detector per @xmath93 events & 282&274 & 274 & 294 & 302stopped for 300fb@xmath29 & 8830 & 2762 & 1052 & 494 & 236 for the gauge mediation model , we take the model points similar to that for the study in @xcite . namely , we fix @xmath73 , @xmath94 and @xmath95 , where @xmath96 tev . the mass spectrum and production cross sections are summarized in table . the gravitino mass for these model can be very small , whose minimum value is given by @xmath97 . if the messenger sector couples a fraction of total susy breaking , i.e. @xmath98 , the gravitino mass , which is proportional total susy breaking @xmath62 , can be large . the pattern of the mass spectra are similar for both msugra and gm sample points . the gaugino masses obey the gut relation , and @xmath0 mass is lighter than @xmath99 mass because of the cnlsp assumption . for msugra , this implies @xmath100 , so that @xmath101 mass is about as large as @xmath102 , as it is so in gm models . the relation @xmath103 is realized in our model points . to estimate the number of stopped cnlsp , the production cross section of the susy particles and the velocity distribution of @xmath0 at lhc must be evaluated . they depend on the @xmath102 and @xmath101 masses , and the mass difference @xmath104 . we estimate the production cross section and the decay distribution by using herwig@xcite , where the mass spectrum and branching ratios are interfaced from isajet@xcite to herwig by using isawig@xcite . we generated @xmath93 susy events for each model point . the flying range @xmath105 of the charged stable massive particle may be calculated by integrating the energy loss equation of heavy ionizing particle ( bethe - bloch equation ) . the result is a function of @xmath106 , with a linear dependence on the mass of the particle @xmath107 . in this paper , the stopping power of the stopper detector is calculated using the data in @xcite . see also the detailed discussion in ref . @xcite . the maximum length of a particle track through the stopper detector @xmath108(max ) , which depends on cnlsp direction , is calculated assuming that the track of the cnlsp goes straight from the production point . we regard the particle is stopped in the stopper detector if the flying range @xmath105 satisfies : @xmath109 where @xmath110g@xmath111@xmath33 is the density of the stopper detector . the number of stopped cnlsp in a stopper per @xmath93 events , and the number of stopped cnlsp for @xmath112 in the two stoppers , are listed in table [ masssugra ] and [ mass ] . in fig . [ fig : position ] we also show the simulated stopped positions in a stopper detector . the position distribution is uniform in the detector . tev gm point . here the @xmath113 axis is the beam direction , and the @xmath114 axis is the vertical direction . the origin @xmath115 is the interaction point and we assume the stopper detector is at 8.5m@xmath11612 m , @xmath117m@xmath1187.5 m , and @xmath117m@xmath1197.5 m ( cf . fig . [ fig : stopper ] ) . the big square is the projection on the @xmath114-@xmath113 plane , the top rectangle is the projection on the @xmath120-@xmath113 plane , and the right rectangle is the projection on the @xmath120-@xmath114 plane . , width=453 ] distribution of @xmath0 for @xmath121 tev and @xmath122 tev in gm models and for point @xmath123 in msugra models for @xmath93 susy events . light ( dark , gray ) shaded histograms are the number of stopped events in the stopper scaled by factor of 10 , for point @xmath123 and for gm models with @xmath124 tev respectively . , width=264 ] in fig . [ fig : betagamma ] , we show the @xmath125 distribution of the cnlsps for a few model points . we find that cnlsp tends to be less relativistic when the susy scale is large , because momentum of parent squarks and gluino is reduced . in fig . [ fig : betagamma ] , the peak position of the @xmath125 is at @xmath126 for @xmath127 tev . because of that , the number of cnlsp in the smaller @xmath125 region is increased as the susy scale is increased . on the other hand , as @xmath128 increases , cnlsps with smaller @xmath125 are stopped in the detector while the number of events in the smaller @xmath125 bins are kinetically suppressed . for instance , for @xmath122 and 80 tev , cnlsps in the bins between @xmath129 and @xmath130 are stopped in the stopper detector , respectively . altogether , the number of stopped @xmath0 for @xmath131 to @xmath132 tev for @xmath93 susy events is roughly constant as we can see in table . [ mass ] . the production cross section reduces when the gluino mass is increased because the parton distribution functions of gluon and quarks are quite small for @xmath133 . for 300fb@xmath29 , the number of events stopped at the assumed two 4 kton stoppers ranges from 8000 events to 30 events in the table . we will see in the next section that accumulation of @xmath134 cnlsps are enough to measure the lifetime with @xmath51(10)% accuracy . we will also estimate the resolution of @xmath40 lepton energy @xmath135 arising from the decay @xmath136 through the end point of the tau jet energy . statistically @xmath137 stopped cnlsps are enough to measure the end point with a few gev error . in the upgrade of lhc ( slhc ) , integrated luminosity of 3000fb@xmath29 is proposed , therefore the number of stopped events ranges from @xmath138 to @xmath139 for the model points presented in table [ mass ] . in this section we study the two body decay of the cnlsp in the stopper detector . both in msugra and gm models , the stau can be the cnlsp and decays into the gravitino @xmath1 and the @xmath40lepton . the cnlsp decay width into a gravitino and a lepton is given by @xcite @xmath140^{3/2}\;. \nonumber\\ & = & ( 68~\mathrm{days})^{-1 } \left(\frac{{m_{\tilde{\tau}}}}{100~\mathrm{gev}}\right)^5 \left(\frac{10~\mathrm{gev}}{{m_{\tilde{g}}}}\right)^2 \times \nonumber\\ & & \left(1- \frac{{m_{\tilde{g}}}^2 + m_\tau^2 } { { m_{\tilde{\tau}}}^2 } \right)^4 \left [ 1-\frac{4 { m_{\tilde{g}}}^2 m_\tau^2 } { ( { m_{\tilde{\tau}}}^2-{m_{\tilde{g}}}^2-m_\tau^2)^2 } \right]^{3/2}\;. \label{eq : gamma32}\end{aligned}\ ] ] we show the dependence of the stau lifetime on the gravitino mass in fig . [ fig : lifetime ] . in the case of gravitino lsp , as a function of the gravitino mass , for the stau mass 100 , 200 , and 400 gev ( from top to bottom ) . ] in general , the two body decay @xmath141 can be triggered by a single tau jets initiating from the position where @xmath0 is stopped . the tau energy is monochromatic and expressed as @xmath142 here @xmath22 is the invisible particle in the @xmath0 decay , in this case @xmath39 . if one can measure both of the lifetime and the mass of the stau , the gravitino mass can be determined _ assuming _ that eq.([eq : gamma32 ] ) is correct . then the total susy breaking scale @xmath143 is also determined , which is very important to understand the hidden sector physics . the lhc main detectors can determine the mass of @xmath0 through the stau velocity measurement @xmath144 , e.g. , in the muon system of the cms detector . however , the measurement of the lifetime may not be easy at the main detectors if the lifetime is too much longer than the detector size . the cross section is typically @xmath27 pb or less , therefore we have at most @xmath145 event at hand for @xmath146 . thus , only 10 events or less decay inside the detector if @xmath147 km . some of the cnlsp are stopped in the main detector , but measuring the decay precisely in the main detector would be challenging during the beam time . on the other hand a stopper detector @xcite can measure the position and the time where a cnlsp is stopped , and its decays without dead - time . the lifetime measurement will be discussed in sec.[subsec : lifetime ] . see @xcite also on the idea to measure the lifetime by triggering muons from the decays of the cnlsp stopped in the surrounding rock . to _ predict _ the cnlsp lifetime , one has to determine the gravitino mass independently . this is possible through the extraction of @xmath135 from the energy distribution of the tau jet from the cnlsp decay , because @xmath135 is a function of @xmath148 and @xmath149 , as can be seen in eq . ( [ etau ] ) . @xmath149 is expressed as a function of @xmath135 as follows ; @xmath150 fig . [ fig : etau ] shows the dependence of the reconstructed lsp mass eq.([eq : mx ] ) on the tau energy @xmath151 for several values of @xmath148 . as can be seen from the figure , it is crucial to measure the @xmath151 as precise as possible , especially for small @xmath149 , in order to determine the mass @xmath149 . hence , the stopper detector should offer a reliable measurement of @xmath135 . we will discuss the @xmath152 measurement and the @xmath149 reconstruction in sec . [ subsec : etau ] and sec . [ subsec : mx ] , respectively . [ see eq.([eq : mx ] ) ] on the tau energy @xmath151 , for @xmath153 , 150 , 200 , 250 , and 300 gev ( solid lines ) . dashed lines show contours of the lifetime in the case of the gravitino lsp ( @xmath24 ) . ] one can study the supersymmetric version of gravity interaction by studying the consistency between the observation and prediction of the decay rate . if the gravitino is the lsp , the decay width is given by eq.([eq : gamma32 ] ) . now if one can independently determine the gravitino mass by means of kinematics , as described above , eq.([eq : gamma32 ] ) can be used in the other way round , which leads to the measurement of the supergravity planck scale @xcite @xmath154^{3/2}\;. \label{eq : mp}\end{aligned}\ ] ] comparison of the obtained @xmath155 with the planck scale of the einstein gravity @xmath156 @xcite would be a crucial test of the supergravity . prospects of the `` planck scale '' measurement will be discussed in sec.[subsec : mpl ] . it should be noted that the undetectable particle @xmath22 may not be the gravitino @xmath1 . any particle which couples weakly to @xmath0 can be particle @xmath22 . if the decay width into @xmath157 is larger than the decay width into @xmath158 , i.e. @xmath159 , , i.e. , @xmath160 . ] the cnlsp lifetime may be different from the supergravity prediction obtained from measured @xmath148 and tau jet energy distribution . inconsistency between the measured and predicted lifetime immediately means a discovery of a new sector that may not be accessible otherwise . one of well motivated examples of such a non - sugra decay is @xmath161 where @xmath2 is the axino , superpartner of the axion . the cnlsp @xmath0 decay into axino is studied in ref . @xcite for hadronic , or ksvz axion models @xcite . in this paper we adopt the set up in ref . @xcite for the axino interaction , which we briefly describe here . in ksvz axion model , the peccei - quinn ( pq ) mechanism @xcite is realized in an action with new heavy quarks . when these heavy quarks are integrated out , anomalous terms involving the axion and gauge bosons are generated at low energy effective action . when the axion interaction is supersymmetrized , its fermionic superpartner , the axino @xmath2 , must be introduced . the axino mass @xmath162 can range between the ev and gev scale depending on the model and susy breaking scheme @xcite , and we regard it as a free parameter in this paper . the coupling of the axino to the bino and the photon/@xmath163-boson at the scale below the peccei - quinn scale @xmath164 is given by the lagrangian @xmath165\tilde{b}(\cos\theta_wf_{\mu\nu } -\sin\theta_wz_{\mu\nu}).\ ] ] . , width=340 ] the action does not contain direct @xmath166 coupling and also strongly suppressed by the pq scale @xmath167 . the two body decay @xmath161 is induced by the one loop diagram shown in fig.[fig : twobody ] . the loop integral has a logarithmic divergence . this is because the effective vertex ( [ eq : axinobino ] ) is applicable only if the momentum is smaller than the heavy ( s)quark masses , whereas the loop momentum exceeds that scale . in the full theory , one calculates two loop diagrams with the heavy ( s)quarks , which leads to a finite result @xcite . here we regulate the logarithmic divergence with the cut off @xmath168 @xcite , and the effective @xmath169 coupling is parameterized as @xmath170 here , @xmath171 and we take @xmath172 hereafter . the parameter @xmath173 is an order one parameter to represent the uncertainty coming from the cut off procedure mentioned above . in this paper , we regard this as a free parameter . the two body decay width is given as @xcite @xmath174 in this subsection , we estimate the statistical error in the cnlsp lifetime measured by the stopper detector . the analysis is model independent and applicable to any long - lived charged particle stopped in the stopper . for each cnlsp stopped in the stopper detector , the stopping time @xmath175 and the decaying time @xmath176 will be recorded . the lifetime is then measured by fitting the temporal distribution of the decaying events @xmath177 , where @xmath178 . here , we use a maximum likelihood fitting and adopt the following procedure : first we define a time @xmath179 such that @xmath180 and @xmath181 , where @xmath182 is total number of stopped event . is defined by @xmath183 , where @xmath184 and @xmath185 is the decay time of the @xmath186-th event : @xmath187 . ] ( for large @xmath182 , this @xmath179 is already a good estimator of the lifetime . ) we then calculate the @xmath188 distribution as a function of lifetime @xmath40 : @xmath189 where @xmath190 is the poisson distribution , @xmath191 is the number of events in the @xmath192-th bin , and @xmath193 is the predicted average number of events in that bin : @xmath194 here , we take 1 bin @xmath195 . an example of the time distribution and the best fit curve for mc distribution are shown in fig . [ fig : lifetime_measure100 ] for @xmath196 . for @xmath196 . in the left plot , the solid line is the generated events , and dashed line is the best fit . ] the @xmath197 confidence interval can be estimated by the range @xmath198 . [ fig : lifetime_measure100 ] shows that the error of the lifetime is @xmath199 for @xmath196 . we have done the same analysis for @xmath200 and found @xmath201 . so far , we assumed that most of the stopped cnlsps decay within the experimental time scale . we now consider the case where the lifetime is longer . suppose that one observes @xmath200 stopped cnlsps and only 10 events decaying within 1 year . in such a case , the lifetime is estimated from the number of the decaying events . for instance , 95% interval of the mean @xmath202 of poisson variable for @xmath203 is @xmath204 $ ] @xcite . using @xmath205 , a 95% interval @xmath206 years is obtained . for much longer lifetime or smaller @xmath182 , only a lower bound of the lifetime is obtained . we have assumed that the background is negligible . the background from cosmic neutrino and hard neutrino produced from the main detector interaction point is small ( cf . @xcite ) , however careful study on the accidental background is necessary when statistics is low . this is beyond of the scope of this paper . in this subsection , we estimate the uncertainty of the tau energy determination . then , in sec.[subsec : mx ] we discuss the kinematical reconstruction of the lsp mass . schematically , the procedure is as follows : @xmath207 when @xmath0 decays into @xmath40 and invisible particle @xmath22 , the tau energy @xmath135 is monochromatic ( see eq.([etau ] ) ) . @xmath151 can be obtained by fitting the @xmath40 jet energy distribution : @xmath208 among the decay products of the @xmath40 lepton we omit the neutrinos and muons . in order to see the prospects of @xmath151 measurement with a finite number of events , we generate the events from @xmath40 decay by using the tauola @xcite , and we perform a maximum likelihood fitting of low statistics ( `` experimental '' ) event sets by high statistics ( `` theoretical '' ) distributions . in practice , the observed jet energy distribution depends not only on the @xmath151 , but also on the detector resolution @xmath209 and the tau polarization @xmath210 : @xmath211 to obtain theoretical predictions for @xmath212 , we have generated tau decay events with high statistics run of tauola for the parameter space @xmath213 30 125 gev and @xmath214 , while fixing @xmath215 @xmath216 @xmath217 . events for each of the parameter sets @xmath218 30 , 31 , 32 , ... 109 , 125 gev and @xmath219 ( i.e. , @xmath220 parameter sets @xmath221 events ) , and interpolated the distribution between those parameter points . ] for each single @xmath40 decay generated by tauola , the jet energy @xmath222 is calculated , and then smeared by a gaussian fluctuation with a variance @xmath223 . , 70 , 90 gev from left to right . the dashed lines show the spectrum without energy resolution effect , and the solid lines show the spectrum with detector energy resolution @xmath224 @xmath225 . ] fig [ fig : taujet ] shows examples of the jet energy distribution . in order to see a realistic situation , we generate a small number of tau lepton decays for a fixed parameter set of ( @xmath151 , @xmath210 , @xmath226 ) , and then fit the result by the `` theoretical '' distribution obtained above . here and hereafter , we assume that the energy resolution will be known in advance . examples are shown in fig . [ fig : lnlfit:1000 ] and fig . [ fig : lnlfit:100 ] for total number of decaying tau leptons @xmath227 and @xmath228 , respectively . the energy of @xmath40 jets which are not contained in the detector may not be measured precisely . the @xmath229 corresponds to the number of well contained events . gev , @xmath230 and @xmath231 ( solid histogram ) , and the best fit distribution ( dotted histogram ) . only the bins between the vertical lines are used for the fit ( see text ) . ( b ) @xmath232 projected onto the @xmath151 axis . ( c ) contour plots of @xmath233 , 4 , 9 projected onto the ( @xmath210 , @xmath151 ) plane . ] [ fig : lnlfit:1000 ] but with 100 decaying tau leptons . ] [ fig : lnlfit:100 ] some comments are in order here . ( i ) we take 1 bin = 10 gev . ( ii ) for the maximum likelihood fitting we take only the bins with number of events @xmath234 , and the bins above the peak energy , which in the examples of fig . [ fig : lnlfit:1000](a ) and [ fig : lnlfit:100](a ) correspond to the bins between the two vertical lines . ( iii ) we then calculate the @xmath188 distribution in the parameter space of ( @xmath151 , @xmath210 ) as follows @xmath235 where @xmath190 is the poisson distribution , @xmath236 is the number of events in the @xmath192th bin for the low statistics run , and @xmath237 is the predicted number of events in the @xmath192th bin as a function of the parameter set @xmath238 , normalized by the total number of decaying tau leptons . here , we assume that the total number of stopped @xmath0 is known , i.e. , the total number of events is _ not _ taken as a free parameter for the fit . the @xmath197 confidence interval can be estimated by the range @xmath198 projected onto the @xmath151 axis [ see figs . [ fig : lnlfit:1000](b ) and [ fig : lnlfit:100](b ) ] . one can see that the primary tau energy can be determined within an error of a few gev . from fig . [ fig : lnlfit:1000](c ) one can also see that the polarization is hardly determined by the energy distribution analysis even with @xmath239 . this is because the sensitivity to the polarization becomes very weak once the finite detector resolution is taken into account ( cf . [ fig : taujet ] ) . to estimate the statistical error in @xmath135 measurement we have generated the event sets with same statistics @xmath240 @xmath241 and repeat the fit on @xmath135 and @xmath242 to obtain the best fit value of the @xmath135 , @xmath243 . in figure [ fig : deletau ] , we show the distribution of @xmath243 for @xmath244 gev , @xmath245 , and @xmath246 . for @xmath247 , 70 and 90 gev ( from left to the right ) , for @xmath239 ( left ) and @xmath248 ( right ) . ] from the variances of these distributions , we estimate the @xmath249 uncertainty of @xmath151 as @xmath250 now we can estimate the uncertainty of the lsp mass . in fig . [ fig : mgmg ] we plot the range of reconstructed lsp mass @xmath251 where @xmath252 and @xmath253 . uncertainty of @xmath151 , for @xmath239 and 100 , and @xmath254 and 300 gev . solid lines represent the case without an error of @xmath148 , and dashed lines include 1% uncertainty of @xmath148 . vertical dotted lines represent the stau lifetime of 1 month , 1 year , and 10 years in the case of gravitino lsp ( @xmath39 ) . ] note that the reconstructed lsp mass depends on not only @xmath151 but also the measured cnlsp mass @xmath255 . in fig . [ fig : mgmg ] we show the range of lsp mass for @xmath256 ( solid lines ) and @xmath257 ( dashed lines ) , the latter corresponding to 1% uncertainty of the cnlsp mass . the stau mass determination from time of flight was discussed in @xcite . for cms detector the mass resolution is estimated as 1020% in each event and less than 1% for @xmath258 events . as can be seen in fig . [ fig : mgmg ] , the kinematical reconstruction of the lsp mass is possible if @xmath149 is sufficiently large , @xmath259 for @xmath260 and @xmath261 for @xmath262 . otherwise one can get only an upper bound on the mass @xmath149 . finally , if the lsp is the gravitino ( @xmath39 ) , the uncertainty of the reconstructed gravitino mass @xmath263 translates into an uncertainty of the supergravity planck scale , which is obtained by substituting @xmath263 for @xmath60 in eq.([eq : mp ] ) . to take into account the error of @xmath148 , we also substitute @xmath255 for @xmath148 . eq.([eq : mp ] ) then becomes @xmath264 which is shown in fig . [ fig : mgmpl ] for @xmath256 ( solid lines ) and @xmath265 ( dashed lines ) . here we do not include the uncertainty of the lifetime measurement , which simply affects the measured planck scale as @xmath266 . neglecting the @xmath40lepton mass , eq . ( [ calmpl ] ) is simplified as @xmath267 uncertainty of @xmath151 , for @xmath239 and 100 , and @xmath268 , 200 , and 300 gev . solid lines represent the case without an error of @xmath148 measurement , and dashed lines include 1% uncertainty of @xmath148 . vertical dotted lines represent the stau lifetime of 1 month , 1 year , and 10 years . ] as discussed in the previous section , the kinematical reconstruction of the gravitino mass is possible only if @xmath60 is sufficiently large , @xmath269 for @xmath260 and @xmath270 for @xmath262 . for smaller values of the gravitino mass , one can get only a lower bound on the planck scale @xmath271 . one can see whether the determined ` planck scale ' is inconsistent with the planck scale of the einstein gravity , @xmath272 gev . in other words , one can test the assumption of the decay @xmath273 by comparing the observed lifetime with the predicted lifetime . for example , if the nlsp dominantly decays into axino , the lifetime would be far shorter compared with gravitino assumption for a fixed mass @xmath149 , leading to a smaller value of fitted @xmath271 . suppose for instance one measures @xmath274 gev and @xmath275 gev . when @xmath25 , the lifetime becomes @xmath276 for @xmath277 and @xmath278 [ cf . ( [ eq : axinogamma ] ) ] . if one uses eq . ( [ calmpl ] ) , a `` planck scale '' @xmath279 would be obtained , thereby falsifying the gravitino assumption . we now discuss msugra model points discussed in section [ sec : modelpoints ] , together with cosmological constraints . in the early universe , the @xmath0 cnlsp has been in thermal equilibrium until its decoupling , @xmath280 . if such a particle has a very long lifetime , as discussed in this paper , its decay during or after the big - bang nucleosynthesis ( bbn ) , @xmath281 mev , may spoil the successful prediction of bbn @xcite . in the models with @xmath0 nlsp and @xmath1 lsp , this leads to severe constraints on the parameter space of ( @xmath148 , @xmath60 ) , in particular to upper bounds on the gravitino mass for a given stau mass @xcite . furthermore , it has been recently pointed out that a heavy charged particle can form a bound state with light elements during the bbn , which can lead to new effects and/or severer constraints @xcite . in this paper we do not discuss these effects because it is difficult to evaluate the net effect quantitatively and it still awaits detailed analysis . we should also mention that those bbn constraints may disappear if there is entropy production between the stau decoupling ( @xmath282 ) and the bbn ( @xmath283 mev ) , because the stau abundance is diluted before its decay @xcite . keeping in mind the possibilities of severer bounds and also a possible loophole , let us discuss the cases of @xmath268 gev and @xmath284 gev , corresponding to the msugra model points @xmath285 and @xmath123 . according to the latest analyses @xcite including the effects of the hadronic decay @xcite , the bounds on the gravitino are @xmath286@xmath287 for @xmath288 and @xmath289@xmath290 for @xmath291.he . if one adopts it , the constraints become severer ( cf . @xcite ) . ] the ranges of upper bounds correspond to the uncertainties of various bounds from primordial light elements . for @xmath274 gev ( model point @xmath285 ) , as can be seen in table [ masssugra ] , one could collect more than 1000 staus for 300 fb@xmath29 . the bound @xmath286@xmath292 gev then suggests that the measurement of the gravitino mass and the `` planck scale '' may become possible if one assumes conservative bbn bounds and if the gravitino mass is sufficiently large ( cf . [ fig : mgmg ] and [ fig : mgmpl ] ) . for @xmath293 gev ( model point @xmath123 ) , the measurement would become easier if one could collect the same number of staus ( cf . figs . [ fig : mgmg ] and [ fig : mgmpl ] ) . however , from table [ masssugra ] we find that expected number of stopped cnlsp is around 30 for 300 fb@xmath29 . this is because gluino mass is above 2 tev for this point and production cross section is small . one needs slhc ( @xmath294fb@xmath29 ) to collect @xmath134 events . in the case of axino lsp , the bbn bound is much weaker because the lifetime of the cnlsp stau becomes much shorter [ see eq.([eq : axinogamma ] ) ] . hence , the axino mass measurement is plausible for sufficiently large @xmath295 . when @xmath296 , it is difficult to determine the gravitino mass from @xmath40 energy measurement at stopper detector . axino is a motivated candidate which couples weakly to the mssm particle with comparable strength to the gravitino . in this section we therefore compare the decay @xmath297 with the decay @xmath298 . for simplicity , we will assume the nlsp is pure ` right - handed ' stau , @xmath299 throughout this section . extension to the case with mixing with @xmath300 is straightforward , and the mixing angle dependence is expected to be small . the gravitino is a spin-3/2 particle . however , in the limit where @xmath301 , the effective interaction to mssm particles would be reduced to that of spin 1/2 particle , goldstino @xmath302 . the effective action relevant to the @xmath303 decay is given as follows , @xmath304\tilde{b } ( \cos\theta_wf_{\mu\nu } -\sin\theta_wz_{\mu\nu } ) . \label{eq : actiongv}\ ] ] the action is similar to that of axino given in eqs.([eq : axinobino ] ) and ( [ eq : axinostau ] ) except the coupling coefficients . the relative weight of the two terms in eq.([eq : actiongv ] ) are fixed by the supergravity , while for the axino the coefficient in eq.([eq : axinostau ] ) is induced from eq.([eq : axinobino ] ) by the radiative corrections . note that the term proportional to @xmath305\tilde{b}f_{\mu\nu}$ ] is a non renormalizable coupling of the photon to gravitino or axino and induces significantly different @xmath306 , @xmath40 distribution . the axino three body decay @xmath307 proceeds through the diagrams shown in fig.[fig : threebody ] , where the hatched triangle express the effective vertex shown in eq . ( [ eq : axinostau ] ) . on the other hand , the relevant diagrams for the three body decay into goldstino @xmath308 , are given in fig.[fig : threebody_gra ] . the diagram corresponding to the top right of fig . [ fig : threebody ] does not exist for the goldstino case . the difference of the actions and the relevant diagrams will appear as the deviation of the decay distributions . in the appendix , we list the three body decay differential width into gravitino / axino in the limit where the gravitino / axino mass can be neglected compared to @xmath148 . the formula for the massive gravitino and axino are given in ref . @xcite . . , width=377 ] in the goldstino limit . , width=377 ] the three body decay @xmath309 should be visible in the stopper detector if it has an ability to measure charged tracks , and also segmented into small units . the position where @xmath0 decays in the detector is the position where the charged track by the @xmath310 , @xmath44 and @xmath311 from @xmath40 is initiated . for hadronic tau decays , a @xmath312 is always in the decay products , sometimes with photons coming from @xmath43 decays . photons are converted into electron after passing @xmath313 one radiation length @xmath47 . therefore for the iron based detector with @xmath314g/@xmath33 , the photon shower starts @xmath315 cm from the decay point . in summary , the three body decay of @xmath316 is identified as a charged track ( which might be associated with collinear em showers ) + an isolated hard em shower pointing back to the point where charged track is started . if the segmentation is not good enough , the efficiency to discriminate em showers from @xmath312 would be reduced . as can be seen in the appendix , the three body decay amplitude can be written as a function of the angle between photon and @xmath40 , @xmath36 , and @xmath317 . to be conservative we assume the energy resolution for the isolated photon shower is @xmath318 and ignore the angular resolution of the photon momentum . ( in the following @xmath319 denotes the photon energy after taking into account of this finite resolution effect . ) note that the shortest length of the detector is 3.5 m , which corresponds to 1750g/@xmath35 for @xmath320g/@xmath321 . the em showers are likely contained in the stopper because we only need 200g/@xmath35 to fully absorb them . we need to require several cuts for the accepted events . * experimentally , the angle between @xmath40 and @xmath306 must be large enough to avoid the overlap between @xmath40 decay products and prompt @xmath306 . we only use the events where @xmath322 . we do not lose sensitivity to the differece between the two decays @xmath323 by cutting these events . in the collinear region , the amplitude is dominated by the contributions from qed vertex , which is common for both of the decays . * the three body decay amplitude suffers soft and collinear singularity . because we only adopt simple leading order calculation , we require @xmath324 10 gev and @xmath325 gev . we define the ratio @xmath326 the dependence of @xmath105 on the mssm parameter is quite different between @xmath327 and @xmath39 . while @xmath328 is one loop process controlled by the parameter @xmath173 , the three body decay contains a tree level contribution which depends on non - renormalizable axino-@xmath99-gauge coupling ( top left of fig . [ fig : threebody ] ) . when @xmath173 is small , the tree level contribution plays a dominant role in the three body decay into axino . this can be seen in fig . [ fig : br ] , where the ratio @xmath329 is plotted as a function of the mass difference @xmath330 . we also fix @xmath331 gev but @xmath332 is insensitive to @xmath333 . @xmath329 is enhanced when the mass difference between @xmath334 and @xmath148 is small relative to the @xmath0 mass . this is because the two body decay of axino is suppressed by @xmath335 . the branching ratio ranges from 8% to 0.5% for the model parameters given in the figure . as a function of @xmath330 . the solid , dashed , dot - dashed , short - dashed curves correspond to the case where @xmath336 gev respectively . lines which increase toward smaller mass difference corresponds to @xmath327 . the @xmath39 lines do not show significant bino mass dependence . , width=264 ] on the other hand , the branching ratio @xmath337 is well below 1% for the parameter given in fig.[fig : br ] . @xmath338 @xmath339 for @xmath340 gev respectively . when the three body decay branching ratio turns out be above 5% , 200 stopped @xmath0 is enough to see the 3 @xmath89 deviation from the gravitino assumption . we estimate the number of events @xmath341 stopped in the detector , which is required to find 3@xmath89 deviation from the gravitino scenario as follows ; @xmath342 the @xmath343 as a function of @xmath330 is given in fig.[fig : total ] . each curve increases as @xmath330 is increased up to the value where @xmath344 coincides with @xmath345 . when @xmath346 gev , @xmath137 ( @xmath347 ) stopped @xmath0 are enough to see the deviation from the measurement of @xmath105 for @xmath348 respectively . deviates more than 3 sigma from the prediction for @xmath349 . the lines corresponds to different stau mass as in fig.[fig : br ] . , width=264 ] in some @xmath36 intervals for @xmath350 decay . left panel : @xmath327 @xmath351 , right panel : @xmath39 . @xmath153 gev , and @xmath352 gev . , title="fig:",width=188 ] in some @xmath36 intervals for @xmath350 decay . left panel : @xmath327 @xmath351 , right panel : @xmath39 . @xmath153 gev , and @xmath352 gev . , title="fig:",width=188 ] gev and right for @xmath353 gev , title="fig:",width=257 ] gev and right for @xmath353 gev , title="fig:",width=264 ] not only the branching ratio , but also the decay distribution contain the information of the invisible particle . the axino decay distribution can be enhanced at the region where @xmath354 is large and @xmath355 is small , namely hard photon and @xmath40 is back to back @xcite . this occurs when @xmath173 is small and relative importance of the direct axino - bino - gauge boson coupling is enhanced . in fig . [ fig : egamdist ] , left panel shows the @xmath317 distribution for different @xmath355 intervals . here we fix @xmath356 , @xmath352 gev , and @xmath153 gev . the distribution has significantly hard component for @xmath357 when compared with that of gravitino in the right panel . the enhancement of back - to - back events is a signature of an axino , and clear @xmath40 and @xmath306 separation is not required to distinguish them . we now estimate the the number of cnlsp decays in the stopper to see the 3 @xmath89 deviation between @xmath358 and @xmath359 from the decay distribution . we define @xmath360 like function from the differential decay width ; @xmath361 here , @xmath362 is the number of event in a @xmath192-th bin for @xmath363 when the number of stopped @xmath0 is @xmath364 . @xmath365 and @xmath366 is the bin size in @xmath317 and @xmath36 . we divide the @xmath317 and @xmath36 into 8 bins and 3 bins respectively , for the range @xmath367 and @xmath368 ; @xmath369 the photon energy resolution of the detector is assumed @xmath370 . we apply the cut @xmath371 , @xmath372 gev and @xmath373 gev . on the other hand , @xmath374 is the number of event in a @xmath192-th bin for @xmath375 with @xmath376 normalized so that the total number is same to that of axino three body decay . namely , we do not use the information for @xmath377 in our fit . from @xmath378 , we define @xmath379 , the number of stopped @xmath0 required to see the 3@xmath89 deviation between @xmath363 and @xmath375 as follows , @xmath380 we show the @xmath381 as a function of @xmath330 in fig . [ fig : dist ] . the sensitivity is significantly increased from the estimate using the branching ratio only . for @xmath382 gev , the deviation may be visible for @xmath137 events for @xmath383 ( @xmath347 events for @xmath356 ) . finally , we estimate sensitivity at our model points in section 3 . for simplicity we assume @xmath384 . in general @xmath303 is mixed with @xmath300 . the mixing angle is defined as @xmath385 for model points discussed in table [ mass ] , the angle is @xmath386 . the effect of the mixing angle in the axino decay is small because the amplitude of @xmath387 is suppressed by both by the small @xmath388 factor and smaller hypercharge , and can be safely ignored . in fig.[fig : lambda ] the expected sensitivity at the stopper detector is shown . here , long - dashed ( long - short - dashed , dashed ) line corresponds to the required statistics for @xmath383 ( 0.75 , 1 ) for different @xmath389 , while the upper and lower solid lines correspond to the number of stopped cnlsp for 300 fb@xmath29 and 3000 fb@xmath29 . one can address the difference between axino and gravitino for @xmath390tev or less for @xmath391 fb@xmath29 and @xmath356 ( 0.75 ) . note that the expected reach in @xmath389 is essentially determined by the susy production cross section , as they decrease steeply with increasing @xmath389 . we therefore show the gluino masses scale on the top of the figure . luminosity for gm points with @xmath392tev . dashed lines are number of required stopped events to see 3-@xmath89 deviation in the @xmath393 distribution of @xmath394 from those expected for the decay @xmath395 . , width=264 ] in this paper we investigate the physics of the long - lived charged next lightest susy particle ( cnlsp ) , which may be explored at a massive stopper detector placed next to the cms detector at the lhc . we assume the cnlsp is the lighter scalar tau , @xmath9 , which decay into @xmath396 where @xmath22 is a invisible particle . a natural candidate of the particle @xmath22 is a gravitino but we also consider the case where @xmath22 is axino @xmath2 . in this paper , we assume very large stopper detector next to the cms detector , with total mass of 8 kton . the stopper must have a capability to measure the position where the cnlsp stopped , and also the energy of the @xmath40 decay products . if the detector can be highly segmented , it is also possible to identify the @xmath40 decay products separately . the number of stopped nlsp ranges from @xmath397 to @xmath134 for gluino and squark with mass below 2 tev . if the size of the detector should be smaller , the number of the events must be scaled down linearly . we estimate the statistical error for @xmath135 , which can be determined from the end point of the tau jet energy . we assume that the energy resolution of the detector is 150%/@xmath398 . @xmath399 ( @xmath400 ) can be achieved for the statistics of the order of @xmath401 . when lsp mass is above @xmath402 , one can constrain the gravitino mass both from above and below . from the lifetime and gravitino mass measurement , the supergravity planck scale @xmath271 can be measured . if many cnlsp can be accumulated , one can even study the three body decay of the cnlp @xmath50 . when the dominant decay mode of the @xmath9 is @xmath403 , we may be able to see the deviation of the decay branching ratio and the distribution from the @xmath39 case . finally we comment on the strategy to proceed this experiment . susy particles will be found in the early stage of the lhc experiment if susy scale is @xmath404 . if lsp is gravitino or axino and the nlsp is charged and long - lived , it would also be recognized easily . the detector proposed in this paper may be placed after the existence of long - lived cnlsp is observed , roughly at the same time to the high luminosity run of the lhc , or the proposed super lhc run . lhc experiment has a great potential to explore new physics in tev regions . it is important to explore new possibilities that can be done with lhc . in this paper we have proposed a large additional detector in the cms cavern when the long - lived cnlsp is found . this requires a significant modification of the cms experiment . the reward is low systematics study of the cnlsp decay which primarily serve for the study of the gravitino sector . the determination of gravitino mass either from the lifetime or ( independently ) from @xmath152 measurement would give us a direct information of the the total susy breaking scale . together with high precision determination of the mssm sector expected with the cnslp momentum information , the nature of the interaction of the mssm sector and hidden sector can be studied in detail . we hope this paper is useful for further , and more realistic studies . we thank j. ellis , h. u. martyn and f. d. steffen for discussions . this work is supported in part by the grant - in - aid for science research , ministry of education , science and culture , japan ( no.16081207 , 18340060 ) for mmn . in this appendix we list the differential decay width of @xmath303 decay into @xmath405 where @xmath22 is either gravitino or axino in limit where @xmath406 . formulas in this appendix are obtained by taking massless limits @xmath407 of the formulas in ref . @xcite . the @xmath303 decay width to @xmath408 is given as follows ; @xmath409 ^ 2 } \sum_{\rm spins } \vert { \cal m}(\tilde{\tau}_r\rightarrow \tau\gamma { \tilde{g}})\vert^2\ ] ] where @xmath410 and @xmath411 \cr & & + \frac{1-{x_{\gamma}}}{2-{x_{\gamma}}(1-{\cos\theta } ) } \left ( -4 + 2x-\frac{4}{{x_{\gamma } } } \right ) + \frac{4(1-{x_{\gamma}})^2}{\{2-{x_{\gamma}}(1-{\cos\theta})\}^2 } \cr & & + \frac{1}{\ { { x_{\gamma}(1+\cos\theta)-2a_{\tilde{b}}+a_{\tilde{b}}x_{\gamma}(1-\cos\theta ) } \}^2}\biggl [ \cr & & ~~ 2{a_{\tilde{b}}}^2(1-{x_{\gamma } } ) \left\{{x_{\gamma}}^2 - 2{x_{\gamma}}+\frac{4}{{x_{\gamma}}}-({x_{\gamma}}^2 - 2{x_{\gamma}}-2){\cos\theta}\right\ } \cr & & ~~ + { a_{\tilde{b}}}(1+{\cos\theta})(4 - 3{x_{\gamma}}+{x_{\gamma}}{\cos\theta})({x_{\gamma}}^2-{x_{\gamma}}-2 ) \cr & & ~~ -(1+{\cos\theta})^2{x_{\gamma}}({x_{\gamma}}^2-{x_{\gamma}}-2 ) \biggr ] .\end{aligned}\ ] ] here @xmath412 and @xmath36 is the angle between @xmath40 and @xmath306 . k. hamaguchi , y. kuno , t. nakaya and m. m. nojiri , phys . d * 70 * ( 2004 ) 115007 [ arxiv : hep - ph/0409248 ] . j. l. feng and b. t. smith , phys . d * 71 * ( 2005 ) 015004 [ erratum - ibid . d * 71 * ( 2005 ) 0109904 ] [ arxiv : hep - ph/0409278 ] . a. de roeck , j. r. ellis , f. gianotti , f. moortgat , k. a. olive and l. pape , arxiv : hep - ph/0508198 . w. buchmuller , k. hamaguchi , m. ratz and t. yanagida , phys . b * 588 * ( 2004 ) 90 [ arxiv : hep - ph/0402179 ] ; arxiv : hep - ph/0403203 . a. brandenburg , l. covi , k. hamaguchi , l. roszkowski and f. d. steffen , phys . b * 617 * ( 2005 ) 99 [ arxiv : hep - ph/0501287 ] . cms technical proposal , cern / lhcc/94 - 38 ( 1994 ) . atlas technical proposal cern / lhcc/94 - 43 ( 1994 ) ; + atlas liquid argon tdr , cern / lhcc/96 - 41 ( 1996 ) . s.grohmann , cern , st - cv , `` concepts of refrigeration circuit design for cms tracker cooling , '' at the homepage of ts / cv / dc section of cern ; + ` http://st-support-cooling-electronics.web.cern.ch/st-support-cooling-electronics/ ` s. eidelman _ et al . _ [ particle data group ] , phys . b * 592 * ( 2004 ) 1 . r. k. bock , t. hansl - kozanecka and t. p. shah , nucl . instrum . * 186 * ( 1981 ) 533 . w. w. m. allison _ et al . _ [ soudan-2 collaboration ] , nucl . . meth . a * 376 * ( 1996 ) 36 . ` http://hepunx.rl.ac.uk/soudan2/ ` h. u. martyn , eur . j. c * 48 * ( 2006 ) 15 [ arxiv : hep - ph/0605257 ] . d. e. kaplan , g. d. kribs and m. schmaltz , phys . d * 62 * ( 2000 ) 035010 [ arxiv : hep - ph/9911293 ] ; + z. chacko , m. a. luty , a. e. nelson and e. ponton , jhep * 0001 * ( 2000 ) 003 [ arxiv : hep - ph/9911323 ] . w. buchmuller , k. hamaguchi and j. kersten , phys . b * 632 * ( 2006 ) 366 [ arxiv : hep - ph/0506105 ] . g. f. giudice and r. rattazzi , phys . * 322 * ( 1999 ) 419 [ arxiv : hep - ph/9801271 ] . t. goto and m. yamaguchi , phys . lett . b * 276 * ( 1992 ) 103 ; + e. j. chun , j. e. kim , and h. p. nilles , phys . b * 287 * ( 1992 ) 123 [ arxiv : hep - ph/9205229 ] ; + e. j. chun and a. lukas , phys . b * 357 * ( 1995 ) 43 [ arxiv : hep - ph/9503233 ] ; + and references therein . s. jadach , j. h. kuhn and z. was , comput . commun . * 64 * ( 1990 ) 275 ; + m. jezabek , z. was , s. jadach and j. h. kuhn , comput . commun . * 70 * ( 1992 ) 69 . + s. jadach , z. was , r. decker and j. h. kuhn , comput . commun . * 76 * ( 1993 ) 361 . i. hinchliffe and f. e. paige , phys . d * 60 * ( 1999 ) 095002 [ arxiv : hep - ph/9812233 ] . i. v. falomkin , g. b. pontecorvo , m. g. sapozhnikov , m. y. khlopov , f. balestra and g. piragino , nuovo cim . a * 79 * ( 1984 ) 193 [ yad . * 39 * ( 1984 ) 990 ] ; + m. y. khlopov and a. d. linde , phys . lett . b * 138 * ( 1984 ) 265 ; + j. r. ellis , j. e. kim and d. v. nanopoulos , phys . b * 145 * ( 1984 ) 181 . m. pospelov , arxiv : hep - ph/0605215 ; + k. kohri and f. takayama , arxiv : hep - ph/0605243 ; + m. kaplinghat and a. rajaraman , phys . d * 74 * ( 2006 ) 103004 [ arxiv : astro - ph/0606209 ] . + r. h. cyburt , j. ellis , b. d. fields , k. a. olive and v. c. spanos , arxiv : astro - ph/0608562 .

since the mode - coupling theory ( mct ) of the liquid - glass transition has been proposed in the mid - eighties @xcite , considerable work has been done in order to compare its predictions with experimental data and computer simulation results @xcite . most of these tests of the mct are based on the use as fitting functions of the quasi - universal laws that are obtained from an asymptotic analysis of the family of non - linear equations to which the mode - coupling equations belong and that are valid only in the vicinity of the predicted so - called ideal glass transition singularity . for experimental data , this is essentially the only practicable approach , since typical glass - formers are molecular liquids for which a first - principle study is out of reach and since most experiments , like dynamic light - scattering spectroscopy @xcite , probe complex mechanisms in which various types of dynamical processes are entangled . in the case of computer simulations , the situation appears more comfortable , since simple systems can be considered for which all the data needed for an extensive comparison with the theory can be directly obtained with high precision . however , it turns out that one still has to rely on asymptotic results of the theory at some point . indeed , one finds that the mct is not able to predict accurately the location of the putative ideal glass transition : it has thus to be determined from the simulation results , usually with a fit to an asymptotic law , in order to calibrate the comparison between theory and simulation @xcite . such a use of the asymptotic laws does not go without difficulties , originating in the fact that the ideal glass transition is never observed as such : it is usually argued to be avoided because of the existence of so - called `` activated processes '' not taken into account by the theory , whose effect is to restore the ergodicity of the system in the predicted ideal glassy state . thus the transition is expected to be replaced by a smooth crossover between two dynamical regimes , one dominated by the mode - coupling mechanism and the other dominated by activated processes . a first problem with the occurrence of these additional processes is that it hinders a clear separation of time scales between the fast transient dynamics and the structural relaxation , with which mct exclusively deals , and thus makes the domain where the asymptotic regime of mct is expected more difficult to resolve . in addition , the study of the corrections to the asymptotic laws shows that these laws are valid only in a very small domain around the ideal glass transition @xcite : it can thus be expected that the dynamics in this domain will be markedly affected by crossover effects . finally , from a more pragmatic point of view , since the asymptotic laws of the theory apply to rather specific aspects of the dynamics ( for instance , in the @xmath0 relaxation regime , since the shape of the relaxation itself is non - universal , the main quantitative asymptotic prediction deals with the variation of the relaxation time with temperature ) , only selected parts of the available experimental information are usually involved in most experimental tests of the mct on molecular glassformers , a situation which can be considered quite unsatisfactory . for these reasons , it appears desirable to have a means to test the mct on experimental data which is not limited to the use of the critical predictions of the theory and which is able to give an overall account of the dynamics . this can precisely be achieved within the so - called schematic approach to experimental data . schematic mode - coupling models are simple sets of mode - coupling equations retaining the essential non - linear structure of the equations of the mct in one or two equations and with only a few parameters . they played a major role in the development of the theory to investigate the universal properties of the general mode - coupling equations and to demonstrate the potentialities of the theory @xcite . recently , they have been found useful as quantitative models for numerical fits of experimental data covering a wide time or frequency range . the basic idea behind this schematic approach to experimental data is that , since all equations of the mode - coupling type share distinctive universal properties , meaningful informations are likely to be obtained if a suitable minimal set of equations of this family is used as an effective mode - coupling model to describe experimental data . one interest of the method is then that the universal properties of the mode - coupling equations are used in a milder sense than in their usual direct applications : it is thus possible to numerically take into account corrections to the asymptotic laws that can be important if the system is too far from its putative ideal glass transition . in addition , one can tentatively describe the full dynamics , including the non - universal @xmath0 relaxation and the fast transient dynamics ( although through rather crude approximations ) , and thus possibly overcome the problem of insufficient separation of time scales . for these reasons , the schematic approach appears as a method of choice to critically investigate the validity of the previous tests of the theory based on the asymptotic laws . of course , the power of the method comes at a price : even a minimal schematic model ( in a sense to be defined latter ) will have numerous adjustable parameters . this makes the approach numerically extremely flexible , and one finds for instance that it is possible to give an account of the experimental dynamics with an ideal mct schematic model even at low temperatures where ideal mct is known inadequate . thus one should not overestimate the physical meaning of the fitted parameters . however , as was found in previous works , by critically looking at the results of a schematic analysis , one can address from a renewed point of view the problem of testing the predictions of the mct on experimental data . in the present paper , the third of a series devoted to the schematic approach to experimental data @xcite , we aim to address various issues that have been overlooked in previous works , in order to better settle and understand the principles and results of this approach . in the next section , we give an overview of the schematic studies proposed so far and of the models on which they were based . in sec . [ sec3 ] , a set of simple requirements that need to be fulfilled in order to define a suitable minimal schematic model is proposed . in sec . [ sec4 ] , we examine the influence of the choice of the schematic model on the results of the analysis of a given set of experimental data , and sec . [ sec5 ] is devoted to a tentative validation of the method , by analyzing model spectra with well defined properties within the schematic approach . we finally present our conclusions and draw perspectives for the approach . in this section , we give an overview of the schematic studies proposed so far . this gives us the opportunity to settle a few definitions and notations to be used latter . the first schematic study has been proposed by alba - simionesco and krauzman @xcite , where they analyzed low - frequency raman scattering spectra of the fragile glassformer metatoluidine with various two - correlator schematic models . one of them , based on the so - called f@xmath1 model @xcite , is defined by the equations [ f12model ] @xmath2 with initial conditions @xmath3 , @xmath4 , @xmath5 , and has subsequently been used in studies of raman spectra of other glassformers @xcite and , more importantly , of depolarized light - scattering spectra of salol and ckn [ @xmath6 covering a wide frequency range ( more than four decades ) @xcite . in all these works , where other choices for the memory - function @xmath7 have occasionally been considered @xcite , the fitting function was defined , up to an amplitude factor , as the susceptibility associated with the effective correlator @xmath8 another widely used model differs only from the preceding one through the choice @xmath9 an expression first proposed by sjgren @xcite , and the use as the fitting function of the susceptibility associated to @xmath10 , with an adjustable amplitude factor . it has been used to study depolarized light - scattering spectra of glycerol @xcite and orthoterphenyle ( otp ) @xcite , as well as coherent neutron - scattering and brillouin spectra of @xmath11 @xcite . an extended mct version of this model has also been used recently to study depolarized light - scattering , dielectric - loss and incoherent neutron - scattering spectra of propylene carbonate @xcite . finally , self - intermediate scattering functions obtained by molecular dynamics simulations of a nickel - zirconium model have been analyzed by teichler @xcite with a single - correlator model with @xmath7 defined as @xmath12 at variance with the works mentioned above , a retarded damping term involving four adjustable parameters has been used in the present case instead of the simple markovian approximation made in eq . involving the single parameter @xmath13 . looking at the equations , one sees that these models involve _ a priori _ a large number of fitting parameters . to reduce this number , simplifying assumptions are usually made : in refs . , , , and , the only temperature - dependent parameters are chosen to be @xmath14 and @xmath15 , whereas in ref . , these are @xmath15 and @xmath16 , and in ref . , @xmath14 , @xmath15 and @xmath16 change with temperature , the latter depending also on the wave vector modulus in all schematic studies proposed so far , a certain degree of arbitrariness has often been involved in the choice of the details of the model to be used . the most serious case relates perhaps to the use of the memory function in our preceding works @xcite , which , at variance with sjgren s model@xcite , can not be given any precise theoretical foundations . but other minor choices can also be questioned . there is indeed no physical reason to prefer using a simple correlator as the fitting function instead of a composite one like in eq . , the apparent advantage of the former choice , which seems to make the relation between the fitting function and the dynamical variable probed by the experiment transparent , relying on a mere tautology . one can also wonder why , for similar experimental data and schematic equations , the effective vertex entering the second memory function ( @xmath17 or @xmath16 ) has been held constant with temperature in refs . and and not in ref . for instance , where it is @xmath14 which has been kept fixed . in most cases , these various choices have been only validated _ a posteriori _ by the success of the model to fit the data under study and/or the obtention of parameters whose variations match the expectations from the microscopic theory @xcite . there are nevertheless a few difficulties that could be expected , related to the possible existence of non - generic features of the chosen model . indeed , as stated in the introduction , generic mode - coupling equations display remarkable universal properties @xcite , but , when a schematic model is considered , because it is very simple , one can experience severe limitations in the realization of these universal features : certain values of important parameters of the theory can be unattainable or _ ad hoc _ correlations between these parameters can be imposed . an example of the first situation is provided by the f@xmath18 models , @xmath19 , for which @xmath20 is the only possible value of the exponent parameter @xmath21 , whereas it can take any value in the interval @xmath22 $ ] for the general mode - coupling equations . the second situation is realized with the f@xmath1 model , where the exponent parameter and the critical non - ergodicity parameter @xmath23 are always related through @xmath24 . for these reasons , it seems desirable to understand what are the minimal requirements that must be met by a schematic model in order to avoid such biases and possible consequent inconsistencies in the results of its application . we propose here a series of criteria based on a discussion of the asymptotic properties of the mode - coupling equations in the vicinity of the ideal glass transition . for this , we consider the case of an experiment probing the evolution of a normalized correlation function @xmath25 associated to an unspecified dynamical process in a given glassforming liquid , when a thermodynamic parameter @xmath26 ( temperature or density for instance ) is varied . note that the following discussion can easily be extended to situations where more than one observable are considered in a unified framework @xcite , with exactly the same conclusions . by looking at the asymptotic predictions of the theory @xcite , a few important parameters emerge . on the one hand , one finds parameters that characterize the liquid under study and the thermodynamic path along which the experiment is performed . these are @xmath27 , the value of @xmath26 at the ideal glass transition , @xmath28 , the proportionality constant relating the separation parameter @xmath29 defined by the theory to @xmath26 through the relation @xmath30 and @xmath21 , the exponent parameter governing numerous aspects of the critical dynamics . on the other hand , one finds parameters that are specific to the correlation function @xmath25 : @xmath31 , the non - ergodicity parameter at the transition , and @xmath32 , the so - called critical amplitude . thus we identify five parameters that are expected to be of special relevance . the first three of them are involved in the description of the relative evolution of the two characteristic time scales of the dynamics in the ergodic phase , @xmath33 and @xmath34 characterizing the @xmath35 regime and the @xmath0 relaxation respectively . this can be seen from the ratio @xmath36 where @xmath37 and @xmath38 are uniquely determined by @xmath21 @xcite . in addition , @xmath21 alone determines the shape of the dynamics in the @xmath35 regime . the two remaining parameters , @xmath31 and @xmath32 , are related to the amplitudes of the @xmath0 relaxation and of the @xmath35 process , as shown for instance by the scaling relation @xmath39 describing the @xmath35 process and the onset of @xmath0 relaxation in the ergodic phase close to the singularity . but in fact , this parameter counting is only valid when the full mode - coupling equations describing the dynamics of the system are known . two differences appear when the only available information is the function @xmath25 at different @xmath26 values . first , it can be seen that @xmath32 never appears alone in the asymptotic laws , but always inside products @xmath40 . for this reason , the relevant parameter is rather the physical amplitude @xmath41 , which appears for instance in the relation between @xmath26 and the non - ergodicity parameter in the ideal glassy state , @xmath42 this remark would be irrelevant if it was possible to know @xmath28 from the data at hand , but this is not the case . indeed , according to the theory , the shape of the @xmath0 relaxation is non - universal . this means that there exists an undetermined proportionality constant relating @xmath34 , the value emerging from the theoretical calculation , to the phenomenological @xmath0 relaxation time @xmath43 , which can be defined for instance after @xmath44 and which is the only type of value that can be accessed from the knowledge of @xmath25 alone . such a problem does not occur for the @xmath35 process which is universal , i.e. it depends only on @xmath21 . this means that the ratio can only be known up to an undetermined factor and thus that the value of @xmath28 can not be obtained . this is in fact an important result , since it shows that one can know the absolute distance ( in terms of the separation parameter @xmath29 ) of a system to its ideal glass transition only if the mode - coupling equations describing the full dynamics are known . moving now back to the schematic approach to experimental data , one finds thus that , in order to analyze the evolution of the correlation function @xmath25 with an effective schematic correlator @xmath45 , the model must be able to reproduce the values of @xmath21 , @xmath31 and @xmath46 characterizing the system and the observable under investigation . a point on the transition surface of the schematic model must exist where one will have @xmath47 the last equality is not a strong constraint since @xmath48 and @xmath49 relate to decoupled aspects of the description of the dynamics , the first one being defined at the transition point , whereas the second one describes the approach of this point . one can thus expect that the interplay of these two aspects will allow to compensate possible non - generic features relative to @xmath48 ( like in the case of the f@xmath1 model where @xmath50 ) . the first two constraints are on the contrary very strong and need to be taken into account . one has in particular to worry about the accessible domain for the values of @xmath51 and @xmath52 and to avoid the introduction of arbitrary correlations between these parameters . eventually , one finds thus that a schematic model needs to have at least three adjustable coupling constants , at least one of them changing with temperature . indeed , with a single coupling constant , only one pair ( @xmath51,@xmath52 ) is possible . with two coupling constants , a relation linking @xmath51 and @xmath52 exists , and for a given value of @xmath51 , only a finite number of @xmath52 values will exist . one needs three coupling parameters to be able to decouple @xmath51 and @xmath52 . we can now look at the schematic models introduced in the previous section at the light of these new results . it is easy to see that these conditions are matched by the model used in ref . and defined by eqs . and , since it relies on four coupling parameters ( @xmath14 , @xmath15 , @xmath17 and @xmath53 ) to obtain the values of @xmath51 , @xmath52 and @xmath54 . this is not the case of the model used by teichler @xcite , which has only two coupling parameters @xmath53 and @xmath55 , and one indeed finds inconsistencies in the results : for instance , the exponent parameter @xmath21 appears to change with the considered observable . it seems thus likely that the success of the calculation is essentially to be credited to the flexibility of the fitting procedure ( in particular through the choice of a four - parameter retarded friction ) and to the consideration of observables with high plateau values which probably turn out to be compatible with the value of the real exponent parameter . about the other studies , one finds in fact that the present discussion is inoperant , because only part of the dynamics is considered : the @xmath0 relaxation is not present in refs . and , and the fast transient dynamics is not included in refs . , , and , with the effect of releasing in both cases the constraint on the value of @xmath52 . an interesting consequence of these constraints is that they restrain the possibility to give a precise physical meaning to the separate components of a schematic model . this is especially clear when a composite correlator like eq . indeed , the relations between the components of the fitting function are expected to be mainly imposed by the need to reproduce the values of @xmath21 , @xmath31 and @xmath46 : it is then hardly conceivable that the value of the coupling constant @xmath53 or the effect of the details of the expression for @xmath45 will reflect any physical reality , for instance in terms of the light scattering mechanism . this is of course less obvious when the fitting function is a pure correlator , because of the tautological relation between the fitting function and the dynamical variable probed by the experiment , but , for the previous reasons , we believe that in general a schematic model has to be considered in its globality as a minimal phenomenological model allowing to give an account of the dynamics of a system in terms of mode - coupling effects , and nothing more has to be expected from it . to conclude this part , it is worth stressing that the present discussion refers only to the asymptotic predictions of the theory and that problems originating in pre - asymptotic corrections or the existence of activated processes have been ignored . moreover , only a few numbers have been considered . in particular , the precise shape of the relaxation ( particularly in the @xmath0 relaxation domain ) has not been taken into account , whereas it can make an _ a priori _ well - suited model totally inefficient for the schematic approach . for these reasons , the preceding remarks are only guidelines aimed at rationalizing and optimizing the use of schematic models , but they offer no guarantee of success . in the present section , we propose a study of depolarized light - scattering spectra of a fragile glassformer by means of a variety of schematic models . the aim of this work is twofold . first , in all previous schematic analysis , only one model has ever been used at a time for a given set of experiments and no systematic investigation of the influence of the choice of the model on the results has been proposed so far . secondly , in the previous section , a few parameters have been isolated that are expected to play a special role in the description of the dynamics of a given observable . the use of different schematic models to analyze the same set of experimental data , depending of their ability to provide consistent estimates for these values , should thus allow us to confirm or infirm the corresponding discussion . the data we have studied are the ckn depolarized light - scattering spectra measured by cummins _ _ @xcite and already analyzed with a schematic model in ref . . the reason ckn has been chosen ( and not salol for instance ) for this study is the low intensity of its @xmath0 peak , which is slightly smaller than the microscopic peak : this guarantees that , when the spectra are studied with a composite effective correlator like in eq . , the amplitudes of the contributions of both terms will not be too different and thus this allows one to obtain a better contrast between the outcome of different schematic analysis . to limit the influence of the avoidance of the ideal glass transition , only rather high temperatures , between 383 and 468 k , have been considered . compared to our previous work , the two lower temperatures ( 363 and 373 k ) have been excluded , leading to small differences in the results ( within the error bars of the method ) . note that at the temperatures we have considered , one can expect the spectra to be free of the influence of the spurious experimental artifacts evidenced in ref . . only schematic models closely related to the one used in ref . and with the same number of parameters have been considered . as in this previous work , they have been used to perform simultaneous fits of the experimental data at the nine selected temperatures : the whole dynamical scenario developing as temperature varies is thus encoded in the only temperature dependent parameters of the schematic models , which are the effective vertices entering in the expression of the memory function @xmath7 . we first stick to the two - correlator f@xmath1 model . there are two routes by which the model can be modified . on the one hand , different second memory functions can be chosen . we have considered two cases : [ varmem1 ] @xmath56 on the other hand , one can vary the expression of the effective correlator as a function of @xmath57 and @xmath10 . we have considered four possibilities : [ varcorreff ] @xmath58 in previous works @xcite , the choice of a specific expression was originally motivated by assumptions on the light - scattering mechanism . however , from the discussion of the preceding section , it seems very unlikely that this question is of any relevance : indeed , since individually each correlator has no real physical meaning , the precise way it enters the calculation is clearly unimportant . * 7c model & i & ii & iii & iv & v & vi + @xmath59 & & + @xmath45 & & & & & & + @xmath17 & 8.6 & 3.6 & 5.9 & 2.2 & & + @xmath16 & & & & & 14.2 & 7.5 + @xmath53 & 0.30 & 0.29 & 0.31 & 0.22 & 0.32 & 0.49 + @xmath60 ( k ) & 388.5 & 388.9 & 388.5 & 388.8 & 390.1 & 390.8 + @xmath21 & 0.712 & 0.704 & 0.700 & 0.698 & 0.705 & 0.701 + @xmath61 & 0.448 & 0.452 & 0.446 & 0.449 & 0.424 & 0.429 + @xmath48 & 0.779 & 0.696 & 0.887 & 0.799 & 0.657 & 0.707 + @xmath62 & 0.150 & 0.183 & 0.113 & 0.133 & 0.230 & 0.150 + @xmath54 & 0.563 & 0.548 & 0.544 & 0.530 & 0.580 & 0.501 a total of six models ( numbered from i to vi , see table [ globalt ] for definition ) obtained by various combinations of the previous expressions have been used . model i , defined by eqs . and , was used in ref . , whereas model vi , using eqs . and , is close to the one used in refs . the fits are always very good ( see fig . [ fitexp ] for spectra and fits at two representative sample temperatures ) and of equal quality for all models . the corresponding results for the various coupling constants and relevant mode - coupling parameters are reported in table [ globalt ] and fig . [ vert1 ] . each model provides its own series of effective vertices @xmath14 and @xmath15 and values of the coupling constants @xmath17 or @xmath16 and @xmath53 , distinct from those obtained with the other models . the description of the dynamics in terms of effective mode - coupling parameters is thus clearly non - unique , as expected from the non - universality of the @xmath0 relaxation . for all models , a behavior similar to the one found in ref . is recovered , interpreted as follows : at the highest temperatures the effective vertices are linear functions of temperature and show the progressive evolution of the system towards an ideal glass transition point , but , before this point is reached , this behavior breaks down , evidencing the phenomenon of avoidance of the transition and the inadequateness of the present ideal mct approach at low temperature . the reader is referred to previous work for a discussion of the consequences of this observation @xcite . looking at the trajectories in the @xmath63 plane , one sees that the change in behavior of the effective vertices between the high and low temperature domains is less visible when sjgren s model , eq . , is used : in this case , the trajectories seem to behave rather regularly in the full temperature domain and to bend gently to follow the transition line , at variance with what is found from the use of eq . , where one can easily separate a domain where @xmath14 does not vary much from a domain where it decreases rather rapidly . we attribute this qualitative difference between the two models to the way the temperature dependence is entered in the schematic model at the level of the second memory function @xmath59 . indeed , at a given point @xmath63 of the ergodic domain , one can qualitatively quantify the strength of the mode - coupling term in eq . , @xmath64 , by considering the value of @xmath59 for representative values @xmath65 and @xmath66 of the correlation functions @xmath57 and @xmath10 , respectively @xcite . when eq . is used with constant @xmath16 , the corresponding factor @xmath67 depends only _ implicitly _ on temperature through the variation of @xmath65 and @xmath66 with the position of the representative point of the system in the control parameter space @xmath63 , whereas , when eq . is used with constant @xmath17 , the factor @xmath68 shows an additional _ explicit _ temperature dependence , since @xmath59 is defined directly in terms of the coupling constants @xmath14 and @xmath15 , and in fact this explicit dependence dominates the implicit one . if one now compares the evolution of these factors for both models when one explores the @xmath63 plane , one easily finds that , in order to have similar increases in the strength of the mode - coupling terms , one has to consider trajectories in the case of sjgren s model that lie above the corresponding trajectories for eq . and that come rather close to the transition line , since in its vicinity only are found significant increases in @xmath65 and @xmath66 , hence the difference in the results obtained with these two models : the catastrophic behavior of the vertices due to the avoidance of the transition is less visible in the case of sjgren s model , because it is by construction already constrained to follow rather closely the transition line . once this subtle non - generic feature has been recognized , one is able to rationalize the similarities and differences between the results of the various schematic analysis proposed so far . since the light - scattering spectra of glycerol have been analyzed with the same parameterization of sjgren s model as the one used here for model v and vi , it is no surprise that the trajectory found by franosch _ @xcite is similar to those obtained here with these two models . this is the case of the study of @xmath11 as well , despite the choice of a temperature - dependent @xmath16 : above @xmath69 , the numerical fits result in a slow increase of this parameter when temperature is decreased , which converts , as one would expect , to a trajectory in the @xmath63 plane that is not different from what is obtained when @xmath16 is taken constant @xcite . the case of otp provides another illustration of this point @xcite . indeed , in this work , the relative position of the vertex trajectory and the transition line is imposed from the start through the choice of a temperature - independent @xmath14 , a constraint that is compatible with the results obtained above for models i to iv , and , as could be anticipated from the previous discussion , this coincides with a rapidly increasing @xmath16 when temperature is decreased , i.e. a strong explicit temperature - dependence of the second memory function . the same seems to hold for the extended mode - coupling analysis of gtze and voigtmann @xcite , where a neat intersection of the ideal glass transition line by the trajectory in the parameter space of the schematic model is associated with values of @xmath16 increasing rapidly with decreasing temperature . these observations appear at first sight rather anecdotic . they have nevertheless at least one important physical implication . indeed , comparing the results for glycerol and @xmath11 , two liquids with intermediate fragility , to those for salol , ckn , and otp , that are all very fragile , it is tempting to propose a connection between the fragility of a glassformer and the shape of the effective vertex trajectory that describes its dynamics . the previous discussion rules out such a tentative connection and shows that the analogies and differences found between the various fluids simply originate in the details of the implementation of the schematic approach . we now turn to the quantitative analysis of our results . for each schematic model , the values of the various parameters characterizing the ideal glass transition have been obtained by extrapolation of the linear dependence of the effective vertices with temperature at high temperature , which we have identified as the signature of the domain where the ideal mct is adequate ( in the next section , support will be given to this analysis that differs from the one used in refs . and ) . they are reported in table [ globalt ] . all six models provide consistent estimates for @xmath69 , @xmath21 , @xmath61 and @xmath54 . this consistency tends to validate the discussion proposed in the previous section about the relevant variables to describe a given observable in terms of mode - coupling parameters . this is clearly a non - trivial observation , since the values of the different parameters entering the calculation vary notably from one model to another ( for instance , @xmath70 changes by a factor of two between model iii and v ) . in order to investigate stronger changes to the schematic model , we have considered other memory functions @xmath7 , keeping @xmath71 as in . one of the most studied schematic models after f@xmath1 is f@xmath72 defined as @xcite @xmath73 we have studied two models based on this memory function . they will be denoted i and iv , since they are directly analogous to the preceding i and iv f@xmath1 models . we observe a clear deterioration of the quality of the fits compared to f@xmath1 , but it remains nevertheless acceptable ( see fig . [ fitexp ] ) . in fact , the use of the f@xmath72 model is made difficult by one of its non - generic properties . indeed , when a b - transition line has a termination point or when two transition lines meet , strong corrections to the generic asymptotic laws appear and noticeably affect the shape of the dynamics @xcite . these corrections can make it impossible to use a schematic model to fit experimental data if the description of the system under study requires to come close to such singularities . in the f@xmath1 model , a termination point is found at @xmath74 , but here @xmath75 , which is quite far from the values characterizing the typical glassformers , of the order of 0.7 . thus no problem has to be expected _ a priori_. but in the f@xmath72 model , one finds that the a- and b - transition lines meet when @xmath76 , @xmath77 , with @xmath78 . this is rather close to the values usually associated to experimental systems and one can thus anticipate difficulties with this model , which we actually see . * 3c model & i & iv + @xmath17 & 1.79 & 0.62 + @xmath53 & 0.11 & 0.050 + @xmath60 ( k ) & 389.3 & 389.7 + @xmath21 & 0.688 & 0.670 + @xmath61 & 0.443 & 0.439 + @xmath48 & 0.541 & 0.444 + @xmath62 & 0.254 & 0.373 + @xmath54 & 0.489 & 0.472 the results we have obtained with the f@xmath72 models are reported in table [ f13tab ] and in fig . [ f13fig ] . the behavior of the effective vertices is the same as with f@xmath1 , but slightly less regular . critical parameters have been obtained from high temperatures as before . the agreement with the previous results for the values of @xmath69 and @xmath61 is very good , but less good for the values of @xmath21 and @xmath54 . however one can hardly conclude from these results , since these parameters characterize the @xmath35 process , in which the non - generic limitations of the f@xmath72 model precisely show up . as it was stressed in the previous section , when using the schematic approach to experimental data , one has to avoid any arbitrary correlation between important parameters of the theory introduced in an _ ad hoc _ way by the choice of a specific schematic model . in all models studied so far , one can identify such a correlation in the fact that , along the b type transition line , an increase of @xmath23 is systematically associated to a decrease of @xmath21 . this could have introduced a bias in the schematic calculation that would have made the avoidance of the ideal glass transition line a purely numerical artifact of the method . indeed , despite the absence of a true ergodicity breaking at @xmath69 in real systems , one expects that the increase of the non - ergodicity parameter below @xmath69 survives under the form of an increase of the @xmath0 peak height of the susceptibility . the vertex drift found with the f@xmath1 model could thus be simply traced back to the need to give an account of this evolution , the obtained trajectories corresponding precisely to an increase of @xmath23 ( and thus of @xmath61 ) @xcite . it would then reflect a misconception and limitation of the schematic approach itself and not more generally a failure of the underlying ideal mct as it was argued in ref . 15 . to show that our calculations are indeed free of this artifact , we have investigated a model in which the variations of @xmath21 and @xmath23 can be decoupled . we chose the f@xmath79 model , the transition line of which is made of two branches along which @xmath21 varies between @xmath20 and 1 : between the points @xmath80 , @xmath81 and @xmath82 , @xmath83 , @xmath21 and @xmath23 grow simultaneously , whereas between the points @xmath82 , @xmath83 and @xmath84 , @xmath85 , @xmath21 decreases and @xmath23 increases . clearly , we have chosen this somewhat far - fetched model only for its mathematical properties and one could hardly give any physical motivation for the specific f@xmath79 functional form . however , we stress that , from the point of view of the criteria of the preceding section , this model is _ a priori _ free from any non - generic features and thus a perfectly acceptable schematic model : first , it does not show any limitation on the accessible values of @xmath21 , at variance with all other f@xmath86 different from f@xmath1 ; secondly , in the domain of @xmath21 we are interested in ( around 0.7 ) , the influence of the higher - order a@xmath87 singularity from which its special mathematical features originate is expected to be negligible . we have performed schematic analysis with this model in the vicinity of each branch of the transition line , in an attempt to find if the coupling of the evolutions of @xmath21 and @xmath23 plays any role in our calculations . the corresponding fits are quite poor : they only give a qualitative account of the evolution of the dynamics and it would seem unreasonable to extract quantitative values for the critical parameters of the theory from the results . however , a clear pattern manifests itself : close to both transition branches , the behavior of the effective vertices corresponds to a decrease of the exponent parameter , irrespective of the evolution of @xmath23 . in fact , this could have been anticipated from the fact that , below the value of @xmath69 we have determined , the maximum of the @xmath0 peak lies outside the available frequency window , and this eliminates the requirement to reproduce its amplitude at these low temperatures . however it is reassuring to show by an independent calculation that our results are free from this potential bias , that should certainly be taken into account if data where available on a wider frequency range . to summarize , we have shown in this section that different schematic models applied to a same set of experimental data provide qualitatively similar results . we have found that they are in good quantitative agreement for the determination of the values of the parameters that have been distinguished in the previous section , provided no non - generic features , the importance of which has been emphasized for the f@xmath72 model , interfere with their determination . in the previous section , we have tried to validate the schematic approach and its results showing that different models provide consistent results when applied to a same set of experimental data . however , on various aspects , this study is not quite satisfactory . indeed , despite favorable indications gained from the use of the f@xmath79 model , the meaningfulness of the anomalous behavior of the effective vertices at low temperature remains to be assessed . in our previous work @xcite , it has been attributed to the existence of relaxation processes that are not taken into account in the ideal mct and that make it impossible to describe consistently the dynamics of the studied systems with a schematic model based on this theory . to strengthen this point of view , it would be necessary to show that the schematic approach is actually able to give an account of a complex purely mode - coupling dynamics with smoothly varying fitting parameters , despite of the extreme simplicity of the schematic equations compared to the complexity of the equations of the original problem . moreover , it would be interesting to analyze data for which the values of the essential parameters @xmath21 , @xmath31 et @xmath46 are known in advance , in order to check to what extent the schematic approach is able to provide quantitative estimates for them . a possibility to perform such a test is to calculate series of curves from mode - coupling equations known in advance and to analyze them as one would do for any set of experimental data . it remains then to compare the results obtained within the schematic approach with the properties that can be directly determined from the starting equations . we report the results of such a test in the present section . recently , franosch _ et al . _ have proposed a detailed study of the mct predictions for the collective dynamics of a simple liquid , with special emphasis on the asymptotic laws valid in the vicinity of the ideal glass transition and their corrections @xcite . because their model has been very well and extensively characterized , it appears clearly as the model of choice to be the basis of the present evaluation of the schematic approach . besides it has already been used in a similar spirit for a critical study of the method consisting of the use of numerical fits of the @xmath35 correlator for experimental tests of the mct @xcite . the model is based on the mode - coupling equations for the brownian dynamics of the hard - sphere fluid with particle diameter @xmath88 , the thermodynamic state of which is determined by a single parameter , the compacity @xmath89 , related to the number density @xmath90 through the relation @xmath91 . one starts with the generalized langevin equations for the time evolution of the fourier components of the normalized density - density correlation functions @xmath92 , [ hsmodel ] @xmath93 with the initial condition @xmath94 . @xmath95 is chosen equal to @xmath96 , where @xmath97 is some arbitrary molecular time scale and @xmath98 is the structure factor of the fluid . the mode - coupling memory kernel , defined as a double integral in fourier space ( thus a six - dimensional integral ) , is approximated after moving to bipolar coordinates , by the riemann sum of the resulting two - dimensional integral on a grid of @xmath99 equidistant values of the wave vector modulus , with a grid step @xmath100 such that the @xmath101 values vary between @xmath102 and @xmath103 . one obtains @xmath104 ^ 2 \phi_k(t)\phi_p(t),\ ] ] with @xmath105 , @xmath106 , @xmath107 and @xmath108 , @xmath109 , @xmath110 . @xmath111 is the direct correlation function related to the structure factor through the ornstein - zernike relation @xmath112 . as in previous works@xcite , the static quantities have been computed within the percus - yevick approximation , @xmath113 and @xmath114 have been used , and the length and time units have been chosen such that @xmath115 and @xmath116 . one finds then that the ideal glass transition of the model occurs at a compacity @xmath117 comprised between @xmath118 and @xmath119 . this model aims to represent the dynamics of a hard - sphere colloidal suspension and has been shown to compare very favorably at a quantitative level with the experimental data of van megen _ et al _ @xcite . the equations have been solved following the algorithm described in ref . , then the associated susceptibilities have been computed by fourier transform using filon s algorithm . we concentrate on the susceptibilities , for which it has been shown that the range of validity of the asymptotic laws is usually smaller than for the corresponding correlation functions @xcite , in order to be as close as possible to the situations in which the schematic approach is usually applied to experimental data . five values of the compacity such that the distances to the ideal glass transition are @xmath120 have been considered . larger @xmath121 values have been excluded , since they would correspond to spectra covering a much wider frequency range compared to the experimental situation . we have analyzed spectra for three values of the wave vector modulus : @xmath122 , corresponding to the main peak of the structure factor , @xmath123 , corresponding to the first minimum after the main peak , and @xmath124 , where certain pre - asymptotic corrections seem to be rather small . having now our reference spectra to be used as input like experimental data would be , each series of curves at a given @xmath125 value has been fitted with the two correlator f@xmath1 model referred to as model iv in the previous section . for compatibility reasons , no inertial term is included , i.e. one has @xmath126 note that an inertial term would be required both in the microscopic theory and in a schematic model to study a molecular glassformer , the dynamics of which is newtonian , and to reproduce its oscillatory dynamics in the raman band or in the boson - peak region . however , for the goal we wish to achieve here , i.e. an assessment of the shape of the vertex trajectory and the calculation of the critical parameters of the theory from model theoretical spectra , the fact that it has been consistently omitted in the original model and in the schematic equations is irrelevant . no amplitude factor is needed , since the data are already normalized ( to check this point , a few calculations have been done with such a factor , showing that it remains equal to 1 ) . as above , only the effective vertices @xmath14 and @xmath15 are allowed to vary with @xmath127 . this means in particular that constant values of @xmath128 and @xmath129 are imposed , despite the fact that the @xmath95 values of the original equations actually change with compacity through @xmath98 . the quality of the fits is good , comparable to what is obtained when experimental light - scattering spectra are analyzed , as illustrated in fig . [ fiths ] . values obtained for the various adjustable mode - coupling parameters are reported in table [ hstab ] and in fig . [ hsvert ] . * 7c @xmath125 & & & + & schem . & theor . & schem . & theor . & schem . & theor . + @xmath17 & 20.3 & & 1.8 & & 0.58 + @xmath53 & 0.102 & & 0.329 & & 0.503 + @xmath130 & 0.5153 & 0.5159 & 0.5155 & 0.5159 & 0.5154 & 0.5159 + @xmath21 & 0.678 & 0.735 & 0.676 & 0.735 & 0.679 & 0.735 + @xmath131 & 0.846 & 0.849 & 0.414 & 0.419 & 0.268 & 0.269 + @xmath132 & 0.306 & 0.323 & 0.738 & 0.642 & 0.603 & 0.582 + @xmath133 & 2.48 & 1.54 & 1.36 & 1.54 & 1.81 & 1.54 + @xmath134 & 0.849 & 0.779 & 1.513 & 1.548 & 1.433 & 1.403 as usual , the effective vertices are the important parameters of the calculation . to each series of curves corresponds a different set of effective vertices , stressing the fact that the study of a single observable is not enough to give an unequivocal account of the dynamics of the system . here again , this is a consequence of the non - universality of the @xmath0 relaxation as described in the mct framework . the important observation is that the behavior of the effective vertices is perfectly regular : @xmath14 and @xmath15 are approximately linear with @xmath135 ( and thus @xmath14 is linear with @xmath15 ) . nothing similar to what has been found with experimental data close to the transition line shows up in the results . this tends to confirm the assumption that the anomalous behavior found from experiments and attributed to the avoidance of the ideal glass transition is not connected to a technical limitation of the schematic approach , but actually , at a more fundamental level , to the inadequateness of ideal mct in a temperature domain starting noticeably above the extrapolated ideal glass transition point . we can now determine the critical parameter values by extrapolating the linear behavior we have found and compare the results with those obtained directly from the original equations , in order to see if the procedure we have used previously is meaningful and if the corresponding results are quantitative . the values are reported in table [ hstab ] . of the three parameters whose importance has been stressed in section [ sec3 ] , @xmath131 is clearly the one for which the agreement between the original value and the schematic calculation is the best . the determination of @xmath134 is less precise but remains satisfactory . on the contrary , a notable disagreement on the value of the crucial exponent parameter @xmath21 is found . the difference we find is significant and shows unambiguously that the schematic approach does not allow _ a priori _ to obtain the exact value of the exponent parameter , at least when the studied dynamics only covers a frequency range of the order of four or five decades . from this point of view , the schematic approach suffers from the same difficulties as the use of the @xmath35 correlator , which has been shown not to be able to give the value of @xmath21 in the same conditions @xcite . it has however to be stressed that , even if it is not perfect , the schematic approach turns out to perform much better than the latter approach . surprisingly , it is found that the three series of curves do not provide values scattered around the exact value @xmath136 , but instead , the same underestimated value is consistently found at all three wave numbers . it seems thus very likely that no improvement would be obtained if the three observables were analyzed in a unified framework , in the spirit of the most recent schematic calculations @xcite . the origin of the present problem is strictly connected to the existence in the analyzed data of corrections to the asymptotic behavior predicted by the theory and not to a deficiency of the schematic approach . indeed , when spectra corresponding to larger values of @xmath121 ( i.e. corresponding to compacity values closer to the critical one ) are analyzed , a slight change in the behavior of the effective vertices occurs around @xmath137 and the trajectory described in the @xmath63 plane bends to indicate the point of the transition line corresponding to the correct value of @xmath21 . we stress , to avoid any ambiguity , that this change of behavior has nothing to do with the one found from experimental data : its amplitude is very small and it occurs in a domain where the points representative of the system in the @xmath63 plane are extremely close to the ideal glass transition line , meaning that the dynamics spreads over more than eight decades in frequency and corresponding for a real glassformer like ckn to a regime where the behavior of the effective vertices is already strongly affected by the existence of activated processes . in this paper we have proposed a detailed study of the schematic mode - coupling approach to experimental data . our aim was twofold . first , we wanted to settle the whole approach on firmer theoretical grounds , through a better understanding of the parameterization it involves and a clarification of the nature of the results it can provide . indeed , the schematic approach involves a certain degree of arbitrariness in the choice of the models to be used and the introduction of numerous adjustable parameters , which are never considered as good things when it comes to testing a theory . thus in order to credibilize the whole approach , one has to address these _ a priori _ problematic issues and show that they are actually harmless . hence this investigation , whose outcome is the formulation of simple rules that need to be taken into account in order to define minimal schematic models suitable for use in this context and the isolation of a few parameters which are particularly relevant for the description of the dynamics and for which the schematic approach should provide estimates . following these criteria , it turns out that most previous schematic studies are based on well designed schematic models and use parameterizations which are close to the minimal one , in particular by letting only effective vertices vary with temperature . secondly , our goal was a validation of the approach , through the investigation of its robustness and of its potential quantitative character by a variety of numerical tests . the results are rather satisfactory . indeed , we have been able to validate the results obtained in ref . and the way they were analyzed . we have shown that they have a limited , if any , dependency in the schematic model we use and we have been able to identify and rule out many potential biases that could have affected our results . most properties of the dynamics whose relevance have been emphasized in the theoretical discussion of the approach appear to be quantitatively evaluated , with the notable exception of the exponent parameter @xmath21 , for which the method is not too bad , but still far from perfect . it has thus to be stressed that the precise and unambiguous determination of this parameter , which actually plays a very important role in the theory , remains elusive , by any of the methods that are available so far : all calculations indeed seem to be too sensitive to the presence of significant pre - asymptotic corrections ( as is shown here or , in the case of the use of the @xmath35 correlator , by ref . ) and very probably to the existence of crossover effects due to the non - ideal behavior of real and simulated glassformers @xcite . in conclusion , we believe that the present work paves the way to new applications of the schematic approach . indeed , the interest of the method to analyze experimental data has been emphasized in all previous works and we think its relevance has been further strengthened here . however , at this level , it was only a kind of simple phenomenological mode - coupling approach and the main issue was its ability to consistently describe a set of experimental data with a simple model . now , from our results , it seems possible to go one step further and to use this approach as an effective means to obtain precise estimates of transition temperatures , non - ergodicity parameters and critical amplitudes to be used for instance in quantitative tests of the mct on simple systems that are amenable to a full first - principle treatment . the authors are indebted to prof . w. gtze and dr . m. fuchs for providing them the codes for some parts of the calculations . we are also grateful to prof . cummins and dr . g. li for the use of their experimental data . dr . g. tarjus is acknowledged for fruitful discussions . w. gtze , in _ liquids , freezing and glass transition , les houches 1989 _ , edited by j .- hansen , d. levesque , and j. zinn - justin ( north holland , amsterdam , 1991 ) , p. 287 ; w. gtze and l. sjgren , rep . prog . phys . * 55 * , 241 ( 1992 ) ; w. gtze , j. phys . : condens . matter * 11 * , a1 ( 1999 ) . schematic models have been used earlier to fit experimental data @xcite , but only to give an account of the shape of the @xmath0 relaxation , and not of the evolution with temperature of the full dynamics as in the more recent schematic approaches . g. li , w. m. du , x. k. chen , h. z. cummins , and n. j. tao , phys . rev . a * 45 * , 3867 ( 1992 ) ; h. z. cummins , w. m. du , m. fuchs , w. gtze , s. hildebrand , a. latz , g. li , and n. j. tao , phys . e * 47 * , 4223 ( 1993 ) . n. v. surotsev , j. a. h. wiedersich , v. n. novikov , e. rssler , and a. p. sokolov , phys . b * 58 * , 14888 ( 1998 ) ; j. gapiski , w. steffen , a. patkowski , a. p. sokolov , a. kisliuk , u. buchenau , m. russina , f. mezei , and h. schober , j. chem . phys . * 110 * , 2312 ( 1999 ) ; h. c. barshilia , g. li , g. q. shen , and h. z. cummins , phys e * 59 * , 5625 ( 1999 ) . @xmath65 and @xmath66 can be defined in various ways . a rigorous definition based on a bifurcation theory analysis has been proposed by gtze and sjgren @xcite , but more simple _ ad hoc _ choices can be used , like taking the values of the non - ergodicity parameters at the point of the transition line such that the corresponding exponent parameter @xmath21 maximizes @xmath141 ( remember that @xmath29 is negative in the ergodic domain ) . w. van megen and p. n. pusey , phys . a * 43 * , 5429 ( 1991 ) ; w. van megen , s. m. underwood , and p. n. pusey , phys . rev . lett . * 67 * , 1586 ( 1991 ) ; w. van megen and s. m. underwood , phys . rev . a * 47 * , 248 ( 1993 ) ; phys . * 70 * , 2766 ( 1993 ) ; phys . rev . e * 49 * , 4206 ( 1994 ) . a very well documented illustration of this fact is provided by the simulation work of w. kob and h. c. andersen [ phys . * 73 * , 1376 ( 1994 ) ; phys . e * 51 * , 4626 ( 1995 ) ; * 52 * , 4134 ( 1995 ) ] , where rather different values of @xmath21 are obtained depending on the method of analysis and the considered observable .

electron scattering is one of the most important processes in determining the emergent thermal spectrum from models of optically thick accretion discs around black holes and neutron stars . electron scattering opacity generally dominates absorption opacity in the atmospheres of the innermost regions of such discs @xcite . in the case where coherent ( thomson ) scattering is a good approximation , the resulting local thermal spectrum of some annulus in the disc is generally harder than a blackbody with the same effective temperature , due to incomplete thermalization at the scattering photosphere . however , incoherent ( compton ) scattering in the disc surface layers by thermal electrons can reduce this spectral hardening by increasing the energy exchange between the photons and the plasma @xcite . in addition to the disc atmosphere itself , many models invoke a powerful corona above the disc consisting of high temperature or nonthermal electrons that compton upscatter disc photons to produce the energetically significant hard x - rays that exist in certain classes of active galactic nuclei and in certain black hole x - ray binary accretion states ( e.g. @xcite ) . these hard x - rays in turn interact with the relatively cool disc atmosphere to produce reflection spectra that are widely observed in many black hole sources ( e.g. @xcite ) . in this work we explore turbulent comptonization , which is the effect of comptonization by bulk , turbulent electron motions on photon spectra @xcite . in sources with radiation pressure dominated accretion flows , bulk velocities may exceed thermal electron velocities , a phenomenon first pointed out in @xcite . the turbulent speeds @xmath0 on the outer scale of an mhd turbulent cascade will be of order the alfvn speed , and the ratio of this to the root mean square electron thermal velocity is therefore @xmath1 here @xmath2 , @xmath3 , and @xmath4 are the gas , radiation , and magnetic pressures , respectively , and @xmath5 is the ratio of the proton to electron mass . stratified shearing box simulations of magnetorotational turbulence generally have disc atmospheres that are supported by magnetic fields rather than thermal pressure @xcite . hence the first factor generally exceeds unity in an otherwise radiation pressure dominated disc . bulk speeds on the outer scale of the turbulence will therefore exceed the electron thermal speeds whenever the radiation pressure to gas pressure ratio exceeds the ratio of the proton to electron mass ratio , and even smaller depending on how magnetically supported is the disc atmosphere . in this regime , bulk comptonization by the turbulence may dominate thermal comptonization in determining the shape of the spectrum emitted by a local patch of the disc . this regime is commonly reached for near eddington accretion on black holes of all mass scales . indeed , the inner disc solution of the standard geometrically thin model of @xcite gives vertically averaged radiation to gas pressure ratios of approximately @xmath6 where @xmath7 is the ratio of turbulent stress to thermal pressure , @xmath8 is the black hole mass , @xmath9 is the radiative efficiency of the disc as a whole , @xmath10 is the luminosity in eddington units , and @xmath11 is the radius in the disc in units of the gravitational radius @xmath12 . hence the radiation to gas pressure ratio of the innermost disc will generally exceed the proton to electron mass ratio for near - eddington accretion even for stellar mass black holes , and certainly for supermassive black holes . on the other hand , energy exchange between the photons and the plasma is generally dominated by true absorption opacity in standard disc models for the most supermassive black holes @xcite , which may reduce bulk comptonization by turbulence in these sources . we are also motivated by observational evidence in many sources for a cooler , more optically thick medium that also compton upscatters disc photons in high luminosity sources , but not to the high energies usually associated with the more traditional external corona . the soft x - ray excess of the type 1 seyfert ngc 5548 was fit by @xcite with a warm ( @xmath13 kev ) optically thick ( @xmath14 ) comptonizing medium . similar parameters were obtained for the soft x - ray excess of the type 1 seyfert mrk 509 by @xcite : @xmath15 kev and scattering optical depth @xmath16 to @xmath17 . @xcite fit the soft x - ray excess of the extreme narrow line seyfert 1 re j0134 + 396 assuming extra comptonization within the disc atmosphere itself with @xmath18 kev and @xmath19 . such a spectral decomposition is consistent with the lack of short time scale variability in the soft x - ray excess compared to the hard x - rays in this source @xcite . @xcite fit the soft x - ray excesses of eleven bright quasars with comptonization in a medium with @xmath20 kev and @xmath21 . @xcite fit soft intermediate state spectra of the black hole x - ray binary gx 339 - 4 with a comptonizing medium with @xmath22 kev and @xmath23 . thermal comptonization fits to the soft x - ray spectra of a number of ultraluminous x - ray sources have @xmath24 kev and @xmath25 @xcite . such fits have been nicely confirmed with _ nustar _ and _ xmm - newton _ spectra of the ultraluminous x - ray sources ngc 1313 x-1 and x-2 , which require warm @xmath26 kev , optically thick @xmath27 comptonizing media @xcite . @xcite and @xcite suggest that one way of forming these optically thick cool coronae is by a radiation pressure driven outflow . the alternative we consider here is bulk comptonization by turbulence in the disc atmosphere itself . comptonization by bulk motions in the accretion flow has also been considered by others . @xcite considered bulk comptonization in converging flows and shocks . starting from this seminal work , bulk comptonization by radial flows has been calculated in detail by numerous authors @xcite . @xcite included bulk comptonization in their monte carlo calculations of photon spectra from radiation mhd simulations of super - eddington accretion flows , and found that it produced significant spectral hardening which resembled spectra of ultra - luminous x - ray sources . here we focus on smaller scale bulk comptonization by turbulence within the disc atmosphere itself . turbulent comptonization has also been invoked in other areas of astrophysics . @xcite and @xcite used then current limits on cosmic microwave background temperature anisotropies to constrain possible turbulent energy on cosmological scales prior to recombination . @xcite considered comptonization by alfvnic turbulence in a relativistic outflow as a model for the spectrum of gamma - ray bursts . the purpose of this paper is to determine how photon spectra produced by turbulent comptonization depend on properties of the turbulence itself , and how to resolve and interpret this effect in radiation mhd simulations . the structure of this paper is as follows . in section [ sec_general ] we show that the macroscopic physical origins of turbulent comptonization energy exchange are work due to radiation pressure and viscous dissipation due to the radiation viscous stress tensor , and we discuss why this requires us to treat divergenceless turbulence separately from turbulence with non - zero divergence . in section [ sec_energy_exchange ] we discuss the consequences of this for correctly implementing radiative transport in simulations , and derive the appropriate radiation energy equation in both lab and fluid frame variables . in section [ sec_incompressible ] we address the conjecture of @xcite that turbulent comptonization can be treated as thermal comptonization by solving the kompaneets equation with an equivalent wave " temperature critically dependent on the photon mean free path . we show this is true only for divergenceless turbulence , derive the exact wave temperature with an analytic solution of the radiative transfer equation , and use this result to discuss how the wave temperature depends on the power spectrum of the turbulence . to provide physical insight , we also perform an intuitive , heuristic calculation of the wave temperature which well approximates the analytic solution . in section [ sec_compressible ] we consider bulk comptonization by turbulence with non - zero divergence . we show that comptonization by turbulence whose wavelengths are short relative to the photon mean free path can be treated as thermal comptonization with an equivalent temperature given by the full turbulent power . in the limit of extremely optically thick turbulence , we show how the evolution of local photon spectra can be understood in terms of compression and expansion of the strongly coupled photon and gas fluids . in section [ sec_discussion ] we discuss our results , and we summarize our findings in section [ sec_conclusion ] . in order to determine how photon spectra produced by turbulent comptonization depend on properties of the turbulence itself , it is useful to understand the distinct macroscopic physical origins of net energy exchange due to this effect . we show that these are simply the work done by radiation pressure and radiation viscous dissipation , and discuss the major consequences of this . we limit our consideration in this paper to non - relativistic velocities . before proceeding , we define terms and quantities that will be used repeatedly . by bulk comptonization we mean comptonization by bulk motions in general , and by turbulent comptonization we mean bulk comptonization specifically due to turbulence . we denote the characteristic photon mean free path @xmath28 , where @xmath29 is the electron density and @xmath30 is the thompson cross section . we denote the typical length scale for bulk velocity variations @xmath31 , such as the wavelength if there is a well - defined spatial period . unless otherwise stated , by the terms optically thin and thick we mean @xmath32 and @xmath33 , where @xmath34 and @xmath35 are the minimum and maximum length scales in the turbulent cascade , respectively , not referring to the optical depth that a photon would need to travel to escape the medium . net energy exchange due to bulk comptonization is simply the net energy exchange between gas mechanical energy and radiation . inside the photosphere , the mechanical energy loss rate per unit volume to radiation is @xmath36 where @xmath37 is the lab frame radiation pressure tensor . this can also be written as @xmath38 where @xmath39 is the trace of the radiation pressure tensor , @xmath40 is the radiation viscous shear stress tensor , and @xmath41 is the velocity shear tensor . we see that the energy exchange is separated into two pieces , one due to only a diverging velocity field and another due to a shearing velocity field in the presence of a radiation viscous shear stress tensor . the first piece has contributions from two effects , ordinary work done by radiation pressure , and radiation viscous dissipation . the former effect is first order in velocity since it is due to the contribution to @xmath42 that is zeroth order in velocity . energy exchange due to viscous effects , on the other hand , is second order in the velocity , as @xmath43 and the relevant contribution to @xmath42 must themselves be at least first order in velocity since they are a consequence of the velocity field . @xcite conjectured that turbulent comptonization can be treated as thermal comptonization by solving the kompaneets equation with an equivalent wave " temperature critically dependent on the photon mean free path . this is physically intuitive for divergenceless turbulence since in this case energy exchange is entirely due to radiation viscous dissipation and is therefore second order in velocity . in section [ sec_incompressible ] we prove this conjecture for an arbitrary , divergenceless velocity field of uniform density , derive the exact expression for the wave temperature , and show how it is related to the standard expression for radiation viscosity in the optically thick limit . the connection between bulk comptonization and the radiation viscous shear stress tensor is useful because it correctly suggests that the wave temperature can be written in terms of @xmath43 . since pressure work , on the other hand , is an effect first order in velocity , and since comptonization by a velocity field with non - zero divergence is a combination of pressure work and radiation viscous dissipation , it is not surprising that in this case bulk comptonization can not be treated as thermal comptonization , as we show in section [ sec_compressible ] . but in the optically thin limit , i.e. when the mean free path is significantly larger than the largest length scale in the turbulence , energy exchange that is first order in velocity vanishes since photons are equally likely to downscatter as they are to upscatter . bulk comptonization by a velocity field with non - zero divergence is then solely due to radiation viscous dissipation , and in section [ sec_compressible ] we show that it may be treated as thermal comptonization by solving the kompaneets equation with a wave temperature equivalent to the full bulk power . in the optically thick case , i.e. when the mean free path is significantly smaller than the smallest scale in the turbulence , the lowest order energy exchange is the work done by radiation pressure to compress the gas , since it is first order in velocity and since radiation viscous effects are suppressed by a factor @xmath44 ( section [ sec_incompressible ] ) . @xcite assumed that effects first order in velocity always vanish on average for turbulent eddies , but in the optically thick limit photons trapped in a converging ( diverging ) region undergo systematic upscattering ( downscattering ) . in the extremely optically thick limit in which the photon and gas fluids are strongly coupled , velocity convergence corresponds to compression in which gas mechanical energy is transferred locally to the photons . in section [ sec_compressible ] we show that in this process a locally thermal photon distribution remains thermal and only changes temperature , completely analogous to the evolution of the cosmic microwave background radiation under the expansion of the universe . unlike energy exchange due to viscous dissipation , this process is reversible . the effect of this process on the emergent spectrum of the disc will depend primarily on how effectively photons are able to escape from such regions to the observer . self - consistent radiation mhd simulations of turbulent , radiation pressure dominated accretion flows now exist , both in local vertically stratified shearing box geometries @xcite and in global simulations @xcite . although these simulations use frequency integrated equations , the emergent radiation spectrum can be computed , including the effects of bulk comptonization , using post - processing monte carlo simulations . indeed , this has already been done by @xcite . however , in order for such calculations to be self - consistent , the frequency integrated radiation mhd equations used in the simulations themselves must include energy exchange due to bulk comptonization . we now discuss the consequences of the macroscopic physical origins of such energy exchange detailed in section [ sec_general ] for ensuring this effect is captured in simulations . we then proceed to derive the appropriate frequency integrated source terms due to compton scattering for the gas and radiation energy equations in both lab frame and fluid frame variables . using these results , we discuss the extent to which bulk comptonization is captured by existing radiation mhd simulation codes . the decomposition of bulk comptonization energy exchange into pressure work and radiation viscous dissipation shows that radiation mhd schemes that neglect contributions to the viscous stress tensor that are first order in velocity can not capture bulk comptonization energy exchange due to a shearing velocity field or any optically thin velocity field with non - zero divergence . as these effects are second order in velocity , we also note that a necessary , but not sufficient , condition for capturing these effects is inclusion of energy terms second order in velocity . without such terms , turbulence in this form , instead of exchanging energy with photons , will eventually cascade down to the gridscale ( or viscous or resistive scale if the code has explicit viscosity or resistivity ) , and increase the internal energy of the gas . gas internal energy may then be exchanged with photons through thermal comptonization . the omission of viscous dissipation by radiation therefore does not prevent the eventual transfer of turbulent energy to radiation , but it may have other physical effects that can in turn affect radiation spectra . to derive the appropriate frequency integrated source terms due to comptonization , we start with the zeroth moment of the radiative transfer equation , correct to order @xmath45 and @xmath46 @xcite , @xmath47+\frac{v_i}{c } n^i\biggr\}\cr + \left({18\over5}+{17\over5}\epsilon{\partial\over\partial\epsilon } + { 11\over20}\epsilon^2{\partial^2\over\partial\epsilon^2}\right ) \left(n^{ij}{v_iv_j\over c^2}-{v^2\over3c^2}n\right ) . \label{eqpsaltis}\end{aligned}\ ] ] here @xmath48 is the photon energy , and the various angle - averaged moments are defined in terms of the energy and direction ( @xmath49 ) dependent photon occupation number @xmath50 by @xmath51 in principle , the energy equation is obtained by writing eq . ( [ eqpsaltis ] ) in terms of moments of the specific intensity and then integrating over all frequencies . unfortunately , we can not integrate over terms multiplied by @xmath52 without prior knowledge of the spectrum . for the purpose of simulations , then , we make two approximations . first , we observe that the fractional energy change per scattering off of non - relativistic electrons is small , so that only regions inside the photosphere contribute to comptonization . since the stimulated scattering terms are already order @xmath52 and in these regions departures from isotropy are order @xmath53 , we make the following approximation for these terms : @xmath54 second , we assume that the spectrum can be approximated by a bose - einstein distribution with temperature @xmath55 . with these approximations , eq . ( [ eqpsaltis ] ) yields @xmath56 this is the correct energy equation , but in order for it to capture second order energy exchange , the substituted value of @xmath57 must be calculated with a moment closure scheme that does not neglect contributions to the radiation viscous stress tensor that are first order in velocity . for example , it is not adequate to calculate @xmath57 by substituting @xmath58 into the first moment equation . here . ] this is equivalent to flux limited diffusion in the diffusion regime , such as that implemented in @xcite . these do , however , capture energy exchange due to pressure work in the optically thick regime . to show this , we substitute into eq . ( [ integrated1 ] ) the standard closure relation , @xmath59 which gives @xmath60 we see that energy exchange second order in velocity is not present . furthermore , we see that energy exchange due to a converging velocity field is indeed the work done by radiation pressure to compress the gas , @xmath61 . the m1 closure scheme @xcite , implemented in , e.g. , @xcite , also captures first order energy exchange but not second order energy exchange . this scheme assumes that there exists a frame in which @xmath62 . the lab frame radiation pressure tensor can then be expressed in terms of the energy density and flux @xcite : @xmath63 where @xmath64 and @xmath65 to show that second order energy exchange is not captured , we consider the case of a non - zero radiation viscous shear stress tensor due to a non - relativistic velocity field in an otherwise homogeneous medium . the lowest order contribution to the flux must be first order in velocity . in this scheme , then , the radiation viscous shear stress tensor is zero to first order in velocity , and hence second order energy exchange , which requires a contribution first order in velocity ( eq . [ energy_exchange ] ) , is not captured . another way to understand why both flux - limited diffusion and the m1 closure scheme fail to capture second order energy exchange is to observe that they both bridge generic optically thick conditions with optically thin conditions , while optically thin turbulence does not fall into either category . in optically thin turbulence , the turbulence length scales are optically thin ( @xmath66 ) , but conditions are otherwise optically thick ( @xmath67 ) , since we are far enough inside the photosphere that photons must scatter many times before escaping . it seems that only a more sophisticated approach , such as explicitly solving the transfer equation as done by @xcite , can capture this effect . we note that @xcite add an artificial viscosity to the m1 closure scheme in order to address a numerical problem associated with artificial shocks in their simulations . they assume a kinematic radiation viscosity given by @xmath68 which , as they acknowledge , underestimates the actual viscosity in the optically thick limit by a factor of @xmath69 ( eq . [ scattering_nu ] ) . since we show ( section [ sec_incompressible ] ) that the viscosity is smallest in the optically thick limit ( eq . [ scattering_nu_k ] ) , this artificial viscosity always underestimates the actual radiation shear stress tensor , and therefore only partially captures secord order energy exchange . for completeness , we also write eq . ( [ integrated1 ] ) in terms of fluid frame radiation variables , indicated by subscript zero : @xmath70 since @xmath71 eq . ( [ integrated2 ] ) simplifies to @xmath72 @xcite ( hereafter s04 ) conjectured that comptonization by turbulence can be treated as thermal comptonization by solving the kompaneets equation with an equivalent wave " temperature given by @xmath73 where @xmath74 is the temperature corresponding to the total electron energy density @xmath75 at wavenumber @xmath76 . equation ( [ twave1 ] ) generalizes s04 eq . ( 8) , @xmath77 which gives @xmath78 for a kolmogorov spectrum , @xmath79 , with maximum wavelength @xmath80 . equation ( [ twave1 ] ) is a weighting scheme of the form @xmath81 with the weighting function @xmath82 given by @xmath83 this scheme roughly gives full weight to wavelengths less than @xmath84 and zero weight to wavelengths greater than @xmath84 . for a periodic velocity field , since the modes are discrete , eq . ( [ eq_scheme_form ] ) is more clearly written @xmath85 where @xmath86 is the temperature of mode @xmath87 . for the remainder of this section we use eq . ( [ twave2 ] ) since it is more useful for applications to radiation mhd simulations , but note that all results also hold for eq . ( [ eq_scheme_form ] ) , as this is just the continuum limit . we also define @xmath88 = @xmath89 , the optical depth divided by @xmath90 across a mode with wavenumber @xmath76 . ( [ twave3 ] ) , for example , can then be written , @xmath91 the structure of this section is as follows . in section [ sec_exact ] we build on s04 by proving that comptonization by an arbitrary velocity field of uniform density can indeed be treated as thermal comptonization , provided that the field is divergenceless and the photon escape probability is sufficiently small . the exact equivalent wave temperature is of the form eq . ( [ twave2 ] ) , with the weighting function @xmath82 given by @xmath92 where @xmath93 . \label{eq : qdefinition}\ ] ] the limiting cases of @xmath82 are @xmath94 we plot @xmath95 and @xmath82 in fig . [ mcdata1 ] . since log scaling is used , the curve for @xmath95 disappears for @xmath96 . we see that @xmath95 does roughly approximate @xmath82 . in section [ sec_heur ] , to provide physical insight into the exact solution , we also find a heuristic , approximate expression for @xmath78 , also plotted in fig . [ mcdata1 ] , with a simple physical model motivated by ideas put forth in s04 . finally , in section [ sec_spec ] we use the correct expression for @xmath78 to discuss how the wave temperature depends on the power spectrum of the turbulence and determine which turbulent wavelengths contribute most to comptonization . to derive eqs . ( [ twave2 ] ) and ( [ eq : analyticsol ] ) , we first find an expression for the wave temperature in terms of the second moment of the occupation number , since it is the latter which gives rise to the radiation shear stress tensor . we start with the zeroth moment of the radiative transfer equation , eq . ( [ eqpsaltis ] ) , correct to order @xmath45 and @xmath46 . note that terms of order @xmath46 are second order in velocity since for comptonized photons @xmath97 . consider a solution for the angle dependent occupation number @xmath98 constructed with the expansion @xmath99 where @xmath100 is formally @xmath87th order in velocity . we make the ansatz that @xmath101 is independent of position and angle , so that to second order in velocity eq . ( [ eq_expansion_1 ] ) is @xmath102 to second order in velocity the zeroth moment equation then simplifies to @xmath103 we also invoke the first moment of the transfer equation , @xcite eq . we multiply this by @xmath104 , so that in steady state , to second order it becomes @xmath105 substituting eq . ( [ eq_first_moment ] ) into eq . ( [ eq_zeroth_simplified ] ) gives @xmath106 we average this over space to obtain @xmath107 this is the kompaneets equation with a bulk comptonization contribution to the temperature given by @xmath108 to evaluate eq . ( [ eq_twave_n ] ) , we need to derive @xmath109 to first order in velocity . we start with the radiative transfer equation , eq . ( a1 ) of @xcite , correct to order @xmath45 and @xmath46 , and substitute in eq . ( [ eq_expansion_2 ] ) . in steady state , to first order in velocity , the transfer equation then simplifies to @xmath110 if the density is constant and the velocity field is divergenceless with sinusoidal mode decomposition @xmath111 then the solution is given by ( appendix [ transverse_solution ] ) @xmath112 where @xmath82 is given by eq . ( [ eq : analyticsol ] ) . finally , we evaluate eq . ( [ eq_twave_n ] ) by plugging in eq . ( [ eq_nij ] ) to get @xmath113 since the modes are sinusoidal , eq . ( [ eq_twave4 ] ) simplifies to @xmath114 which is equivalent to eq . ( [ twave2 ] ) . if , however , one solves the kompaneets equation with an escape probability term , @xmath115 , we must also add such a term to the radiative transfer equation , which may affect the occupation number second moment . for an optically thick system , for example , we must require that this term be small compared to the term that sets the diffusion time scale . this term comes from the term @xmath116 , which is approximated by @xmath117 the diffusion timescale is set by the second derivative term , @xmath118 which gives the condition @xmath119 we now discuss the extent to which , in this case , radiation viscous dissipation can be described by a coefficient of kinematic viscosity @xmath120 , which is usually defined by @xmath121 where @xmath122 is the fluid mass density . first we find the radiation shear stress tensor corresponding to eq . ( [ eq_nij ] ) by frequency integrating and subtracting off the scalar radiation pressure , which gives @xmath123 incidentally , we note that eq . ( [ eq_twave_n ] ) can then alternatively be written in terms of the radiation viscous shear stress tensor , @xmath124 according to eqs . ( [ eq_kinematic ] ) and ( [ eq_vis_p ] ) , although we can not in general define a kinematic viscosity coefficient , for any single velocity mode the kinematic radiation viscosity is given by @xmath125 so that the radiation viscous shear stress tensor can be written @xmath126 however , in the optically thick limit ( i.e. @xmath127 ) , the kinematic viscosity is independent of @xmath128 , @xmath129 so that in this limit the kinematic viscosity is well - defined for an arbitrary ( divergenceless ) velocity field . we also note that in this limit eq . ( [ eq_twave4 ] ) agrees with the heating temperature " in @xcite , @xmath130 they also mention that this can be understood in terms of the kinematic radiation viscosity given by eq . ( [ scattering_nu ] ) . from the more commonly cited value , @xmath131 , in , e.g. , @xcite , @xcite , and @xcite . the reason for the discrepancy is that the more commonly cited value , first derived by @xcite , assumes pure absorption , while eq . ( [ scattering_nu ] ) is correct for pure scattering ( @xcite ; @xcite ) . ] to provide physical insight into our analytic solution given by eqs . ( [ twave2 ] ) and ( [ eq : analyticsol ] ) , we now find a heuristic , approximate expression for @xmath132 with a simple physical model motivated by ideas put forth in s04 . to do this , we first consider the wave temperature of a single mode with wavenumber @xmath133 , which we denote @xmath134 . s04 suggested that photons can only sample turbulent velocities on scales @xmath135 , since longer turbulent wavelengths will advect photons back and forth with the flow without allowing the photons to sample " their velocities . a rough interpretation of this reasoning leads to @xmath136 if one then assumes that the wave temperature of an arbitrary field is the sum of the wave temperatures of its modes , then eq . ( [ twave1 ] ) follows . but a more subtle interpretation of this reasoning suggests that @xmath134 is determined by the second moment of the distribution of velocity differences between subsequent scatterings , @xmath137 . in the long wavelength limit , velocity differences between subsequent scatterings are negligible and so the wave power does not contribute to comptonization . in the short wavelength limit , velocity differences allow photons to sample the full power of the wave . this model of comptonization suggests that we define @xmath138 where @xmath139 before proceeding , we note that defining @xmath137 is potentially tricky because the distribution of @xmath140 for a photon is dependent on its current location , in effect introducing correlations into subsequent @xmath140 s . in other words , subsequent @xmath140 s are not independent . but if the escape probability is low enough , a condition we quantify below , then the set of @xmath140 s encountered by a photon before it escapes is indistinguishable from a set of @xmath140 s independently drawn from the position - averaged @xmath140 distribution . the order of @xmath140 s is different in the two cases , but the total photon energy change does not depend on the order because the fractional photon energy change per scattering is small for @xmath141 . with these potential problems accounted for , we proceed to calculate @xmath142 . first we find the proportionality constant between @xmath142 and @xmath137 by evaluating both sides in the short wavelength limit . in this limit , the full wave power must contribute , so @xmath143 . to evaluate @xmath137 in this limit , let @xmath144 be any normalized distribution of velocities . then , @xmath145 therefore , @xmath146 we now calculate @xmath147 for the position - averaged @xmath140 for a single , divergenceless ( i.e. transverse ) mode with wavelength @xmath148 , @xmath149 let @xmath150 be the probability density that a photon travels a displacement @xmath151 between scatterings . then , at a given position @xmath152 , @xmath153 averaging over all positions in a volume @xmath154 , this is @xmath155 for a single mode , eq . ( [ vel_field ] ) gives @xmath156 and @xmath157 the probability that a photon with mean free path @xmath158 travels a distance between @xmath159 and @xmath160 is @xmath161 . let @xmath162 , where @xmath163 is the angle between the photon propagation direction and the @xmath164-axis , so that @xmath165 . then , expressing @xmath166 in spherical polar coordinates and invoking axisymmetry about the @xmath164-axis , equation ( [ eq : dv2deltar ] ) becomes @xmath167 ^ 2 e^{-s / l}.\end{aligned}\ ] ] this is easily evaluated by performing the integral over @xmath164 first , giving @xmath168 . \\ & = 2 \sigma_{{\bf v}}^2\left[1 - \tau_k \tan^{-1}\left(\frac{1}{\tau_k}\right)\right].\end{aligned}\ ] ] eq . ( [ fheur ] ) then gives @xmath169 the limiting cases are @xmath170 by comparison , the exact solution for a single mode is determined by eq . ( [ eq : analyticsol ] ) . our heuristic result is plotted in fig . [ mcdata1 ] , and is remarkably close to the exact solution . in particular , it is a much better approximation than the rough weighting function , eq . ( [ twave3 ] ) , also shown in fig . [ mcdata1 ] . our model based on the second moment of the velocity difference distribution therefore captures the essential physics of comptonization by a single mode . before proceeding to define @xmath171 for an arbitrary velocity field , we quantify the condition that the escape probability be low enough , presupposed by our derivation of @xmath147 . since the distribution of velocity differences , as a function of position , repeats every quarter wavelength , our results should be valid provided photons travel a distance @xmath172 in the @xmath164 direction that is greater than @xmath173 before escaping . for an ( optically thick ) random walk , @xmath174 , where @xmath175 is the average number of scatterings . since @xmath176 , where @xmath177 is the escape probability per unit time during the average time @xmath178 between subsequent scattering events , @xmath179 for optically thick modes . up to a factor of unity , this agrees with eq . ( [ eq_escape_constraint1 ] ) . we now use our model to compute @xmath171 for an arbitrary velocity field in terms of @xmath180 . proceeding analogously to the single mode case , we define @xmath181 where @xmath182 is the temperature corresponding to the full electron energy density of the velocity field . to simplify this , we compute @xmath183 ^ 2 p_{\delta { \bf r}}(\delta { \bf r } ) d^3 \delta { \bf r } d^3 { \bf r } \notag\\ & = \frac{1}{v}\int \sum_i \left({\bf v}_i({\bf r}+\delta { \bf r})-{\bf v}_i ( { \bf r})\right)^2 p_{\delta { \bf r}}(\delta { \bf r } ) d^3 \delta { \bf r } d^3 { \bf r } \notag\\ % & = \frac{1}{v}\int \sum_i \left ( \delta { \bf v}_i(\delta { \bf r},{\bf r } ) % \right)^2 p_{\delta { \bf r}}(\delta { \bf r } ) d^3 \delta { \bf r } % d^3 { \bf r}\notag \\ & = \sum_i \langle ( \delta { \bf v}_i)^2 \rangle . \label{deltavcalc}\end{aligned}\ ] ] to get from line 2 to line 3 we made use of the orthogonality for distinct sinusoidal modes . that is , for two distinct modes , @xmath184 and @xmath185 , @xmath186 , and any displacement @xmath187 , @xmath188 then , since @xmath189 , eqs . ( [ fheur ] ) , ( [ theur ] ) , and ( [ deltavcalc ] ) give @xmath190 or , alternatively , @xmath191 eq . ( [ theur2 ] ) is the same as the exact solution for the wave temperature , eq . ( [ twave2 ] ) , except that here the heuristic weighting function is used . note that in our heuristic derivation it is the orthogonality of distinct modes that allows us to express the wave temperature of an arbitrary velocity field as a sum over the wave temperatures of its modes . unsurprisingly , orthogonality was used analogously to prove eq . ( [ twave2 ] ) . therefore , our model based on the second moment of the velocity difference distribution captures the essential physics of comptonization by an arbitrary , divergenceless velocity field . we now analyse the dependence of @xmath132 on the power spectrum of the turbulence . for the remainder of this section , we write @xmath76 in units of @xmath192 for clarity ( i.e. @xmath193 and @xmath194 denote optically thick and thin scales , respectively ) . if the turbulence is completely optically thin on all scales ( @xmath195 ) , then @xmath196 , independent of the energy spectral index , @xmath197 . however , if some scales in the turbulent cascade are optically thick , then @xmath197 will affect @xmath132 . for the case where all scales in the turbulence are optically thick ( @xmath198 ) , eq . ( [ eq : analyticsol ] ) implies that @xmath199 integrating this over an energy spectrum @xmath200 then gives @xmath201 for a broad power spectrum , i.e. @xmath202 , this simplifies to @xmath203 this is illustrated in fig . [ ueff_p ] . note that @xmath204 drops significantly for @xmath205 , because then the energy bearing modes are on the largest scales . these are the most optically thick and therefore the most downweighted in their contribution to bulk comptonization . we next analyse whether @xmath204 , for a given spectral index @xmath197 and range of modes , @xmath206 , is dominated by small or large scales . in other words , we examine which turbulent modes in a given spectrum contribute most to bulk comptonization . the relative contribution of a scale with wavenumber @xmath76 is @xmath207 where , from eq . ( [ eq : analyticsol ] ) , @xmath208 for optically thick ( @xmath209 ) scales , and @xmath210 for optically thin ( @xmath211 ) scales . now consider an underlying power spectrum with some @xmath212 and @xmath213 . we see that for @xmath214 the exponent in eq . ( [ contribution ] ) is always positive , and so small scales contribute most to bulk comptonization , regardless of @xmath212 and @xmath213 . this is physically intuitive ; for @xmath214 , the turbulent power is concentrated on small scales . since the weighting factor @xmath82 also favors small scales , they of course contribute most . for @xmath215 , the exponent is always negative , and so large scales always contribute most . in this case , the turbulent power is so concentrated on large scales that they contribute more even though @xmath82 favors small scales . for @xmath216 , we first consider the part of the spectrum with @xmath194 ( if it exists ) . since @xmath210 , small scales contribute more than large scales for these modes . now consider the part of the spectrum with @xmath193 ( if it exists ) . here , large scales contribute more than small scales . therefore , looking at the entire power spectrum , it is intermediate scales that contribute most , assuming it is broad enough to include regions of both small and large @xmath76 . if it is not sufficiently broad , then whether small or large scales contribute most depends on @xmath212 and @xmath213 relative to @xmath217 ( the optically thin to thick transition wavenumber ) . these results are depicted in fig . [ ueff_dist ] . the curve in this figure shows the values of @xmath197 and @xmath76 such that the derivative of @xmath218 is zero , using the full analytic expression for @xmath82 from eq . ( [ eq : analyticsol ] ) . to connect this figure to our discussion , draw a horizontal line from @xmath212 to @xmath213 at a given value of @xmath197 . if the line lies in the lower ( upper ) region , then small ( large ) scales contribute most . if the line straddles the two regions , then for the part of the line that lies in the lower region small scales contribute most , and for the part that lies in the upper region large scales contribute the most . in this case , then , for the entire spectrum it is the scales that straddle the curve which contribute most . note that for @xmath214 and @xmath215 a spectrum can never straddle the curve , whereas for a kolmogorov spectrum ( @xmath219 ) , e.g. , it can , if @xmath220 . @xcite conjectured that comptonization by turbulence can be treated as thermal comptonization by solving the kompaneets equation with an equivalent wave " temperature . in section [ sec_incompressible ] we proved this under certain conditions , one of which is that the turbulence be divergenceless . in this section we investigate comptonization by velocity fields with non - zero divergence , an effect that usually can not be treated as thermal comptonization , with the aim of understanding how it impacts radiation spectra in generic , turbulent regions of stratified accretion disc atmospheres . while we do not have an exact solution for comptonization by an arbitrary velocity field with non - zero divergence , we have solutions for two limiting cases . the trivial case is the optically thin case , i.e. when the mean free path is significantly larger than the largest length scale in the turbulence , @xmath221 . electron velocities sampled " by photons are uncorrelated and so compton scattering should depend on only the total spatial average distribution of electron velocities . this is , therefore , the one case where comptonization by a velocity field with non - zero divergence can be treated as thermal comptonization , by solving the kompaneets equation with @xmath132 given by @xmath222 note that although we formally write @xmath132 as a function of @xmath84 , in this limit @xmath132 is independent of @xmath84 . energy exchange first order in velocity vanishes since photons are equally likely to downscatter as they are to upscatter . bulk comptonization is then solely due to radiation viscous dissipation ( see section [ sec_general ] ) . as viscous effects are second order in velocity , it is unsurprising that they can be characterized by a temperature . we also note that in this limit the wave temperature is the same as that for a divergenceless velocity field , eq . ( [ eq_limits ] ) . optically thin bulk comptonization is therefore a single phenomenon that depends on only the mean square speed of the velocity field . to arrive at this result with a more formal approach , we start with the zeroth moment of the radiative transfer equation , eq . ( [ eqpsaltis ] ) . in the limit of optically thin turbulence the radiation variables must be homogeneous and isotropic , so that @xmath223 , and @xmath224 . then , averaging eq . ( [ eqpsaltis ] ) over the largest scale @xmath35 gives @xmath225\biggr\}. \label{eqkomp}\end{aligned}\ ] ] this is the kompaneets equation , with the contribution from the velocity field to the comptonization temperature given by eq . ( [ compressible_thin ] ) . in the optically thick case , i.e. when the photon mean free path is significantly smaller than the smallest scale in the turbulence , the lowest order energy exchange is the work done by radiation pressure to compress the gas , since it is first order in velocity and since radiation viscous effects are suppressed by a factor @xmath44 ( section [ sec_incompressible ] ) . we focus on the extremely optically thick case , which we define as the limit in which photon diffusion is negligible relative to photon advection , so that the photon and gas fluids are strongly coupled . if we define @xmath226 , the photon number density at energy @xmath48 , and @xmath227 , the total photon number density , then the advection and diffusion fluxes are given by @xmath228 and @xmath229 respectively . the extremely optically thick limit is then given by @xmath230 in this case , velocity convergence corresponds to compression in which gas mechanical energy is transferred locally to the photons . we expect that photons with wavelength @xmath231 are effectively compressed at a rate given by the velocity difference across @xmath231 , @xmath232 so that , for example , a locally thermal photon distribution remains thermal and only changes temperature , completely analogous to the evolution of the cosmic microwave background radiation under the expansion of the universe . this is equivalent to a fractional energy change per scattering given by @xmath233 since @xmath234 we now confirm that eq . ( [ upscattering_fraction ] ) correctly describes extremely optically thick compton scattering in a converging or diverging flow , both by providing a heuristic argument and by deriving it from the radiative transfer equation . before proceeding , we note that the study of photon upscattering by a converging velocity field can be traced back to @xcite and @xcite , who , along with later authors , made detailed spectral calculations for specific velocity fields in shocks and spherically accreting systems . in fact , eq . ( [ upscattering_fraction ] ) can equivalently be stated as the upscattering timescale @xmath235 given in @xcite . in this section , by contrast , we have been investigating how this effect manifests itself locally in a generic , turbulent region of a stratified disc atmosphere , with the goal of resolving and interpreting it in spectral calculations of radiation mhd simulations . we have been focusing on the extremely optically thick case , in which the photon and gas fluids are strongly coupled , because the physics is both relevant and intuitive . in the moderately optically thick case , on the other hand , i.e. @xmath236 such as photon upscattering in a radiation pressure dominated shock @xcite , diffusion competes with advection so that photon distributions at neighboring fluid elements mix . photon upscattering in such a converging flow may not be viewed as simply the compression of a photon fluid strongly coupled to the gas , and photon upscattering in which a photon thermal distribution is not preserved can occur . to heuristically derive eq . ( [ upscattering_fraction ] ) , consider a disturbance converging in the @xmath237 direction given by @xmath238 for a 3d random walk , the average distance between scatterings traveled by a photon in the direction of convergence is @xmath239 . since the fractional energy change per scattering for low energy photons is @xmath53 , at @xmath240 this gives @xmath241 in agreement with eq . ( [ upscattering_fraction ] ) . we now derive eq . ( [ upscattering_fraction ] ) with the radiative transfer equation , eq . ( [ eqpsaltis ] ) . if we ( 1 ) omit stimulated scattering terms to facilitate comparison with simulations , ( 2 ) substitute in the standard closure relations for the first moment in the optically thick limit , @xmath242 and @xmath243 and ( 3 ) substitute in the photon number density @xmath244 , then the radiative transfer equation to second order in velocity and first order in @xmath245 becomes @xmath246 neglecting stimulated scattering , this is @xcite eq . ( 18 ) , cast in the physically revealing form of a fokker - plank equation . the terms inside the divergence operator correspond to spatial drift ( i.e. photon advection ) and spatial diffusion , respectively . the next term corresponds to energy drift due to photon upscattering ( downscattering ) in the presence of a converging ( diverging ) velocity field . the remaining terms correspond to energy drift and diffusion due to thermal comptonization . note that even though @xcite start with a zeroth moment equation correct only to first order in @xmath53 , their resulting equation is the same because with the standard closure relation ( eq . [ closure1 ] ) the second order terms cancel . the fractional energy change per scattering given by eq . ( [ upscattering_fraction ] ) follows from the bulk upscattering term . in the extremely optically thick limit the spatial diffusion term is negligible , and so photons are advected with the velocity field and upscatter according to eq . ( [ upscattering_fraction ] ) . the effect of this process on the emergent spectrum of the disc will depend primarily on how effectively photons are able to escape from converging ( or diverging ) regions to the observer . we have investigated how photon spectra produced by turbulent comptonization depend on the properties of the turbulence itself , and how to resolve and interpret this effect in radiation mhd simulations . our principle results are as follows . first we address the conjecture of @xcite that turbulent comptonization can be treated as thermal componization by solving the kompaneets equation with an equivalent wave " temperature . for comptonization by divergenceless turbulence , which is due to the radiation viscous shear stress tensor and is therefore an effect second order in velocity , this conjecture holds provided that the density is uniform and the escape probability is sufficiently small ( eq . [ eq_escape_constraint1 ] ) . the correct wave temperature can be expressed in terms of the radiation viscous shear stress tensor ( eqs . [ eq_twave_n ] and [ eq_twave_p ] ) and is in turn given by eq . ( [ twave2 ] ) . this is a weighting over the velocity power at each scale in the turbulence , with the weighting function given by eq . ( [ eq : analyticsol ] ) . optically thin scales have unit weight factors , implying that the power at these scales contributes fully to the wave temperature . optically thick scales , on the other hand , are downweighted by a factor proportional to one over the square of the optical depth across those scales . this downweighting arises because bulk comptonization is sensitive to the difference in bulk fluid velocity between subsequent scatterings , as demonstrated by our heuristic argument in section [ sec_incompressible ] . these velocity differences are much smaller than the velocity amplitude on some scale if the photon mean free path is much smaller than that scale . we note that our expression for the wave temperature of divergenceless turbulence assumes a uniform density . furthermore , it is non - local in that it is expressed in terms of the mode decomposition of the entire velocity field . although a generic velocity field has density variations , if the largest turbulent length scale @xmath35 is smaller than the scale on which the density varies then we can define the wave temperature for regions of order @xmath35 . if the density also varies on smaller scales , then we can define the wave temperature for smaller regions by omitting contributions from larger modes . assuming these modes would be significantly downweighted anyway , this procedure would likely capture nearly all comptonized power in the turbulence . for turbulence with non - zero divergence , bulk comptonization energy exchange is due to both work done by radiation pressure and radiation viscous dissipation . since pressure work is an effect first order in velocity , it can not be treated with an equivalent temperature . but in the optically thin limit , i.e. when the mean free path is significantly larger than the largest length scale in the turbulence , energy exchange first order in velocity vanishes since photons are equally likely to downscatter as they are to upscatter . bulk comptonization by a velocity field with non - zero divergence is then solely due to radiation viscous dissipation , and it may be treated as thermal comptonization . the wave temperature is given by the full power in the turbulence , eq . ( [ compressible_thin ] ) , directly analogous to the optically thin limit of divergenceless turbulence . in the optically thick case , i.e. when the mean free path is significantly smaller than the smallest scale in the turbulence , the lowest order energy exchange is the work done by radiation pressure to compress the gas , since it is first order in velocity and since radiation viscous effects are suppressed by a factor @xmath44 . the extremely optically thick limit , defined as the limit in which spatial photon diffusion is negligible relative to photon advection , is given by eq . ( [ eq_extremely_optically_thick ] ) . in this limit , the photon and gas fluids are strongly coupled , so that velocity convergence corresponds to compression in which gas mechanical energy is transferred locally to the photons . photons upscatter with fractional energy change given by eq . ( [ upscattering_fraction ] ) . in this process a locally thermal photon distribution remains thermal and only changes temperature , analogous to the evolution of the cosmic microwave background radiation under the expansion of the universe . unlike energy exchange due to viscous dissipation , this process is reversible . the effect of this process on the emergent spectrum of the disc will depend primarily on how effectively photons are able to escape from such regions to the observer . for moderate optical depths , such as photon upscattering in a radiation pressure dominated shock @xcite , diffusion competes with advection so that photon distributions at neighboring fluid elements mix . the effect of photon upscattering in a converging region on local spectra is more complex and depends on the details of the flow . it is important to note that optically thick comptonization by a velocity field with non - zero divergence may be very sensitive to the time dependence of the velocity field . this is important because usually post - processing monte - carlo simulations invoke time - independent atmospheres with the assumption that they approximately capture the effects of interest . for example , consider comptonization in a converging region . a time - independent velocity field results in the accumulation of photons and subsequent upscattering to high energies at the point of zero velocity . if the region is near the photosphere , the emergent spectrum will be strongly upscattered . but in a time - dependent velocity field photons have a limited time to upscatter before converging regions become diverging regions , and so the effect on the emergent spectrum is significantly different . real accretion flows are spatially stratified . simulations of magnetorotational turbulence generally indicate that this turbulence dominates the fluid velocities in the midplane regions of the accretion flow . this turbulence is largely incompressible ( divergenceless ) , although it generally excites compressible spiral acoustic waves @xcite . sufficiently far from the midplane , magnetic forces always dominate thermal pressure forces , and support the flow vertically against the tidal gravitational field of the compact object . such regions are dominated by parker instability dynamics , and exhibit considerable compressive behavior ( i.e. the flow has non - zero divergence ) with large density fluctuations @xcite . in order for radiation mhd simulations to properly account for energy exchange due to turbulent comptonization so that post - processing monte carlo simulations of photon spectra are self - consistent , they must include energy terms second order in velocity and use a moment closure scheme that correctly captures contributions to the radiation stress tensor that are first order in velocity . the appropriate energy equation source terms in lab and fluid frame variables are given by eqs . ( [ integrated1 ] ) and ( [ eq_intergrated2 ] ) , respectively . flux - limited diffusion and the m1 closure scheme are insufficient because they neglect the lowest order contribution to the radiation stress tensor . in this case , instead of exchanging energy with photons , turbulence will eventually cascade down to the gridscale ( or viscous or resistive scale if the code has explicit viscosity or resistivity ) and increase the internal energy of the gas . gas internal energy may then be exchanged with photons through thermal comptonization . the omission of viscous dissipation by radiation therefore does not prevent the eventual transfer of turbulent energy to radiation , but it may have other physical effects which can in turn affect radiation spectra . radiation pressure dominated accretion flows may have bulk velocities that exceed thermal electron velocities . turbulent comptonization , which is the effect of comptonization by bulk , turbulent electron motions on photon spectra , should be significant in such sources . for turbulence with zero divergence , bulk comptonization is equivalent to a radiation viscous shear stress tensor acting on the fluid . the resulting comptonized spectrum can be calculated by solving the kompaneets equation with an equivalent wave " temperature , which is a weighted sum over the power present at each scale in the turbulent cascade ( eq . [ twave2 ] ) . bulk comptonization by turbulence with non - zero divergence is due to both pressure work and radiation viscous dissipation . in the limit of optically thin turbulence , pressure work has negligible effect on photon spectra and radiation viscous dissipation alone can be treated as thermal comptonization with a temperature equivalent to the full turbulent power ( eq . [ compressible_thin ] ) . in the limit of extremely optically thick turbulence ( eq . [ eq_extremely_optically_thick ] ) , radiation viscous dissipation is suppressed , and the evolution of local photon spectra can be understood in terms of compression and expansion of the strongly coupled photon and gas fluids . photons upscatter with fractional energy change given by eq . ( [ upscattering_fraction ] ) . in this process a locally thermal photon distribution remains thermal and only changes temperature . the effect of this process on the emergent spectrum of the disc will depend primarily on how effectively photons are able to escape from such regions to the observer . modeling turbulent comptonization will ultimately require detailed analysis of radiation mhd simulations and post - processing monte - carlo simulations . by exploring how photon spectra produced by turbulent comptonization depend on the properties of the turbulence itself , we have laid the groundwork necessary to make sure this effect is both captured and correctly interpreted in these simulations . we thank aristotle socrates for useful discussions . we also thank donald pakey for forwarding errata to the paper by @xcite , and li xue and feng yuan for hosting ob at xiamen university and shanghai astronomical observatory where part of this work was carried out . we gratefully acknowledge support by nasa astrophysics theory program grant nnx13ag61 g , and the international space science institute ( issi ) in bern . bachetti m. et al . , 2013 , apj , 778 , 163 blaes o. m. , hirose s. , krolik j. h. , 2007 , apj , 664 , 1057 blaes o. m. , krolik j. h. , hirose s. , shabaltas n. , 2011 , apj , 733 , 110 blandford r. d. , payne d. g. , 1981a , mnras , 194 , 1033 blandford r. d. , payne d. g. , 1981b , mnras , 194 , 1041 chan k.l . , jones b.j.t . , 1975 , apj , 200 , 454 colpi m. , 1988 , apj , 326 , 223 collinson j. s. , ward m. j. , done c. , landt h. , elvis m. , & mcdowell j. c. , 2015 , mnras , in press [ arxiv:1502.05850 ] davis s. w. , blaes o. m. , hirose s. , krolik j. h. , 2009 , apj , 703 , 569 davis s. w. , blaes o. m. , hubeny i. , turner n. j. , 2005 , apj , 621 , 372 davis s. w. , done c. , blaes o. m. , 2006 , apj , 647 , 525 done c. , davis s. w. , jin c. , blaes o. m. , ward m. , 2012 , mnras , 420 , 1848 gladstone j. c. , roberts t. p. , done c. , 2009 , mnras , 397 , 1836 guan x. , gammie c. f. , 2011 , apj , 728 , 130 haardt f. , maraschi l. , 1991 , apj , 380 , 51 heinemann t. , papaloizou j. c. b. , 2009a , mnras , 397 , 52 heinemann t. , papaloizou j. c. b. , 2009b , mnras , 397 , 64 hirose s. , krolik j. h. , stone j. m. , 2006 , apj , 640 , 901 hirose s. , krolik j. h. , blaes o. m. , 2009 , apj , 691 , 16 hubeny i. , blaes o. m. , krolik j. h. , agol e. , 2001 , apj , 559 , 680 jiang y .- f . , stone j. m. , davis s. w. , 2013 , apj , 778 , 65 jiang y .- f . , stone j. m. , davis s. w. , 2014a , apj , 784 , 169 jiang y .- f . , stone j. m. , davis s. w. , 2014b , apj , 796 , 106 kawashima t. , ohsuga k. , mineshige s. , yoshida t. , heinzeller d. , matsumoto r. , 2012 , apj , 752 , 18 kolehmainen m. , done c. , daz trigo m. , 2011 , mnras , 416 , 311 laor a. , netzer h. , 1989 , mnras , 238 , 897 lightman a. p. , white t. r. , 1988 , apj , 335 , 57 levermore c. d. , 1984 , j. quant . transfer , 31 , 149 magdziarz p. , blaes o. m. , zdziarski a. a. , johnson w. n. , & smith d. a. , 1998 , mnras , 301 , 179 masaki i. , 1971 , pub . japan , 23 , 425 mckinney j. c. , tchekhovskoy a. , sadowski a. , narayan r. , 2014 , mnras , 441 , 3177 mehdipour m. et al . , 2011 , a&a , 534 , a39 mihalas d. , mihalas b. , 1984 , foundations of radiation hydrodynamics ( new york : oxford univ . press ) middleton m. , done c. , ward m. , gierliski m. , schurch n. , 2009 , mnras , 394 , 250 miller k. a. , stone j. m. , 2000 , apj , 534 , 398 niedwiecki a. , zdziarski a. a. , 2006 , mnras , 365 , 606 ohsuga k. , mineshige s. s. , 2011 , apj , 736 , 2 payne d. g. , blandford r. d. , 1981 , mnras , 196 , 781 pozdnyakov l. a. , sobol i. m. , & sunyaev r. a. , 1983 , astrophys . space phys . , 2 , 189 psaltis d. , 2001 , apj , 555 , 786 psaltis d. , & lamb f. k. , 1997 , apj , 488 , 881 ross r. r. , fabian a. c. , 1993 , mnras , 261 , 74 sadowski a. , narayan r. , tchekhovskoy a. , abarca , d. , zhu y. , mckinney , j. , 2015 , mnras , 447 , 49 sadowski a. , narayan r. , tchekhovskoy a. , zhu y. , 2015 , mnras , 429 , 3533 shakura n. i. , sunyaev r. a. , 1973 , a&a , 24 337 shimura t. , takahara f. , 1995 , apj , 445 , 780 socrates a. , 2010 , apj , 719 , 784 socrates a. , davis s. w. , blaes o. m. , 2004 , apj , 601 , 405 stobbart a .- m . , roberts t. p. , wilms j. , 2006 , mnras , 368 , 397 straumann n. , 1976 , helv . acta , 49 , 269 svensson r. , zdziarski a. a. , 1994 , apj , 436 , 599 takeuchi s. , ohsuga k. , mineshige s. , 2013 , pasj , 65 , 88 thomas , l.h . , 1930 , q.j . math . , 1 , 329 thompson c. , 1994 , mnras , 270 , 480 titarchuk l. , mastichiadis a. , kylafis n. , 1997 , apj , 487 , 834 weinberg s. , 1971 , apj , 168 , 175 weinberg s. , 1972 , gravitation and cosmology ( new york : wiley ) yoshida t. , isobe n. , mineshige s. , kubota a. , mizuno t. , saitou k. , 2013 , pasj , 65 , 48 zeldovich ya . b. , illarionov a. f. , sunyaev r. a. , 1972 , sov . jetp , 35 , 643 beginning with eq . ( [ eq_transfer_first_order ] ) , we prove that the steady state occupation number second moment for a divergenceless velocity field of uniform density to first order in velocity is given by eq . ( [ eq_nij ] ) . first we find the solution for a single mode given by @xmath247 for this mode , the transfer equation is @xmath248 where we assume @xmath249 and @xmath250 , which we can check later . then , the transfer equation is @xmath251 first we address the @xmath164-dependence . because this equation is linear , it must be that @xmath252 is a superposition of a sine and a cosine , @xmath253 this gives two coupled integral equations for @xmath254 and @xmath255 , @xmath256 and @xmath257 it seems that both @xmath254 and @xmath255 are proportional to one power of @xmath258 . writing @xmath259 and @xmath260 , and @xmath261 , we then obtain @xmath262 and @xmath263 observing that letting @xmath264 and @xmath265 be odd and even , respectively , is a consistent solution , the @xmath266 integral of @xmath265 vanishes , and the two equations can be combined to give a single equation for @xmath264 , @xmath267 we then solve this equation with a series expansion of odd powers of @xmath268 : @xmath269 gives @xmath270 which is just the expansion of @xmath271 , so that @xmath272 substituting this back into the integral equation gives @xmath273 , which completes the solution . so far then , we have @xmath274\sin \left({2\pi z\over\lambda}\right)\biggr\},\end{aligned}\ ] ] where @xmath275 . \label{eq : qexact}\ ] ] note that @xmath276 in the optically thin limit , and @xmath277 in the optically thick limit . this solution is consistent with our assumptions and solves eq . ( [ eq_transfer_first_order ] ) to first order in velocity . the second moment is @xmath278 where @xmath279 is given by eq . ( [ eq : analyticsol ] ) . ( [ eq_transfer_first_order ] ) is linear , the solution for an arbitrary , divergenceless velocity field of uniform density is then given by eq . ( [ eq_nij ] ) .

emission at 511 kev , the characteristic signature of positron annihilation , has been observed in the direction of the galactic centre ( gc ) , since the 1970s . apparently diffuse gamma - ray emission at approximately this energy was first detected in 1970 by the balloon - borne experiments of the rice group ( johnson , harnden , & haymes 1972 ; johnson & haymes 1973 ; haymes et al . 1975 ; preliminary indications were also reported in haymes et al . 1969 ) and was confirmed as positron annihilation emission in 1978 by the balloon - borne experiments of the bell - sandia group ( leventhal , maccallum , & stang 1978 ; leventhal et al . high - energy balloon experiments and space observatories through the early 1990s continued to detect the 511 kev emission ; however the relatively low spatial resolution of these detectors prevented determination of the true location and distribution of the emission ( see purcell et al . 1997 , jean et al . 2003 , or teegarden et al . 2005 for a summary of these early observations ) . specifically , from these data it is unclear whether the emission is truly diffuse , or if it originates either from a single discrete source ( _ e.g . _ sgr a * or 1e1740.7 - 2942,``the great annihilator '' ) or from a small number of discrete but unresolved sources . some early detections suggested time variability in the signal , indicating a small number of discrete sources , but these variations were not confirmed in subsequent observations ( _ e.g . _ purcell et al . 1997 ; teegarden et al . 2005 ) . with the advent of space observatories of increasing sensitivity , spectral coverage , and spatial resolution , the resultant improvement in data quality now provides much stronger constraints on the origin of the emission . in particular , observations of the gc by the spi spectrometre on the satellite integral ( the international gamma - ray astrophysicsl laboratory ) have recently produced the most detailed map of the anomalous 511 kev emission to date ( kndlseder et al . 2005 ; weidenspointner et al . 2006 , 2007 ) . the spi / integral map clearly shows this emission arising from the central @xmath01.5 kpc ( @xmath2@xmath3 ) of the galaxy , with a fainter component of 511 kev flux detected from the remainder of the galactic disk . the smoothness of the emission constrains the number of discrete sources responsible to be in excess of @xmath0 8 ( kndlseder et al . 2005 ) , as do point source limits of @xmath4 ph @xmath5 s@xmath6 from searches with the ibis imager on integral ( de cesare et al . most recently , weidenspointner et al . ( 2008 ) have detected a longitudinal asymmetry in the disk component of the emission , which matches the asymmetry in the distribution of the brightest low - mass x - ray binaries ( lmxbs ) detected by the ibis instrument on integral , suggesting that these @xmath7 objects may account for much of the disk component of the 511 kev emission . diffuse emission from the galactic centre , unaccounted for by known sources , has also been observed at several other wavelengths , notably high - frequency radio ( 2163 ghz ) with wmap ( finkbeiner 2004a ) and soft x - rays ( 110 kev ) with _ e.g. _ heao and chandra ( worrall et al . 1982 , muno et al . the origin and nature of the diffuse emission is puzzling in each case . in the case of the diffuse x - ray component , a population of discrete but unresolved point sources was originally postulated as a possible source of the emission ( see _ e.g. _ skibo et al . 1997 , valinia & marshall 1998 ) . with the advent of high - resolution x - ray imaging using chandra , a large new population of low x - ray luminosity sources has indeed been identified in the central 2@xmath3@xmath80.8@xmath3of the gc ( 300@xmath8120 pc at a gc distance of 8.5 kpc ; wang et al . 2002 , muno et al . 2003 , muno et al . 2006 ) . many of these @xmath04200 discrete x - ray sources are likely to be accreting binaries , including high and low mass x - ray binaries and cataclysmic variables . however , it is worth noting that these detected sources only account for 10% of the previously observed diffuse 2 - 10 kev emission ( ebisawa et al . 2001 , muno et al . 2003 ) , so an even larger population of still fainter sources ( with @xmath9@xmath10erg@xmath10s@xmath6 ) could exist at the gc ( but see muno et al . 2004 who argue that the spectral characteristics of the remaining diffuse x - ray emission are inconsistent with a stellar x - ray source origin ) . in any case , it is not clear how much these new populations might contribute to the diffuse flux at lower ( _ e.g . _ wmap ) or higher ( _ e.g . _ integral ) energies . in this paper , we consider the hypothesis that jet outflows from x - ray binary systems ( xrb ) , which are accretion - powered mass - transferring binaries containing a black hole ( bh ) or neutron star ( ns ) , are the main source of the 511 kev emission in the gc . this idea has been explored previously by various authors , notably ramaty & lingenfelter ( 1979 ) , prantzos ( 2004 ) , kndlseder et al . ( 2005 ) , and guessoum et al . these previous studies generally focused on the large - scale discrete jet ejections produced by the class of luminous black hole xrbs known as `` microquasars '' . however , it is now thought that luminous black hole xrbs much more commonly produce lower - luminosity `` steady '' jets , and that these outflows are `` on '' for a substantially greater fraction of the xrb duty cycle than the large - scale ejection events ( gallo et al . we note that neutron star x - ray binaries are , in principle , also possible sources of jet positrons . the fraction of low magnetic field ns xrbs with outflows could be as high as @xmath0 100% ( fender 2006 ; migliari & fender 2006 ) . however , they show a different scaling than black hole xrbs between x - ray luminosity and radio luminosity ( i.e. @xmath11 in bh systems and @xmath12 in ns systems ; miglari & fender 2006 ) . as a result of this scaling , the jets from quiescent neutron star x - ray binaries are likely to be several orders of magnitude weaker than those from quiescent bh x - ray binaries , if one assumes the same scaling relations continue into quiescence . more likely , though , is that the scaling relation becomes even steeper for neutron stars as they fade more deeply into quiescence than the current radio flux limits allow us to probe ; the emission from the faintest neutron star xrbs is generally dominated by thermal crustal emission from cooling neutron stars , rather than by accretion power ( _ e.g . _ rutledge et al . 2001 ) . also , while some luminous high - mass x - ray binaries ( hmxbs ) have been observed to emit jets ( e.g. cyg x-1 , cyg x-3 ) , these systems are predominantly located in the galactic disk ( grimm et al . thus these canonical bh hmxbs of which there are only a few known in the entire milky way are not a class of objects which can cause an excess of high - energy flux in the bulge relative to the galactic disk . recent integral observations have detected a population of lower - luminosity , x - ray hard , high - mass xrbs including highly - obscured hmxbs and `` supergiant fast x - ray transients '' , a number of which are located in the bulge ( chaty et al . however , there is as yet no evidence for jet outflows in these hmxbs . furthermore , the nature of the compact objects in these systems is currently not known ; but we note that of those canonical hmxbs for which the nature of the compact object has been identified , the majority contain neutron stars rather than black holes . finally , in a detailed study of the spatial distribution of integral - detected xrbs ( including the new highly - obscured systems ) , lutovinov et al . ( 2005 ) find that the angular distribution of hmxbs in the inner galaxy is significantly different from that of lmxbs specifically , lmxbs are clearly the dominant population within the bulge , significantly overabundant as compared with the hmxbs in this region . thus while we can not rule out a high - energy flux contribution in the bulge from these newly discovered low - luminosity hmxbs , we will not consider them further here . therefore in this paper we primarily consider jets from low - mass black hole binaries . it seems plausible that steady , low - luminosity outflows from galactic bh xrbs contribute substantially to the annihilation line emission , and possible also to the diffuse emission observed at other wavelengths . the outline of the paper is as follows . in section 2 , we first review the 511 kev observations of the gc and discuss possible sources for the positrons . in section 3 , we outline a simple model for positron production in ( quiescent ) lmxb jets . in section 4 , we briefly examine the unexplained diffuse gc emission observed at microwave and x - ray wavelengths , and consider whether this emission could also originate in outflows from low - luminosity lmxbs . in section 5 , we discuss the recent association of the disk component of the 511 kev emission with bright lmxbs detected by ibis , pointing out some unresolved problems with this association . we conclude by proposing several observational tests that may help elucidate the nature of the population responsible for the 511 kev emission . since its initial discovery almost 40 years ago , our picture of the annihilation emission from the gc has gradually become clearer . the emission amounts to @xmath13 photons @xmath14 s@xmath6 , coming from a region roughly @xmath15 in radius around the gc . assuming the mean distance to the positron sources is the distance to the gc , 8.5 kpc , this flux corresponds to an integrated luminosity of @xmath16 photons / s emitted within a region 1.5 kpc in radius . after initial suggestions of temporal variability in the flux , extensive observations over the 1990s have ruled this out at any substantial level ( see purcell et al . 1997 for a summary of this evidence ) . the detailed spatial distribution of the emission has become much clearer since observations first by osse on cgro , and then more recently by spi on integral . the most recent analyses of 4 years of spi / integral data ( weidenspointner et al . 2008 ) suggest two main components : a central bulge and an asymmetric disk . the bulge component is reasonably well described by a single gaussian with a fwhm of 6@xmath17 , but even better described by a superposition of two gaussians of fwhm of 3@xmath17 and 12@xmath17 , or alternately a compact , symmetric bulge component from @xmath180.5 kpc , and an extended shell of emission from @xmath191.5 kpc . the disk component is now detected at @xmath20 in 4 years of spi , and appears to be be asymmetric at 3.8@xmath21 significance ( weidenspointner et al . 2008 ) , with 1.8 times more emission at negative longitudes than at positive ones . the total flux from the disk is @xmath22 photons @xmath14 s@xmath6 , _ i.e. _ comparable to the bulge flux . there is also marginal evidence for emission more than 1.5 kpc from gc ( kndlseder et al . 2005 ) , possibly extending out as much as 40@xmath17/5 kpc ( bouchet et al . 2008 ) , though detection here is hampered by the very uneven exposure maps of integral away from the gc . on the other hand there has been no significant detection of the emission more than 4050 degrees away from the gc ( teegarden et al . 2005 ; weidenspointner et al . 2007 ) . detailed analysis of the annihilation line profile , the positronium continuum below 511 kev , and the spectrum of the diffuse gamma - ray background at higher energies place further limits on the origin of the line . integral and comptel measurements of the diffuse background at 110 mev indicates that the positrons must be produced at relatively low energy , since otherwise inflight annihilation of relativistic positrons off background electrons would produce a visible bump in the spectrum . this constraint on the injection energy , first identified by agaronyan & atoyan ( 1981 ) , places an upper limit of @xmath03 mev on the mass of a light dark matter candidate producing positrons through annihilation in a neutral medium ( beacom & yksel 2006 ; note that in this case internal bremsstrahlung in the annihilation process also contributes to the high - energy emission ) . allowing for a partially ionized medium ( and for production mechanisms without associated internal bremsstrahlung emission ) , the more general limit on the injection energy is @xmath048.5 mev ( sizun , cass , & schanne 2006 , 2007 ) . furthermore , low - energy positrons rarely annihilate directly ; instead the majority form positronium , a short - lived bound system composed of an electron - positron pair . positronium then decays either from the singlet ( `` para - positronium '' , spins anti - parallel ) state , via the emission of two photons at 511 kev , or from the triplet ( `` ortho - positronium '' , spins parallel ) state by the emission of three photons , leading to a continuum below 511 kev . detailed analysis of the line profile and line - to - continuum ratio can constrain the properties of the medium in which the positrons propagate and annihilate . the recent analyses of churazov et al . ( 2005 ) and jean et al . ( 2006 ) indicate that the dominant emission source can not be located in the very hot or very cold components of the ism , but that annihilation probably occurs in the warm neutral and ionized medium . given these constraints and the expected lifetimes for positrons in different components of the ism , the implication is that positrons are either produced in , and annihilate in , the warm medium , or are produced in the hot medium but propagate further and annihilate in the warm medium at the edge of hot bubbles . in the former case they would travel only 50100 pc ( @xmath23 at gc ) from their source before annihilating , and in even in the latter they would annihilate within @xmath24 pc of their source ( jean et al . thus , in the absence of large - scale magnetic fields ( which could cause the positrons to propagate even further from their sources ; prantzos 2006 ) the spatial extent of the observed emission implies sources distributed over a similar area on the sky . since its discovery , many possible sources have been invoked to explain the 511 kev emission from the galactic centre . a number that were initially consistent with early detections are now disfavoured , given more accurate maps and flux measurements . some sources , _ e.g. _ cosmic rays interacting with the intergalactic medium ( ramaty et al . 1970 ; lingenfelter & ramaty 1982 ) , or populations of young stellar objects such as pulsars ( sturrock 1971 ) , or supernovae or wolf - rayet stars ( via the radioactive nuclei they produce clayton 1973 ; ramaty & lingenfelter 1979 ; signore & vedrenne 1988 ; woosley & pinto 1988 ; lingenfelter & ramaty 1989 ; milne et al . 2002 ) , would produce a more flattened distribution on the sky , inconsistent with the bulge component . some other possible candidates , _ e.g. _ sgr a * ( lingenfelter & ramaty 1982 ; rees 1982 ; ozernoy 1989 ; ramaty et al . 1992 ) , a single grb or hypernova at galactic centre ( lingenfelter & hueter 1984 ) , or a population of classical novae ( purcell et al . 1997 ) , should appear as a single point source and/or show temporal variability ( although models where stronger emission of positrons from sgr a * occured at some time in the past may be possible see cheng , chernyshov & dogiel 2006 and 2007 ; and totani 2006 ) . while some of these sources may account for part of the annihilation flux ( _ e.g . _ the radioactive decay of @xmath25al from young massive stars should account for 20 - 30% of the 511 kev emission in the central part of the galactic disk ; bouchet et al . 2008 ) , the majority of the 511 kev flux now seems most likely to come from a distributed population of faint sources that traces the old stellar bulge . we refer the reader to _ e.g. _ kndlseder et al . ( 2005 ) for further discussion of the alternatives . of sources associated with old stellar populations , type ia sne are too rare to account for much of the flux . hypernovae / grb progenitors ( cass et al . 2004 ) or grbs ( parizot et al . 2005 ; bertone et al . 2006 ) might produce enough positrons , but it is not clear they could necessarily reproduce the smoothly distributed emission seen by spi . another possibility worth mentioning is pulsars . rotationally powered pulsars are well known to produce high energy particles . in recent years , it has become clear through the observation of a synchrotron nebula around the pulsar 1957 + 20 ( stappers et al . 2003 ) that even millisecond pulsars ( msps ) can have significant pair - dominated winds . given their considerably larger numbers compared with young pulsars , especially in regions of old stellar populations like the galactic bulge , the pair flux from msps should dominate the total pair flux from pulsars , and it has been shown that reasonable parameter values for pulsar numbers and pair flux rates can reproduce the observed annihilation line rate ( wang , pun & cheng 2006 ) . thus msps are a plausible source of the 511 kev line emission . however , pulsars are expected to produce very high energy ( @xmath110 mev ) positrons ; but as noted earlier , the integral and comptel measurements of the diffuse 1 - 10 mev background indicate that the positrons responsible for the observed annihilation line must be produced at relatively low energy , as otherwise a visible bump in the spectrum from inflight annihilation of relativistic positrons would be observed ( agaronyan & atoyan 1981 ) . thus we will not consider msps further in this paper . finally , a large number of more exotic possibilities have been suggested to explain the 511 kev emission . they include annihilation of an mev - scale particle , such as `` light '' dark matter ( boehm et al . 2004 ) ; sterile neutrinos or pseudo - scalar relics with masses @xmath26 mev ( picciotto & pospelov 2005 ) ; decays of axions ( hooper & wang 2004 ) , sub - gev scale neutralinos ( gunion et al . 2006 ; bird et al . 2006 ) , or other supersymmetric particles ( takahashi & yanagida 2006 ) ; q - balls ( kasuya and takahashi 2005 ) , mirror matter ( foot & silagadze 2005 ) , moduli ( kawasaki & yanagida 2005 ; kasuya and kawsaki 2006 ) , superconducting cosmic strings ( ferrer & vachaspati ) , droplets of superconducting quark matter ( oaknin & zhitnitsky 2004 ; mcneil , forbes , & zhitnitsky 2008 ) , tev - scale particles with an excitation at the mev scale ( finkbeiner & weiner 2007 ; pospelov & ritz 2007 ) or small , accreting black holes ( titarchuk & chardonnet 2006 ) . over the past decade , multiwavelength observations of x - ray binaries have clearly shown that these systems produce relativistic jet outflows , analogous to those observed in agn and quasars ( see fender 2006 for a review ) . there are known accreting binary populations in the gc , and there is increasing evidence that all accreting sources containing black holes and neutron stars have a _ compact , steady _ jet outflow during their `` hard '' states , when their x - ray luminosity is relatively low ( indeed , there is evidence that these compact jets are also present during the `` quiescent '' state when the source x - ray luminosity is extremely low ; gallo et al . 2006 , 2007 ) . several ( perhaps all ) bh xrbs also have _ transient _ powerful , large - scale , extended jets , which are associated with the transition between the `` hard '' or `` quiescent '' states and the x - ray luminous `` high / soft '' state . these large - scale ejection events are episodic but their duty cycle is poorly known ; however , it is clear that most bh xrbs spend less time in the transitional regime where the large - scale jets are emitted than in the lower x - ray luminosity hard ( or quiescent ) states . we therefore focus here on the low - luminosity steady jets . extending the results of deep x - ray surveys of the central galactic bulge ( wang et al . 2002 , muno et al . 2003 , muno et al . 2006 ) to the entire bulge , we would expect roughly @xmath27 x - ray point sources similar to those detected by the chandra surveys . while a large fraction of these are likely to be cataclysmic variables ( see _ e.g. _ muno et al . 2003 ) , a few percent of the sources are likely to be quiescent low - mass bh x - ray binaries , sources which trace the older stellar population of the galaxy and thus are observed to be concentrated in the inner galaxy . such quiescent systems might be good candidates for the jet positron source , consistent with known point sources down to @xmath28ergs / s . these jets could be sources of high - energy electrons and protons , and also , as they interact with the ism , of positrons . alternatively , this could be done by low energy cosmic ray injection in jets with associated acceleration to high energies in the jet - ism interaction sites ( _ e.g . _ heinz & sunyaev 2003 ; fender , maccarone & van kesteren 2005 ; heinz & grimm 2005 ) . these two distinct components would in turn produce diffuse emission across the electromagnetic spectrum via positron annihilation , synchrotron emission , and perhaps heating of the ism . this solution is preferable to many of the more exotic scenarios , which require fine tuning to avoid constraints on positron production at energies above @xmath010 mev , and can not simultaneously produce both low - energy positrons and high - energy electrons and protons . ( light dark matter , for instance would produce the former , whereas conventional heavy dark matter would generate the latter via pion production . ) the jets in low / hard state or quiescent x - ray binaries are thought to be mildly relativistic ( see gallo , fender , pooley 2003 ) . while there are some arguments that the data on xrbs do allow for higher jet velocities in the low / hard states ( heinz & merloni 2004 ) , additional indirect evidence supports the former assertion . the giant jet ejections that take place at transitions from hard to soft states are well explained by a jet speed that increases as the state transition proceeds , leading to a shock of the faster jet material against the recently ejected slower material ( vadawale et al . 2003 ; fender , belloni & gallo 2004 ) . furthermore , the jets in low luminosity active galactic nuclei are typically two - sided , while those in high luminosity agn are typically one - sided , consistent with the idea that doppler boosting is a more important effect in high luminosity agn than low luminosity ones . we thus take a jet speed of 0.7c , meaning that the kinetic luminosity of the jet is about @xmath29 . this gives @xmath30 g / sec for a jet power of about @xmath31 @xmath10erg@xmath10s@xmath6 , typical in the quiescent systems whose x - ray luminosities are about @xmath32 @xmath10erg@xmath10s@xmath6 . theoretical studies suggest a range of 1000 - 10000 for the number of quiescent bh xrbs in the entire galaxy ( romani 1992 ; portegies zwart , verbunt & ergma 1997 ) . from this range , we derive an estimate of @xmath33 ergs / sec of total kinetic power input into the galactic interstellar medium ( ism ) from quiescent black hole xrbs . since roughly 1/3 of the stellar mass of the milky way is in the bulge , and low mass xrbs are good tracers of stellar populations ( gilfanov 2004 ) , this results in an estimate of @xmath0 3003000 bh in the bulge itself . we can then estimate that the kinetic power injected by quiescent bh x - ray binaries into the bulge is about @xmath34 ergs / s , and that the total mass injected into the bulge ism from these jets is about @xmath35 g / s , yielding a total proton injection rate of about @xmath36 per second . if we see @xmath37 photons / s/@xmath5 , then this means we see @xmath16 511 kev photons per sec . if we assume that 3/4 of the annihilations go through the 3-photon positronium channel , then we need 2@xmath8@xmath16 positrons / s produced to get the @xmath16/s 511 kev photons needed ( 75% produce no 511 , 25% produce two 511 photons ) . we thus need a pair to proton ratio of about 40 - 400 to account for the observed 511 kev luminosity . the best attempts at estimating the pair fractions for jets come from active galactic nuclei . it has been shown by sikora & madejski ( 2000 ) that excessive pair fractions in quasar jets would lead to a strong bump in the soft x - rays due to the bulk compton upscattering of the thermal photons from the accretion disk of the quasar , in constrast to observations . on the other hand , for lower luminosity agn , which are more likely to be analogous to the quiescent bh x - ray binaries discussed here , the situation is less clear these still could be pair - dominated as they do not have strong thermal accretion disks to contribute photons for upscattering . comparisons in m87 between the bulk kinetic power observed from jet - intracluster medium interactions and the synchrotron luminosity argue that there must be a large number of leptons per unit kinetic power , and hence that the pair fraction must be large ( reynolds et al . 1996 ; dunn , fabian & celotti 2006 ) . reynolds et al . ( 2006 ) also find that the pairs in these jets are most likely to be predominantly cold ( i.e. the energy spectrum for the electrons has no low energy cutoff , so that the number density is dominated by electrons with low enough energies to satisfy the constraint discussed in section 2.1 , that high energy pairs would produce excessive emission in @xmath0 mev range via in - flight annihilation ) . guessoum , jean , & prantzos ( 2006 ) have estimated the production rate of positrons in `` canonical '' jet - producing microquasars , based on the energetics and models of these luminous xrbs which have been presented in the literature ; we summarize their results here . they use an average value of 10@xmath38 pairs / sec from a luminous microquasar producing `` steady '' jets ( e.g. in the low / hard state , rather than the large - scale episodic jets ) at @xmath39 0.01 - 0.1@xmath40 . using an estimate of @xmath0100 for the total number of microquasars in the galaxy , together with the galactic positional distribution of observationally - confirmed microquasars , they derive an estimate of @xmath0 4.1@xmath810@xmath41 positrons / s for the rate of annihilating positrons in the bulge . they note that the positron production rate they derive is smaller than what is inferred from the integral 511 kev observations . comparing their results with ours , we see that their calculated positron injection rate from luminous microquasars within the bulge a fairly small population , of order @xmath040 sources is an order of magnitude smaller than the @xmath42 positrons / s we calculate as being produced by quiescent bh xrbs in the bulge a population we estimate to have @xmath03000 members . as such , in the scenario we describe in this paper , the quiescent bh xrbs would be the dominant source of the bulge 511 kev emission , although the luminous microquasars would also make a significant contribution . from one year of integral / spi observations , kndlseder et al . ( 2005 ) found that the spatial distribution of 511 kev luminosity shows a bulge - to - disk ( b / d ) ratio of 3 - 9 , which is higher than the mass of the galaxy in general . subsequently , using two years of integral / spi data , weidenspointner et al . ( 2007 ) revised this ratio downward to a range of 1 - 4 . since there exist mechanisms for producing 511 kev emission in the galactic plane which involve young stars not present in the bulge , any viable mechanism for producing the bulk of the 511 kev emission in the bulge must be one in which a larger amount of annihilation per unit mass comes from the bulge than from the galactic plane . at first glance , this seems to represent a problem for our model , in which we suggest that 1/3 of the quiescent bh xrbs are located in the bulge ; if the remaining 2/3 are simply assumed to be in the galactic disk , then the 511 kev luminosity b / d ratio would be 0.5 , well below the lower end of the range determined by weidenspointner et al . ( 2007 ) . however , this simple calculation neglects two important factors , highlighted by guessoum et al . first , a significant portion of the galactic lmxb population is located in the galactic halo , distinct from the bulge and disk lmxb populations . grimm et al . ( 2002 ) find that 25% of the total number of lmxbs in the galaxy are located in the halo . they also find that 1/3 of bright lmxbs are located in the galactic bulge , consistent with the expectation from the stellar mass distribution of the galaxy discussed above . thus if we assume that the quiescent black hole xrbs described here have a similar distribution to canonical luminous low mass xrbs , we expect @xmath01/3 in the bulge , @xmath01/4 in the halo , and the remainder to be located in the disk . the halo sources are physically located well outside of the bulge and the plane ; thus they will not significantly contribute to the 511 kev luminosity in the spatially - constrained bulge and inner disk areas ( within @xmath43 1.5 kpc of the galactic centre ) over which the 511 kev emission has been detected ( weidenspointner et al . 2008 ) . removing the halo sources results in a b / d luminosity ratio of 0.8 still below the lower limit of weidenspointner et al . ( note that guessoum et al . also made an estimate for the number of luminous microquasars in the halo , and removed those sources before calculating their bulge - to - disk luminosity ratio . ) the second factor , as discussed by kndlseder et al . ( 2005 ) , is that the scale height of xrbs in the galactic disk is about 700 pc ( jonker & nelemans 2004 ) , considerably larger than that of the gas in the galaxy , and that this can have profound effects on the inferred rate of positron production in different parts of the galaxy . if positrons are injected into gas poor parts of the galaxy , they may travel fairly large distances before annihilating , yielding a situation where the distribution of locations for positron annihilation is not the same as the distribution of locations for positron production . the largest effect would be to transfer positrons from production sites several gas scale heights above the galactic plane into the halo or bulge before they undergo annihilation . following guessoum et al . ( 2006 ) , if we estimate that 50% of positrons produced in the disk escape into the halo and/or propagate along field lines towards the bulge , while all positrons produced in the bulge are retained and annihilate therein ( see _ e.g. _ jean et al . 2006 ) , then we must reduce the disk 511 kev luminosity by 1/2 . combining this factor with the bulge and disk population estimates above , we now find a b / d 511 kev luminosity ratio for the quiescent bh xrb population of @xmath01.6 , well within the range for the b / d ratio found by weidenspointner et al . finally , we note that some positrons which escape from the disk may end up annihilating in the bulge , rather than in the halo , resulting in a still larger b / d ratio . therefore , it seems plausible that the 511 kev emission could come from the jets of quiescent ( x - ray faint ) black hole x - ray binaries . the requirements to make this happen _ ( i ) _ that the jets have @xmath44 positrons per proton ; _ ( ii ) _ that there are on the order of 3000 quiescent bh x - ray binaries in the bulge ; and _ ( iii ) _ that the pairs are mostly cold are all within the range of reasonably standard assumptions and well within the range of observational constraints . we first introduce a few standard assumptions about jet kinetic power from accreting binaries , and its relationships with observables such as radio and x - ray luminosities . the kinetic power input into a relativistic jet is assumed to be a constant fraction of the accretion power : @xmath45 this follows in a straightforward manner from the standard mechanisms for producing jets by extracting either the spin energy of the central black hole ( blandford & znajek 1977 ) , or the rotational energy of the accretion disk ( blandford & payne 1982 ) , provided that the height to radius relation for the accretion disk does not change as a function of accretion rate ( _ e.g . _ livio , ogilvie & pringle 1999 ; meier 2001 ) . based on the results of standard synchrotron theory ( _ e.g . _ blandford & knigl 1979 ; falcke & biermann 1995 ; heinz & sunyaev 2003 ) , it is normally assumed that the observed radio luminosity from a jet is proportional to its kinetic luminosity to the 1.4 power : @xmath46 next , it is assumed that black holes at low accretion rates ( in the hard or quiescent states ) are radiatively inefficient , because they can advect energy across their event horizons , leading to a scaling of the bolometric luminosity on the square of the mass accretion rate ( narayan & yi 1994 ) @xmath47 whereas neutron stars , with their solid surfaces , can not advect matter or energy , leading to a linear proportionality between mass transfer rate and luminosity : @xmath48 the resulting expectations are that black holes should have a radio luminosity which depends on the x - ray luminosity to the 0.7 power , assuming that the radio emission comes from the jet , and the x - ray emission comes from an advection - dominated accretion flow : @xmath49 similarly , neutron stars should have a radio luminosity which scales as the x - ray luminosity to the 1.4 power . both these correlations are indeed observed ; see _ e.g. _ corbel et al . ( 2003 ) and gallo , fender & pooley ( 2003 ) for the case of black holes , and migliari and fender ( 2006 ) for the case of neutron stars . the kinetic energy input into the ism from black hole x - ray transients in quiescence can then be estimated based on their x - ray luminosities . the characteristic x - ray luminosities of quiescent bh x - ray transients are @xmath32 ergs / sec . jets are likely to take away a substantial fraction of kinetic power in the low / hard state ( _ e.g . _ malzac , merloni & fabian 2004 ) , but , given that state transitions between the radiatively efficient high / soft state in which little or no jet power is seen , and the radiatively inefficient low / hard state in which there is a steady jet , do not correspond with abrupt luminosity changes , the jet power can not be much greater than the radiated power at the state transition ( maccarone 2005b ) . therefore , the best estimate of the jet kinetic power is that it scales with @xmath50 , as predicted by the model , with the normalization fixed such that the jet power and the x - ray luminosity are equal at the state transition luminosity which is typically about 2% of the eddington luminosity ( maccarone 2003 ) . such arguments have been used to estimate that the total kinetic power input into the ism from _ bright _ xrbs is about @xmath51 ergs / sec ( fender , maccarone & van kesteren 2005 ; heinz & grimm 2006 ) , while the kinetic power input from _ quiescent _ black hole x - ray binaries may be @xmath33 ergs / sec ( as derived in section 3.2 ; see also maccarone 2005a ) . ( note that we use this power estimate based on the theoretical numbers of the bh xrb population rather than using the luminosity function derived from observed bright bh x - ray transients , as the latter option would involve extrapolating the luminosity function to x - ray luminosities 36 orders of magnitude lower than what has been directly measured for the bright sources . ) since the total kinetic power in jets from quiescent bh x - ray binaries is likely to be of the same order as that from the most luminous systems , and the quiescent systems will be more strongly concentrated in the bulge than will the luminous xrbs ( a population which includes hmxbs as well as lmxbs ) , it is to be expected that a substantial fraction of the bulge positrons come from quiescent black hole x - ray binaries . finally , we note two other diffuse components in the galactic centre that may be related to the annihilation radiation . + _ x - ray : _ + in the bulge core , diffuse chandra emission amounts to @xmath52ergs / s , with a spectrum equivalent to an 8 kev thermal plasma over the 17x17 sq . arcmin chandra field ( muno et al . 2004 ) . the observed spectral shape is indicative of a truly diffuse component rather than originating from a collection of unresolved discrete sources . the flux limit of the muno et al . ( 2003 ) x - ray survey is @xmath010@xmath53 ergs / s . normal stars have luminosities of @xmath54ergs / s or less , and so at least @xmath55 normal stars would be needed to account for the diffuse emission component . however , @xmath55 stars are all the ( solar mass ) stars in the bulge , not just those within the central 40 square parsecs covered by the muno et al . chandra survey it has been argued that this plasma can be thermally maintained by low energy cosmic ray heating ( yusef - zadeh et al . could these cosmic rays possibly come from the x - ray binary jets themselves ? the energetics should be adequate . one question is whether this fraction of the energy actually comes out in cosmic rays , rather than just leading to bulk kinetic motions of the gas at the collisional interface of the jets . + _ microwave : _ + wmap finds a likely synchrotron background with an integrated luminosity of about 2@xmath56 ergs / sec in the 23 - 61 ghz frequency range from a region 2030@xmath17 in radius around the galactic centre region ( finkbeiner 2004a ) . the total power emitted at these frequencies is estimated to be between 15@xmath57 erg s@xmath6 . assuming a distance of 8.5 kpc , this corresponds to a flux level between 3000 and 15000 jy . we note that there are large uncertainties in this estimate , given the need to subtract many other contributions , including thermal dust , spinning dust , thermal bremsstrahlung emission and synchrotron from electrons accelerated in sne shocks , from the microwave maps . moreover , use of the wmap internal linear combination ( ilc ) dust extinction template leaves much uncertainty in possible spatial variations of the composition - dependent frequency sensitivity of the dust emissivity . nonetheless , the signal has generated interest given its possible connection to exotic processes such as the annihilation of massive wimps ( finkbeiner 2004b ; hooper , finkbeiner & dobler 2007 ) , or states of ` excited ' dark matter ( finkbeiner & weiner 2007 ) . the reported haze flux level can not be produced from core emission from individual black hole x - ray binaries as previously discussed , no more than about 3000 quiescent bh xrbs are thought to be in the entire gc region ( _ e.g . _ romani 1992 ; portegies zwart , verbunt & ergma 1997 ) , meaning that the flux of the typical source would have to be @xmath01 to several jy . sources above even 100 mjy would be well - known radio sources at lower frequencies ( since they would likely be flat spectrum sources , or sources whose flux density drops with increasing frequency ) , and an excess of such sources would be clear from existing radio surveys . the typical flux of a quiescent low mass x - ray binary in the radio at the galactic centre distance should be @xmath58 1020 @xmath59jy ( see _ e.g. _ gallo et al . 2006 for the faintest known system of this sort ; hynes et al . 2004 for v404 cyg , the brightest one ) . given @xmath03000 bh with an average quiescent radio flux level of @xmath0100 @xmath59jy , the integrated luminosity of those outflows at wmap wavelengths would be far less than that of the wmap gc haze ( even if the haze fluz has been significantly overestimated ) . this may support the interpretation of the haze emission source as relativistic electrons , produced by annihilating 100 gev dark matter particles ( finkbeiner 2004b ; hooper , finkbeiner & dobler 2007 ) . it thus seems reasonable that most or all of the galactic centre 511 kev emission line could come from low luminosity bh x - ray binaries . we note that this scenario is not required , since the positron - to - proton ratios for x - ray binary jets are largely unconstrained by observations . what about the contribution from high luminosity ( `` canonical '' ) xrbs ? we need to estimate the fraction of the overall 511 kev line flux that can be accounted for by the well - studied canonical jet sources , and correspondingly the percentage of the flux which needs to be accounted for by the low luminosity population . however , if we only consider black hole xrbs , since ( as previously discussed ) the ns jet flux is too small , then at any given time there are only a few active luminous hard state black holes which would contribute to the 511 kev emission . but in this case , if a significant fraction of the overall 511 kev flux originated from a handful of sources , and if the positrons do not propagate and annihilate far from their production sites , we would expect to see these individual sources in the 511 kev flux maps - the 511 kev line emission would not appear to be as smooth as it does . there is no convincing evidence for time variability of the 511 kev line . it is clear that the mechanism proposed here summed emission from hundreds to thousands of sources of roughly equal importance should predict low variability levels in the 511 kev emission . it is less clear whether this would be true from many of the other mechanisms proposed which rely more heavily on a smaller number of sources . however , temporal variability is not useful for distinguishing the total number of sources if the characteristic positron lifetime is @xmath27 years as suggested by ferrer & vachaspati ( 2005 ) . only spatial variability would be useful in this case . weidenspointner et al . ( 2008 ) suggest that an asymmetry in the 511 kev emission is correlated with an asymmetry in the lmxb distribution around the inner galaxy . the asymmetry argument is that the disk 511 kev flux is asymmetric , and so is the hard lmxb population , with the same ratio between halves ( positive and negative galactic longitudes ) of the disk . furthermore , the number of objects ( @xmath070 ) times a plausible mean flux per source ( @xmath60 photons/@xmath5/s each ) gives the right total disk flux ( @xmath61 photons/@xmath5/s ) ; thus the luminous hard lmxbs may be the source of the disk flux , if not the ( @xmath13 photons/@xmath5/s ) bulge flux ( note that the latter corresponds to a larger luminosity , since the bulge is further away ; overall it should be 3 - 9 times the disk luminosity ) . while the positional asymmetries in the known luminous x - ray binary distribution do appear to be in the same sense as the asymmetry in the 511 kev emission , it is not clear that the asymmetry in the bulge component of the x - ray binary population is real . the inner 20@xmath3of the galaxy show only a 1.5@xmath21 asymmetry - 30 ( negative longitude ) versus 18 ( positive longitude ) sources . in other words , given a total of 48 sources , the probability of having @xmath6230 sources at negative latitudes is 5.6% . we note that at least eight of the x - ray binaries in the integral catalog ( 5 at negative longitudes , 3 at positive ) used by weidenspointner et al . ( 2008 ) in their determination of the asymmetry are `` foreground '' sources , with known distances which are not consistent with being in the galactic bulge ( jonker & nelemans 2004 ) . if the 511 kev emission is physically associated with the population of hard xrbs , as is suggested by the asymmetry , then the observed enhancement of 511 kev flux in the central region of the galactic plane is simply a projection effect along our line of sight towards the gc , which coincidentally results in an apparent concentration of 511 kev line emission within @xmath63@xmath64@xmath65@xmath3 . thus in this scenario , the 511 kev enhancement is not physically associated with the galactic bulge itself , and the morphology and intensity of the 511 kev map is dependent upon the longitudinal location of bright , hard xrbs . however , we note that grs 1915 + 105 , which is almost certainly the strongest emitter of positrons from jets outside of the innermost region of the bulge , is on the other side of the galaxy ( at positive galactic longitude ) from the bulk of the integral sources and the observed 511 kev enhancement ( which are at negative galactic longitudes ) . the asymmetry interpretation has at least 4 distinct problems : 1 . the lmxb luminosity function is steep near the lower end ; changing the sensitivity limit of the ibis catalog would change the positive / negative longitude number ratio . integrating flux down the luminosity function , the asymmetry also depends on the catalogue limit , in that the total flux asymmetry is actually the reverse of the number asymmetry , at least at high luminosities . a few sources ( e.g. grs 1915 + 105 ) are responsible for much of the flux ; if this is always true , then the numbers of objects in the catalog may not mean much ; a few bright sources could determine how many positrons are produced at any given time ( i.e. the total flux need not correlate with the number of sources over the ibis limit ) . 4 . since lmxb x - ray emission is time - variable , there is no indication that any current asymmetry in numbers or flux would be constant in time . 1a shows the cumulative ( 20 - 100 kev ) luminosity function for all 72 galactic lmxbs within @xmath6410@xmath3 latitude of the plane in the catalogue of bird et al . ( 2007 ) , divided up into objects at positive and at negative longitudes ( solid and dashed lines respectively ) . note this scale height excludes high - latitude luminous lmxbs such as sco x-1 . 1b shows the integrated flux contributed by binaries over a given flux limit for each of these samples . clearly the longitudinal asymmetry in number counts is sensitive to the lower flux limit of the catalogue ; here , this is set to the integral / ibis 20 - 100 kev detection limit . the current integrated flux is dominated by a small number of bright sources , with the majority of the sources in the integral catalogue being just at the flux detection limit . given the recently identified large population of x - ray faint xrbs discussed above , it is clear that there are many lower luminosity sources below the integral detection threshold . the asymmetry in the source number counts could thus easily be an artifact of the flux bias of the integral sample . therefore the source numbers used to determine the positional asymmetry could be greatly changed by a slight change in the flux cutoff ( to either higher or lower luminosities ) . cutting objects below 10 mcrab , for instance , would produce no asymmetry at all , while cutting at a high flux limit would produce more sources at positive longitudes . so overall , the observed positional asymmetry seen in the integral catalogue , with its instrumentally - determined bias towards luminous sources , may not be significant even if the observed current 20 - 100 kev flux were to correlate exactly with the 511 kev flux . furthermore , the integrated flux from lmxbs in the catalogue is dominated by objects at positive longitudes at all flux limits ( fig.1b ) , although admittedly this is mainly due to the strong contribution from the single brightest source , grs 1915 + 105 . thus the asymmetry in the number of sources does not agree with the asymmetry in the total flux from the integral catalogue sources there is more integrated flux where there are fewer sources ( positive longitudes ) , and less integrated flux where there are more sources ( negative longitudes ) . the fact that the current `` snapshot '' distribution of the integrated flux does not match the 511 kev asymmetry may in itself not be problematic for the weidenspointner et al . ( 2008 ) hypothesis , since the timescale for positrons to annihilate and produce 511 kev emission is @xmath66 years . however , we note that a large fraction of the sources observed by integral and used to `` find '' the positional distribution asymmetry are transients . in addition , those x - ray binaries which we call `` persistent '' merely means that the sources have been consistently luminous over the @xmath035 years of observational x - ray astronomy , which provides very little information about the persistence of their luminosity over the @xmath27 year timescale of photon interaction . thus the current apparent positional asymmetry is merely a snapshot in time , and does not constitute evidence that the total absolute numbers of 511 kev - producing binaries in the direction of the inner galaxy including those currently in quiescence , which are likely to be the majority at negative longitudes is larger than at positive longitudes . over the long lifetime of positron interaction , the integrated flux from all sources which become active during that time may indeed be larger at negative longitudes than positive ones , resulting in the observed asymmetry in the 511 kev emission . to determine this , however , we would need to be certain that there is a true asymmetry in the distribution of all positron - producing xrbs in the inner galaxy - not just during the current epoch , when only a small fraction of such xrbs are observed to be active and luminous . we would have to assume that the asymmetric distribution of currently luminous xrbs - a sample of @xmath070 sources - reflects an underlying asymmetry in the overall distribution of all xrbs - on the order of @xmath03000 bh binaries - which would contribute to the 511 kev emission over @xmath67 years . while this is certainly possible , it can not be considered conclusive , as it is dependent upon both the arbitrary lower flux cutoff of the integral catalogue ( set by the instrument sensitivity rather than by any intrinsic property of the sources themselves ) and the arbitrary time of observation ( _ e.g . _ the modern era of high - energy astronomy ) . a much deeper census of the black hole x - ray binary population , which explores down to the quiescent luminosity of these sources ( @xmath68 erg / s ) , is needed in order to search for a true asymmetry in the physical distribution of jet - emitting xrbs in the inner galaxy . such a census would require both a deep x - ray survey and infrared follow - up observations to identify stellar counterparts to a representative sample of the x - ray sources , in order to conclusively determine which sources contain black holes . it would then be possible to assess whether or not the asymmetry in the 511 kev flux is due to an intrinsic spatial asymmetry in the number of xrbs which episodically exhibit large x - ray luminosities integrated over the timescale for 511 kev production . if stellar sources do account for 511 kev emission , it is unlikely that they would show up as point sources at higher spatial resolution at other wavelengths than the x - ray , as the positrons likely propagate a significant distance ( @xmath050 - 250 pc , which is @xmath00.5@xmath3 1.5@xmath3at the gc ; jean et al . 2006 but see also prantzos et al . 2006 , which suggested a larger distance ) before annihilating . for the brightest of the low - luminosity sources ( @xmath31 erg / s ) , we could perhaps detect a flat near - ir spectrum which would not be consistent with the flux of a mass donor star alone , but which instead would denote the presence of another component - _ e.g. _ a jet outflow . it is not clear if there is enough sensitivity in the mid - ir for ground - based observations _ e.g. _ at 4 and 8 microns to detect the flux from these sources . similarly , for sources with this x - ray luminosity , high frequency ( 8.6 ghz ) radio emission might be detectable ( _ e.g . _ with the vla ) to search for the flat spectrum signature associated with the steady jets from hard state x - ray binaries . the authors wish to thank tony bird , sera markoff , and philipp podsiadlowski for helpful discussions . we also thank mark rayner for his preliminary work on extracting osse observations of x - ray binaries from the heasarc archives . we thank the referee for a close reading of the manuscript and useful suggestions . jet gratefully acknowledges support from pparc , the aspen center for physics , the us doe ( grant ast-0307859 ) , the us nsf ( contract de - fg02 - 04er41316 ) , and nserc canada . beacom , j. f. , yksel , h. 2006 , physical review letters , 97 , 071102 bertone , g. , kusenko , a. , palomares - ruiz , s. , pascoli , s. , & semikoz , d. 2006 , physics letters b , 636 , 20 bird , c. , kowalewski , r. , & pospelov , m. 2006 , modern physics letters a , 21 , 457 bouchet , l. , jourdain , e. , roques , j. p. , strong , a. , diehl , r. , lebrun , f. , & terrier , r. 2008 , apj , 679 , 1315 cass , m. , cordier , b. , paul , j. , & schanne , s. 2004 , apj , 602 , l17 chaty , s. , rahoui , f. , foellmi , c. , tomsick , j. a. , rodriguez , j. , & walter , r. 2008 , a&ap , 484 , 783 de cesare , g. , bazzano , a. , capitanio , f. , del santo , m. , lonjou , v. , natalucci , l. , ubertini , p. , & von ballmoos , p. 2006 , advances in space research , 38 , 1457 dobler , g. , & finkbeiner , d. p. 2007 , arxiv e - prints , 712 , arxiv:0712.2238 finkbeiner , d. p. 2004a , apj , 614 , 186 finkbeiner , d. p. 2004b , arxiv astrophysics e - prints , arxiv : astro - ph/0409027 finkbeiner , d. p. , & weiner , n. 2007 , phys . rev . d , 76 , 083519 foot , r. , & silagadze , z. k. 2005 , international journal of modern physics d , 14 , 143 haymes , r. c. , ellis , d. v. , fishman , g. j. , glenn , s. w. , & kurfess , j. d. 1969 , apj , 157 , 1455 haymes , r. c. , walraven , g. d. , meegan , c. a. , hall , r. d. , djuth , f. t. , & shelton , d. h. 1975 , apj , 201 , 593 heinz , s. , & grimm , h. j. 2005 , apj , 633 , 384 jean , p. , et al . 2003 , a&ap , 407 , l55 jean , p. , kndlseder , j. , gillard , w. , guessoum , n. , ferrire , k. , marcowith , a. , lonjou , v. , & roques , j. p. 2006 , a&ap , 445 , 579 johnson , w. n. , iii , & haymes , r. c. 1973 , apj , 184 , 103 johnson , w. n. , iii , harnden , f. r. , jr . , & haymes , r. c. 1972 , apj , 172 , l1 jonker , p. g. , & nelemans , g. 2004 , mnras , 354 , 355 kndlseder , j. , et al . 2005 , a&ap , 441 , 513 leventhal , m. , maccallum , c. j. , & stang , p. d. 1978 , apj , 225 , l11 leventhal , m. , maccallum , c. j. , huters , a. f. , & stang , p. d. 1980 , apj , 240 , 338 lingenfelter , r. e. , & hueter , g. j. 1984 , american institute of physics conference series , 115 , 558

the burrows - wheeler transform ( bwt ) is a widely used preprocessing technique in lossless data compression @xcite . it brings every word into a form which is likely to be easier to compress @xcite . its compression performance is almost as good as ppm ( prediction by partial matching ) schemes @xcite while its speed is comparable to that of lempel - ziv algorithms @xcite . therefore , bwt based compression schemes are a very reasonable trade - off between running time and compression ratio . in the classic setting , the bwt maps a word of length @xmath0 to a word of length @xmath0 and an index ( comprising @xmath3 bits ) . thus , the bwt is not bijective and hence , it is introducing new redundancies to the data , which is cumbersome and undesired in applications of data compression or cryptography . instead of using an index , a very common technique is to assume that the input has a unique end - of - string symbol @xcite . even though this often simplifies proofs or allows speeding up the algorithms , the use of an end - of - string symbol introduces new redundancies ( again @xmath3 bits are required for coding the end - of - string symbol ) . we discuss bijective versions of the bwt which are one - to - one correspondences between words of length @xmath0 . in particular , no index and no end - of - string symbol is needed . not only does bijectivity save a few bits , for example , it also increases data security when cryptographic procedures are involved ; it is more natural and it can help us to understand the bwt even better . moreover , the bijective variants give us new possibilities for enhancements ; for example , in the bijective bwt different orders on the letters can be used for the two main stages . several variants of the bwt have been introduced @xcite . an overview can be found in the textbook by adjeroh , bell , and mukherjee @xcite . one particularly important variant for this paper is the sort transform ( st ) , which is also known under the name schindler transform @xcite . in the original paper , the inverse of the st is described only very briefly . more precise descriptions and improved algorithms for the inverse of the st have been proposed recently @xcite . as for the bwt , the st also involves an index or an end - of - string symbol . in particular , the st is not onto and it introduces new redundancies . the bijective bwt was discovered and first described by scott ( 2007 ) , but his exposition of the algorithm was somewhat cryptic , and was not appreciated as such . in particular , the fact that this transform is based on the lyndon factorization went unnoticed by scott . gil and scott @xcite provided an accessible description of the algorithm . here , we give an alternative description , a proof of its correctness , and more importantly , draw connections between scott s algorithm and other results in combinatorics on words . further , this variation of the bwt is used to introduce techniques which are employed at the bijective sort transform , which makes the main contribution of this paper . the forward transform of the bijective st is rather easy , but we have to be very careful with some details . compared with the inverse of the bijective bwt , the inverse of the bijective st is more involved . * outline . * the paper is organized as follows . in section [ sec : prelim ] we fix some notation and repeat basic facts about combinatorics on words . on our way to the bijective sort transform ( section [ sec : lst ] ) we investigate the bwt ( section [ sec : bwt ] ) , the bijective bwt ( section [ sec : lbwt ] ) , and the sort transform ( section [ sec : st ] ) . we give full constructive proofs for the injectivity of the respective transforms . each section ends with a running example which illustrates the respective concepts . apart from basic combinatorics on words , the paper is completely self - contained . throughout this paper we fix the finite non - empty alphabet @xmath4 and assume that @xmath4 is equipped with a linear order @xmath5 . a _ word _ is a sequence @xmath6 of letters @xmath7 , @xmath8 . the set of all such sequences is denoted by @xmath9 ; it is the free monoid over @xmath4 with concatenation as composition and with the empty word @xmath10 as neutral element . the set @xmath11 consists of all non - empty words . for words @xmath12 we write @xmath13 if @xmath14 or if @xmath15 is lexicographically smaller than @xmath16 with respect to the order @xmath5 on the letters . let @xmath17 be a non - empty word with letters @xmath18 . the _ length _ of @xmath19 , denoted by @xmath20 , is @xmath0 . the empty word is the unique word of length @xmath21 . we can think of @xmath19 as a labeled linear order : position @xmath22 of @xmath19 is labeled by @xmath18 and in this case we write @xmath23 , so each word @xmath19 induces a labeling function @xmath24 . the first letter @xmath25 of @xmath19 is denoted by @xmath26 while the last letter @xmath27 is denoted by @xmath28 . the _ reversal _ of a word @xmath19 is @xmath29 . we say that two words @xmath12 are _ conjugate _ if @xmath30 and @xmath31 for some words @xmath32 , i.e. , @xmath15 and @xmath16 are cyclic shifts of one another . the @xmath33-fold concatenation of @xmath19 with itself is denoted by @xmath34 . a word @xmath15 is a _ root _ of @xmath19 if @xmath35 for some @xmath36 . a word @xmath19 is _ primitive _ if @xmath35 implies @xmath37 and hence @xmath38 , i.e. , @xmath19 has only the trivial root @xmath19 . the _ right - shift _ of @xmath39 is @xmath40 and the @xmath22-fold right shift @xmath41 is defined inductively by @xmath42 and @xmath43 . we have @xmath44 for @xmath45 . the word @xmath41 is also well - defined for @xmath46 and then @xmath47 where @xmath48 . we define the _ ordered conjugacy class _ of a word @xmath49 as @xmath50 = ( w_1 , \ldots , w_n)$ ] where @xmath51 . it is convenient to think of @xmath50 $ ] as a cycle of length @xmath0 with a pointer to a distinguished starting position . every position @xmath22 , @xmath8 , on this cycle is labeled by @xmath52 . in particular , @xmath25 is a successor of @xmath27 on this cycle since the position @xmath1 is a successor of the position @xmath0 . the mapping @xmath53 moves the pointer to its predecessor . the ( unordered ) conjugacy class of @xmath19 is the multiset @xmath54 . whenever there is no confusion , then by abuse of notation we also write @xmath50 $ ] to denote the ( unordered ) conjugacy class of @xmath19 . for instance , this is the case if @xmath19 is in some way distinguished within its conjugacy class , which is true if @xmath19 is a lyndon word . a _ lyndon word _ is a non - empty word which is the unique lexicographic minimal element within its conjugacy class . more formally , let @xmath50 = ( w , w_2 , \ldots , w_n)$ ] , then @xmath55 is a lyndon word if @xmath56 for all @xmath57 . lyndon words have a lot of nice properties @xcite . for instance , lyndon words are primitive . another interesting fact is the following . [ thm : lyndon ] every word @xmath58 has a unique factorization @xmath59 such that @xmath60 is a non - decreasing sequence of lyndon words . an alternative formulation of the above fact is that every word @xmath19 has a unique factorization @xmath61 where @xmath62 for all @xmath22 and where @xmath63 is a strictly increasing sequence of lyndon words . the factorization of @xmath19 as in fact [ thm : lyndon ] is called the _ lyndon factorization _ of @xmath19 . it can be computed in linear time using duval s algorithm @xcite . suppose we are given a multiset @xmath64 of lyndon words enumerated in non - decreasing order @xmath65 . now , @xmath66 uniquely determines the word @xmath67 . therefore , the lyndon factorization induces a one - to - one correspondence between arbitrary words of length @xmath0 and multisets of lyndon words of total length @xmath0 . of course , by definition of lyndon words , the multiset @xmath68 of lyndon words and the multiset @xmath69 , \ldots , [ v_s]}\right\}}$ ] of conjugacy classes of lyndon words are also in one - to - one correspondence . we extend the order @xmath5 on @xmath4 as follows to non - empty words . let @xmath70 be the infinite sequences obtained as the infinite power of @xmath19 . for @xmath71 we write @xmath72 if either @xmath73 or @xmath74 and @xmath75 for @xmath76 , @xmath77 with @xmath78 , and infinite sequences @xmath79 ; phrased differently , @xmath72 means that the infinite sequences @xmath80 and @xmath81 satisfy @xmath82 . if @xmath15 and @xmath16 have the same length , then @xmath83 coincides with the lexicographic order induced by the order on the letters . for arbitrary words , @xmath83 is only a preorder since for example @xmath84 and @xmath85 . on the other hand , if @xmath86 and @xmath87 then @xmath88 . hence , by the periodicity lemma @xcite , there exists a common root @xmath89 and @xmath90 such that @xmath91 and @xmath92 . also note that @xmath93 whereas @xmath94 for @xmath78 . intuitively , the _ context of order @xmath2 _ of @xmath19 is the sequence of the first @xmath2 letters of @xmath19 . we want this notion to be well - defined even if @xmath95 . to this end let @xmath96 be the prefix of length @xmath2 of @xmath97 , i.e. , @xmath96 consists of the first @xmath2 letters on the cycle @xmath50 $ ] . note that our definition of a context of order @xmath2 is left - right symmetric to the corresponding notion used in data compression . this is due to the fact that typical compression schemes are applying the bwt or the st to the reversal of the input . an important construction in this paper is the _ standard permutation _ @xmath98 on the set of positions @xmath99 induced by a word @xmath100 @xcite . the first step is to introduce a new order @xmath101 on the positions of @xmath19 by sorting the letters within @xmath19 such that identical letters preserve their order . more formally , the linear order @xmath101 on @xmath102 is defined as follows : @xmath103 if @xmath104 let @xmath105 be the linearization of @xmath106 according to this new order . now , the standard permutation @xmath98 is defined by @xmath107 . [ exa : start ] consider the word @xmath108 over the ordered alphabet @xmath109 . we have @xmath110 . therefore , the positions in @xmath19 are @xmath111 . for instance , the label of position @xmath112 is @xmath113 . its lyndon factorization is @xmath114 . the context of order @xmath115 of the prefix @xmath116 of length @xmath117 is @xmath118 and the context of order @xmath115 of the factor @xmath119 is @xmath120 . for computing the standard permutation we write @xmath19 column - wise , add positions , and then sort the pairs lexicographically : [ cols="^,^,^",options="header " , ] this yields the standard permutation @xmath121 the conjugacy class @xmath50 $ ] of @xmath19 is depicted in figure [ sfg : w ] ; the @xmath22-th word in @xmath50 $ ] is written in the @xmath22-th row . the last column of the matrix for @xmath50 $ ] is the reversal @xmath122 of @xmath19 . the _ burrows - wheeler transform _ ( bwt ) maps words @xmath19 of length @xmath0 to pairs @xmath123 where @xmath124 is a word of length @xmath0 and @xmath22 is an index in @xmath125 . the word @xmath124 is usually referred to as the burrows - wheeler transform of @xmath19 . in particular , the bwt is not surjective . we will see below how the bwt works and that it is one - to - one . it follows that only a fraction of @xmath126 of all possible pairs @xmath123 appears as an image under the bwt . for instance @xmath127 where @xmath128 is not an image under the bwt . for @xmath58 we define @xmath129 where @xmath130 $ ] and @xmath131 . now , the burrows - wheeler transform of @xmath19 consists of the word @xmath132 and an index @xmath22 such that @xmath133 . note that in contrast to the usual definition of the bwt , we are using right shifts ; at this point this makes no difference but it unifies the presentation of succeeding transforms . at first glance , it is surprising that one can reconstruct @xmath134 from @xmath135 . moreover , if we know the index @xmath22 of @xmath19 in the sorted list @xmath134 , then we can reconstruct @xmath19 from @xmath135 . one way of how to reconstruct @xmath134 is presented in the following lemma . for later use , we prove a more general statement than needed for computing the inverse of the bwt . [ lem : contextretrieval ] let @xmath136 . let @xmath137 = { \left\{\mathinner{w_1 , \ldots , w_n}\right\ } } \subseteq \sigma^+$ ] be a multiset built from conjugacy classes @xmath138 $ ] . let @xmath139 satisfy @xmath140 and let @xmath141 be the sequence of the last symbols . then @xmath142 where @xmath143 denotes the @xmath144-fold application of @xmath145 and @xmath146 . by induction over the context length @xmath144 , we prove that for all @xmath147 we have @xmath148 . for @xmath149 we have @xmath150 and hence , the claim is trivially true . let now @xmath151 . by the induction hypothesis , the @xmath152-order context of each @xmath153 is @xmath154 . by applying one right - shift , we see that the @xmath144-order context of @xmath155 is @xmath156 . the list @xmath157 meets the sort order induced by @xmath2-order contexts . in particular , @xmath158 is sorted by @xmath152-order contexts . let @xmath159 be a stable sort by @xmath144-order contexts of the right - shifts @xmath160 . the construction of @xmath159 only requires a sorting of the first letters of @xmath161 such that identical letters preserve their order . the sequence of first letters of the words @xmath162 is exactly @xmath124 . by construction of @xmath163 , it follows that @xmath164 . since @xmath157 is built from conjugacy classes , the multisets of elements occurring in @xmath158 and @xmath165 are identical . the same holds for the multisets induced by @xmath166 and @xmath159 . therefore , the sequences of @xmath144-order contexts induced by @xmath167 and @xmath168 are identical . moreover , we conclude @xmath169 which completes the induction . we note that in general @xmath170 since the sort order of @xmath157 beyond @xmath2-order contexts is arbitrary . moreover , for @xmath171 the property @xmath172 does not need to hold ( even though the multisets of @xmath173-order contexts coincide ) . note that in lemma [ lem : contextretrieval ] we do not require that all @xmath174 have the same length . applying the bwt to conjugacy classes of words with different lengths has also been used for the _ extended bwt _ @xcite . the bwt is invertible , i.e. , given @xmath175 where @xmath22 is the index of @xmath19 in @xmath134 one can reconstruct the word @xmath19 . we set @xmath176 . let @xmath177 and @xmath178 . now , by lemma [ lem : contextretrieval ] we see that @xmath179 in particular , @xmath180 only depends on @xmath124 and @xmath22 . in the special case of the bwt it is possible to compute the @xmath22-th element @xmath153 of @xmath134 by using the inverse @xmath181 of the permutation @xmath163 : @xmath182 this justifies the usual way of computing the inverse of @xmath175 from right to left ( by using the restriction of @xmath181 to the cycle containing the element @xmath22 ) . the motivation is that the ( required cycle of the ) inverse @xmath181 seems to be easier to compute than the standard permutation @xmath163 . [ exa : bwt ] we compute the bwt of @xmath108 from example [ exa : start ] . the lexicographically sorted list @xmath134 can be found in figure [ sfg : m ] . this yields the transform @xmath183 where @xmath178 is the last column of the matrix @xmath134 and @xmath19 is the @xmath22-th row in @xmath134 . the standard permutation of @xmath124 is @xmath184 now , @xmath185 gives us the following sequence of positions starting with @xmath186 : @xmath187 applying the labeling function @xmath188 to this sequence of positions yields @xmath189 i.e. , we have successfully reconstructed the input @xmath19 from @xmath175 . now we are ready to give a comprehensive description of scott s bijective variant of the bwt and to prove its correctness . it maps a word of length @xmath0 to a word of length @xmath0without any index or end - of - string symbol being involved . the key ingredient is the lyndon factorization : suppose we are computing the bwt of a lyndon word @xmath16 , then we do not need an index since we know that @xmath16 is the first element of the list @xmath190 . this leads to the computation of a multi - word bwt of the lyndon factors of the input . the bijective bwt of a word @xmath19 of length @xmath0 is defined as follows . let @xmath191 with @xmath192 be the lyndon factorization of @xmath19 . let @xmath193 where @xmath194 and where the multiset @xmath195 $ ] . then , the bijective bwt of @xmath19 is @xmath196 . the _ s _ in @xmath197 is for _ scottified_. note that if @xmath19 is a power of a lyndon word , then @xmath198 . in some sense , the bijective bwt can be thought of as the composition of the lyndon factorization @xcite with the inverse of the gessel - reutenauer transform @xcite . in particular , a first step towards a bijective bwt can be found in a 1993 article by gessel and reutenauer @xcite ( prior to the publication of the bwt @xcite ) . the link between the gessel - reutenauer transform and the bwt was pointed out later by crochemore et al . a similar approach as in the bijective bwt has been employed by mantaci et al . @xcite ; instead of the lyndon factorization they used a decomposition of the input into blocks of equal length . the output of this variant is a word and a sequence of indices ( one for each block ) . in its current form , the bijective bwt has been proposed by scott @xcite in a newsgroup posting in 2007 . gil and scott gave an accessible version of the transform , an independent proof of its correctness , and they tested its performance in data compression @xcite . the outcome of these tests is that the bijective bwt beats the usual bwt on almost all files of the calgary corpus @xcite by at least a few hundred bytes which exceeds the gain of just saving the rotation index . [ lem : cycleretrieval ] let @xmath191 with @xmath192 be the lyndon factorization of @xmath19 , let @xmath193 , and let @xmath199 . consider the cycle @xmath200 of the permutation @xmath163 which contains the element @xmath1 and let @xmath201 be the length of @xmath200 . then @xmath202 . by lemma [ lem : contextretrieval ] we see that @xmath203 . since @xmath204 is primitive it follows @xmath205 for some @xmath206 . in particular , the lyndon factorization of @xmath19 ends with @xmath207 . let @xmath208 be the subsequence of @xmath209 which consists of those @xmath210 which come from this last factor @xmath207 . the sequence @xmath208 contains each right - shift of @xmath204 exactly @xmath211 times . moreover , the sort - order within @xmath208 depends only on @xmath212-order contexts . the element @xmath213 is the first element in @xmath208 since @xmath204 is a lyndon word . in particular , @xmath214 is the first occurrence of @xmath215 within @xmath208 . suppose @xmath216 is the first occurrence of @xmath217 within @xmath208 . let @xmath218 be the indices of all occurrences of @xmath217 in @xmath208 . by construction of @xmath163 , we have @xmath219 and therefore @xmath220 is the first occurrence of @xmath221 within @xmath208 . inductively , @xmath216 always refers to the first occurrence of @xmath217 within @xmath208 ( for all @xmath222 ) . in particular it follows that @xmath223 and @xmath224 . [ thm : bijbwt ] the bijective bwt is invertible , i.e. , given @xmath225 one can reconstruct the word @xmath19 . let @xmath199 and let @xmath191 with @xmath226 be the lyndon factorization of @xmath19 . each permutation admits a cycle structure . we decompose the standard permutation @xmath163 into cycles @xmath227 . let @xmath228 be the smallest element of the cycle @xmath229 and let @xmath230 be the length of @xmath229 . we can assume that @xmath231 . we claim that @xmath232 , @xmath233 , and @xmath234 . by lemma [ lem : cycleretrieval ] we have @xmath235 . let @xmath236 denote the restriction of @xmath163 to the set @xmath237 , where by abuse of notation @xmath238 denotes the set of all elements occurring in @xmath239 . let @xmath240 . the word @xmath241 can be obtained from @xmath124 by removing all positions occurring in the cycle @xmath242 . this yields a monotone bijection @xmath243 such that @xmath244 and @xmath245 for all @xmath246 . in particular , @xmath247 has the same cycle structure as @xmath236 and @xmath248 is the sequence of the minimal elements within the cycles . by induction on the number of lyndon factors , @xmath249 appending @xmath250 to the last line allows us to reconstruct @xmath19 by @xmath251 moreover , @xmath252 and @xmath233 . we note that this formula for @xmath19 only depends on @xmath124 and does not require any index to an element in @xmath209 . [ exa : bijbwt ] we again consider the word @xmath108 from example [ exa : start ] and its lyndon factorization @xmath253 where @xmath254 , @xmath255 , @xmath256 , @xmath257 , @xmath258 , and @xmath259 . the lists @xmath260 , \ldots , [ v_6])$ ] and @xmath209 are : r|*5p1.75mm| & + & @xmath261 & & & & + & @xmath261 & @xmath261 & @xmath262 & & + & @xmath262 & @xmath261 & @xmath261 & & + & @xmath261 & @xmath262 & @xmath261 & & + & @xmath261 & @xmath262 & @xmath262 & & + & @xmath262 & @xmath261 & @xmath262 & & + & @xmath262 & @xmath262 & @xmath261 & & + & @xmath262 & @xmath263 & & & + & @xmath263 & @xmath262 & & & + & @xmath262 & @xmath263 & & & + & @xmath263 & @xmath262 & & & + & @xmath262 & @xmath263 & @xmath262 & @xmath263 & @xmath263 + & @xmath263 & @xmath262 & @xmath263 & @xmath262 & @xmath263 + & @xmath263 & @xmath263 & @xmath262 & @xmath263 & @xmath262 + & @xmath262 & @xmath263 & @xmath263 & @xmath262 & @xmath263 + & @xmath263 & @xmath262 & @xmath263 & @xmath263 & @xmath262 + r|*5p1.75mm| & + & @xmath261 & & & & + & @xmath261 & @xmath261 & @xmath262 & & + & @xmath261 & @xmath262 & @xmath261 & & + & @xmath261 & @xmath262 & @xmath262 & & + & @xmath262 & @xmath261 & @xmath261 & & + & @xmath262 & @xmath261 & @xmath262 & & + & @xmath262 & @xmath262 & @xmath261 & & + & @xmath262 & @xmath263 & & & + & @xmath262 & @xmath263 & & & + & @xmath262 & @xmath263 & @xmath262 & @xmath263 & @xmath263 + & @xmath262 & @xmath263 & @xmath263 & @xmath262 & @xmath263 + & @xmath263 & @xmath262 & & & + & @xmath263 & @xmath262 & & & + & @xmath263 & @xmath262 & @xmath263 & @xmath262 & @xmath263 + & @xmath263 & @xmath262 & @xmath263 & @xmath263 & @xmath262 + & @xmath263 & @xmath263 & @xmath262 & @xmath263 & @xmath262 + hence , we obtain @xmath264 as the sequence of the last symbols of the words in @xmath209 . the standard permutation @xmath163 induced by @xmath124 is @xmath265 the cycles of @xmath163 arranged by their smallest elements are @xmath266 , @xmath267 , @xmath268 , @xmath269 , @xmath270 , and @xmath271 . applying the labeling function @xmath188 to the cycle @xmath272 ( starting with the second element ) yields the lyndon factor @xmath174 . with this procedure , we reconstructed @xmath273 from @xmath199 . [ sec : st ] the _ sort transform _ ( st ) is a bwt where we only sort the conjugates of the input up to a given depth @xmath2 and then we are using the index of the conjugates as a tie - breaker . depending on the depth @xmath2 and the implementation details this can speed up compression ( while at the same time slightly slowing down decompression ) . in contrast to the usual presentation of the st , we are using right shifts . this defines a slightly different version of the st . the effect is that the order of the symbols occurring in some particular context is reversed . this makes sense , because in data compression the st is applied to the reversal of a word . hence , in the st of the reversal of @xmath19 the order of the symbols in some particular context is the same as in @xmath19 . more formally , suppose @xmath274 for @xmath275 then in the sort transform of order @xmath276 of @xmath19 , the order of the occurrences of the letters @xmath52 is not changed . this property can enable better compression ratios on certain data . while the standard permutation is induced by a sequence of letters ( i.e. , a word ) we now generalize this concept to sequences of words . for a list of non - empty words @xmath277 we now define the _ @xmath2-order standard permutation _ @xmath278 induced by @xmath66 . as for the standard permutation , the first step is the construction of a new linear order @xmath101 on @xmath102 . we define @xmath103 by the condition @xmath279 let @xmath105 be the linearization of @xmath102 according to this new order . the idea is that we sort the line numbers of @xmath280 by first considering the @xmath2-order contexts and , if these are equal , then use the line numbers as tie - breaker . as before , the linearization according to @xmath101 induces a permutation @xmath278 by setting @xmath281 . now , @xmath282 is the position of @xmath174 if we are sorting @xmath66 by @xmath2-order context such that the line numbers serve as tie - breaker . we set @xmath283 where @xmath284 . now , we are ready to define the sort transform of order @xmath2 of a word @xmath19 : let @xmath285 ) = ( w_1 , \ldots , w_n)$ ] ; then @xmath286 , i.e. , we first sort all cyclic right - shifts of @xmath19 by their @xmath2-order contexts ( by using a stable sort method ) and then we take the sequence of last symbols according to this new sort order as the image under @xmath287 . since the tie - breaker relies on right - shifts , we have @xmath288 , i.e. , @xmath289 is the reversal mapping . the @xmath2-order sort transform of @xmath19 is the pair @xmath290 where @xmath22 is the index of @xmath19 in @xmath285)$ ] . as for the bwt , we see that the @xmath2-order sort transform is not bijective . next , we show that it is possible to reconstruct @xmath285)$ ] from @xmath291 . hence , it is possible to reconstruct @xmath19 from the pair @xmath290 where @xmath22 is the index of @xmath19 in @xmath285)$ ] . the presentation of the back transform is as follows . first , we will introduce the _ @xmath2-order context graph _ @xmath292 and we will show that it is possible to rebuild @xmath285)$ ] from @xmath292 . then we will show how to construct @xmath292 from @xmath291 . again , the approach will be slightly more general than required at the moment ; but we will be able to reuse it in the presentation of a bijective st . let @xmath293 , \ldots , [ u_s ] ) = ( v_1 , \ldots , v_n)$ ] be a list of words built from conjugacy classes @xmath294 $ ] of non - empty words @xmath210 . let @xmath139 be an arbitrary permutation of the elements in @xmath66 . we are now describing the edge - labeled directed graph @xmath2-order context graph of @xmath157 which will be used later as a presentation tool for the inverses of the st and the bijective st . the vertices of @xmath295 consist of all @xmath2-order contexts @xmath96 of words @xmath19 occurring in @xmath157 . we draw an edge @xmath296 from context @xmath297 to context @xmath298 labeled by @xmath22 if @xmath299 and @xmath300 . hence , every index @xmath301 of @xmath157 defines a unique edge in @xmath295 . we can also think of @xmath302 as an additional implicit label of the edge @xmath296 , since @xmath303 . a _ configuration _ @xmath304 of the @xmath2-order context graph @xmath295 consists of a subset of the edges @xmath305 and a vertex @xmath263 . the idea is that ( starting at context @xmath263 ) we are walking along the edges of @xmath295 and whenever an edge is used , it is removed from the set of edges @xmath305 . we now define the transition @xmath306 from a configuration @xmath307 to another configuration @xmath308 with output @xmath309 more formally . if there exists an edge in @xmath310 starting at @xmath297 and if @xmath311 is the unique edge with the smallest label @xmath22 starting at @xmath297 , then we have the single - step transition @xmath312 if there is no edge in @xmath310 starting at @xmath297 , then the outcome of @xmath313 is undefined . inductively , we define @xmath314 and for @xmath315 and @xmath316 we have @xmath317 for some configuration @xmath318 . hence , the reversal @xmath319 is the label along the path of length @xmath320 starting at configuration @xmath307 . in particular , if @xmath321 holds , then it is possible to chase at least @xmath322 transitions starting at @xmath307 ; vice versa , if we are chasing @xmath323 transitions then we obtain a word of length @xmath323 as a label . we note that successively taking the edge with the smallest label comes from the use of right - shifts . if we had used left - shifts we would have needed to chase largest edges for the following lemma to hold . the reverse labeling of the big - step transitions is motivated by the reconstruction procedure which will work from right to left . [ lem : smallestedgechasing ] let @xmath136 , @xmath324 , \ldots , [ v_s])$ ] , @xmath325 , and @xmath326 . let @xmath310 consist of all edges of @xmath327 . then @xmath328 & \ \ \vdots \\ ( \mathcal{c}_s , c_s ) & \stackrel{v_s}{\to } ( \mathcal{c}_{s+1},c_s ) . \end{aligned}\ ] ] let @xmath329 . consider some index @xmath22 , @xmath330 , and let @xmath331 , \ldots , [ v_{i-1}])$ ] . suppose that @xmath332 consists of all edges of @xmath327 except for those with labels @xmath333 for @xmath334 . let @xmath335 . we write @xmath336 and @xmath337 , i.e. , @xmath138 = ( u_{t+1 } , \ldots , u_{t+q})$ ] . starting with @xmath338 , we show that the sequence of transitions @xmath339 is defined . more precisely , we will see that the transition @xmath340 walks along the edge @xmath341 and hence indeed is labeled with the letter @xmath342 . consider the context @xmath343 . by induction , we have @xmath344 and no edge with label @xmath345 for @xmath346 occurs in @xmath347 while all other labels do occur . in particular , @xmath348 for @xmath349 is an edge in @xmath347 ( where @xmath350 only holds for @xmath351 ; we will consider the case @xmath352 below ) . suppose there were an edge @xmath353 with @xmath354 . then @xmath355 and hence , @xmath356 has the same @xmath2-order context as @xmath357 . but in this case , in the construction of @xmath358 we used the index in @xmath66 as a tie - breaker . it follows @xmath359 which contradicts the properties of @xmath347 . hence , @xmath348 is the edge with the smallest label starting at context @xmath343 . therefore , @xmath360 and @xmath340 indeed walks along the edge @xmath348 . it remains to verify that @xmath361 , but this is clear since @xmath362 . [ lem : reconstructingg ] let @xmath136 , @xmath324 , \ldots , [ v_s])$ ] , @xmath363 , and @xmath364 . then it is possible to reconstruct @xmath295 from @xmath124 . by lemma [ lem : contextretrieval ] it is possible to reconstruct the contexts @xmath365 . this gives the vertices of the graph @xmath295 . write @xmath366 . for each @xmath367 we draw an edge @xmath368 . this yields the edges of @xmath295 . the @xmath2-order st is invertible , i.e. , given @xmath290 where @xmath22 is the index of @xmath19 in @xmath285)$ ] one can reconstruct the word @xmath19 . the construction of @xmath19 consists of two phases . first , by lemma [ lem : reconstructingg ] we can compute @xmath369))$ ] . by lemma [ lem : contextretrieval ] we can compute @xmath370 from @xmath290 . in the second stage , we are using lemma [ lem : smallestedgechasing ] for reconstructing @xmath19 by chasing @xmath371 where @xmath305 consists of all edges in @xmath369))$ ] . efficient implementations of the inverse transform rely on the fact that the @xmath2-order contexts of @xmath285)$ ] are ordered . this allows the implementation of the @xmath2-order context graph @xmath292 in a vectorized form @xcite . [ exa : st ] we compute the sort transform of order @xmath372 of @xmath108 from example [ exa : start ] . the list @xmath373)$ ] is depicted in figure [ sfg : m2 ] . this yields the transform @xmath374 where @xmath375 is the last column of the matrix @xmath373)$ ] and @xmath19 is the @xmath22-th element in @xmath373)$ ] . next , we show how to reconstruct the input @xmath19 from @xmath123 . the standard permutation induced by @xmath124 is @xmath376 note that @xmath163 has four cycles @xmath377 , @xmath378 , @xmath379 , and @xmath380 . we obtain the context of order @xmath372 of the @xmath33-th word by @xmath381 . in particular , @xmath382 , @xmath383 , @xmath384 , @xmath385 , @xmath386 , @xmath387 , @xmath388 , and @xmath389 . with @xmath124 and these contexts we can construct the graph @xmath390)$ ] . the vertices of @xmath327 are the contexts and the edge - labels represent positions in @xmath124 . the graph @xmath327 is depicted below : ( 0,0 ) node[circle , draw , inner sep=2.5pt ] ( ba ) @xmath391 ; ( 3,0 ) node[circle , draw , inner sep=2.5pt ] ( aa ) @xmath392 ; ( 6,0 ) node[circle , draw , inner sep=2.5pt ] ( ca ) @xmath393 ; ( 9,0 ) node[circle , draw , inner sep=2.5pt ] ( cb ) @xmath394 ; ( 0,3 ) node[circle , draw , inner sep=2.5pt ] ( bb ) @xmath395 ; ( 3,3 ) node[circle , draw , inner sep=2.5pt ] ( ab ) @xmath396 ; ( 6,3 ) node[circle , draw , inner sep=2.5pt ] ( bc ) @xmath119 ; ( 9,3 ) node[circle , draw , inner sep=2.5pt ] ( cc ) @xmath397 ; ( aa ) node[above ] @xmath1 ( ba ) ; ( ab ) .. controls ( 1.25,1.75 ) .. node[sloped , above ] @xmath372 ( ba ) ; ( ab ) node[left ] @xmath398 ( aa ) ; ( ab ) node[above right ] @xmath399 ( ca ) ; ( ba ) .. controls ( 1.75,1.25 ) .. node[sloped , above ] @xmath117 ( ab ) ; ( ba ) node[left ] @xmath112 ( bb ) ; ( bb ) node[above ] @xmath115 ( ab ) ; ( bc ) node[above ] @xmath400 ( ab ) ; ( bc ) node[sloped , above ] @xmath401 ( cb ) ; ( bc ) .. controls ( 8,2 ) .. node[sloped , above ] @xmath402 ( cb ) ; ( bc ) .. controls ( 8.5,2.5 ) .. node[sloped , above ] @xmath403 ( cb ) ; ( ca ) node[left ] @xmath404 ( bc ) ; ( cb ) .. controls ( 7,1 ) .. node[sloped , above ] @xmath405 ( bc ) ; ( cb ) node[right ] @xmath406 ( cc ) ; ( cb ) .. controls ( 6.5,0.5 ) .. node[sloped , above ] @xmath407 ( bc ) ; ( cc ) node[above ] @xmath408 ( bc ) ; we are starting at the context @xmath409 and then we are traversing @xmath327 along the smallest edge - label amongst the unused edges . the sequence of the edge labels obtained this way is @xmath410 the labeling of this sequence of positions yields @xmath411 . since we are constructing the input from right to left , we obtain @xmath108 . the bijective sort transform combines the lyndon factorization with the st . this yields a new algorithm which serves as a similar preprocessing step in data compression as the bwt . in a lot of applications , it can be used as a substitute for the st . the proof of the bijectivity of the transform is slightly more technical than the analogous result for the bijective bwt . the main reason is that the bijective sort transform is less modular than the bijective bwt ( which can be grouped into a ` lyndon factorization part ' and a ` gessel - reutenauer transform part ' and which for example allows the use of different orders on the alphabet for the different parts ) . for the description of the bijective st and of its inverse , we rely on notions from section [ sec : sorttransform ] . the bijective st of a word @xmath19 of length @xmath0 is defined as follows . let @xmath67 with @xmath192 be the lyndon factorization of @xmath19 . let @xmath412 , \ldots , [ v_s ] ) = ( u_1 , \ldots , u_n)$ ] . then the bijective st of order @xmath2 of @xmath19 is @xmath413 . that is , we are sorting the conjugacy classes of the lyndon factors by @xmath2-order contexts and then take the sequence of the last letters . the letter _ l _ in @xmath414 is for _ [ thm : bijst ] the bijective st of order @xmath2 is invertible , i.e. , given @xmath415 one can reconstruct the word @xmath19 . let @xmath191 with @xmath192 be the lyndon factorization of @xmath19 , let @xmath416 , and let @xmath417 . by lemma [ lem : reconstructingg ] we can rebuild the @xmath2-order context graph @xmath418 , \ldots , [ v_s ] ) ) = ( w_1 , \ldots , w_n)$ ] from @xmath124 . let @xmath310 consist of all edges in @xmath327 . then by lemma [ lem : smallestedgechasing ] we see that @xmath419 & \ \ \vdots \\ ( \mathcal{c}_s , c_s ) & \stackrel{v_s}{\to } ( \mathcal{c}_{s+1},c_s ) . \end{aligned}\ ] ] we can not use this directly for the reconstruction of @xmath19 since we do not know the lyndon factors @xmath174 and the contexts @xmath420 . the word @xmath204 is the first element in the list @xmath412 , \ldots , [ v_s])$ ] because @xmath204 is lexicographically minimal and it appears as the first element in the list @xmath260 , \ldots , [ v_s])$ ] . therefore , by lemma [ lem : contextretrieval ] we obtain @xmath421 . the reconstruction procedure works from right to left . suppose we have already reconstructed @xmath422 for @xmath423 with @xmath424 being a ( possibly empty ) suffix of @xmath425 . moreover , suppose we have used the correct contexts @xmath426 . consider the configuration @xmath318 defined by @xmath419 & \ \ \vdots \\ ( \mathcal{c}_j , c_j ) & \stackrel{v_j}{\to } ( \mathcal{c}_{j+1},c_j ) \\ ( \mathcal{c}_{j+1},c_{j+1 } ) & \stackrel{w'}{\to } ( \mathcal{c}',c ' ) \end{aligned}\ ] ] we assume that the following invariant holds : @xmath427 contains no edges @xmath428 with @xmath429 . we want to rebuild the next letter . we have to consider three cases . first , if @xmath430 then @xmath431 yields the next letter @xmath261 such that @xmath432 is a suffix of @xmath425 . second , let @xmath433 and suppose that there exists an edge @xmath434 starting at @xmath435 . then there exists a word @xmath436 in @xmath437 , \ldots , [ v_s]$ ] such that @xmath438 . if @xmath439 then from the invariant it follows that @xmath440 . this is a contradiction , since @xmath441 is minimal among the words in @xmath437 , \ldots , [ v_s]$ ] . hence , @xmath442 and the invariant still holds for @xmath443 . the last letter @xmath261 of @xmath441 is obtained by @xmath444 the third case is @xmath433 and there is no edge @xmath434 starting at @xmath435 . as before , @xmath441 is minimal among the ( remaining ) words in @xmath437 , \ldots , [ v_s]$ ] . by construction of @xmath327 , the unique edge @xmath445 with the minimal label @xmath323 has the property that @xmath446 . in particular , since @xmath441 is minimal , the invariant for @xmath448 is established . in this case , the last letter @xmath261 of @xmath441 is obtained by @xmath449 we note that we can not distinguish between the first and the second case since we do not know the length of @xmath425 , but in both cases , the computation of the next symbol is identical . in particular , in contrast to the bijective bwt we do not implicitly recover the lyndon factorization of @xmath19 . we note that the proof of theorem [ thm : bijst ] heavily relies on two design criteria . the first one is to consider @xmath412 , \ldots , [ v_s])$ ] rather than @xmath450 , \ldots , [ v_1])$ ] , and the second is to use right - shifts rather than left - shifts . the proof of theorem [ thm : bijst ] yields the following algorithm for reconstructing @xmath19 from @xmath451 : 1 . compute the @xmath2-order context graph @xmath452 and the @xmath2-order context @xmath297 of the last lyndon factor of @xmath19 . 2 . start with the configuration @xmath304 where @xmath305 contains all edges of @xmath327 and @xmath453 . 3 . if there exists an outgoing edge starting at @xmath263 in the set @xmath305 , then * let @xmath454 be the edge with the minimal label @xmath323 starting at @xmath263 . * output @xmath455 . * set @xmath456 and @xmath457 . * continue with step ( 3 ) . 4 . if there is no outgoing edge starting at @xmath263 in the set @xmath305 , but @xmath458 , then * let @xmath459 be the edge with the minimal label @xmath323 . * output @xmath455 . * set @xmath460 and @xmath461 . * continue with step ( 3 ) . the algorithm terminates as soon as @xmath462 . the sequence of the outputs is the reversal @xmath122 of the word @xmath19 . we consider the word @xmath108 from example [ exa : start ] and its lyndon factorization @xmath253 where @xmath254 , @xmath255 , @xmath256 , @xmath257 , @xmath258 , and @xmath259 . for this particular word @xmath19 the bijective burrows - wheeler transform and the bijective sort transform of order @xmath372 coincide . from example [ exa : bijbwt ] , we know @xmath463 and the standard permutation @xmath163 . as in example [ exa : st ] we can reconstruct the @xmath372-order contexts @xmath464 of @xmath465,\ldots,[v_6])$ ] : @xmath466 , @xmath467 , @xmath468 , @xmath385 , @xmath469 , @xmath470 , and @xmath471 . with @xmath124 and the @xmath372-order contexts we can construct the graph @xmath472,\ldots,[v_6]))$ ] : ( 0,0 ) node[circle , draw , inner sep=2.5pt ] ( ba ) @xmath391 ; ( 3,0 ) node[circle , draw , inner sep=2.5pt ] ( aa ) @xmath392 ; ( 9,0 ) node[circle , draw , inner sep=2.5pt ] ( cb ) @xmath394 ; ( 0,3 ) node[circle , draw , inner sep=2.5pt ] ( bb ) @xmath395 ; ( 3,3 ) node[circle , draw , inner sep=2.5pt ] ( ab ) @xmath396 ; ( 6,3 ) node[circle , draw , inner sep=2.5pt ] ( bc ) @xmath119 ; ( 9,3 ) node[circle , draw , inner sep=2.5pt ] ( cc ) @xmath397 ; ( aa ) .. controls ( 4,-0.5 ) and ( 4,+0.5 ) .. node[right ] @xmath1 ( aa ) ; ( aa ) node[above ] @xmath372 ( ba ) ; ( ab ) node[left ] @xmath398 ( aa ) ; ( ab ) .. controls ( 1.25,1.75 ) .. node[sloped , above ] @xmath399 ( ba ) ; ( ba ) .. controls ( 1.75,1.25 ) .. node[sloped , above ] @xmath117 ( ab ) ; ( ba ) node[left ] @xmath112 ( bb ) ; ( bb ) node[above ] @xmath115 ( ab ) ; ( bc ) .. controls ( 7,1 ) .. node[sloped , above ] @xmath400 ( cb ) ; ( bc ) node[sloped , above ] @xmath401 ( cb ) ; ( bc ) .. controls ( 8,2 ) .. node[sloped , above ] @xmath402 ( cb ) ; ( bc ) .. controls ( 8.5,2.5 ) .. node[sloped , above ] @xmath403 ( cb ) ; ( cb ) .. controls ( 6.5,0.5 ) .. node[sloped , above ] @xmath404 ( bc ) ; ( cb ) .. controls ( 6.0,0.0 ) .. node[sloped , above ] @xmath405 ( bc ) ; ( cb ) node[right ] @xmath406 ( cc ) ; ( cb ) .. controls ( 5.5,-0.5 ) .. node[sloped , above ] @xmath407 ( bc ) ; ( cc ) node[above ] @xmath408 ( bc ) ; we are starting with the edge with label @xmath1 and then we are traversing @xmath327 along the smallest unused edges . if we end in a context with no outgoing unused edges , then we are continuing with the smallest unused edge . this gives the sequence @xmath473 after which we end in context @xmath392 with no unused edges available . then we continue with the sequences @xmath474 and @xmath475 . the complete sequence of edge labels obtained this way is @xmath476 and the labeling of this sequence with @xmath188 yields @xmath477 . as for the st , we are reconstructing the input from right to left , and hence we get @xmath108 . we discussed two bijective variants of the burrows - wheeler transform ( bwt ) . the first one is due to scott . roughly speaking , it is a combination of the lyndon factorization and the gessel - reuternauer transform . the second variant is derived from the sort transform ( st ) ; it is the main contribution of this paper . we gave full constructive proofs for the bijectivity of both transforms . as a by - product , we provided algorithms for the inverse of the bwt and the inverse of the st . for the latter , we introduced an auxiliary graph structure the @xmath2-order context graph . this graph yields an intermediate step in the computation of the inverse of the st and the bijective st . it can be seen as a generalization of the cycle decomposition of the standard permutation which in turn can be used as an intermediate step in the computation of the inverse of the bwt and the bijective bwt . * acknowledgments . * the author would like to thank yossi gil and david a. scott for many helpful discussions on this topic as well as alexander lauser , antonio restivo , and the anonymous referees for their numerous suggestions which improved the presentation of this paper .

the problem of a single particle interacting with its environment is encountered frequently in condensed matter physics , and in many variations . such scenarios appear in the form of genuine defects in materials , like magnetic impurities in metals and the associated ubiquitous kondo problem.@xcite impurity problems are also often used as a stepping stone to gain insight into the behavior of strongly correlated systems , such as studying a single hole doped into a mott insulator helps understanding some properties of high - temperature superconductors.@xcite in recent years , impurity problems have also started to gain some attention in the context of ultracold atomic gases . amongst others , the fermionic polaron problem has been studied thoroughly in connection with strongly imbalanced fermi gases close to a feshbach resonance,@xcite where a single `` spin - down '' atom interacting with a fermi sea of `` spin - up '' atoms appears as a polaronic quasiparticle and eventually forms a two - particle bound - state with one of the majority atoms if the interactions are strong enough.@xcite more generally , such polaronic problems appear when we consider the nature of the threshold of spectral functions of gapped excitations above many - body ground states.@xcite here we are interested in a seemingly similar problem , where a mobile impurity is coupled to a continuum of gapless excitations at a quantum critical point , instead of gapless particle - hole excitations at a fermi surface . in particular , we want to study the situation where bosons in an optical lattice are tuned to the superfluid - mott insulator transition,@xcite and a single atom of a different species is coupled to the bosons via a density - density interaction . this scenario is especially interesting due to the recent experimental implementation of quantum - gas microscopes,@xcite which allow tracking the motion of the impurity through the bosonic bath . indeed , a recent combined experimental and theoretical work has addressed a very similar situation in an effectively one - dimensional system.@xcite we are going to focus on the other experimentally relevant , two dimensional situation instead . from our perspective , this case also happens to be the most interesting scenario , because in @xmath0 spatial dimensions the critical behavior at the sf - mi transition is described by a non - trivial fixed point without well - defined quasiparticle excitations.@xcite recently there has been a renewed interest in this particular transition as experiments succeeded in observing the so - called higgs amplitude mode.@xcite the main motivation for our work is the prospect of studying aspects of quantum critical transport in a clean , well - defined experimental setting . calculating transport coefficients of systems close to a quantum phase transition remains a challenging problem in theoretical condensed matter physics . the main reason is that all traditional methods for calculating transport properties rely on a quasiparticle picture of the low - energy excitations above the ground - state , which breaks down at a quantum critical point . indeed , strongly coupled conformal field theories ( cft ) that describe physical properties of physical systems close to quantum critical points typically have no well defined quasiparticle excitations . some progress has been made recently using the so called ads / cft correspondence , which circumvents this problem by mapping the strongly coupled cft to a dual , weakly coupled gravitational theory , where transport coefficients can be computed reliably.@xcite in this paper our main interest is to calculate the diffusion constant @xmath1 of the impurity as a function of temperature in the quantum critical regime , shown in fig . [ fig : phase ] . this diffusion constant can be measured directly in experiments with quantum - gas microscopes by evaluating the mean square displacement @xmath2 of the impurity atom from its initial position after a time @xmath3 @xmath4 where the brackets denote an average over many experimental realizations and the time @xmath5 has to be larger than the typical collision time @xmath6 , below which the impurity propagates ballistically . several theoretical works have addressed similar questions using the holographic correspondence in the context of a heavy quark moving through a quark gluon plasma.@xcite here we will use a conventional method based on the boltzmann equation , however , which turns out to provide an adequate description at low temperatures . the quantum criticality of the bulk critical point has a single characteristic time which controls transport and relaxational processes close to equilibrium @xmath7.@xcite we will show here that the situation with impurity motion is not as universal , and the impurity time scale is determined by an interplay between several couplings which are formally irrelevant at the critical point . one of the these couplings is the inverse effective mass , @xmath8 , of the mobile impurity ; there is the associated energy scale @xmath9 , where @xmath10 the velocity of the bulk bosonic excitations at the critical point . the value of @xmath11 controls the quantum wave packet spread of the impurity at short times . the other `` irrelevant '' parameters couple the impurity to the bulk excitations , and associated energy scales determine the diffusion of the impurity at long times even at the lowest temperatures . one of our interesting results is that the diffusion constant of the impurity does _ not _ depend on @xmath11 at low temperatures , however . the rest of the paper is organized as follows . in section [ sec : ft ] we derive an effective low energy theory for a mobile impurity coupled to bosons at the superfluid mott - insulator ( sf - mi ) transition and we show that the presence of three - body interactions changes the theory significantly at the particle - hole symmetric point . in section [ sec:2p ] we study perturbative self - energy corrections to the impurity propagator and calculate its diffusion constant as a function of temperature , neglecting three - body interactions . we repeat these calculations including three - body interactions in section [ sec:3p ] and show that this changes the results qualitatively at low temperatures . finally we perform a renormalization group analysis of the effective field theory in sec . [ sec : rg ] . we start by deriving the effective low - energy theory for a bose - hubbard model at the sf - mi critical point in @xmath0 spatial dimensions , following ref . . a convenient starting point is to consider particle- and hole- excitations on top of a mott insulating state , which are described by the lagrangian @xmath12 here the fields @xmath13 and @xmath14 represent particle- and hole excitations on top of a state with an integer filling of bosons . the respective energy gaps for these excitations are denoted by @xmath15 and @xmath16 , their effective masses by @xmath17 and @xmath18 . the last term in eq . creates or annihilates particles and holes in pairs , as required by the conservation of the total number of bosons , and we do not explicitly show higher order terms in the fields @xmath19 and @xmath20 that are allowed by symmetry . ultimately we are interested in the experimentally relevant situation where the transition happens at a constant density , i.e. at the tip of the mott - lobe . this happens only if the gaps @xmath15 and @xmath16 are equal , such that particles and holes condense at the same time when the gap closes . for the moment we will consider the general case , however , and parametrize the gaps as @xmath21 now we add a single , mobile impurity atom to the bosons , the dynamics of which is described by the free particle lagrangian @xmath22 where the field @xmath23 represents the impurity and the chemical potential @xmath24 has to be adjusted such that the density of impurities is zero . this is important only at finite temperatures , however , where we need to ensure that no artificial , thermally excited impurities exist . at zero temperature we can safely set @xmath25 . within a microscopic bose - hubbard like model the coupling between the impurity and the bosons is an onsite density - density interaction , which is a pure two - body interaction in the simplest case . since the local boson density is given by the difference between particle- and hole- densities ( up to the constant average number of bosons that we neglect in the following , as it only renormalizes the impurity s chemical potential ) , the interaction term in our effective field theory takes the form @xmath26 where the subscript @xmath27 of the coupling constant @xmath28 indicates that this term derives from a two - body interaction . furthermore we include a term that descends from a three - body interaction @xmath29 because in the case of particle - hole symmetry ( @xmath30 ) this interaction generates couplings that are _ more _ relevant in an rg sense than those deriving from two - body interactions only , as will be shown below . to avoid confusion we note that a three - body interaction in the bose - hubbard model of the form @xmath31 , where @xmath32 denotes the number of bosons on site @xmath33 and @xmath34 is the impurity creation operator , also generates a term of the form in addition to . for brevity we will always refer to the term in as the three - body interaction , however . eventually our system is described by the total lagrangian @xmath35 in order to arrive at the more familiar low energy theory for the sf - mi critical point we define the fields @xmath36 and integrate out the field @xmath37 . after a rescaling of @xmath38 and defining @xmath39 as well as @xmath40 we arrive at our final result @xmath41 the terms shown are the leading order terms within a gradient expansion . for the case of particle - hole symmetry ( @xmath42 , i.e. @xmath43 ) the dynamical critical exponent of the bosons is @xmath44 and the critical point is described by the well known @xmath45 cft in @xmath46 dimensions . even more interesting , the direct interaction term between the bosons and the impurity @xmath47 vanishes if only two - body interactions are present . this somewhat counterintuitive result ( in the sense that the particle - hole symmetric interaction term vanishes if the @xmath48-theory is particle - hole symmetric ) can be understood from the fact that the two - body interaction term @xmath49 is invariant under a particle - hole transformation ( @xmath50 ) _ only _ in combination with @xmath51 . this is not true for the three - body interaction term @xmath52 , however , and that is why the parametrically smaller @xmath53 generates a term that is more relevant than the one generated by two - body interactions at the particle - hole symmetric point . the leading order terms that are consistent with the symmetry @xmath54 = \mathcal{l}[\psi^ * ; -\delta ,- u_2,u_3 ] \label{syms}\ ] ] are the ones shown in eq . . also note that the two - body interaction @xmath28 couples the impurity density @xmath55 to the time - like component of the bosonic current - density @xmath56 . we stress again that we focus solely on the particle - hole symmetric case ( @xmath30 ) in the rest of the paper . a tree - level scaling analysis of the lagrangian using the exact critical exponents of the bosonic cft shows that both interactions @xmath28 and @xmath53 are irrelevant in @xmath0 spatial dimensions at the p / h - symmetric point . indeed , performing a scaling with @xmath44 gives the scaling dimensions @xmath57 = ( d-1+\eta)/2 $ ] and @xmath58 = d/2 $ ] for the fields , as well as @xmath59 = d+1 - 1/\nu$ ] and @xmath60 = d$ ] for the composite bosonic density- and current operators per definition . here @xmath61 and @xmath62 denote the anomalous dimension and the correlation length exponent of the bosonic cft , which take the values @xmath63 and @xmath64 at the quantum critical point in @xmath0 spatial dimensions.@xcite from these scaling dimensions we get @xmath65 = 1-d$ ] and @xmath66=1/\nu - d$ ] , thus both couplings are irrelevant in @xmath0 . also note that the curvature of the impurity dispersion is an irrelevant perturbation , as the scaling dimension of the impurity mass @xmath11 is @xmath67=1 $ ] . we will perform a more elaborate rg analysis in sec . [ sec : rg ] . in this section we analyze the field theory at the particle - hole symmetric point ( @xmath30 ) , assuming that three body interactions are absent ( i.e. @xmath68 ) . the lagrangian thus takes the simplified form @xmath69 note that we dropped the chemical potential since it is not important at zero temperature , as argued above . as a first step towards an understanding of the impurity s properties we calculate its self - energy perturbatively in the interaction strength @xmath28 at zero temperature in the following . since @xmath28 is irrelevant , we expect the perturbative calculation to give qualitatively correct results . the terms up to second order are shown in fig . the first order tadpole contribution turns out to be identically zero . note that the boson loop in the second order diagram corresponds to the time - like component of the bosonic current - current correlation function of the free bosonic theory due to the particular form of the interaction vertex in . at @xmath70 the form of the bosonic current - current correlator of the interacting @xmath48-theory at the critical point in @xmath0 is fixed by conformal invariance and is given by @xmath71 here @xmath72 is a numerical factor that is renormalized by the boson interaction @xmath73 and takes the value @xmath74 for the free bosonic theory . the functional form of @xmath75 is independent of @xmath73 , however . for this reason our results for the impurity self - energy below are perturbative in @xmath28 , but valid to arbitrary order in the boson interaction @xmath73 . at zero temperature the imaginary part of the retarded self - energy up to second order in @xmath28 is given by @xmath76 where @xmath77 denotes the impurity dispersion . at @xmath78 the integral can be evaluated exactly and we obtain @xmath79 \notag \\ & = & - \frac{u_2 ^ 2 \kappa}{3 \pi c^4 } \ , \omega^3 + \mathcal{o}(\omega^4)\end{aligned}\ ] ] for finite momenta we can derive approximate analytic expressions , which take the form @xmath80 with @xmath81 . the qualitative change of the self - energy at @xmath82 can be understood from kinematic considerations . the only lifetime limiting process for the impurity at zero temperature happens to be the excitation of a bosonic `` phonon '' mode . due to energy- and momentum conservation this process is prohibited as long as the group velocity of the impurity @xmath83 is smaller than the `` sound '' velocity @xmath10 of the bosons . for this reason the impurity spectral function has a sharp delta - function peak for all momenta @xmath84 , which is reflected by the fact that the imaginary part of the self energy scales as @xmath85 close to the quasiparticle pole . on the other hand , for @xmath86 the impurity can scatter strongly by exciting the bosons . consequently the imaginary part takes a non - zero on - shell value , implying that there is no well defined impurity excitation anymore . the appearance of this kinematic constraint can be readily seen from eq . , where the imaginary part of the current correlator is non - zero only if @xmath87 . it is important to note that this threshold behavior for creating bosonic excitations at @xmath88 holds to all orders in perturbation theory in the impurity - boson coupling @xmath28 . higher order diagrams with more boson propagators describe processes where the impurity excites multiple boson modes , which in turn have an even higher energetic threshold . from a theoretical perspective it is an interesting question , if the impurity s properties above this kinematic threshold at @xmath82 are described by a rg fixed point . this is not the case , however , and we will come back to this problem in sec . [ sec : rg ] . an important transport coefficient that can be directly measured in experiments is the diffusion constant @xmath1 of the impurity . here we calculate @xmath89 as a function of temperature @xmath90 , impurity mass @xmath11 and the mott gap @xmath91 using the boltzmann equation , which gives qualitatively correct results as long as the impurity has a well defined quasiparticle peak . this condition is satisfied in the temperature regime @xmath92 ( we set @xmath93 from now on ) , where the kinematic threshold discussed in the previous section does nt matter . let us quickly estimate the experimentally relevant temperature regime before proceeding with the boltzmann calculation . typical temperatures in experiments@xcite in units of the hubbard interaction @xmath94 are on the order of @xmath95 . using the theoretical results for the hopping amplitude @xmath96 as well as the sound velocity @xmath10 at the critical point of the 2d bose - hubbard model@xcite , which take the values @xmath97 and @xmath98 , together with the band mass @xmath99 of the impurity , we obtain @xmath100 , which is typically well below the kinematic threshold . note that @xmath101 is a relatively large energy scale . we are interested in a situation without external forces but with an initial density gradient , thus the boltzmann equation takes the form @xmath102 \ , \label{boltzequ}\ ] ] with @xmath103 as the wigner distribution function of the impurity at momentum @xmath104 , position @xmath105 and time @xmath3 . the collision integral has the standard form @xmath106 = - \sum_{\k ' } \big ( w_{\k,\k ' } f_\k - w_{\k',\k } \ , f_{\k ' } \big ) \ , \label{collint}\ ] ] where the transition rates @xmath107 can be calculated using fermi s golden rule and are given by @xmath108 \ . \label{wkk}\end{aligned}\ ] ] here @xmath109 denotes the time - like component of the retarded bosonic current - current correlation function , @xmath110 is the bose - einstein distribution function and @xmath111 denotes the unit step function . details of the derivation are given in appendix [ app : boltz ] . in typical experimental situations the impurity is initially localized , released at some time @xmath112 and then is allowed to propagate through the bosonic bath . in principle the boltzmann equation allows to calculate the full time evolution of the impurity distribution function . at short times collisions do nt play a role and the impurity propagates ballistically @xmath113 where @xmath6 denotes the typical time between collisions . if we start at @xmath112 with the particles localized in a gaussian wavefunction @xmath114 then the short time evolution of the wigner distribution function according to eq . is free particle behavior @xmath115 at long times @xmath5 collisions dominate and the impurity propagates diffusively . the wigner distribution then takes the form @xmath116 with a maxwell - boltzmann distribution of the momenta . indeed , in the long time limit , where the wigner distribution function of the impurity is close to equilibrium , we can derive a diffusion equation from the boltzmann equation for arbitrary linear collision integrals of the form . again , details can be found in appendix [ app : boltz ] . the corresponding impurity diffusion constant @xmath1 takes the simple form @xmath117 with a transport relaxation time @xmath6 which is given by @xmath118 f^\text{boltz}_\k \ , \label{reltime}\ ] ] where @xmath119 and @xmath120 is the equilibrium boltzmann distribution function . using the explicit form of the transition rates @xmath107 in eq . we can cast the expression for the diffusion constant @xmath89 in a scaling form @xmath121 the corresponding scaling function @xmath122 can be calculated numerically and is shown in fig . [ fig : phij3d ] . here we used the one loop result for the bosonic current correlator @xmath123 at finite temperature , where the dispersion relation of the bosonic modes @xmath124 acquires a temperature dependent mass term @xmath125 which follows from a @xmath126 expansion in the limit @xmath127 of the bosonic cft.@xcite note that this temperature dependent gap is an artifact of the @xmath127 limit and it is not the scope of this paper to go beyond this limitation . at a fixed temperature @xmath90 , the diffusion constant @xmath1 takes its minimal value in the quantum critical regime directly above the quantum critical point , where the mott gap vanishes ( @xmath128 ) . this is consistent with the naive expectation that the phase space for scattering processes between the impurity and the bosonic modes is large , if the gap of the bosonic modes is small . the scaling function in the quantum critical regime for @xmath128 is shown in more detail in fig . [ fig : phij ] . note that @xmath122 approaches a finite value @xmath129 at zero temperature and scales linearly for small @xmath130 , as @xmath131 interestingly , the fact that the scaling function takes a constant value at @xmath70 and @xmath128 implies that the diffusion constant at the critical point does not depend on the impurity mass @xmath11 at very low temperatures . indeed , the curvature of the impurity dispersion , _ i.e. _ the inverse mass @xmath132 , is an irrelevant perturbation , and here there is no conspiracy between the two irrelevant couplings @xmath132 and @xmath28 that potentially could change the @xmath90 dependence of @xmath1 from that expected from its @xmath133 dependence . only at higher temperatures the mass starts to play a role , leading to a crossover from a @xmath134 behavior at very low temperatures to a @xmath135 dependence at temperatures @xmath136 . note that this crossover temperature is around the typical temperatures in experiments , where @xmath137 , thus we expect that the @xmath135 behavior at larger temperatures should be easily accessible . in order to compare our results directly to experiments it would be natural to express the diffusion constant in units of the hopping amplitude @xmath96 ( or more precisely @xmath138 , if the lattice constant @xmath139 and planck s constant are not set to unity ) . assuming that the hubbard interaction @xmath28 between the impurity and the bosons equals the hubbard-@xmath94 in the bulk , the diffusion constant at a typical temperature @xmath140 above the critical point takes the value @xmath141 . the corresponding transport relaxation time @xmath6 takes the value @xmath142 , showing that it is experimentally promising to reach the diffusive regime at times @xmath5 . we note here that in certain parameter regimes a direct comparison of experimental results with our calculations might be difficult due to the harmonic confinement in experiments . this confinement effectively leads to an averaging of the diffusion constant when the impurity explores different regions of the trap . in most cases this is not a problem , however . only in the mott regime at temperatures well below the mott gap , where the diffusion constant is large and the impurity propagates almost ballistically , it explores larger regions of the trap and averaging could become important . such confinement problems can be circumvented by increasing the temperature slightly , since @xmath1 and @xmath6 are strongly temperature dependent and decrease by an order of magnitude if the temperature is raised from @xmath143 to @xmath144 . on a formal level , the fact that the scaling function at the critical point @xmath145 is constant can be traced back to the behavior of the current correlator at small frequencies , which scales as @xmath146 for @xmath147 . it is important to stress that this behavior , which is related to the fact that @xmath148 is an odd function of frequency and expected to be an analytic at @xmath149 , is likely valid in general and not related to our one - loop approximation of the current correlator . indeed it has been argued that the current - current correlation function takes a hydrodynamic form for @xmath150 @xmath151 the imaginary part of which obviously scales as @xmath152 for small frequencies . this has also been confirmed by calculations using the ads / cft correspondence.@xcite in the mott regime away from the critical point , where the excitation gap @xmath91 is finite , the scaling function diverges exponentially as the temperature approaches zero @xmath153 this behavior can be understood from the fact that the impurity does not scatter with bosonic excitations at temperatures well below the mott gap and thus propagates ballistically , which translates to a diverging diffusion constant . we also note that the mobility @xmath154 of the impurity is independent of its mass at the critical point at low enough temperatures . the mobility is defined as @xmath155 , where @xmath156 is the impurity s terminal drift velocity in response to an external force @xmath157 . it is related to the diffusion constant @xmath1 via the einstein relation @xmath158 and thus has the same dependence on the impurity mass as the diffusion constant . in the superfluid phase the impurity diffusion constant @xmath1 can be calculated in the same way as in sec . [ sec : boltz2 ] , using a @xmath126 expansion in the limit @xmath127 for the bosonic current correlation function @xmath123 . the only difference is that eq . for the boson mass term has to be replaced by@xcite @xmath159 in contrast to the mott regime , where the energy scale is set by the mott gap @xmath91 , the energy scale @xmath160 in the superfluid phase is set by the helicity modulus at zero temperature @xmath161.@xcite in the @xmath126 expansion , the helicity modulus is non - zero only at @xmath70 , and the o(@xmath162 ) symmetry is fully restored at any non - zero temperature . however , for the @xmath163 case of interest here , the helicity modulus is non - zero for a range of low @xmath90 : the helicity modulus @xmath164 is proportional to the superfluid density @xmath165 , and close to the quantum critical point , @xmath166 is a universal function of @xmath167 and @xmath90 . nevertheless , we expect the @xmath126 expansion to provide a reasonable description of the impurity atom motion , because the latter couples only to quantities which are invariant under the u(1 ) global symmetry of the superfluid . furthermore , the gap in eq . ( [ mbsuperfluid ] ) is much smaller than @xmath90 , and so is washed out by thermal excitations . in the following we restrict our discussion to the low temperature regime , where the superfluid density is essentially constant and independent of temperature . again , we can define a scaling function @xmath122 for the impurity diffusion constant , which takes the same form as eq . . the scaling function in the superfluid regime is shown in fig . [ fig : phijsf ] , together with the results at the critical point and in the mott regime . note that the temperature dependence of the diffusion constant in the superfluid phase is qualitatively the same as at the critical point , with the only difference that @xmath1 takes numerically slightly smaller values in the superfluid phase than at the critical point . this can be understood from that fact the impurity couples efficiently to gapless phonons in the superfluid . we discussed in sec . [ sec : ft ] that three - body interactions are more relevant than two - body interactions at the particle - hole symmetric quantum critical point . indeed , the behavior of the impurity s transport coefficients will be dominated by three - body interactions at low enough temperatures , albeit the three - body interaction being parametrically smaller than the two - body interaction . in this section we thus repeat the analysis of sec . [ sec:2p ] for the case of three - body interactions @xmath53 . for the sake of simplicity we set the two - body interaction to zero in this section and start our discussion from the lagrangian in eq . with @xmath30 and @xmath168 . in reality both interactions will be present , however , which will lead to a crossover in the temperature dependence of the diffusion constant . the associated crossover scale depends on the relative strength of the two interactions , which will be discussed briefly at the end of this section . again , we start by performing a perturbative analysis of the impurity s self - energy at zero temperature , including terms up to second order in @xmath53 . the two corresponding self - energy diagrams are analogous to the ones shown in fig . note also that mixed terms @xmath169 vanish by symmetry . we neglect the first order tadpole contribution as it only renormalizes the chemical potential , which we set to zero in order to ensure a vanishing density of impurities . the second order contribution to the self - energy can be computed by replacing the current - current correlator @xmath75 in eq . with the bosonic polarization function @xmath170 . again we want to obtain a result for the impurity self - energy which is non - peturbative in the boson coupling @xmath73 , thus we use the scaling form @xmath171 for the bosonic polarization function at zero temperature , which follows from the scaling dimensions of the fields discussed in sec . [ sec : ft ] . the corresponding integral to eq . for the imaginary part of the impurity self - energy can be evaluated exactly at @xmath78 . in the limit of small frequencies @xmath172 we obtain @xmath173 using the exact value for the correlation length exponent@xcite @xmath174 , the imaginary part of the self - energy scales as @xmath175 at low frequencies , i.e. the impurity spectral function again has a sharp delta - function peak . note that this is not true in mean - field theory , where @xmath176 and the the imaginary part of the self - energy scales as @xmath152 at small frequencies , which would result in an algebraic singularity instead of a delta - function peak in the spectral function . at finite momenta we expect a similar behavior of @xmath177 as described in sec . [ sec : pert2 ] for the case of two - body interactions , where the self - energy acquires a finite imaginary part beyond the kinematic threshold at @xmath178 . we calculate this imaginary part explicitly in sec . [ sec : rg ] . for the computation of the impurity s diffusion constant the same arguments hold as in sec . [ sec : boltz2 ] and we restrict our discussion here to the critical point where the mott gap vanishes ( @xmath128 ) . the transition rates @xmath107 have the same form as in eq . , with @xmath179 replaced by @xmath180 . using the scaling form of the bosonic polarization function at finite temperature @xmath181 we can express the impurity diffusion constant in the scaling form @xmath182 unfortunately we can not reliably calculate the scaling function @xmath183 in the equation above , because neither the mean - field approximation with @xmath184 , nor the leading order @xmath126 approximation where @xmath185 are good approximations for the polarization function at finite temperature . in fact , from eq . [ bospolariz ] one can see that @xmath186 is close to being constant at zero temperature , because the correlation length exponent almost takes the value @xmath187 at the critical point . however , it is reasonable to assume that the scaling function @xmath188 in eq . will again take a constant value at @xmath189 , because @xmath190 is an odd function of frequency and expected to scale linearly @xmath191 for small frequencies , similar to the case discussed in sec . [ sec : boltz2 ] . based on this assumption the diffusion constant scales as @xmath192 at low temperatures . if both , two - and three - body interactions are present , there will be a crossover in the temperature dependence of the diffusion constant . summing up the two inverse relaxation times using matthiessen s rule , we obtain @xmath193 with a constant of order one . we thus expect a crossover from a @xmath194 to a @xmath195 behavior at temperatures @xmath196 , provided that the mass - dependent crossovers happen at a higher temperature . the simple tree level - scaling at the end of sec . [ sec : ft ] showed that both @xmath28 and @xmath53 are irrelevant couplings at the particle - hole symmetric point ( @xmath30 ) , with @xmath53 being less irrelevant than @xmath28 . in this section we perform a more elaborate zero - temperature rg analysis of the three - body interaction @xmath53 . our aim is to understand how a finite imaginary part of the self - energy is generated beyond the kinematic threshold discussed in sec . [ sec : pert2 ] , thus we are going to analyze the flow of the impurity propagator at a fixed external momentum . the two - body interaction @xmath28 will be set to zero throughout this section . we are using a functional rg formulation to derive our flow equations.@xcite in order to keep the analysis transparent we use simple parametrizations for the propagators and vertices , however . since the coupling between the impurity and the bosons is a four point coupling and the frg flow - equations have a one loop structure , we employ the standard trick and consider the flow in the symmetry broken phase in order to generate a non - trivial frequency- and momentum - dependence for the impurity propagator while maintaining a simple parametrization of the interaction vertices.@xcite this approach is similar to the background field method used in field - theoretic rg approaches . our starting point is to choose a simple truncation of the effective action @xmath197 $ ] at momentum scale @xmath198 in the symmetry broken phase @xmath199 \notag \\ & & + \frac{g_\lambda \phi_\lambda}{2 } \sum_{q , k } \big ( \psi^*_q \psi_{q - k } \psi_k + \text{c.c . } \big ) \notag \\ & & + \frac{g_\lambda}{4 } \sum_{q , k , p } \psi^*_k \psi^*_{q - k } \psi_{q - p } \psi_p + \sum_q b^*_q \big ( -b_{\l , q}^{-1 } \big ) b_q \notag \\ & & + \sum_{q , k , p } u_\lambda(\omega_q ) \ , \psi^*_{k } b^*_{q - k } b_{q - p } \psi_{p } \notag \\ & & + \phi_\lambda \sum_{q , k } u_\lambda(\omega_q ) \left ( b^*_q b_{q - k } \psi_k + \text{c.c . } \right ) \ . \label{gammaeff}\end{aligned}\ ] ] here @xmath200 is the expectation value of the bosonic field @xmath38 , the subscript @xmath201 denotes scale dependent quantities , we use the shorthand notation @xmath202 for sums over bosonic matsubara frequencies and momenta , and we dropped the subscript on the three - body coupling @xmath203 for notational brevity . note that even though @xmath200 is zero at the critical point , the properly rescaled field expectation value @xmath204 will be non - zero and is related to the anomalous dimension of the bosons . the effective action in eq . is a straightforward generalization of eq . for @xmath30 and @xmath168 to the symmetry broken phase . the important renormalizations appear in the propagators , which we parametrize as ( we set the velocity of bosonic excitations to unity ( @xmath205 ) from now on ) @xmath206 where @xmath207 and @xmath208 are wave - function renormalizations and @xmath209 denotes the sign function . note two important modifications of the impurity propagator @xmath210 . first , we expanded the dispersion relation around a fixed momentum @xmath211 and its corresponding on - shell frequency @xmath212 in order to study the flow of the propagator at a fixed external momentum . the group velocity @xmath213 is allowed to flow , which in turn captures renormalizations of the impurity s effective mass . second , we include a damping term @xmath214 , which will flow to non - zero values above the kinematic threshold . as will become clear later , generating a finite imaginary self - energy @xmath214 is the only possible way to cure the mass - shell singularity that arises from the large phase space for scattering processes between the ( almost ) linearly dispersing impurity mode and the linearly dispersing critical bosonic modes above the kinematic threshold . also note that we did nt include a chemical potential in the impurity propagator . during all our calculations we enforce a vanishing density of impurities by neglecting the contribution of the impurity propagator s pole ( or branch cut , if @xmath215 ) in all diagrams . this procedure is tantamount to adjusting a chemical potential such that the density of impurities is zero . apart from the impurity propagator also the three - body coupling @xmath216 in will acquire an important non - analytic frequency dependence during the flow above the kinematic threshold , which can be parametrized as @xmath217 flow equations for all scale dependent quantities can now be derived in the usual way from the exact flow equation for the effective action@xcite @xmath218 = \frac{1}{2 } \text{tr } \left [ \frac{1}{\gamma^{(2)}_\lambda[\phi ] + r_\lambda } \ , \partial_\lambda r_\lambda \right ] \label{wequ}\ ] ] here @xmath219 $ ] is shorthand for the second functional derivative of the effective action with respect to the fields , i.e. @xmath220)_{kq } = \delta^2 \gamma /(\delta \phi_{k,\alpha } \delta \phi_{q,\beta})$ ] has a @xmath221 matrix structure coming from the field index @xmath222 with @xmath223 , besides its usual frequency / momentum - space matrix structure . the trace in eq . is taken with respect to all momentum , frequency and field indices . @xmath224 denotes the regulator function @xmath225 and we choose an optimized litim cutoff for both , the bosons as well as the impurity@xcite @xmath226 as customary , we neglect derivatives of @xmath227 and @xmath228 in the scale derivative of the regulator @xmath229 . the explicit form of our flow equations for all scale - dependent parameters is shown in appendix [ app : a ] . note that the flow equations for the four boson parameters , @xmath230 , @xmath231 , @xmath232 and @xmath233 are decoupled from the impurity flow equations . this is because we only consider the case of a vanishing impurity density where all diagrams with internal impurity loops evaluate to zero and thus do not give a contribution to the flow equations . for this reason the properties of the bosons at the critical point are described by the usual wilson - fisher fixed point with critical exponents in the universality class of the 3d xy - model . without discussing the standard flow equations for the bosons much further , we now turn to the impurity flow equations , shown in fig . [ figflow ] in a very schematic way . the equations for the impurity propagator and the three - point vertex are given by @xmath234 \ , \label{flowequb } \\ \partial_\l ( u_\l \phi_\l ) & = & - u_\l g_\l \phi_\l \sum_p \big [ 2 \ , \mathcal{g}_p^3 + 3 \ , \mathcal{g}_p^2 \mathcal{f}_p + 6 \ , \mathcal{g}_p \mathcal{f}_p^2+\mathcal{f}_p^3 \big ] \ , \dot{r}^\psi_{\l , p } \notag \\ & & - 2 \ , u_\l^2 \phi_\l \sum_p \big [ ( \mathcal{g}_p-\mathcal{f}_p)^2 \mathcal{b}_{p } \,\dot{r}^\psi_{\l , p } + ( \mathcal{g}_{p}-\mathcal{f}_{p } ) \mathcal{b}_{p}^2 \ , \dot{r}^b_{\l , p } \big ] \notag \\ & & - 3 \ , u_\l^2 g_\l \phi_\l^3 \sum_p \big[2 ( \mathcal{g}_p-\mathcal{f}_p)^3 \mathcal{b}_{p } \ , \dot{r}^\psi_{\l , p } + ( \mathcal{g}_{p}-\mathcal{f}_{p})^2 \mathcal{b}_{p}^2 \ , \dot{r}^b_{\l , p } \big ] \notag \\ & & - 2 \ , u_\l^3 \phi_\l^3 \sum_p \big[(\mathcal{g}_p-\mathcal{f}_p)^2 \mathcal{b}_{p}^2 \ , \dot{r}^\psi_{\l , p } + 2 ( \mathcal{g}_{p}-\mathcal{f}_{p } ) \mathcal{b}_{p}^3 \ , \dot{r}^b_{\l , p } \big ] \ , \label{flowequu}\end{aligned}\ ] ] where @xmath235 is the scale derivative of the regulator and @xmath236 , @xmath237 and @xmath238 denote the regularized normal and anomalous boson green s functions as well as the regularized impurity green s function , respectively . @xmath239 from the explicit form of the flow equations in app . [ app : a ] it is clear that the three - body interaction @xmath240 is always irrelevant . this is mainly due to the screening of the three - body interaction by bosonic particle - hole excitations , giving rise to a contribution to the @xmath241-function of the interaction @xmath240 which is linear in @xmath240 and negative , independent of the impurity s properties . even though @xmath240 always flows to zero , a qualitative change in the rg - flow occurs if the group velocity @xmath242 of the impurity is above the kinematic threshold for creating particle - hole excitations . in perturbation theory this happens for @xmath243 , as discussed in sec . [ sec : pert2 ] . in our rg procedure the threshold velocity is slightly renormalized to @xmath244 in units with @xmath205 , where @xmath245 , @xmath246 and @xmath247 denote the fixed - point values of the rescaled variables ( see appendix [ app : a ] ) . indeed , for @xmath248 the rg generates a finite damping term @xmath214 for the impurity . typical rg - flows of the impurity parameters in the regime @xmath248 are shown in fig . [ fig : paramflow ] . we do nt find a rg fixed - point that controls the flow of @xmath214 above the kinematic threshold . instead , @xmath214 flows to a finite , non - universal value while @xmath240 flows to zero . in fig . [ fig : gamma ] we show the value of @xmath214 at the end of the flow ( i.e. at @xmath249 ) as a function of the initial velocity @xmath250 , showing the sharp onset of @xmath214 for @xmath248 . formally , generating a flow for @xmath214 is intimately linked to the appearance of a non - integrable mass - shell singularity above the kinematic threshold . in this paper we have analyzed the problem of a mobile impurity atom interacting with bosons at the superfluid - mott insulator critical point in @xmath0 spatial dimensions . the effective low energy theory describing this systems is qualitatively different if three - body interactions are present . we calculated the diffusion constant @xmath1 of the impurity as a function of temperature and showed that @xmath1 does not depend on the impurity mass at low temperatures . furthermore we showed that the zero - temperature properties of the impurity change if its group velocity exceeds the kinematic threshold for creating particle - hole excitations . this behavior is not controlled by a rg fixed - point , however . instead the imaginary part of the impurity s self - energy flows to a non - universal value while the interaction flows to zero . we gratefully acknowledge interesting discussions with i. bloch , m. endres , t. fukuhara , c. gross , s. hild and in particular p. strack . this research was supported by the national science foundation under grant dmr-1103860 , and by u.s . army research office award w911nf-12 - 1 - 0227 . mp is supported by the erwin schrdinger fellowship j 3077-n16 of the austrian science fund ( fwf ) . in this appendix we derive the collision integral for the boltzmann equation in sec . [ sec : boltz2 ] as well as the diffusion equation , which describes the propagation of the impurity at long times . our collision integral will be perturbative in @xmath28 , but we want to make sure that effects to all order in the boson coupling @xmath73 are accounted for , similar to the perturbative calculation in sec . [ sec : pert2 ] . this can be done without invoking the keldysh formalism by considering an effective model where the impurity is coupled to a bath of phonons . as will be shown in the following , the spectral density of the phonons needs to be chosen such that it matches the spectral function of the bosonic current - current correlation function . the collision integral for the boltzmann equation can then be derived straightforwardly from the effective phonon model using fermi s golden rule . we start from the effective action describing the impurity coupled to a bath of harmonic oscillators @xmath251 where @xmath252 denotes the frequency of the phonon mode @xmath253 , @xmath254 are coupling constants and we use the shorthand notation @xmath255 . in order to match this effective action to our original action we integrate out the phonons and obtain @xmath256 where we have defined the spectral density of the phonons via @xmath257 on the other hand , if we formally integrate out the bosons in our original action to second order in @xmath28 , we get the same result as in eq . [ seffj ] with @xmath258 now that we have matched the spectral density of the effective phonon bath to our original model we can straightforwardly employ fermi s golden rule and obtain the collision integral in eq . . close to equilibrium we can solve the boltzmann equation by first employing a fourier transform with respect to the spatial coordinate @xmath105 @xmath259 \ , \label{boltzequq}\ ] ] and making the ansatz @xmath260 here @xmath261 and @xmath262 describes the deviation of the momentum distribution from equilibrium . this is basically a leading order expansion in spherical harmonics where the angular dependence of @xmath263 on the momentum @xmath104 only enters through the @xmath264 term . note that the impurity density distribution is given by @xmath265 using the ansatz we can derive a diffusion equation from the boltzmann equation for arbitrary linear collision integrals of the form as follows . taking an angular average of the boltzmann equation we obtain @xmath266 \notag \\ & = & i[f^0_k(\q ) ] \ , \equiv \ , 0 \ . \label{equb1}\end{aligned}\ ] ] moreover , taking an angular average over the boltzmann equation multiplied by @xmath267 we get @xmath268 \ , \label{equb2}\ ] ] where we have used @xmath269 to replace @xmath270 with @xmath271 . the rhs of the above equation can be evaluated straightforwardly for a collision integral of the form @xmath272 = \label{equb3 } \\ & & \ \ \ \ - \frac{q^2 k^2}{2 } \sum_{\k ' } \big [ w_{\k,\k ' } \tilde{f}_k - w_{\k',\k } \frac{k'}{k } \cos \theta_{\k,\k ' } \ , \tilde{f}_{k ' } \big ] \ . \notag\end{aligned}\ ] ] again , using eq . to replace @xmath270 with @xmath271 in eq . , summing over @xmath104 and neglecting the @xmath273 term , we finally arrive at the diffusion equation @xmath274 with the diffusion constant specified in eqs . and . the explicit expression of the scaling function for the diffusion constant in eq . is @xmath275 e^{-z } \ , \label{scalingfunc}\end{aligned}\ ] ] where @xmath276 is the scaling function of the current correlator @xmath277 and @xmath278 . all of the results derived above also apply to the case of three - body interactions if the current correlator @xmath75 is replaced by the density correlator @xmath170 as well as @xmath28 by @xmath53 . it has been shown in sec . [ sec : rg ] that the flow equations for the boson parameters decouple from the impurity flow equations . indeed , using the rescaled variables @xmath279 and defining the anomalous dimension of the bosons via @xmath280 the flow equations for the boson parameters are given by ( @xmath281 ) @xmath282 in order to bring the flow equations into an analytically tractable form we expanded them in @xmath283 . this basically amounts to an expansion in a small anomalous dimension , which works well in @xmath0 spatial dimensions . at the wilson - fisher fixed point the rescaled variables take the values @xmath284 note that the value of the anomalous dimension at the critical point @xmath285 agrees pretty well with the expected value for a 3d xy - transition , where @xmath286.@xcite for the impurity flow equations we define the rescaled variables and the anomalous dimension of the impurity as @xmath287 and the flow equations take the form @xmath288 \ , \\ \partial_s \tilde{u}_s ' & = & - \eta _ * \tilde{u}_s ' - \frac{\tilde{u}_s ' \tilde{g}_*}{16 \pi } \bigg [ \frac{3}{(1+\tilde{r}_*)^{5/2}}-\frac{115 \ , \tilde{g } _ * \tilde{\phi}_*^{2}}{4 ( 1+\tilde{r}_*)^{7/2}}\bigg]-\text{re } \bigg [ \frac{(\tilde{u}_s'+i \tilde{u}_s'')^2}{2 \pi } \bigg(\frac{\tilde{v}_s^2+i \tilde{\gamma}_s ( \alpha - i \tilde{\gamma}_s)}{\tilde{v}_s^2 \alpha^3 \big ( ( \alpha - i \tilde{\gamma}_s)^2-\tilde{v}_s^2\big)^{1/2 } } -\frac{i \tilde{\gamma}_s}{\tilde{v}_s^2 \alpha^3 } \bigg ) \bigg ] \notag \\ & & + \frac{3 \tilde{g } _ * \tilde{\phi}_*^{2}}{2 \pi } \text{re}\bigg [ ( \tilde{u}_s'+i \tilde{u}_s'')^2 \bigg(\frac{3 i \tilde{\gamma}_s ( \alpha - i \tilde{\gamma}_s)^3 - 3 \tilde{v}_s+\tilde{v}_s^2 ( 4 \alpha^2 - 9 i \tilde{\gamma}_s \alpha-6 \tilde{\gamma}_s^2)}{4 \tilde{v}_s^2 \alpha^5 \big ( ( \alpha - i \tilde{\gamma}_s)^2-\tilde{v}_s^2\big)^{3/2 } } - \frac{3 i \tilde{\gamma}_s}{4 \tilde{v}_s^2 \alpha^5 } \bigg)\bigg ] \ , \end{aligned}\ ] ] @xmath289-\text{im } \bigg [ \frac{(\tilde{u}_s'+i \tilde{u}_s'')^2}{2 \pi } \bigg(\frac{\tilde{v}_s^2+i \tilde{\gamma}_s ( \alpha - i \tilde{\gamma}_s)}{\tilde{v}_s^2 \alpha^3 \big ( ( \alpha - i \tilde{\gamma}_s)^2-\tilde{v}_s^2\big)^{1/2 } } -\frac{i \tilde{\gamma}_s}{\tilde{v}_s^2 \alpha^3 } \bigg ) \bigg ] \notag \\ & & + \frac{3 \tilde{g } _ * \tilde{\phi}_*^{2}}{2 \pi } \text{im}\bigg [ ( \tilde{u}_s'+i \tilde{u}_s'')^2 \bigg(\frac{3 i \tilde{\gamma}_s ( \alpha - i \tilde{\gamma}_s)^3 - 3 \tilde{v}_s+\tilde{v}_s^2 ( 4 \alpha^2 - 9 i \tilde{\gamma}_s \alpha-6 \tilde{\gamma}_s^2)}{4 \tilde{v}_s^2 \alpha^5 \big ( ( \alpha - i \tilde{\gamma}_s)^2-\tilde{v}_s^2\big)^{3/2 } } - \frac{3 i \tilde{\gamma}_s}{4 \tilde{v}_s^2 \alpha^5 } \bigg)\bigg ] \ , \\ \partial_s \tilde{v}_s & = & -\eta_b \tilde{v}_s+\frac { \tilde{\phi}_*^{2}}{2 \pi } \text{re } \bigg[(\tilde{u}_s'+\tilde{u}_s'')^2 \bigg ( \frac{\tilde{v}_s ( \alpha - i \tilde{\gamma}_s)^3+i \tilde{\gamma}_s \tilde{v}_s^3}{\tilde{v}_s^2 \alpha^3 \big ( ( \alpha - i \tilde{\gamma}_s)^2-\tilde{v}_s^2\big)^{3/2}}-\frac{1}{v \alpha^3 } \bigg)\bigg ] \ , \\ \partial_s \tilde{\gamma}_s & = & ( 1-\eta_b ) \tilde{\gamma}_s - \frac{\tilde{u}_s''}{8 \pi } \ , \frac{2 ( 1+\tilde{r}_*)-3 \ , \tilde{g } _ * \tilde{\phi}_*^{2}}{2 ( 1+\tilde{r}_*)^{5/2}}+ \frac{\tilde{\phi}_*^{2}}{4 \pi } \text{im } \bigg [ ( \tilde{u}_s'+i \tilde{u}_s'')^2\bigg ( \frac{\tilde{v}_s^2+i \tilde{\gamma}_s ( \alpha - i \tilde{\gamma}_s)}{\tilde{v}_s^2 \alpha^3 \big ( ( \alpha - i \tilde{\gamma}_s)^2-\tilde{v}_s^2\big)^{1/2 } } -\frac{i \tilde{\gamma}_s}{\tilde{v}_s^2 \alpha^3 } \bigg ) \bigg ] \ , \end{aligned}\ ] ] where we have defined @xmath290 and fixed all boson parameters at their respective critical values . again we have expanded boson loops in @xmath283 and moreover we have taken @xmath291 for the sake of simplicity , which amounts to neglecting the small curvature of the impurity dispersion in . this does not change the results qualitatively . we also mention that the term @xmath292 cancelled exactly with the @xmath293 contribution in the flow equation for the three - body interaction @xmath216 . the particular structure with real- and imaginary - parts in the impurity flow equations originates from the non - analytic term @xmath294 in the boson propagator . after setting the external boson frequencies to zero , all diagrams with internal impurity lines depend on the external impurity frequency @xmath295 via the damping term @xmath296 . we can extract the contributions of analytic and non - analytic terms by taking limits @xmath297 from above and below and adding or subtracting the resulting expressions , which gives rise to the particular form of the impurity flow equations above . c. p. herzog , a. karch , p. kovtun , c. kozcaz and l. g. yaffe , jhep * 0607 * , 013 ( 2006 ) . j. casalderrey - solana and d. teaney , phys . d * 74 * , 085012 ( 2006 ) . s. s. gubser , phys . d * 74 * , 126005 ( 2006 ) . j. de boer , v. e. hubeny , m. rangamani and m. shigemori , jhep * 0907 * , 094 ( 2009 ) . d. t. son and d. teaney , jhep * 0907 * , 021 ( 2009 ) . s. caron - huot , p. m. chesler and d. teaney , phys . d * 84 * , 026012 ( 2011 ) . d. tong and k. wong , arxiv:1210.1580 [ hep - th ] .

the global financial crisis that started in 2007 has stimulated an extensive literature on numerous credit - related themes such as risk assessment , financial contagion , regulatory indebtedness or liquidity ratios , misuse of derivatives ( see for example @xcite ) . as the crisis unfolded , the understanding of the credit exposure at an individual bank level , as well as at an aggregate systemic standpoint became increasingly important . a given bank was perceived safer or riskier based not only on its balance sheet figures or growth estimates , but also on the financial strength of the insurance companies who would step in in the case of a debtor s default , the financial distress of the peers with whom the bank had more interbank exposure , liquidity dry - ups which could trigger massive asset fire - sales , or even the assigned probability to a lender of last resort type of bailout from the central government @xcite . insurance companies played a key function in the network : they had the commitment to enter the scene if a given asset failed to make promptly payments . however once it became clear that insurance companies would be not be able to compensate for every risky asset they had insured , it also became apparent that a financial distress event could propagate across the entire financial system @xcite . banks would have to start recognizing massive asset losses , and if a bank defaulted , other interconnected institutions could follow , potentially triggering a bankruptcy cascade @xcite . people fearing for its savings could line up and withdraw their deposits causing a so - called bank run , as was for example observed in the united states , united kingdom , iceland , spain , or brazil @xcite . this context is pertinent to the present work , as nowadays much of the discussion continues to be how to define and regulate higher capital requirements , while at the same time weighting the trade - offs involved , namely higher costs , or lending shifting to the shadow banking ( @xcite ) . this paper contributes to this academic discussion by using a network - based framework to examine the structure and inner dynamics behind the links between financial institutions in different countries . the application of network theory to the financial markets @xcite gave place to a new set of methodologies to study interaction in interconnected agents , institutions or financial products ( e.g. portfolios , stock indices , derivatives ) . however its literature is fairly recent , and to our knowledge none has studied the banking system structure using actual ( non - simulated ) data from banks financial statements across several countries . to the extent that actual interbank data ( e.g. loans , repos , swaps ) is not publicly available , we argue that relevant implications can still be derived upon the common behavior of certain key balance sheet items . as we have learned from the crisis , two banks can be exposed to each other s risk , even if there is no financial transaction between them : if a bank fails , market conditions ( liquidity , stock prices , mergers or acquisitions expectations ) are all likely to experience cascade consequences . in general , financial institutions , be them commercial banks , investment banks , development banks , broker - dealers , or credit unions , fall under the supervision of each country s central bank . these institutions , henceforth referred to as banks , must usually submit their quarterly or monthly financial statements to the corresponding central bank regulator . this article takes advantage of that rich , homogeneous and regular available information , and uses balance sheet data from 2005 to 2010 from five emerging countries : argentina , brazil , mexico , south africa , and taiwan . we then construct banking functional networks from the leverage dependence across financial institutions , and analyze the structure of this networks . we find configurations that tend to group into large clusters at relatively low correlation levels . a modular structure characterized by one large bank community , some small ones and isolated banks is also found in all countries . in addition , we propose and simulate a model of corporate and interbank loans that generates rich and diverse balance sheet growth . and when we construct simulated banking functional networks according to their leverage dependence , we find results that are consistent with the empirical networks . despite the absence of actual interbank contracts , these findings translate into relevant policy implications in terms of contagion and concentration , as well as suggest potential avenues for future research . the paper is divided as follows . section ii describes the data and the methodology , section iii describes the main results of the paper , section iv simulates a model of interbank loans and compares with empirical data , and section v concludes . we study the interaction between financial institutions using balance sheet data from 2005 to 2010 from five developing countries : argentina , brazil , mexico , south africa , and taiwan . we retrieve total assets and total liabilities for each financial institution sampled at a monthly or quarterly basis , whichever is more frequent in its reporting . let @xmath0 and @xmath1 be the total assets and liabilities , respectively , for bank @xmath2 at time @xmath3 . for each country , the number of banks @xmath4 , fluctuates in time , and thus @xmath5 is a stochastic birth and death process . throughout the observed period 2005 - 2010 the existing banks at 2005 may fall bankrupt , merge with others , or simply survive until 2010 . on the other side , new banks can appear in between , say in 2006 , thus modifying @xmath5 as well . as the purpose of this work is to study the interaction between banks , we shall remove from the dataset those banks with incomplete observations . for example , if a bank stops or starts reporting data in 2007 , its observations are removed from that country . let @xmath6 be the banks that satisfy the previous condition , i.e. the effective number of banks studied . the following table shows both the number of banks in 2005 and 2011 ( @xmath7 and @xmath8 ) , as well as the number of banks present throughout the time investigated ( @xmath6 ) . table 1 also includes the number of banks births ( deaths ) , @xmath9 ( @xmath10 ) , during the period 2005 - 2011 . . financial system size of the five countries studied in the period 2005 - 2011 . @xmath11one bank from brazil and one from south africa were excluded because of missing data in some time periods . [ cols="^,^,^,^,^,^ " , ] for instance in brazil 49 banks were created and 28 disappeared between 2005 and 2011 . only two of these new banks have gone out of the banking system before 2011 . except for brazil , the rest of the countries satisfy the following equation : @xmath12 we divide the countries into two groups according to their size @xmath6 , one composed by argentina and brazil , and the other composed by mexico , south africa and taiwan . the first group represents a large size banking system with @xmath13 and the second a smaller banking system with @xmath14 . also notice the different dynamics between brazil ( also mexico ) and taiwan : whereas in the former country the number of banks increased over time , in the latter decreased . by contrast , argentina and south africa exhibit very small fluctuations in @xmath5 for the period studied . the construction of a network of associations out of balance sheet variables requires selecting an interaction metric that meets two general properties . first , it should be robust to spurious correlations so that we avoid as much as possible to use a non - stationary variable that may increase or decrease simply following a global trend . and second , it should be relevant to describe an economic process . in other words , if two banks are associated with a given variable , it should mean that they are subject to certain equivalent market or economic conditions , and therefore may share the same risk . we argue that balance sheet aggregates such as assets are inappropriate to define a network , and instead propose to use the leverage ratio . consider for instance total assets , which are regarded as a proxy of bank size , and are subject to a large number of global or macroeconomic circumstances . under normal market conditions , the assets of a financial institution are expected to grow with time , as a result of general growth of the economy , inflation , balance sheet effects from expansionary monetary policy , population ( customer base ) growth , increased access to banking services ( banking penetration ) , higher value of the securities held by the bank ( stocks or debt ) . it seems that almost any pair of banks , regardless whether they are contractually linked or not , will yield a positive assets correlation , and thus will incorrectly suggest an overlinked network . we instead propose using the degree of indebtedness , henceforth referred to as the leverage ratio , as the metric on which to base the interactions . we define the leverage as follows @xmath15 where liabilities and equity refer to the balance sheet variables of a given bank which must satisfy the usual accounting identity : @xmath16 , and as measured at book value . therefore , the leverage ratio of bank @xmath2 , @xmath17 , indicates what fraction of equity and debt this bank is using to finance its assets . consider for instance panel ( a ) and panel ( b ) of fig . 1 , which shows an example where the leverage increases and decreases , respectively . the former may happen , for instance , through a credit quality shock , such that the value of the existing claims decreases and the bank increases its loan loss provision . such loan write - offs translate into a reduction in total equity , thus increasing the ratio . the leverage can also rise if , for example , the bank simply issues debt to finance a potential acquisition or expand its credit portfolio . what regards to panel ( b ) , a bank may deleverage by prepaying debt with its own cash , where both total assets and liabilities decrease . it is also possible for the shareholders to conduct a capital increase , with proceeds either paying down debt or cushioning its assets . these examples illustrate only a few of the many different scenarios why a leverage ratio may fluctuate within a given financial institution . finally , to address the stationary concern fig . 2 depicts the median ( results remain unchanged with the average ) leverage ratio for each country as a function of time . contrary to what happens with liabilities and assets ( fig . a.1 ) , the leverage ratio does not exhibit any strong time dependence . each country exhibits its own leverage range , with more or less fluctuations over time . taiwan , for example , is characterized by exhibiting a large median leverage ratio ( @xmath18 ) while argentina and brazil exhibit a lower ratio ( @xmath19 ) . leverage levels across countries are expected to differ due to different bank capital regulation . each country may impose its own restrictions on hedging operations , ability to buy or sell foreign currency , place debt in the national or international markets ( and at which currency ) , banks may be subject to a particular government program ( e.g. subsidized mortgage or consumer credit programs with predetermined interest rates ) , restrictions on dividend distribution . the central bank may also have in place a monetary and/or inflation target rate , which in turn modifies the reference rate at which banks lend each other and hence the rates at which households or corporates access to credit . countries also differ in their accounting consolidation practices or taxation . overall these factors limit the construction of leverage - based networks at the country level , as opposed to aggregated inter - country networks . in particular , we construct country - level networks of leverage dependence as follows . 1 . compute the leverage time series for all banks from 2005 to 2010 in a given country . after accounting for mergers and acquisitions , banks with incomplete data points are excluded ( either because they appeared later or disappeared sooner than the 2005 - 2010 range ) . 2 . compute the leverage correlation matrix between all pairs of banks . 3 . establish links between banks ( nodes ) whenever the correlation exceeds a given threshold @xmath20 . the structure of the network is studied for different values of @xmath20 . finally , notice that the leverage ratio also presents the important advantage of not depending on the currency , thus allowing us to use the same variable across countries . furthermore , any difference in regulation or local practice of an improved indebtedness ratio ( e.g. capital ratio after tier 1 / 2 adjustments ) is neutral within a country ( for it applies to all ) and not relevant across countries , since this paper does not compare absolute leverage levels , but rather the interaction that can be derived from common leverage behaviors . we next illustrate an example using data from the argentine banking system . given two time series , corresponding to the leverage evolution of two banks , @xmath21 and @xmath22 with @xmath23 , we compute the pearson correlation coefficient . where this coefficient takes value @xmath24 [ -1 , 1 ] , and is equal to 1 when the two time series are identical ( @xmath25 for all @xmath3 ) . the correlation matrix is composed by the @xmath26 pairs of banks correlation coefficients ( interactions ) . figure 3 ( a ) shows the leverage ratio evolution for six ( from a total of 75 ) different argentine banks . three colors are used to illustrate the different behaviors observed throughout the sample . for example the middle ( bottom ) two leverage curves in blue ( green ) correspond to two banks that exhibit an increasing ( decreasing ) leverage evolution . another type of leverage evolution is shown in red , with a concave - shaped leverage behavior . the dynamics of the argentine banking system is found to be very rich from an indebtedness perspective . as we shall see later , these rich dynamics lead to a complex banking network . it also follows that banks c and d have a similar leverage increasing pattern . the same is found between banks e and f with practically a scale factor of difference between them , and to banks a and b , although with a different type of evolution . each of these pairs of banks exhibit a large correlation coefficient between its corresponding leverage time series , e.g. @xmath27 , and @xmath28 for the pair of banks a - b , c - d , and e - f respectively . however , if we compute the leverage correlation between bank c ( or d ) and e ( or f ) this value will be negative , thus stating that the two banks exhibit an opposite behavior . in this last case the four correlation coefficients range between -0.35 ( for c - e ) and -0.65 ( d - e ) . the remaining possible interactions between the six banks fluctuate around zero . the complete characterization of interdependence between the 75 argentine banks is captured by the correlation matrix in panel b , with each column and row representing a bank . a special algorithm is used to order the banks for a better visualization , such that light yellow `` spots '' correspond to group of banks that share a similar behavior among them ( large @xmath29 coefficient ) , while negative correlations are represented in dark red color . although the leverage dependence does not imply an interbank transaction ( e.g. as would be a repo agreement ) , a strong and recurring synchronization is still relevant in the understanding of concentration and systemic risk . it is reasonable to think that a highly leveraged network where many banks are simultaneously leveraged does not imply the same contagion risk as does a network where only a small set of banks are moderately leveraged . we can also think of the degree of negative and positive correlations as another determinant of how robust the banking system is . in this sense , identifying and studying which banks form clusters that tend to leverage ( deleverage ) all together can also provide useful policy implications at a time where stricter banking regulation is at the core of the debate ( @xcite ) . interestingly , from a risk - sharing point of view , it would be optimal for the financial institutions not to share identical leverage dynamics , such that a shock affecting the indebtedness of a given bank is not be followed by its peers . for a banking system to remain healthy and resilient to financial contagion , its dynamics needs to be heterogeneous . that is , greater heterogeneity helps to isolate individual shocks by setting a less interconnected financial system . further , this can be understood as the system s endogenous solution to reduce its risk from financial shocks . notice that the economic implications behind the positive and negative dependence scenarios are quite different . a positive correlation between two banks implies that they both increase and decrease their debt - to - equity ratio simultaneously . in this case , we would expect both banks to share ( or target ) similar growing . alternatively , banks balance sheets could be affected by exogenous policies along the theories of balance sheet effects ( @xcite ) , bank lending channels ( @xcite ) , and bank runs ( @xcite ) . on the other hand , a negative correlation implies that when one bank increases its leverage , the opposite is true for the other bank , and vice versa . a potential explanation for this case is related to the literature of fly - to - safety , i.e. in periods of financial turmoil households withdraw their deposits from a small bank and transfer them to a bigger ( perceived as safer ) institution . thus liabilities decrease in the small bank , but increase in the other bank , resulting in a negative leverage correlation between them . as documented in @xcite , this effect was observed in brazil in late 2008 , where depositors ran from smaller banks to larger banks , the latter ones perceived as too - big - to - fail institutions with implicit guarantees from the central bank . an alternative mechanism comes from what the literature calls the depositors discipline : depositors may punish banks with poor performance either by demanding higher interest rates or by withdrawing their deposits . @xcite find evidence in argentina , chile , and mexico of the depositors discipline mechanism , in particular that bank deposits growth falls as risk exposure increases . also @xcite and @xcite document that depositors favor big banks , although it is hard to differentiate whether depositors base their decisions on too - big - to - fail sentiments or bank fundamentals . also , @xcite report that during the 2001 argentine convertibility crisis deposit withdrawals were more pronounced in banks with higher risk taking , thus providing further evidence of opposite changes in balance sheet accounts across banks . . , scaledwidth=100.0% ] we now characterize the complete argentine network . the upper panel of fig . 4 depicts the graphical representation of the argentine financial network with @xmath30 links , or with an average degree @xmath31 equal to @xmath32 . recall that each node represents a bank , and the @xmath30 highest banks pair correlations are connected by an undirected edge . put alternatively , an edge ( or link ) between two nodes ( or vertices ) is added whenever its leverage pair correlation exceeds a certain threshold value @xmath20 . the same banks showed in fig . 3 are highlighted in the network representation of fig . 4 shows clusters of banks , where a cluster is defined as a community of interconnected ( linked ) banks . and each color represents a different leverage ( temporal ) dependence . in particular , the network can be characterized by four homogeneous groups : a large cluster of banks that essentially have an increasing leverage evolution ( e.g. banks c and d ) ; three small groups with various leverage behaviors ( decreasing , concave - shaped , and a mixed of both ) ; and finally a large set of isolated nodes ( 41 from 75 ) , represented with white circles . interestingly , we also find that the modular structure observed in the argentine banking system is similar to other countries . despite having different size , regulation , or even average indebtedness level , the leverage dynamics are consistent across countries . bottom panel of fig . 4 shows the brazilian banking network using the same methodology as well as the same @xmath33 used in argentina . it can be observed that brazil s network is similar to argentina s , namely both are characterized by a large community of banks , another set of small clusters of banks , and many isolated banks . as the correlation threshold decreases ( and the number of links @xmath30 increases ) , both the isolated and the small - cluster nodes start merging into one large interconnected cluster . this merger - type of behavior between different groups can be observed in fig . 5 , where the fraction of nodes that belongs to the largest cluster ( blue nodes as per fig . 4 ) is shown as a function of the correlation threshold . as we lower the threshold @xmath20 , the largest cluster starts absorbing smaller communities . big jumps in this graph indicate that two large clusters have merged . for example , in brazil s case a small change in the correlation threshold around 0.6 causes the two largest clusters ( each one representing approximately 40% of the nodes ) to be either merged or divided . in argentina this similar and sudden concentration occurs at a greater correlation threshold ( @xmath34 ) . we also find that in both countries 90% of the banks belong to a unique cluster at a relatively small correlation threshold @xmath35 , thus suggesting that the financial networks may become suddenly too concentrated . finally , we replicate the analysis to the smaller banking systems of mexico , south africa and taiwan . ( a ) shows the leverage - based networks of these countries for an equal number of links per node , i.e. for @xmath36 . similar network properties are observed across these countries . in particular , although taiwan s banking system is more volatile ( large fluctuations in @xmath5 ) , the banking network structure is very similar to the south african one , and both have almost the same total number of banks ( @xmath37 ) . this result is strengthen by the similar relationship between the size of the largest cluster and the correlation threshold shown in panel b. the mexican network , on the contrary , seems to exhibit certain differences with respect to the other two . for @xmath38 the largest cluster does not have significant weight ( right bottom panel ) , as it only accounts for 30% of the total nodes in the network . as the correlation threshold decreases , a second cluster grows until is finally merged with the largest one ( left bottom panel ) . thus we observe that the consolidation of the mexican banking network is characterized by the merger of two large clusters , whereas in the other countries there is only one large cluster which then absorbs either individual nodes or very small clusters . in summary , our proposed method to examine the banking system from a leverage perspective yields networks composed by a large group of banks that are densely interconnected ( blue cluster in fig . 4 and 6 ) , a few smaller clusters , and many isolated banks . interestingly , this topology is homogeneous across countries , as depicted in fig . 4 and 6 . as the interaction criterion becomes more and more lax ( smaller @xmath20 , in absolute value ) , banks start re - grouping into a unique financial cluster . furthermore , the network structure is not dominated by one hub , as it occurs with many nature networks ( @xcite ) . on the contrary , these banking networks exhibit a modular structure where each module is densely connected . only a few banks function as _ bridges _ , or connectors between different modules . the existing structure also suggests a somewhat fragile property : a small modification to these connector banks can have sensible effects to the configuration of clusters , as well as to the isolation or propagation of a shock across the network . on the one hand , highly interconnected banks ( hubs ) are not desirable , since a failure in a hub can be spread out the network , potentially causing an aggregate system breakdown . but on the other hand , some degree of interconnectedness is positive , as it helps to reduce the exposure of an affected node by sharing and loading off risk with its peers . therefore we argue that the existence of dense network structures may be desirable to some extent in order to increase risk sharing and reduce hub exposure . in this sense , in order to reduce risk and increase shock resilience , regulators should recognize a basic trade - off : as connectivity and clustering increases , individual ( idiosyncratic ) risk may decrease due to better risk sharing , but global or systemic risk could increase thanks to faster channels of propagation . the degree of optimal bank connectivity , and how can it be reached , are still active questions in the literature . in this section we intend to contrast the empirical results from section [ sec : results ] with simulations from a model of corporate and interbank loans . we first start by describing the set up of the model . we propose that the corporate sector requests loans from the banking system according to a given random process , and links between banks arise due to insufficient liquidity to meet these loans . this lending effect generates rich balance sheet and leverage dynamics that yield networks that are overall consistent with the empirical results . our model builds on the interbank model proposed in @xcite . for simplicity we assume perfect information in the economy , although as explained by @xcite imperfect information can result in an additional source of financial risk . further , although we think of agents in the real sector as the implicit source of interbank links , the model does not generate balance sheet dynamics from the agents side of the economy . see @xcite for an example of an agent - based interbank network model . see also @xcite for a review of ace ( agent - based computational economics ) modeling in economics . the model is structured as follows . 1 . the banking system is composed by @xmath4 banks . assets are defined as the sum of liquidity @xmath39 , iliquid assets @xmath40 , corporate loans @xmath41 , and interbank loans @xmath42 . liabilities are defined as the sum of deposits @xmath43 and interbank debt @xmath44 . moreover , any bank @xmath45 at time @xmath3 must satisfy : @xmath46 @xmath47 2 . let assets @xmath48 at time @xmath49 be drawn from a uniform distribution , e.g. from 5,000 to 50,000 . then for each bank set a conservative equity - to - assets ratio selected at random from the interval 0.10 to 0.35 . and finally , let liquidity @xmath50 at @xmath51 be a fixed share of assets , thus featuring that a given bank can be solvent to provide loans or take additional debt , but still illiquid to do so . these steps so far are sufficient to characterize the banking system at the initial period . 3 . next assume the corporate sector requests loans from the banking system according to a poisson process of rate @xmath52 , i.e. the length of time between loan requests is exponentially distributed . let @xmath53 be the fixed amount of the loan , and @xmath54 be the interest rate of these loans . each loan is supplied by only one bank @xmath45 , where @xmath45 is chosen at random with equal probability among the @xmath4 banks . however , if bank @xmath45 is selected and @xmath55 , it lacks enough liquid assets to fulfill the entire loan . to cover the difference , bank @xmath45 enters into the interbank market and borrows at rate _ r@xmath56 _ ( with @xmath57 ) from bank @xmath58 , where @xmath58 is also selected randomly across the @xmath59 remaining banks . 4 . similarly , if @xmath60 then bank @xmath58 can not supply the interbank loan . if this occurs then a new bank is chosen from the remaining @xmath61 banks . as soon as @xmath53 is supplied to the corporate sector , liquidity @xmath62 as well as household deposits @xmath63 increase in a subset of banks @xmath64 by the amount @xmath65 , where @xmath66 are random weights such that @xmath67 , and @xmath68 . the intuition is that the corporate sector distributes the loan in the economy , for example in the form of wages , and households in turn deposit these proceeds randomly across a small number of banks . notice that this generates balance sheet ( assets ) growth , which will further enrich the leverage dynamics . although the @xmath30 banks are drawn randomly each time a loan takes place , @xmath45 s weights are fixed but re - scaled , i.e. @xmath69 @xmath70 . the rationale is that some banks tend to have a higher ( or lower ) share of deposits that is stable over time . in addition , we propose that in each period there is a ( low ) probability of a negative shock to a random bank @xmath45 . when the shock event is triggered , deposits and liquidity decrease in fixed amount , e.g. @xmath71 . 7 . finally , the corporate and interbank loans are paid back after @xmath72 periods . therefore at time @xmath73 bank equity @xmath74 will increase by the net amount of interest earned between the loan granted to the corporate sector and , if any , the funds borrowed from the interbank market . contrary to the loan being initially distributed across the banking system , the funds for repayment come exogenously from the real sector . in other words , we implicitly assume the corporate sector is always productive and therefore is able to generate proceeds elsewhere . notice that the complete set of bank interactions resulting from the model can be described in the form of a directed network . in this graph nodes are given by the @xmath4 banks , and the directed links between nodes are given by the direction of the interbank loan . therefore each point in time can be fully characterized by an adjacency matrix @xmath75 , where @xmath75 is a @xmath76 matrix and in general will not be symmetric . an entry @xmath77 ( i.e. edge from @xmath45 to @xmath58 ) implies bank @xmath58 has borrowed from bank @xmath45 . however , we will not use this information for analyzing the data , we will only apply the methodology described before for the simulated leverage time series to study the generated functional networks . we now simulate the model setting @xmath78 to be comparable to the actual size of the argentine and brazil network . this allows , without loss of generality , to more easily contrast the results with those in section 3 . in fact , results remain qualitatively similar if we modify parameters such as @xmath4 , @xmath53 , @xmath79 , @xmath80 , etc . for instance , consider panel ( a ) and ( b ) of fig . [ fig : simulacion1 ] which show the simulated dynamics of average assets and leverage over time , respectively . after some transient period ( not shown ) the average leverage become stationary ( panel ( b ) ) . moreover , the stationary value is 6 which is very similar to the value observed in argentina and brazil ( see fig . 2 ) . the average assets grows over time ( panel ( a ) ) , obtaining a growth of 4.5 for the time period studied . this last value is compatible to the brazil assets growth ( fig . now , that the average assets and leverage are similar to the empirical one , we study the individual bank behavior applying the network methodology described before for the individual simulated leverage time series . panel ( c ) in fig . 7 shows the graphical representation of the simulated leverage network for a correlation threshold @xmath81 . the topology is very similar to the network in fig . in particular , the network is characterized by a large interconnected cluster , together with smaller communities of banks , and many isolated banks . furthermore , not only banks tend to merge with the largest cluster as @xmath20 decreases , but also it presents the discontinuity property that leads to a unique large cluster ( fig . 7 ( d ) ) the proportion of nodes that belongs to the largest cluster jumps from 0.4 to over 0.8 at around @xmath35 . the result shown in panel ( d ) is very similar to that obtained in fig . 5 . and 5,000 periods . _ ( a ) banks assets growth . ( b ) average leverage over time . ( c ) representation of simulated network . ( d ) proportion of banks that belongs to the largest cluster as a function of the correlation threshold.,scaledwidth=105.0% ] therefore the interbank model generates modular structures that resemble the empirical networks . but can we say more about the mechanisms that drive the links between banks ? in particular , we want to know whether the model can shed light on the dynamics behind the leverage dependence . in order to answer this we propose to study two banks , and in particular , the two most correlated banks . to avoid drawing conclusions from one simulation , we instead simulate the model 40 times , and for each of them we identify the two banks with the largest leverage dependence . then compute the total leverage and assets growth , that is , the growth between the last period @xmath82 and @xmath51 . we compute this ratio for both banks in each of the 40 replications . the histogram of these ratios is depicted in panels ( a ) and ( b ) of fig . interestingly , we find that the two banks exhibit large growth in both variables . in the model this takes place when banks receive greater frequency of loan requests . and at the same time , since these banks tend to have sufficient liquidity , they are also more likely to lend to other banks . the model therefore suggests that banks with more aggressive growth strategies may be more interconnected , and hence share additional risks between them . we then translate this measure to the analog of the empirical samples . in panels ( c ) and ( d ) of fig . 8 we compute the leverage and the total assets growth for each bank in argentina and brazil , respectively . where banks are ordered for better visualization . recall we are comparing the simulations against argentina and brazil because they share a similar size @xmath78 . red and green dots identify the two most correlated banks in each country . notice that both banks indeed exhibit a higher leverage growth . a higher assets growth is also found in argentina , and to a less extent in brazil s network . after the recent and still ongoing financial crisis an important bulk of research has turned its eyes into new ways to model financial risk and contagion between financial institutions , and how to improve regulation so as to reduce risk at both the individual and aggregate level . at the time of writing , banking supervision codes are being modified in order to incorporate tighter capital and liquidity requirements such that financial institutions are much better prepared to absorb losses in future credit - related events . this is particularly important since a crisis starting at a given place can very easily amplify or spread across the system through various channels : counterparty risk due to interbank lending , generalized fall in asset prices ( fire sales to boost capital / margin requirements ) , liquidity hoarding or funding constraints ( higher margins , run on repo / money market ) , self - fulfilling prophecies ( bank runs ) , non - government bailout , or underestimation of risk in , and exposure to , derivate securities . this paper studies the banking system within five different countries , and then compares the results across them . we construct networks using bank - level balance sheet data , where links between each pair of banks are assigned upon the leverage ratio dependence . although this interaction does not imply an interbank financial transaction , our findings do illustrate that recurring leverage dynamics ( e.g. cluster of banks that simultaneously leverage or deleverage ) is relevant in the understanding of risk concentration . interestingly , we find that in all countries ( argentina , brazil , mexico , south africa , and taiwan ) the network topology is characterized by a large cluster of interconnected banks , together with other small communities and isolated banks . as the correlation threshold gets lower , the proportion of banks that belongs to the largest cluster increases almost dramatically : in both argentina and brazil , for instance , 90% of the banks group into a unique cluster at a 0.5 threshold . we also find that these networks are not dominated by a hub ( extremely interconnected node ) , but by a modular structure where communities are connected by bridge banks . this novel finding can be thought of as an optimal strategy to diversify risk within a group of banks that share similar characteristics . in other words , this subset restricts the spread of a distress cascade when one bank fails , as well as limits ex - post government assistance to a reduced set of banks . to test the robustness of our results we present a model of corporate and interbank loans . the model generates rich balance sheet and lending dynamics , which certainly do not intend to capture all the intricacies between the financial system and the rest of the economy . our objective is much smaller , and is to generate lending dynamics that can shed light on the complexity of interbank relationships . interestingly , in doing so we obtain leverage networks that closely follow the empirical ones . our simulations also suggest that banks with recent rapid or aggressive growth strategies are more likely to be interconnected , and hence share more risks . understanding risk and leverage using a network framework can also be of interest to policymakers and regulators , who for instance can build upon these tools to locate heavily interconnected banks and evaluate its importance to the aggregate system . when credit events occur , the indebtedness ratio is often the most straightforward and commonly used proxy to assess sustainability . policymakers should enrich this analysis by incorporating additional indicators of direct exposure such as interbank loans or repo agreements , which can also be used to construct more detailed networks . finally , understanding how individual or common shocks trigger financial crises is not only necessary to design countercyclical or containment policies , but also to design policies that prevent or limit the surge of such crises in the first place . as suggested in @xcite , @xcite and @xcite , institutions that are more likely to destabilize the system could be subject to stricter supervision , higher capital requirements or lower interbank exposure . this situation would certainly reduce moral hazard incentives for excessive risk - taking at institutions that are too large to be risked to fail . the g-20 , the financial stability oversight committee and several countries on their own are already in the process of drafting special supervision guidelines for these so - called systematically important financial institutions " or sifis ( @xcite and @xcite ) . we are grateful to staff members from the corresponding central or reserve banks for providing us the data . we thank d. heymann , j.c . hallak , and two anonymous referees for helpful comments . 32 blanchard , o. j. `` the crisis : basic mechanisms and appropriate policies '' , _ imf working papers _ , april , 1- 22 ( 2009 ) . duffie , darrell.innovations in credit risk transfer : implications for financial stability._bis working paper _ 255 ( 2008 ) . haldane , r.m . systemic risk in banking ecosystems . _ nature _ 469:351 - 355 ( 2011 ) . buiter , willem , `` lessons from the 2007 financial crisis '' , background paper submitted to the uk treasury select committee , december 11 ( 2007 ) . gorton , g.b . and metrick , a. . `` securitized banking and the run on repo '' . yale icf working paper no . 09 - 14 ( 2010 ) . brunnermeier , m.k . `` deciphering the liquidity and credit crunch 2007 - 2008 '' . _ journal of economic perspectives _ , 23 , 1 , 77 - 100 ( 2009 ) . klodt h , lehment h , eds . `` the crisis and beyond . lessons from the 2007 financial crisis '' . kiel institute for the world economy , e - books , ( 2009 ) . fender , ingo , and jacob gyntelberg . `` overview : global financial crisis spurs unprecedented policy actions '' . _ bis quarterly review _ , december ( 2008 ) . battiston , s. , gatti , d.d , gallegati , m. , greenwald , b.c.n . , and stiglitz , j.e . `` liaisons dangereuses : increasing connectivity , risk sharing and systemic risk '' . _ journal of economic dynamics and control _ , 36 , 1121 - 1141 ( 2012 ) . shin , hyun song . `` reflections on northern rock : the bank run that heralded the global financial crisis '' . _ journal of economic perspectives _ , 23(1 ) : 101 - 19 ( 2009 ) . basel committee on banking supervision . basel ii : international convergence of capital measurement and capital standards : a revised framework ( 2004 ) . basel committee on banking supervision . basel iii : a global regulatory framework for more resilient banks and banking systems ( 2010 ) . hannoun , h. the basel iii capital framework : a decisive breakthrough . bank for international settlements , ( 2010 ) . s. heise and r. kuhn . `` derivatives and credit contagion in interconnected networks '' . _ the european physical journal b - condensed matter and complex systems _ , 85 , number 4 , 115 ( 2012 ) . bastos e santos , e. and cont , r .. `` the brazilian interbank network structure and systemic risk '' . _ banco central do brasil _ , working paper no . 219 ( 2010 ) nier , e. , yang , j. , yorulmazer , t. , and alentorn , a. `` network models and financial stability '' . _ journal of economic dynamics and control _ , 31(6 ) : 2033 - 2060 ( 2007 ) . iori , g. , jafarey , s. , and padilla , f. `` systemic risk on the interbank market '' . _ journal of economic behaviour and organisation _ , 61(4 ) : 525 - 542 ( 2006 ) . kiyotaki , n. and moore , j .. `` credit cycles '' . _ journal of political economy _ , 105 , 2 , 211 - 248 ( 1997 ) . krugman , p. `` balance sheets , the transfer problem , and financial crisis '' . _ international tax and public finance _ , 6 , 459 - 472 ( 1999 ) . l.f . cspedes , r. chang , a. velasco . `` balance sheets and exchange rate policy '' . _ american economic review _ * 94 * , 1183 - 1193 ( 2004 ) . f.s . mishkin . `` symposium on the monetary transmission mechanism '' . _ the journal of economic perspectives _ * 9 * , 4 , 3 - 10 ( 1995 ) . kishan , t.p . `` bank size , bank capital , and the bank lending channel '' . _ journal of money , credit , and banking _ * 32 * , 1 , 121 - 141 ( 2000 ) . s. morris , h.s . shin.``financial regulation in a system context '' . _ brookings papers on economic activity _ , 229 - 274 ( 2008 ) . r. oliveira , r.f . schiozer , l.a.b . `` too big to fail perception by depositors : an empirical investigation '' . _ banco central do brasil _ , working paper no . 233 ( 2011 ) . martinez peria , s.l . `` do depositors punish banks for bad behavior ? market discipline , deposit insurance , and banking crises '' . _ the journal of finance _ * 56 * , 3 , 1029 - 1051 ( 2001 ) . a.m. maechler , k.m . `` dynamic depositor discipline in us banks '' . _ journal of banking and finance _ * 30 * 7 , 1871 - 1898 ( 2006 ) . m. imai . `` market discipline and deposit insurance reform in japan '' . _ journal of banking and finance _ , 30 , 3433 - 3452 ( 2006 ) . `` using market information in prudential bank supervision : a review of the u.s . empirical evidence '' . _ journal of money , credit and banking _ * 30 * , 3 , 273 - 305 ( 1998 ) . e. levy - yeyati , m.s . martinez peria , s.l . `` market discipline under systemic risk : evidence from bank runs in emerging economies '' . _ the world bank _ , policy research working paper series 3400 ( 2004 ) . s. galiani , d. heymann , m. tommasi.``great expectations and hard times : the argentine convertibility plan '' . _ economa _ * 3 * , 2 , 109 - 147 ( 2003 ) . calvo , a. izquierdo , e. talvi . `` sudden stops , the real exchange rate , and fiscal sustainability : argentina s lessons '' . nber working paper no . 9828 ( 2003 ) . l. damato , t. burdisso , v. cohen . `` the argentine banking and exchange rate crisis of 2001 : can we learn something from financial crises ? '' . banco central de la repblica argentina _ , working paper ( 2002 ) . m. newman , a.l . barabasi , and d.j . watts . `` the structure and dynamics of networks '' . princeton studies in complexity , ( 2006 ) . g. caldarelli . `` scale - free networks : complex webs in nature and technology '' . oxford finance , ( 2007 ) . bank for international settlements . global systemically important banks : assessment methodology and the additional loss absorbency requirement . ( 2011 ) bank of england , the role of macroprudential policy . discussion paper , november ( 2009 ) . bank of england , instruments of macroprudential policy . discussion paper , december ( 2011 ) . m. raberto , f. rapallo , and e. scalas . `` semi - markov graph dynamics '' . plos one 6.8 : e23370 , ( 2011 ) . `` banking panics : the role of the first - come , first - served rule and information externalities '' . _ journal of political economy _ , 107 ( 5 ) , 946968 , ( 1999 ) . m. lengnick , s. krug , and h.w . `` money creation and financial instability : an agent - based credit network approach '' . economics : the open - access , open - assessment e - journal , 7 ( 2013 - 32 ) : 1 - 44 , ( 2013 ) . b. lebaron , and l. tesfatsion . `` modeling macroeconomics as open - ended dynamic systems of interaction agents '' . _ american economic review _ , 98(2 ) : 246 - 50 , ( 2006 ) . gai , p. , haldane , a. , kapadia , s. `` complexity , concentration and contagion '' . _ journal of monetary economics _ , 58(5 ) , 453 - 470 ( 2011 ) . arinaminpathy , n. , kapadia , s. , may , r. m .. `` size and complexity in model financial systems '' . _ proceedings of the national academy of sciences _ , ( 2012 ) . upper , c , `` simulation methods to assess the danger of contagion in interbank markets '' . _ journal of financial stability _ , 7(3 ) , 111 - 125 ( 2011 ) . bair , s.c . `` we must resolve to end too big to fail '' . proceedings , federal reserve bank of chicago , may , 9 - 15 ( 2011 ) . elliot , d. j. and litan , r. e. `` identifying and regulating systemically important financial institutions : the risks of under and over identification and regulation '' . washington : the brookings institution , ( 2011 ) .

cold , dense quark matter is a color superconductor @xcite . for two massless quark flavors ( say , up and down ) , cooper pairs with total spin zero condense in the color - antitriplet , flavor - singlet channel . in this so - called two - flavor color superconductor , the @xmath0 gauge symmetry is spontaneously broken to @xmath1 @xcite . if we choose to orient the ( anti- ) color charge of the cooper pair along the ( anti- ) blue direction in color space , only red and green quarks form cooper pairs , while blue quarks remain unpaired . then , the three generators @xmath2 and @xmath3 of the original @xmath0 gauge group form the generators of the residual @xmath1 symmetry . the remaining five generators @xmath4 are broken . ( more precisely , the last broken generator is a combination of @xmath5 and the generator @xmath6 of the global @xmath7 symmetry of baryon number conservation , for details see ref . @xcite and below ) . according to goldstone s theorem , this pattern of symmetry breaking gives rise to five massless bosons , the so - called nambu - goldstone bosons , corresponding to the five broken generators of @xmath0 . physically , these massless bosons correspond to fluctuations of the order parameter , in our case the diquark condensate , in directions in color - flavor space where the effective potential is flat . for gauge theories ( where the local gauge symmetry can not truly be spontaneously broken ) , these bosons are `` eaten '' by the gauge bosons corresponding to the broken generators of the original gauge group , _ i.e. _ , in our case the gluons with adjoint colors @xmath8 . they give rise to a longitudinal degree of freedom for these gauge bosons . the appearance of a longitudinal degree of freedom is commonly a sign that the gauge boson becomes massive . in a dense ( or hot ) medium , however , even _ without _ spontaneous breaking of the gauge symmetry the gauge bosons already have a longitudinal degree of freedom , the so - called _ plasmon _ mode @xcite . its appearance is related to the presence of gapless charged quasiparticles . both transverse and longitudinal modes exhibit a mass gap , _ i.e. _ , the gluon energy @xmath9 for momenta @xmath10 . in quark matter with @xmath11 massless quark flavors at zero temperature @xmath12 , the gluon mass parameter ( squared ) is @xcite @xmath13 where @xmath14 is the qcd coupling constant and @xmath15 is the quark chemical potential . it is _ a priori _ unclear how the nambu - goldstone bosons interact with these longitudinal gluon modes . in particular , it is of interest to know whether coupling terms between these modes exist and , if yes , whether these terms can be eliminated by a suitable choice of ( t hooft ) gauge . the aim of the present work is to address these questions . we shall show that the answer to both questions is `` yes '' . we shall then demonstrate by focussing on the gluon of adjoint color 8 , how the nambu - goldstone mode affects the spectral density of the longitudinal gluon . our work is partially based on and motivated by previous studies of gluons in a two - flavor color superconductor @xcite . the gluon self - energy and the resulting spectral properties have been discussed in ref . @xcite . in that paper , however , the fluctuations of the diquark condensate have been neglected . consequently , the longitudinal degrees of freedom of the gluons corresponding to the broken generators of @xmath0 have not been treated correctly . the gluon polarization tensor was no longer explicitly transverse ( a transverse polarization tensor @xmath16 obeys @xmath17 ) , and it did not satisfy the slavnov - taylor identity . as a consequence , the plasmon mode exhibited a certain peculiar behavior in the low - momentum limit , which can not be physical ( cf . fig . 5 ( a ) of ref . it was already realized in ref . @xcite that the reason for this unphysical behavior is the fact that the mixing of the gluon with the excitations of the condensate was neglected . it was moreover suggested in ref . @xcite that proper inclusion of this mixing would amend the shortcomings of the previous analysis . the aim of the present work is to follow this suggestion and thus to correct the results of ref . @xcite with respect to the longitudinal gluon . note that in ref . @xcite fluctuations of the color - superconducting condensate were taken into account in the calculation of the gluon polarization tensor . as a consequence , the latter is explicitly transverse . however , the analysis was done in the vacuum , at @xmath18 , not at ( asymptotically ) large chemical potential . the outline of the present work is as follows . in section [ ii ] we derive the transverse and longitudinal gluon propagators including fluctuations of the diquark condensate . in section [ iii ] we use the resulting expressions to compute the spectral density for the gluon of adjoint color 8 . section [ iv ] concludes this work with a summary of our results . our units are @xmath19 . the metric tensor is @xmath20 . we denote 4-vectors in energy - momentum space by capital letters , @xmath21 . absolute magnitudes of 3-vectors are denoted as @xmath22 , and the unit vector in the direction of @xmath23 is @xmath24 . in this section , we derive the gluon propagator taking into account the fluctuations of the diquark condensate . a short version of this derivation can be found in appendix c of ref . @xcite [ see also the original ref . nevertheless , for the sake of clarity and in order to make our presentation self - contained , we decide to present this once more in greater detail and in the notation of ref . @xcite . as this part is rather technical , the reader less interested in the details of the derivation should skip directly to our main result , eqs . ( [ transverse ] ) , ( [ longitudinal ] ) , and ( [ hatpi00aa ] ) . we start with the grand partition function of qcd , [ z ] @xmath25\,\ , , \ ] ] where @xmath26 = \int { \cal d } \bar{\psi } \ , { \cal d } \psi\ , \exp \left [ \int_x \bar{\psi } \left ( i \gamma^\mu \partial_\mu + \mu \gamma_0 + g \gamma^\mu a_\mu^a t_a \right ) \psi \right ] \ , . \label{zquarks}\ ] ] is the grand partition function for massless quarks in the presence of a gluon field @xmath27 . in eq . ( [ z ] ) , the space - time integration is defined as @xmath28 , where @xmath29 is the volume of the system , @xmath30 are the dirac matrices , and @xmath31 are the generators of @xmath32 . for qcd , @xmath33 , and @xmath34 are the gell - mann matrices . the quark fields @xmath35 are @xmath36-component spinors , _ i.e. _ , they carry dirac indices @xmath37 , fundamental color indices @xmath38 , and flavor indices @xmath39 . the action for the gauge fields consists of three parts , @xmath40 where @xmath41 is the gauge field part ; here , @xmath42 is the field strength tensor . the part corresponding to gauge fixing , @xmath43 , and to fadeev - popov ghosts , @xmath44 , will be discussed later . for fermions at finite chemical potential it is advantageous to introduce the charge - conjugate degrees of freedom explicitly . this restores the symmetry of the theory under @xmath45 . therefore , in ref . @xcite , a kind of replica method was applied , in which one first artificially increases the number of quark species , and then replaces half of these species of quark fields by charge - conjugate quark fields . more precisely , first replace the quark partition function @xmath46 $ ] by @xmath47 \equiv \left\ { { \cal z}_q[a ] \right\}^m$ ] , @xmath48 being some large integer number . ( sending @xmath49 at the end of the calculation reproduces the original partition function . ) then , take @xmath48 to be an even integer number , and replace the quark fields by charge - conjugate quark fields in @xmath50 of the factors @xmath46 $ ] in @xmath47 $ ] . this results in @xmath51 = \int \prod_{r=1}^{m/2 } { \cal d } \bar{\psi}_r \ , { \cal d } \psi_r \ ; \exp \left\ { \sum_{r=1}^{m/2 } \left [ \int_{x , y } \bar{\psi}_r(x ) \,{\cal g}_0^{-1 } ( x , y)\ , \psi_r(y ) + \int_x g\ , \bar{\psi}_r(x ) \ , a_\mu^a(x)\,\hat{\gamma}^\mu_a\ , \psi_r(x ) \right ] \right\ } \,\ , . \label{zquarks2}\ ] ] here , @xmath52 labels the quark species and @xmath53 , @xmath54 are @xmath55-component nambu - gorkov spinors , @xmath56 where @xmath57 is the charge conjugate spinor and @xmath58 is the charge conjugation matrix . the inverse of the @xmath59-dimensional nambu - gorkov propagator for non - interacting quarks is defined as @xmath60^{-1 } & 0 \\ 0 & [ g_0 ^ -]^{-1 } \end{array } \right)\,\ , , \ ] ] where @xmath61^{-1}(x , y ) \equiv -i \left ( i \gamma_\mu \partial^\mu_x \pm \mu \gamma_0 \right ) \delta^{(4)}(x - y)\ ] ] is the inverse propagator for non - interacting quarks ( upper sign ) or charge conjugate quarks ( lower sign ) , respectively . the nambu - gorkov matrix vertex describing the interaction between quarks and gauge fields is defined as follows : @xmath62 where @xmath63 and @xmath64 . following ref . @xcite we now add the term @xmath65 and the corresponding charge - conjugate term @xmath66 , where @xmath67 , to the argument of the exponent in eq . ( [ zquarks2 ] ) . this defines the quark ( replica ) partition function in the presence of the gluon field @xmath27 _ and _ the diquark source fields @xmath68 , @xmath69 : @xmath70 \equiv \int \prod_{r=1}^{m/2 } { \cal d } \bar{\psi}_r \ , { \cal d } \psi_r \ ; \exp \left\ { \sum_{r=1}^{m/2 } \left [ \int_{x , y } \bar{\psi}_r(x ) \,{\cal g}^{-1 } ( x , y)\ , \psi_r(y ) + \int_x g\ , \bar{\psi}_r(x ) \ , a_\mu^a(x)\,\hat{\gamma}^\mu_a\ , \psi_r(x ) \right ] \right\ } \,\ , , \label{zquarks3}\ ] ] where @xmath71^{-1 } & \delta^- \\ \delta^+ & [ g^-_0]^{-1 } \end{array } \right)\ ] ] is the inverse quasiparticle propagator . inserting the partition function ( [ zquarks3 ] ) into eq . ( [ zqcd ] ) , the ( replica ) qcd partition function is then computed in the presence of the ( external ) diquark source terms @xmath72 , @xmath73 $ ] . in principle , this is not the physically relevant quantity , from which one derives thermodynamic properties of the color superconductor . the diquark condensate is not an external field , but assumes a nonzero value because of an intrinsic property of the system , namely the attractive gluon interaction in the color - antitriplet channel , which destabilizes the fermi surface . the proper functional from which one derives thermodynamic functions is obtained by a legendre transformation of @xmath74 $ ] , in which the functional dependence on the diquark source term is replaced by that on the corresponding canonically conjugate variable , the diquark condensate . the legendre - transformed functional is the effective action for the diquark condensate . if the latter is _ constant _ , the effective action is , up to a factor of @xmath75 , identical to the effective potential . the effective potential is simply a function of the diquark condensate . its explicit form for large - density qcd was derived in ref . the value of this function at its maximum determines the pressure . the maximum is determined by a dyson - schwinger equation for the diquark condensate , which is identical to the standard gap equation for the color - superconducting gap . it has been solved in the mean - field approximation in refs . @xcite . in the mean - field approximation @xcite , @xmath76 in this work , we are interested in the gluon propagator , and the derivation of the pressure via a legendre transformation of @xmath77 $ ] is of no concern to us . in the following , we shall therefore continue to consider the partition function in the presence of ( external ) diquark source terms @xmath78 . the diquark source terms in the quark ( replica ) partition function ( [ zquarks3 ] ) could in principle be chosen differently for each quark species . this could be made explicit by giving @xmath78 a subscript @xmath52 , @xmath79 . however , as we take the limit @xmath49 at the end , it is not necessary to do so , as only @xmath80 will survive anyway . in other words , we use the _ same _ diquark sources for _ all _ quark species . the next step is to explicitly investigate the fluctuations of the diquark condensate around its expectation value . these fluctuations correspond physically to the nambu - goldstone excitations ( loosely termed `` mesons '' in the following ) in a color superconductor . as mentioned in the introduction , there are five such mesons in a two - flavor color superconductor , corresponding to the generators of @xmath0 which are broken in the color - superconducting phase . if the condensate is chosen to point in the ( anti- ) blue direction in fundamental color space , the broken generators are @xmath81 of the original @xmath0 group and the particular combination @xmath82 of generators of the global @xmath83 and local @xmath0 symmetry @xcite . the effective action for the diquark condensate and , consequently , for the meson fields as fluctuations of the diquark condensate , is derived via a legendre transformation of @xmath77 $ ] . in this work , we are concerned with the properties of the gluons and thus refrain from computing this effective action explicitly . consequently , instead of considering the physical meson fields , we consider the variables in @xmath84 $ ] , which correspond to these fields . these are the fluctuations of the diquark source terms @xmath78 . we choose these fluctuations to be complex phase factors multiplying the magnitude of the source terms , [ deltaphi ] @xmath85 where @xmath86\,\,.\ ] ] the extra factor @xmath87 in front of @xmath88 as compared to the treatment in ref . @xcite is chosen to simplify the notation in the following . although the fields @xmath89 are not the meson fields themselves , but external fields which , after a legendre transformation of @xmath77 $ ] , are replaced by the meson fields , we nevertheless ( and somewhat imprecisely ) refer to them as meson fields in the following . after having explicitly introduced the fluctuations of the diquark source terms in terms of phase factors , the functions @xmath90 are only allowed to fluctuate in magnitude . for the sake of completeness , let us mention that one could again have introduced different fields @xmath91 for each replica @xmath52 , but this is not really necessary , as we shall take the limit @xmath49 at the end of the calculation anyway . it is advantageous to also subject the quark fields @xmath92 to a nonlinear transformation , introducing new fields @xmath93 via @xmath94 since the meson fields are real - valued and the generators @xmath81 and @xmath95 are hermitian , the ( matrix - valued ) operator @xmath96 is unitary , @xmath97 . therefore , the measure of the grassmann integration over quark fields in eq . ( [ zquarks3 ] ) remains unchanged . from eq . ( [ chi ] ) , the charge - conjugate fields transform as @xmath98 the advantage of transforming the quark fields is that this preserves the simple structure of the terms coupling the quark fields to the diquark sources , @xmath99 in mean - field approximation , the diquark source terms are proportional to @xmath100 the transformation ( [ chi ] ) has the following effect on the kinetic terms of the quarks and the term coupling quarks to gluons : @xmath101 where [ maurer1 ] @xmath102 is the @xmath103-dimensional maurer - cartan one - form introduced in ref . @xcite and @xmath104 is its charge - conjugate version . note that the partial derivative acts only on the phase factors @xmath96 and @xmath105 on the right . introducing the nambu - gorkov spinors @xmath106 and the @xmath107-dimensional maurer - cartan one - form @xmath108 the quark ( replica ) partition function becomes @xmath109 \equiv \int \prod_{r=1}^{m/2 } { \cal d } \bar{x}_r \ , { \cal d } x_r \ ; \exp \left\ { \sum_{r=1}^{m/2 } \int_{x , y } \bar{x}_r(x ) \,\left [ \ , { \cal s}^{-1 } ( x , y ) + \gamma_\mu \omega^\mu(x , y ) \ , \right ] \ , x_r(y ) \right\ } \,\ , , \label{zquarks4}\ ] ] where @xmath110^{-1 } & \phi^- \\ \phi^+ & [ g^-_0]^{-1 } \end{array } \right)\,\ , .\ ] ] we are interested in the properties of the gluons , and thus may integrate out the fermion fields . this integration can be performed analytically , with the result @xmath109 \equiv \left [ \,{\rm det } \left ( { \cal s}^{-1 } + \gamma_\mu \omega^\mu \right ) \ , \right]^{m/2 } \,\ , . \label{zquarks5}\ ] ] the determinant is to be taken over nambu - gorkov , color , flavor , spin , and space - time indices . finally , letting @xmath49 , we obtain the qcd partition function ( in the presence of meson , @xmath89 , and diquark , @xmath90 , source fields ) @xmath111 = \int { \cal d } a \ ; \exp\left [ s_a + \frac{1}{2 } \ , { \rm tr } \ln \left({\cal s}^{-1 } + \gamma_\mu \omega^\mu \right ) \ , \right]\,\ , .\ ] ] remembering that @xmath112 is linear in @xmath27 , cf . ( [ maurer2 ] ) with ( [ maurer1 ] ) , in order to derive the gluon propagator it is sufficient to expand the logarithm to second order in @xmath112 , @xmath113 + s_1[\omega,\phi^+,\phi^- ] + s_2[\omega,\phi^+,\phi^-]\,\ , , \label{expandlog}\end{aligned}\ ] ] with obvious definitions for the @xmath114 . the quasiparticle propagator is @xmath115 with @xmath116^{-1 } - \sigma^\pm \right\}^{-1 } \,\,\,\ , , \,\,\,\,\ , \sigma^\pm = \phi^\mp \ , g_0^\mp \ , \phi^\pm \,\,\,\ , , \,\,\,\,\ , \xi^\pm = - g_0^\mp \ , \phi^\pm \ , g^\pm\,\ , .\ ] ] to make further progress , we now expand @xmath117 and @xmath118 to linear order in the meson fields , [ linearomega ] @xmath119 the term @xmath120 in eq . ( [ expandlog ] ) is simply a tadpole source term for the gluon fields . this term does not affect the gluon propagator , and thus can be ignored in the following . the quadratic term @xmath121 represents the contribution of a fermion loop to the gluon self - energy . its computation proceeds by first taking the trace over nambu - gorkov space , @xmath122 \,\ , . \label{s2}\end{aligned}\ ] ] the remaining trace runs only over color , flavor , and spin indices . using translational invariance , the propagators and fields are now fourier - transformed as @xmath123 inserting this into eq . ( [ s2 ] ) , we arrive at eq . ( c16 ) of ref . @xcite , which in our notation reads @xmath124 \,\ , . \end{aligned}\ ] ] the remainder of the calculation is straightforward , but somewhat tedious . first , insert the ( fourier - transform of the ) linearized version ( [ linearomega ] ) for the fields @xmath117 and @xmath125 . this produces a plethora of terms which are second order in the gluon and meson fields , with coefficients that are traces over color , flavor , and spin . next , perform the color and flavor traces in these coefficients . it turns out that some of them are identically zero , preventing the occurrence of terms which mix gluons of adjoint colors 1 , 2 , and 3 ( the unbroken @xmath1 subgroup ) among themselves and with the other gluon and meson fields . furthermore , there are no terms mixing the meson fields @xmath126 with @xmath88 . there are mixed terms between gluons and mesons with adjoint color indices @xmath127 , and between the gluon field @xmath128 and the meson field @xmath88 . some of the mixed terms ( those which mix gluons and mesons of adjoint colors 4 and 5 , as well as 6 and 7 ) can be eliminated via a unitary transformation analogous to the one employed in ref . @xcite , eq . ( 80 ) . introducing the tensors @xmath129 \,\ , , \label{pi11}\end{aligned}\ ] ] cf . ( 78a ) of ref . @xcite , @xmath130\,\ , , \label{pi44diag } \end{aligned}\ ] ] cf . ( 83a ) of ref . @xcite , @xmath131\,\ , . \label{pi55diag}\end{aligned}\ ] ] cf . ( 83b ) of ref . @xcite , as well as @xmath132 \,\ , , \label{pitilde}\end{aligned}\ ] ] cf . ( 78c ) of ref . @xcite , where @xmath133 is the gluon self - energy in a dense , but normal - conducting system , @xmath134\,\ , , \label{pi0}\ ] ] cf . ( 27b ) of ref . @xcite , the final result can be written in the compact form ( cf . ( c19 ) of ref . @xcite ) @xmath135 \ , \pi^{\mu \nu}_{aa}(p ) \ , \left[a_\nu^a(p ) + \frac{i}{g}\ , p_\nu\ , \varphi^a(p)\right ] \,\ , .\ ] ] in deriving eq . ( [ s2final ] ) , we have made use of the transversality of the polarization tensor in the normal - conducting phase , @xmath136 . note that the tensors @xmath137 for @xmath138 and 3 are also transverse , but those for @xmath139 are not . this can be seen explicitly from the expressions given in ref . the compact notation of eq . ( [ s2final ] ) is made possible by the fact that @xmath140 for @xmath141 , and because we introduced the extra factor @xmath87 in eq . ( [ phase ] ) as compared to ref . @xcite . to make further progress , it is advantageous to tensor - decompose @xmath137 . various ways to do this are possible @xcite ; here we follow the notation of ref . first , define a projector onto the subspace parallel to @xmath142 , @xmath143 then choose a vector orthogonal to @xmath142 , for instance @xmath144 with @xmath145 . note that @xmath146 . now define the projectors @xmath147 using the explicit form of @xmath148 , one convinces oneself that the tensor @xmath149 projects onto the spatially transverse subspace orthogonal to @xmath142 , @xmath150 ( reference @xcite also uses the notation @xmath151 for @xmath149 . ) consequently , the tensor @xmath152 projects onto the spatially longitudinal subspace orthogonal to @xmath142 , @xmath153 ( reference @xcite also employs the notation @xmath154 for @xmath152 . ) with these tensors , the gluon self - energy can be written in the form @xmath155 the polarization functions @xmath156 and @xmath157 can be computed by projecting the tensor @xmath137 onto the respective subspaces of the projectors ( [ e ] ) and ( [ bca ] ) . introducing the abbreviations @xmath158 these functions read [ pifunctions ] @xmath159 \,\ , , \\ \pi^{\rm c}_{aa}(p ) & = & \frac{1}{2\ , n^2 \ , p^2}\ , \pi^{\mu \nu}_{aa}(p)\ , { \rm c}_{\mu \nu } = -\frac{1}{p^2}\ , \left [ \pi^{00}_{aa}(p ) + \frac{p_0 ^ 2+p^2}{p_0\,p}\ , \pi^{0i}_{aa}(p)\,\hat{p}_i + \pi^\ell_{aa}(p ) \right ] \,\ , , \\ \pi^{\rm e}_{aa}(p ) & = & \pi^{\mu \nu}_{aa}(p)\ , { \rm e}_{\mu \nu } = \frac{1}{p^2}\ , \left [ p_0 ^ 2 \ , \pi^{00}_{aa}(p ) + 2\,p_0\,p \ , \pi^{0i}_{aa}(p ) \ , \hat{p}_i + p^2 \ , \pi^\ell_{aa}(p ) \right ] \,\ , .\end{aligned}\ ] ] for the explicitly transverse tensor @xmath160 , the functions @xmath161 . the same holds for the hdl polarization tensor @xmath133 . for the other gluon colors @xmath162 , the functions @xmath163 and @xmath157 do not vanish . note that the dimensions of @xmath164 and @xmath157 are @xmath165 $ ] , while @xmath163 is dimensionless . now let us define the functions @xmath166 note that @xmath167 , and @xmath168 , since @xmath148 is odd under @xmath169 . the fields @xmath170 and @xmath171 are dimensionless . with the tensor decomposition ( [ tensor ] ) and the functions ( [ functions ] ) , eq . ( [ s2final ] ) becomes @xmath172\ , \pi^{\rm c}_{aa}(p ) \,n^2 p^2 \ , a_n^a(p ) - a_n^a(-p ) \ , \pi^{\rm c}_{aa}(p ) \,n^2 p^2 \ \left[a_\parallel^a(p ) + \frac{i}{g}\ , \varphi^a(p)\right ] \nonumber \\ & & - \left . \left[a_\parallel^a(-p ) + \frac{i}{g}\ , \varphi^a(-p)\right]\ , \pi^{\rm e}_{aa}(p ) \ , \left[a_\parallel^a(p ) + \frac{i}{g}\ , \varphi^a(p)\right ] \right\}\,\ , . \label{decompose}\end{aligned}\ ] ] in any spontaneously broken gauge theory , the excitations of the condensate mix with the gauge fields corresponding to the broken generators of the underlying gauge group . the mixing occurs in the components orthogonal to the spatially transverse degrees of freedom , _ i.e. _ , for the spatially longitudinal fields , @xmath173 , and the fields parallel to @xmath142 , @xmath174 . for the two - flavor color superconductor , these components mix with the meson fields for gluon colors @xmath175 . the mixing is particularly evident in eq . ( [ decompose ] ) . the terms mixing mesons and gauge fields can be eliminated by a suitable choice of gauge . the gauge to accomplish this goal is the t hooft gauge . the `` unmixing '' procedure of mesons and gauge fields consists of two steps . first , we eliminate the terms in eq . ( [ decompose ] ) which mix @xmath173 and @xmath174 . this is achieved by substituting @xmath176 ( we do not perform this substitution for @xmath177 ; for these gluon colors @xmath178 , such that there are no terms in eq . ( [ decompose ] ) which mix @xmath174 and @xmath173 ) . this shift of the gauge field component @xmath174 is completely innocuous for the following reasons . first , the jacobian @xmath179 is unity , so the measure of the functional integral over gauge fields is not affected . second , the only other term in the gauge field action , which is quadratic in the gauge fields and thus relevant for the derivation of the gluon propagator , is the free field action @xmath180 and it does not contain the parallel components @xmath170 . it is therefore also not affected by the shift of variables ( [ sub ] ) . after renaming @xmath181 , the final result for @xmath121 reads : @xmath182\ , \pi^{\rm e}_{aa}(p ) \ , p^2 \ , \left[a_\parallel^a(p ) + \frac{i}{g}\ , \varphi^a(p)\right ] \right\}\,\ , , \label{s2finalunmix}\end{aligned}\ ] ] where we introduced @xmath183 ^ 2 n^2 p^2}{\pi^{\rm e}_{aa}(p ) } \,\ , .\ ] ] the t hooft gauge fixing term is now chosen to eliminate the mixing between @xmath174 and @xmath184 : @xmath185 \ , \left [ p^2\ , a_\parallel^a(p ) - \lambda\ , \frac{i}{g}\ , \pi^{\rm e}_{aa}(p)\ , \varphi^a(p ) \right ] \,\ , . \label{l_gf}\ ] ] this gauge condition is non - local in coordinate space , which seems peculiar , but poses no problem in momentum space . note that @xmath186 . therefore , in various limits the choice of gauge ( [ l_gf ] ) corresponds to covariant gauge , @xmath187 the first limit we consider is @xmath188 , _ i.e. _ the vacuum . then , @xmath189 , and eq . ( [ l_gf ] ) becomes ( [ l_cg ] ) . the second case is the limit of large 4-momenta , @xmath190 . as shown in ref . @xcite , in this region of phase space the effects from a color - superconducting condensate on the gluon polarization tensor are negligible . in other words , the gluon polarization tensor approaches the hdl limit . the physical reason is that gluons with large momenta do not see quark cooper pairs as composite objects , but resolve the individual color charges inside the pair . consequently , @xmath191 for @xmath190 and , for large @xmath192 , the individual terms in the sum over @xmath192 in eqs . ( [ l_gf ] ) and ( [ l_cg ] ) agree . finally , for gluon colors @xmath177 , @xmath189 , since the self - energy @xmath160 is transverse . thus , for @xmath177 the terms in eqs . ( [ l_gf ] ) and ( [ l_cg ] ) are identical . the decoupling of mesons and gluon degrees of freedom becomes obvious once we add ( [ l_gf ] ) to ( [ s2finalunmix ] ) and ( [ s0 ] ) , @xmath193\ , { \rm a}^{\mu \nu}\ , a_{\perp\,\nu}^a(p ) \right . \nonumber \\ & & \hspace*{2.1 cm } - \ ; a_n^a(-p ) \ , \left [ p^2 + \hat{\pi}^{\rm b}_{aa}(p ) \right ] \ , n^2 \ , a_n^a(p ) \nonumber \\ & & \hspace*{2.1 cm } - \ ; a_\parallel^a(-p ) \ , \left [ \frac{1}{\lambda}\ , p^2 + \pi^{\rm e}_{aa}(p ) \right]\ , p^2 \ , a_\parallel^a(p ) \nonumber \\ & & \hspace*{2.1 cm } + \left . \frac{\lambda}{g^2}\ , \varphi^a(-p)\ , \left [ \frac{1}{\lambda}\ , p^2 + \pi^{\rm e}_{aa}(p ) \right ] \ , \pi^{\rm e}_{aa}(p)\ , \varphi^a(p ) \right\ } \,\ , . \label{sgfs2}\end{aligned}\ ] ] consequently , the inverse gluon propagator is @xmath194\ , { \rm a}^{\mu \nu } + \left [ p^2 + \hat{\pi}^{\rm b}_{aa}(p ) \right ] \ , { \rm b}^{\mu \nu } + \left [ \frac{1}{\lambda}\ , p^2 + \pi^{\rm e}_{aa}(p ) \right ] \ , { \rm e}^{\mu \nu}\,\ , .\ ] ] inverting this as discussed in ref . @xcite , one obtains the gluon propagator for gluons of color @xmath195 , @xmath196 for any @xmath197 , the gluon propagator contains unphysical contributions parallel to @xmath142 , which have to be cancelled by the corresponding faddeev - popov ghosts when computing physical observables . only for @xmath198 these contributions vanish and the gluon propagator is explicitly transverse , _ i.e. _ , @xmath199 . also , in this case the ghost propagator is independent of the chemical potential @xmath15 . the contribution of fadeev - popov ghosts to the gluon polarization tensor is then @xmath200 and thus negligible at @xmath12 . we shall therefore focus on this particular choice for the gauge parameter in the following . note that for @xmath198 , the inverse meson field propagator is @xmath201 and the dispersion relation for the mesons follows from the condition @xmath202 , as demonstrated in ref . @xcite for a three - flavor color superconductor in the color - flavor - locked phase . the gluon propagator for transverse and longitudinal modes can now be read off eq . ( [ glueprop ] ) as coefficients of the corresponding tensors @xmath149 ( the projector onto the spatially transverse subspace orthogonal to @xmath142 ) and @xmath152 ( the projector onto the spatially longitudinal subspace orthogonal to @xmath142 ) . for the transverse modes one has @xcite @xmath203 where we used eq . ( [ pia ] ) . multiplying the coefficient of @xmath152 in eq . ( [ glueprop ] ) with the standard factor @xmath204 @xcite , one obtains for the longitudinal modes @xmath205 where the longitudinal gluon self - energy @xmath206 ^ 2 } { p_0 ^ 2 \ , \pi^{00}_{aa}(p ) + 2\ , p_0\,p\ , \pi^{0i}_{aa}(p ) \ , \hat{p}_i + p^2 \ , \pi^\ell_{aa}(p ) } \ ] ] follows from the definition of @xmath207 , eq.([hatpib ] ) , and the relations ( [ pifunctions ] ) . the longitudinal gluon propagator @xmath208 _ must not be confused _ with the the @xmath209-component of @xmath210 . we deliberately use this ( slightly ambiguous ) notation to facilitate the comparison of our new and correct results with those of ref . @xcite , which were partially incorrect . the results of that paper were derived in coulomb gauge , where the @xmath209-component of the propagator is indeed _ identical _ to the longitudinal propagator ( [ longitudinal ] ) . we were not able to find a t hooft gauge that converged to the coulomb gauge in the various limits discussed above , and consequently had to base our discussion on the covariant gauge ( [ l_cg ] ) as limiting case of eq . ( [ l_gf ] ) . to summarize this section , we have computed the gluon propagator for gluons in a two - flavor color superconductor . due to condensation of quark cooper pairs , the @xmath0 gauge symmetry is spontaneously broken to @xmath1 , leading to the appearance of five nambu - goldstone bosons . in general , these bosons mix with some components of the gauge fields corresponding to the broken generators . to `` unmix '' them we have used a form of t hooft gauge which smoothly converges to covariant gauge in the vacuum , as well as for large gluon momenta , and when the gluon polarization tensor is explicitly transverse . finally , choosing the gauge fixing parameter @xmath211 we derived the gluon propagator for transverse , eq . ( [ transverse ] ) , and longitudinal modes , eq . ( [ longitudinal ] ) with ( [ hatpi00aa ] ) . in this section , we explicitly compute the spectral properties of the eighth gluon . we shall confirm the results of ref . @xcite for the transverse mode and amend those for the longitudinal mode , which have not been correctly computed in ref . in particular , we shall show that the plasmon dispersion relation now has the correct behavior @xmath212 as @xmath10 . furthermore , the longitudinal spectral density vanishes for gluon energies and momenta located on the dispersion branch of the nambu - goldstone bosons , _ i.e. _ , for energies and momenta given by the roots of eq . ( [ ngbosons ] ) . for the eighth gluon , this condition can be written in the form @xmath213 @xcite , since the hdl self - energy is transverse , @xmath214 . we first compute the polarization tensor for the transverse and longitudinal components of the eighth gluon . to this end , it is convenient to rewrite the longitudinal gluon self - energy ( [ hatpi00aa ] ) in the form @xmath215 ^ 2 } { p_0 ^ 2 \ , \tilde{\pi}^{00}(p ) + 2\ , p_0\,p\ , \tilde{\pi}^{0i}(p ) \ , \hat{p}_i + p^2 \ , \tilde{\pi}^\ell(p ) } \,\ , , \label{hatpi00}\end{aligned}\ ] ] with @xmath216 . let us now explicitly compute the polarization functions . as in ref . @xcite we take @xmath12 , and we shall use the identity @xmath217 where @xmath218 stands for the principal value description , in order to decompose the polarization tensor into real and imaginary parts . the imaginary parts can then be straightforwardly computed , while the real parts are computed from the dispersion integral @xmath219 where @xmath220 is a ( subtraction ) constant . if @xmath221 is an odd function of @xmath222 , @xmath223 , eq . ( [ dispint ] ) becomes eq . ( 39 ) of ref . @xcite , @xmath224 and if it is an even function of @xmath222 , @xmath225 , we have instead @xmath226 since the polarization tensor for the transverse gluon modes , @xmath227 , has already been computed in ref . @xcite , we just cite the results . the imaginary part of the transverse hdl polarization function reads ( cf . ( 22b ) of ref . @xcite ) @xmath228 the corresponding real part is computed from eq . ( [ odd ] ) , with the result ( cf . eqs . ( 40b ) and ( 41 ) of ref . @xcite ) @xmath229\,\ , .\ ] ] we have used the fact that the value of the subtraction constant is @xmath230 , which can be derived from comparing a direct calculation of @xmath231 using eq . ( 19b ) of ref . @xcite with the above computation via the dispersion formula ( [ odd ] ) . the imaginary part of the tensor @xmath232 is given by ( cf . ( 36 ) of ref . @xcite ) @xmath233 \right . } \nonumber \\ & + & \left . \theta(p_0 - e_p ) \left [ \left ( 1 - \frac{p_0 ^ 2}{p^2}\ , ( 1+s^2 ) \right ) \ , e(\alpha , t ) - \left ( 1- \frac{p_0 ^ 2}{p^2 } \right)\ , \frac{p}{p_0}\ , \sqrt{1 - \frac{4\ , \phi^2}{p_0 ^ 2 - p^2 } } - s^2 \,\left ( 1- 2\ , \frac{p_0 ^ 2}{p^2 } \right)\ , f(\alpha , t ) \right ] \right\}\,\ , , \end{aligned}\ ] ] where @xmath234 is the value of the color - superconducting gap , @xmath235 , @xmath236 , @xmath237 , @xmath238 $ ] , and @xmath239 , @xmath240 are elliptic integrals of the first and second kind , while @xmath241 and @xmath242 are the corresponding complete elliptic integrals . the real part is again computed from eq . ( [ odd ] ) . the integral has to be done numerically , see appendix a of ref . @xcite for details . the subtraction constant is , for reasons discussed at length in ref . @xcite , identical to the one in the hdl limit , @xmath243 . finally , taking the linear combination @xmath227 completes the calculation of the transverse polarization function @xmath244 . in order to compute the polarization function for the longitudinal gluon , @xmath245 , we have to know the functions @xmath246 , @xmath247 , @xmath248 , and @xmath249 . the first two functions , @xmath246 and @xmath247 have also been computed in ref . the imaginary part of the longitudinal hdl polarization function is ( cf . ( 22a ) of ref . @xcite ) @xmath250 the real part is computed from eq . ( [ odd ] ) , with the result ( cf . eqs . ( 40a ) and ( 41 ) of ref . @xcite ) @xmath251 here , the subtraction constant is @xmath252 . the imaginary part of the function @xmath253 is ( cf . ( 35 ) of ref . @xcite ) @xmath254 \right\}\,\ , .\ ] ] the real part is computed from eq . ( [ odd ] ) , with the subtraction constant @xmath255 . again , the integral has to be done numerically . it remains to compute the functions @xmath248 and @xmath249 . first , one performs the spin traces in eq . ( [ pitilde ] ) to obtain eqs . ( 102b ) and ( 102c ) of ref . @xcite . then , taking @xmath12 , [ pi0ipil ] @xmath256\ , \frac{\epsilon_1 \epsilon_2 - \xi_1 \xi_2 - \phi_1 \phi_2}{2 \ , \epsilon_1 \epsilon_2 } \nonumber \\ & & \hspace*{2.6 cm } \times \ , \left ( \frac{1}{p_0 + \epsilon_1 + \epsilon_2 + i \eta } - \frac{1}{p_0 - \epsilon_1 - \epsilon_2 + i \eta } \right ) \,\ , , \end{aligned}\ ] ] where @xmath257 , @xmath258 is the gap function for quasiparticles ( @xmath259 ) or quasi - antiparticles ( @xmath260 ) with momentum @xmath261 , @xmath262 , and @xmath263 . one now repeats the steps discussed in detail in section ii.a of ref . @xcite to obtain ( for @xmath264 ) [ pi0ipil2 ] @xmath265 \right . \nonumber \\ & & \hspace*{3.2 cm } + \left . \theta(p_0 - e_p ) \left [ e(\alpha , t ) - \frac{p}{p_0}\ , \sqrt{1 - \frac{4\ , \phi^2}{p_0 ^ 2 - p^2 } } - s^2 \ , f(\alpha , t ) \right ] \right\}\,\ , , \\ { \rm im}\ , \tilde{\pi}^\ell(p ) & = & - \pi\ , \frac{3}{2}\ , m_g^2 \ , \theta(p_0 - 2\ , \phi)\ , \frac{p_0 ^ 3}{p^3 } \left\ { \frac { } { } \theta(e_p - p_0 ) \ , \left [ ( 1+s^2)\ , { \bf e}(t ) - 2\ , s^2 \ , { \bf k}(t ) \right ] \right . \nonumber \\ & & \hspace*{2.5 cm } + \left . \theta(p_0 - e_p ) \left [ ( 1+s^2)\ , e(\alpha , t ) - \frac{p}{p_0}\ , \sqrt{1 - \frac{4\ , \phi^2}{p_0 ^ 2 - p^2 } } - 2\ , s^2 \ , f(\alpha , t ) \right ] \right\}\,\ , .\end{aligned}\ ] ] one observes that in the limit @xmath266 , the functions ( [ pi0ipil2 ] ) approach the hdl result @xmath267 applying eq . ( [ ident ] ) to eqs . ( [ pi0ipil ] ) we immediately see that the imaginary part of @xmath248 is _ even _ , while that of @xmath249 is _ odd_. thus , in order to compute the real part of @xmath248 , we have to use eq . ( [ even ] ) , while the real part of @xmath249 has to be computed from eq . ( [ odd ] ) . when implementing the numerical procedure discussed in appendix a of ref . @xcite for the integral in eq . ( [ even ] ) , one has to modify eq . ( a1 ) of ref . @xcite appropriately . finally , one has to determine the values of the subtraction constants @xmath268 and @xmath269 . we again use the fact that @xmath270 and @xmath271 , where the index `` 0 '' refers to the hdl limit . the corresponding constants are determined by first computing @xmath272 and @xmath273 from the dispersion formulas ( [ odd ] ) and ( [ even ] ) . the result of this calculation is then compared to that of a direct computation using , for instance , the result ( [ repi000 ] ) for @xmath274 and then inferring @xmath275 and @xmath273 from the transversality of @xmath133 . the result is @xmath276 and @xmath277 . at this point , we have determined all functions entering the transverse and longitudinal polarization functions for the eighth gluon . in fig . [ fig1 ] we show the imaginary parts and in fig . [ fig2 ] the real parts , for a fixed gluon momentum @xmath278 , as a function of gluon energy @xmath279 ( in units of @xmath280 ) . the units for the imaginary parts are @xmath281 , and for the real parts @xmath282 . for comparison , in parts ( a ) and ( g ) of these figures , we show the results from ref . @xcite for the longitudinal and transverse polarization function of the gluon with adjoint color 1 . in parts ( d ) , ( e ) , and ( f ) the functions @xmath253 , @xmath283 , and @xmath284 are shown . according to eq . ( [ hatpi00 ] ) these are required to determine @xmath285 , shown in part ( b ) . using eq . ( [ pi0088 ] ) , this result is then combined with the hdl polarization function @xmath286 to compute @xmath245 , shown in part ( c ) . finally , the transverse polarization function for gluons of color 8 is shown in part ( i ) . this function is given by the linear combination @xmath287 of the transverse hdl polarization function @xmath288 and the function @xmath232 , both of which are shown in part ( h ) . in all figures , the results for the two - flavor color superconductor are drawn as solid lines , while the dotted lines correspond to those in a normal conductor , @xmath266 ( the hdl limit ) . note that parts ( a ) , ( d ) , ( g ) , ( h ) , and ( i ) of figs . [ fig1 ] and [ fig2 ] agree with parts ( a ) , ( b ) , ( d ) , ( e ) , and ( f ) of figs . 2 and 3 of ref . @xcite . the new results are parts ( e ) and ( f ) of figs . [ fig1 ] and [ fig2 ] , which are used to determine the functions in parts ( b ) and ( c ) , the latter showing the correct longitudinal polarization function for the eighth gluon . in ref . @xcite , this function was not computed correctly , as the effect from the fluctuations of the condensate on the polarization tensor of the gluons was not taken into account . the singularity around a gluon energy somewhat smaller than @xmath289 visible in figs . ( b ) and ( c ) seems peculiar . it turns out that it arises due to a zero in the denominator of @xmath285 in eq . ( [ hatpi00 ] ) , _ i.e. _ , when @xmath213 . as discussed above , this condition defines the dispersion branch of the nambu - goldstone excitations @xcite . therefore , the singularity is tied to the existence of the nambu - goldstone excitations of the diquark condensate . let us now determine the spectral densities for longitudinal and transverse modes , defined by ( cf . ( 45 ) of ref . @xcite ) @xmath290 the longitudinal and transverse spectral densities for gluons of color 8 are shown in figs . [ fig3 ] ( c ) and ( d ) , for fixed gluon momentum @xmath291 and @xmath292 . for comparison , the corresponding spectral densities for gluons of color 1 are shown in parts ( a ) and ( b ) . parts ( a ) , ( b ) , and ( d ) are identical to those of fig . 6 of ref . @xcite , part ( c ) is new and replaces fig . 6 ( c ) of ref . one observes a peak in the spectral density around @xmath293 . this peak corresponds to the ordinary longitudinal gluon mode ( the plasmon ) present in a dense ( or hot ) medium . note that the longitudinal spectral density for gluons of color 8 vanishes at an energy somewhat smaller than @xmath294 . the reason is the singularity of the real part of the gluon self - energy seen in figs . [ fig2 ] ( b ) and ( c ) . the location of this point is where @xmath295 , _ i.e. _ , on the dispersion branch of the nambu - goldstone excitations . finally , we show in fig . [ fig4 ] the dispersion relations for all excitations , defined by the roots of [ disprel ] @xmath296 for longitudinal gluons ( cf . ( 47a ) of ref . @xcite ) , and by the roots of @xmath297 for transverse gluons ( cf . ( 47b ) of ref . let us mention that not all excitations found via eqs . ( [ disprel ] ) correspond to truly stable quasiparticles , _ i.e. _ , the imaginary parts of the self - energies do not always vanish along the dispersion curves . nevertheless , in that case eqs . ( [ disprel ] ) can still be used to identify peaks in the spectral densities , which correspond to _ unstable _ modes ( which decay with a rate proportional to the width of the peak ) . as long as the width of the peak ( the decay rate of the quasiparticles ) is small compared to its height , it makes sense to refer to these modes as quasiparticles . [ fig4 ] corresponds to fig . 5 of ref . in fact , part ( b ) is identical in both figures . [ fig4 ] ( a ) differs from fig . 5 ( a ) of ref . @xcite , reflecting our new and correct results for the longitudinal gluon self - energy . in fig . 5 ( a ) of ref . @xcite , the dispersion curve for the longitudinal gluon of color 8 was seen to diverge for small gluon momenta . in ref . @xcite it was argued that this behavior was due to neglecting the mesonic fluctuations of the diquark condensate . indeed , properly accounting for these modes , we obtain a reasonable dispersion curve , approaching @xmath293 as the momentum goes to zero . in fig . [ fig4 ] ( a ) we also show the dispersion branch for the nambu - goldstone excitations ( dash - dotted ) . this is strictly speaking not given by a root of eq . ( [ disprel ] ) , but by the singularity of the real part of the longitudinal gluon self - energy . however , because this singularity involves a change of sign , a normal root - finding algorithm applied to eq . ( [ disprel ] ) will also locate this singularity . as expected @xcite , the dispersion branch is linear , @xmath298 for small gluon momenta , and approaches the value @xmath289 for @xmath299 . in cold , dense quark matter with @xmath300 massless quark flavors , condensation of quark cooper pairs spontaneously breaks the @xmath0 gauge symmetry to @xmath1 . this results in five nambu - goldstone excitations which mix with some of the components of the gluon fields corresponding to the broken generators . we have shown how to decouple them by a particular choice of t hooft gauge . the unphysical degrees of freedom in the gluon propagator can be eliminated by fixing the t hooft gauge parameter @xmath198 . in this way , we derived the propagator for transverse and longitudinal gluon modes in a two - flavor color superconductor accounting for the effect of the nambu - goldstone excitations . we then proceeded to explicitly compute the spectral properties of transverse and longitudinal gluons of adjoint color 8 . the spectral density of the longitudinal mode now exhibits a well - behaved plasmon branch with the correct low - momentum limit @xmath212 . moreover , the spectral density vanishes for gluon energies and momenta corresponding to the dispersion relation for nambu - goldstone excitations . we have thus amended and corrected previous results presented in ref . @xcite . our results pose one final question : using the correct expression for the longitudinal self - energy of adjoint colors @xmath301 , do the values of the debye masses derived in ref . @xcite change ? the answer is `` no '' . in the limit @xmath302 , application of eqs . ( 120 ) , ( 124 ) , and ( 129 ) of ref . @xcite to eq . ( [ hatpi00aa ] ) yields @xmath303 , and the results of ref . @xcite for the debye masses remain valid . we thank g. carter , d. diakonov , and r.d . pisarski for discussions . we thank r.d . pisarski in particular for a critical reading of the manuscript and for the suggestion to use t hooft gauge to decouple meson and gluon modes . thanks the nuclear theory groups at bnl and columbia university for their hospitality during a visit where part of this work was done . he also gratefully acknowledges continuing access to the computing facilities of columbia university s nuclear theory group . i.a.s . would like to thank the members of the institut fr theoretische physik at the johann wolfgang goethe - universitt for their hospitality , where part of this work was done . the work of i.a.s . was supported by the u.s.department of energy grant no . de - fg02 - 87er40328 . d. bailin and a. love , phys . rept . * 107 * , 325 ( 1984 ) . m. alford , k. rajagopal , and f. wilczek , phys . b * 422 * , 247 ( 1998 ) ; r. rapp , t. schfer , e.v . shuryak , and m. velkovsky , phys . lett . * 81 * , 53 ( 1998 ) . t. schfer and f. wilczek , phys . d * 60 * , 074014 ( 1999 ) . m. le bellac , _ thermal field theory _ ( cambridge , cambridge university press , 1996 ) . rischke , phys . d * 64 * , 094003 ( 2001 ) . miransky , i.a . shovkovy , and l.c.r . wijewardhana , phys . d * 64 * , 096002 ( 2001 ) . gusynin and i.a . shovkovy , nucl . a * 700 * , 577 ( 2002 ) . miransky , i.a . shovkovy , and l.c.r . wijewardhana , phys . b * 468 * , 270 ( 1999 ) . pisarski and d.h . rischke , phys . d * 61 * , 074017 ( 2000 ) . t. schfer and f. wilczek , phys . d * 60 * , 114033 ( 1999 ) . hong , v.a . miransky , i.a . shovkovy , and l.c.r . wijewardhana , phys . d * 61 * , 056001 ( 2000 ) [ erratum - ibid . d * 62 * , 059903 ( 2000 ) ] . pisarski and d.h . rischke , phys . d * 60 * , 094013 ( 1999 ) . r. casalbuoni , z. duan , and f. sannino , phys . d * 62 * , 094004 ( 2000 ) . k. zarembo , phys . d * 62 * , 054003 ( 2000 ) .

the interaction of fluid / fluid interfaces with solid boundaries is of fundamental importance to a variety of wetting and dewetting phenomena . in this paper , we present a computational framework for the inclusion of a general fluid / solid interaction , treated as a temporally and spatially dependent body force , in a direct solver of the navier - stokes equations . this approach allows for computing fluid wetting properties ( such as equilibrium contact angle ) based on first principles , and without restriction to small contact angles . due to the complexities involved in modeling dynamics of fluids on solid substrates , a significant amount of modeling and computational work has been carried out using the long - wave ( lubrication ) approach . still , even within the long - wave approach , a difficulty arises when employing the commonly used no - slip boundary condition at the fluid / solid interface : a non - integrable shear - stress singularity at the moving contact line . simulating dynamic contact lines therefore requires additional ingredients for the model . one option is to include fluid / solid interaction forces with conjoining - disjoining terms which lead to a prewetted ( often called ` precursor ' ) layer in nominally ` dry ' regions . this approach effectively removes the ` true ' contact line , consequently avoiding the associated singularity @xcite . a second approach is to relax the no - slip condition and instead assume the presence of slip at the fluid / solid interface . both slip and disjoining pressure approaches have been extensively used to model a variety of problems including wetting , dewetting , film breakup , and many others ( see e.g. @xcite for reviews ) . while the approach based on the long - wave model has been very successful , it does include limitations inherent in its formulation : in particular , the restriction to small interfacial slopes ( strictly speaking , the slopes much less than unity ) , and therefore small contact angles . we have shown in our earlier work @xcite that , depending on the choice of flow geometry , the comparison between the solutions of the long - wave model and of the navier - stokes equations may be better than expected ; however , still for slopes of @xmath0 , quantitative agreement disappears . therefore , one would like to be able to consider wetting / dewetting problems by working outside of the long - wave limit , while still considering the most important physical effects ; such as fluid / solid interaction forces . these fluid / solid interactions are known to be crucial in determining stability properties of a fluid film ; without their presence , a fluid film on a substrate is stable , since there are no forces in the model to destabilize it . in particular , for thin nanoscale films , fluid / solid interaction forces may be dominant . we note that the approaches based on the derjaguin approximation to include the van der waals or electrostatic interactions into the model in the form of a local pressure contribution ( disjoining pressure ) acting on the fluid / solid interface are derived under the assumption of a flat film @xcite . therefore these approaches can not be trivially extended to the configurations involving large contact angles . in the context of direct simulations of the navier - stokes equations with free interfaces , a large variety of methods are used to track the evolution of the interface . lagrangian methods conform the computational grid to the interface ( e.g. @xcite ) . eulerian methods require a separate mechanism to track the interface location ; these include front tracking methods ( e.g. @xcite ) , and interface capturing methods such as volume of fluid methods and level set methods . the latter two methods easily treat topology changes , and with recent developments have been shown to be effective for simulating surface tension driven flows @xcite . a common feature of volume of fluid methods is that contact angles are imposed geometrically , in that the angle at which the interface intersects the solid substrate is specified as a boundary condition on the interface @xcite . such approaches were used to model the dynamics of non - wetting drops , that could even detach from the substrate @xcite , as well as spreading drops @xcite . the van der waals interaction has been implemented previously in a volume of fluid based solver for the liquid / liquid interaction of colliding droplets @xcite , but to our knowledge has not been considered for flows involving wetting phenomena . a variety of other computational methods have been considered in the context of wetting / dewetting . here we mention phase - field methods that treat two fluids with a diffuse interface by means of a smooth concentration function , which typically satisfies the cahn - hilliard or allen - cahn equations , and is coupled to the navier - stokes equations . jacqmin @xcite describes a phase - field contact angle model that use a wall energy to determine the value of the normal derivative of the concentration on a solid substrate . this model has been used to study contact line dynamics @xcite , and similar models have been considered in the investigation of the sharp interface limit of the diffuse interface model @xcite . lattice - boltzmann methods have also treated the contact angle with a wall energy contribution @xcite . these approaches have explained a variety of phenomena related to spreading of fluids on solid substrates , but do not consider explicitly the stabilizing and destabilizing forces between fluid and solid , as has been done via disjoining pressure within the context of the long - wave model . the liquid / solid interaction is naturally included in molecular dynamics ( md ) simulations @xcite that typically consider lennard jones potential between fluid and solid particles . however , md simulations are , in general , computationally expensive , even when simulating nanoscale systems . one would like to be able to include liquid - solid interaction within the framework of a continuum model . here , we present a novel approach , based on a volume of fluid formulation , which includes the fluid / solid interaction forces into the governing navier - stokes equations , without limitations inherent in the long - wave model . this inclusion allows for arbitrary contact angles to be incorporated based on modeling the underlying physics , in contrast to conventional volume of fluid methods . the presented approach also leads to the regularization of the viscous stress since the fluid film thickness never becomes zero . furthermore , our framework can account for additional physical effects , such as instability and breakup of thin fluid films , that would not be described if fluid / solid interaction forces were not explicitly included . we note here that while film rupture can also occur in phase - field based approaches ( as in @xcite ) , this seems to be due to the presence of a rather thick interface , and not due to the explicit inclusion of destabilizing liquid / solid interaction forces . in the present paper we focus on formulating and discretizing the model , and on discussing issues related to convergence and accuracy . to validate our proposed numerical scheme , we consider two representative examples , involving relaxation and spreading of sessile drops with various contact angles on a substrate . these benchmark cases permit comparison of our results with well established analytical solutions for a particular flow regime . the application of the method to the study of thin film stability including dewetting will be considered in the sequel @xcite . the rest of this paper is organized as follows . we describe the details of the fluid / solid interaction in sec . [ sec : model ] . in sec . [ sec : methods ] , we describe two finite - volume methods for the discretization of the considered fluid / solid interaction forces . the presentation in these two sections applies to any generic fluids . in sec . [ sec : results ] we present a comparison of the two considered discretization methods for equilibrium and spreading drops , for a particular choice of material parameters . in sec . [ sec : conc ] , we give an overview and future outlook . consider a perfectly flat solid substrate covered by two immiscible fluids . for clarity , we will refer to these fluids as the liquid phase ( subscript @xmath1 ) , and the vapor phase ( subscript @xmath2 ) , although the present formulation applies to any two fluids . assume that gravity can be neglected , and also ignore any phase - change effects , such as evaporation and condensation . there are three relevant interfacial energies : the liquid / solid , @xmath3 , the vapor / solid , @xmath4 , and the liquid / vapor , @xmath5 , energies . the contact angle is commonly defined as the angle between the tangent plane of the interface between the liquid and vapor phases and the solid substrate at the point where the interface meets the surface . at equilibrium , the contact angle , @xmath6 , and the surface energies are related by young s equation @xcite : @xmath7 if there is a nonzero contact angle at equilibrium , the liquid partially wets the solid surface ; if the equilibrium configuration is a flat layer covering the whole substrate , then it fully wets the solid surface . it is also possible for the liquid to be non - wetting , where the liquid beads up into a sphere on the surface . the wetting behavior of the system can be characterized by the equilibrium spreading coefficient , defined by : @xmath8 which expresses the difference in energy per unit area between a surface with no liquid , and one with a layer of liquid ( what we call ` dry ' and ` wet ' states , respectively ) . wetting is determined by the sign of @xmath9 ; partial wetting occurs for @xmath10 , and complete wetting for @xmath11 @xcite . the above characterization of the contact angle is straightforward for static configurations and at macroscopic length scales . if these assumptions are not satisfied , definitions of contact angles become more complex . in the literature , a distinction is made between the apparent contact angle , @xmath12 , and the microscopic contact angle , @xmath13 , distinguished by the distance from the contact line at which the measurement is made @xcite . the contact angle resulting from measuring on macroscopic length scales is @xmath12 , while @xmath13 is measured at short length scales which are still long enough so that the continuum limit is appropriate @xcite . the microscopic contact angle , @xmath13 , is often identified with @xmath6 , which is commonly used in the derivation of spreading laws , such as the classical cox - voinov law @xcite . it should be noted that the details of the fluid behavior on nano scales in the vicinity of the fluid fronts and associated contact lines are far from being completely understood @xcite , and the way in which the contact angle arises at small scales is a subject of ongoing research @xcite . the surface energies entering eq . and eq . arise due to the van der waals interaction between the different phases that are relevant on short length scales . three kinds of van der waals interactions are usually considered : interactions between polar molecules , interaction of molecules that have an induced polarization , and the dispersion forces @xcite . the dispersion interaction is relevant for all molecules , and will be the only one that we consider . a common model for approximating the dispersion interaction of two particles , of phases @xmath14 and @xmath15 , centered at at @xmath16 and @xmath17 is the 12 - 6 lennard jones potential @xcite : @xmath18 this potential has a minimum , @xmath19 , at @xmath20 , and for simplicity we assume that @xmath21 is a constant for any two interacting particles . the center distance between the two particles is given by : @xmath22 the powers @xmath23 and @xmath24 in eq . correspond to short range repulsion and long range attraction , respectively . for our purposes , we generalize this formulation and use the following form : @xmath25 here @xmath26 is the scale of the potential well , having units of energy per particle pair interaction . in general , we require only that @xmath27 are integers , such that @xmath28 , for reasons that will become clear shortly . our model considers two fluid phases , a liquid phase and a vapor phase occupying the region @xmath29 , interacting with a flat , half infinite , solid substrate ( subscript @xmath30 ) in the region @xmath31 . consider now a particle located at @xmath32 of phase @xmath14 ( where @xmath14 is either @xmath1 or @xmath2 ) . the interaction energy between this particle and the substrate per unit volume of the substrate is : @xmath33 where @xmath34 is the particle density in the substrate . we derive the force per unit volume following a similar procedure outlined in @xcite . integrating eq . over the region @xmath31 yields the following total interaction of a particle in phase @xmath14 with the substrate : @xmath35\end{gathered}\ ] ] although the van der waals interaction is not strictly additive , the effects due to non - additivity are usually weak @xcite , and we ignore them for simplicity . multiplying by @xmath36 , the particle density in phase @xmath14 , we obtain the van der waals interaction per unit volume of phase @xmath14 : @xmath37\end{gathered}\ ] ] we introduce the following parameters : @xmath38^{q-3}}{\left[(q-2)(q-3)\right]^{p-3}}\right)^{\frac{1}{p - q}}\ ] ] @xmath39^\frac{1}{p - q}\sigma\ ] ] @xmath40 this simplifies eq . to : @xmath41\ ] ] the quantity @xmath42 is referred to as the ` equilibrium film thickness ' and will be considered in more detail below . note that eq . has an identical form as eq . . the force per unit volume on phase @xmath14 that results from the potential is computed by taking the gradient of eq . : @xmath43\hat{y}\ ] ] here @xmath44 refers to the unit vector @xmath45 . we proceed to derive an expression for @xmath6 in terms of @xmath46 and @xmath47 in eq . according to the microscopic arguments outlined in @xcite . combining eqs . and , we obtain the following expression for the equilibrium spreading coefficient : @xmath48 where @xmath9 is the difference in the energy per unit area between a dry system and a wet system . the total energy required to remove a liquid layer originally occupying @xmath49 and replace it with a layer of the vapor phase is given by : @xmath50 if @xmath51 is the smallest distance between particles of the substrate and fluids when they are in contact , eq . ( [ eq : energy_diff ] ) specifies the total energy required to completely remove the liquid from the substrate . even if @xmath51 takes a larger value , eq . consistently describes the energy difference between a wet system , and a dry system which consists of a fluid layer of thickness @xmath51 wetting the solid substrate . performing the integration in eq . , we obtain : @xmath52\ ] ] in order to be in a partial wetting regime , i.e. where there is a non - zero contact angle , we require @xmath10 . based on eq . , we find : 1 . @xmath53 : in this case , the attractive term in eq . dominates . this leads to a negative spreading coefficient only if @xmath54 , so that the vapor phase must experience a greater attraction than the liquid phase in order to be in a partial wetting regime . @xmath55 : in this case , the repulsive term in eq . dominates . thus partial wetting is possible when @xmath56 . the first case states that when @xmath51 is large , the vapor phase must interact more strongly with the substrate than the liquid phase ; the reverse is true when @xmath51 is small . for example , the second case must hold for partial wetting of a droplet surrounded by a vacuum . for comparison , we briefly describe a similar method of contact angle implementation used in the long - wave theory @xcite . in its most basic form , the disjoining pressure modifies the pressure jump across the interface of a thin , flat film , and can be derived as a macroscopic consequence of the van der waals interaction @xcite . for a flat film of thickness @xmath57 , the difference between the pressure in the exterior vapor phase , @xmath58 , and the liquid phase , @xmath59 , is given by : @xmath60.\ ] ] the definition of the parameter @xmath42 is identical to that of eq . . in the long - wave literature , this quantity is identified with an equilibrium film thickness , such that in the case of a drop in the partial wetting regime , there is an additional film of thickness @xmath42 completely wetting the substrate . following the example of disjoining pressure above , we consider situations such that there is a layer of thickness @xmath42 completely covering the surface even when the fluid is partially wetting . in particular , this means that we assume that @xmath61 , so that in eq . , the liquid is removed only to a thickness @xmath42 . this value of @xmath51 has the convenient property that it allows for eq . to be simplified , while the presence of a wetting layer over the whole substrate removes the contact line singularity . plugging @xmath62 into eq . , and substituting in the expression for @xmath9 from eq . , we obtain the following expression for the difference between @xmath47 and @xmath46 as a function of @xmath6 : @xmath63 therefore , only the difference @xmath64 is needed to specify @xmath6 . similar formulations appear in the long - wave literature @xcite . note that since eq . assumes that the fluid occupies a half infinite domain in the wet state , eq . is satisfied exactly only in the limit where @xmath42 is vanishingly small relative to the drop thickness . to summarize , in this section we have formulated a method that allows for the inclusion of fluid / solid interaction forces without the limitations inherent in long - wave theory , such as negligible inertia and small interfacial slope . next we proceed to discuss numerical implementation . equations and present a difficulty in that they diverge as @xmath65 . this can be dealt with by introducing a cutoff length , i.e. forcing @xmath66 when @xmath67 is less than the cutoff . this method has the drawback of making the force discontinuous , and may lead to poor numerical convergence . for this reason , we introduce a shifted potential : @xmath68\ ] ] where @xmath69 is some parameter less than @xmath42 . equation removes the singular portion of the potential near @xmath65 for a small @xmath69 , and is equal to zero at @xmath70 , but otherwise keeps the same functional form . we refer to films occupying @xmath71 as equilibrium films , and @xmath72 as the equilibrium film thickness . the force corresponding to eq . is given by @xmath73\hat{y}\ ] ] we can express the total body force due to the van der waals interaction more compactly by introducing a characteristic function , @xmath74 , which takes the value @xmath75 inside the liquid phase and @xmath76 elsewhere : @xmath77 where the interaction strength explicitly depends on the phase through @xmath74 , so that @xmath78 and @xmath79 . we solve the full two - phase navier - stokes equations subject to the force given by eq . : @xmath80 + \gamma \kappa \delta_s \textbf{n } \\+ \hat{\textbf{f}}_b(y)\end{gathered}\ ] ] @xmath81 the phase dependent density and viscosity are given by @xmath82 and @xmath83 . the velocity field is @xmath84 , and @xmath85 is the pressure . the force due to the surface tension is included as a singular body force in the third term on the right hand side of eq . ( see @xcite ) , where @xmath86 is a delta function centered on the interface , @xmath87 is the curvature , and @xmath88 is the inward pointing unit normal of the interface . letting @xmath89 be the length scale , and @xmath90 the time scale , we define the following dimensionless variables : @xmath91 with these scales , and dropping the tildes , the dimensionless navier - stokes equations are as follows : @xmath92 + \kappa \delta_s \textbf{n } \\ + \textbf{f}_b(y)\end{gathered}\ ] ] @xmath93 where @xmath94\hat{y}\ ] ] here , the dimensionless @xmath67-coordinate is translated , so that @xmath95 , and the equilibrium film occupies @xmath96 . the ( phase - dependent ) dimensionless numbers are the weber number , @xmath97 , the capillary number , @xmath98 , and the force scale for the van der waals interaction , @xmath99 , given by the following : @xmath100 upon nondimensionalization , eq . yields the following expression for @xmath6 : @xmath101 has a corresponding size @xmath102 . the red curve shows the piecewise linear reconstruction of the liquid / vapor interface obtained using the volume of fluid method . ] the system of eqs . - , is solved using the open - source package gerris @xcite , described in detail in @xcite . in gerris , the navier - stokes equations are solved using a finite - volume based projection method , with the interface between the two fluids tracked using the volume of fluid method . the fluid domain is discretized into a quad - tree of square cells ( see fig . [ fig : tree_diagram ] ) ; note that we consider the implementation in two spatial dimensions from now onward . the finite - volume method treats each cell @xmath103 as a control volume , so that the variables associated with @xmath103 represent the volume averaged value of the variable over @xmath103 . each cell has a size @xmath102 , and center coordinates @xmath104 . the volume of fluid method tracks the interface implicitly by defining a volume fraction function , @xmath105 , which gives the fraction of the volume of @xmath103 occupied by the liquid phase . @xmath105 is advected by the fluid flow according to the following equation : @xmath106 the value of @xmath97 and @xmath98 in a cell @xmath103 are then represented using an average : @xmath107 here , as before , subscript @xmath1 corresponds to liquid , and @xmath2 to the vapor phase . the volume of fluid method reconstructs a sharp interface . for each cell , the gradient , @xmath108 , is computed using the mixed - young s method @xcite , so that the unit normal vector is given by @xmath109 . this permits a linear reconstruction of the interface in each cell , according to the equation @xmath110 , where @xmath111 is determined by @xmath112 ( see @xcite for details ) . the exact average of the force over the computational cell , @xmath103 , is given by : @xmath113\mathcal{f}(y ) d\mathcal{c}\,\hat{y}\end{gathered}\ ] ] where @xmath114 is defined by eq . . we detail two possibilities for the discretization of the van der waals force term in eq . . for both of these methods , the force is included explicitly in the predictor step of the projection method . * method i : * our first method proceeds by a simple second order discretization of eq . : d\mathcal{c}\,\hat{y}\\ & = \left[\frac{\mathcal{f}(y_c)}{\delta_c^2}k_{ls } \int\int_\mathcal{c } \chi(x , y)d\mathcal{c } + \frac{\mathcal{f}(y_c)}{\delta_c^2}k_{vs } \int\int_\mathcal{c } ( 1-\chi(x , y))d\mathcal{c}\right]\,\hat{y}\\ \end{split}\ ] ] we identify the average of @xmath74 over @xmath103 with @xmath105 , yielding the following expression : @xmath116 \mathcal{f}(y_c)\,\hat{y}\ ] ] however , we note that the accuracy of this simple method can deteriorate at low mesh resolutions because @xmath117 has a large gradient as @xmath65 . consider a cell such that @xmath118 . to a first approximation , the error in eq . is given by : @xmath119 to show that this can lead to large errors , consider the error when @xmath120 , in a cell with center at @xmath121 , i.e. for a cell along the bottom boundary when @xmath42 is just barely resolved . the error in this cell is : @xmath122 the lower bound of this error can be shown to be : @xmath123 since @xmath124 , we can bound this further by : @xmath125 note that @xmath126 is not the discretization error , but the error found for a fixed grid size . recalling that eq . implies that @xmath127 is proportional to @xmath128 , so if @xmath69 is significantly smaller than @xmath42 the errors are quite large unless @xmath129 is much smaller than @xmath42 . furthermore , @xmath42 is generally small relative to the length scale of the droplet considered in eq . , so that the mesh size required to accurately resolve the right hand side of eq . will increase the computational cost . 0.4 : ( a ) cut cell divided into regions @xmath130 and @xmath131 entirely occupied by the liquid and vapor phases , respectively , and ( b ) illustration of the regions of integration of a cut cell . the linear reconstruction of the interface from the volume of fluid method is shown by the red line.,title="fig:",width=192 ] 0.4 : ( a ) cut cell divided into regions @xmath130 and @xmath131 entirely occupied by the liquid and vapor phases , respectively , and ( b ) illustration of the regions of integration of a cut cell . the linear reconstruction of the interface from the volume of fluid method is shown by the red line.,title="fig:",width=192 ] * method ii : * since eq . gives an exact formula for the force per unit volume in a single phase , much of the simplification used in deriving eq . is unnecessary . for the second method , we take advantage of this fact to more accurately approximate eq . . first , note that it is trivial to integrate eq . when @xmath118 or @xmath132 . moreover , the volume of fluid method gives additional information beyond just the fraction of the cell occupied by the liquid phase ; using the reconstructed interface , we have an approximation of the portion of the cell which is occupied by the liquid phase as well . therefore , we introduce the following alternative method . consider a cell @xmath103 with center @xmath104 , where @xmath133 , so that the volume of fluid method yields a linear reconstruction of the interface in the cell . this interface divides the cell into a region occupied by the liquid phase , @xmath130 , and a region occupied by the vapor phase , @xmath131 ( see fig . [ fig : cut_cell ] ) . we can write a general algorithm to integrate over these regions , yielding the following expression for the average force on cell @xmath103 : @xmath134\hat{y}\ ] ] we briefly outline the general process to perform the integrations over @xmath130 and @xmath131 . let the volume of fluid reconstructed interface in the cell be given by @xmath135 , where @xmath136 is a normal pointing into the liquid phase . we only present the problem of integrating over @xmath130 , since @xmath131 involves an analogous procedure . if @xmath137 , for a small tolerance @xmath138 , the interface is approximately horizontal ; thus the integration of @xmath114 over @xmath130 becomes : @xmath139 on the other hand , if @xmath140 , the interface is approximately vertical , and the integration over @xmath130 becomes : @xmath141 we can now proceed to provide general formulas when @xmath142 and @xmath143 . since @xmath114 is independent of @xmath144 , the sign of the interfacial slope is irrelevant , so , without loss of generality , suppose that @xmath145 . define the following values : @xmath146 if @xmath147 , the integration over @xmath130 can be expressed as follows : @xmath148\end{gathered}\ ] ] on the other hand , if @xmath149 , the integration over @xmath130 is given by : @xmath150\end{gathered}\ ] ] thus , in two dimensions , the general process of integration reduces to integration over two regions . note again that @xmath114 is known exactly , so that the integrals in eqs . - can be computed exactly , in contrast to the large error for method i in eq . . therefore , we conclude that regarding approximation of eq . , method ii is superior . for example , for a single phase fluid , or when the interface is flat , eq . we will consider both method i and method ii in what follows and discuss their performance . our computational domain is rectangular , with @xmath151 and @xmath152 . throughout the remainder of the paper , we will impose a homogeneous neumann boundary condition on all boundaries for the pressure , @xmath85 . for the velocity field , we will apply a homogeneous neumann boundary condition at all boundaries except @xmath153 , where we will apply one of the following two boundary conditions ; the no - slip , no - penetration condition by setting @xmath154 to @xmath76 on the substrate : @xmath155 or a free - slip condition by : @xmath156 the boundary condition for the van der waals force is straightforward . at @xmath153 , we set @xmath157 at @xmath158 , we set @xmath159 the boundary condition for the volume fraction , @xmath112 , is again homogeneous neumann on all boundaries , except on the bottom boundary , where we apply @xmath160 which is equivalent to taking the bottom boundary to always be wetted with the liquid phase . in this section , we consider simulations of the full navier - stokes equations in which contact angles have been imposed using the van der waals force , which is included using methods i and ii described in sec . [ sec : methods ] . our simulation setup consists of a drop on an equilibrium film , initially at rest , which then relaxes to equilibrium under the influence of the van der waals force . we compare methods i and ii for a drop which is initially close to its equilibrium contact angle , as predicted by eq . , as well as for a spreading drop which is initially far from its equilibrium . furthermore , we consider the effect of equilibrium film thickness for small , intermediate , and large contact angles , using method ii . in simulations , droplets are surrounded by an equilibrium film of thickness @xmath161 defined by @xmath162 , where @xmath69 is the amount the force is translated by in eq . . the initial profile is given by the following : @xmath163 here @xmath164 is chosen so that the total area of the circular cap is equal to @xmath165 . the quantity @xmath166 is the initial contact angle of the droplet . in all simulations , we fix the ratio of these length scales so that @xmath167 . various values of the exponents @xmath168 and @xmath169 in eq . can be found in the literature , in particular @xmath170 have been used in the context of the disjoining pressure in thin films @xcite , the former corresponding to the 12 - 6 lennard jones potential . in this paper we will restrict our attention to @xmath171 , which was shown to lead to favorable agreement with experiments involving liquid metal films @xcite . different exponents may affect the structure of the contact line region and the pressure distribution , but do not impact significantly the results presented here . there are four contact angles that we consider in this section . the initial contact angle , @xmath166 , as in eq . , specifies the angle formed by the tangent of the initial circular profile with the equilibrium film at the point where they meet . the ( time dependent ) apparent contact angle is denoted @xmath172 , and the numerical equilibrium angle @xmath173 is the value of @xmath172 when the system under consideration is in equilibrium ; @xmath172 is measured by a circle fit procedure described below . finally , we refer to the imposed contact angle , @xmath6 , which we use to specify @xmath174 via eq . ; the only relevant quantity is the difference between @xmath175 and @xmath127 , but for definiteness , we set @xmath176 and @xmath177 . the value of @xmath6 differs from @xmath173 in simulations because @xmath42 is small but non negligible relative to the thickness of the drop . we measure @xmath172 according to the following procedure . a drop has a circular cap shape that transitions smoothly to the equilibrium film . we define the contact angle in this context in the following way : let @xmath178 represent the points of inflection of the drop profile . we perform a least squares fit of a circle to the profile over the interval @xmath179 . the point at which the fitted circle intersects the equilibrium film is called the contact point ( in three dimensions , this is the contact line ) . @xmath172 is measured as the angle the fitted circle makes with the equilibrium film at the contact point . we first investigate a static drop , with @xmath180 and @xmath181 , which is then allowed to relax to its equilibrium shape . the dimensionless numbers are set to be @xmath182 . we set the material parameters to be the same in both fluid phases in order to restrict our attention to a smaller parameter space , so that we can focus on the properties of methods i and ii . note that since the expression determining @xmath6 , eq . , does not depend on the fluid parameters , this choice does not affect the equilibrium shape . the computational domain is @xmath183 \times [ h_c,1+h_c]$ ] ; the domain is resolved at a constant resolution . a no - slip boundary condition is imposed on the solid boundary , @xmath153 . , @xmath180 , at equilibrium . the pressure is normalized by the ( non - dimensional ) capillary pressure of the equilibrium drop , which is given by the curvature of the interface far from the contact point , where it is approximately circular . the solid black curve shows the volume of fluid reconstructed interface , and the dashed curve is the initial profile ; @xmath184 . ] figure [ fig : eq_drop_profile ] shows such a drop at equilibrium ( with the contact angle imposed using method ii ) . the equilibrium film has negative pressure with very high absolute value , and its thickness at equilibrium differs from @xmath185 by about @xmath186 . with this @xmath42 , there is a noticeable difference between the true equilibrium contact angle found by simulations , @xmath173 , and the imposed angle @xmath6 . the effect of @xmath42 on the equilibrium drop shape will be considered below . figure [ fig : l1_error ] shows the convergence for methods i and ii for a uniform mesh . simulations are run until time @xmath187 , with a constant time step of @xmath188 ; the stability constraint due to the explicit discretization of the surface tension dominates in this velocity regime , and this time step ensures that this constraint is satisfied for all resolutions we consider ( see @xcite ) . for two sets of volume fractions @xmath189 and @xmath190 such that @xmath190 is computed on a quad - tree @xmath191 with mesh - size @xmath192 , the @xmath193 norm is computed according to the following formula : @xmath194 norm of the difference between @xmath105 at @xmath195 and @xmath184 , for a drop at equilibrium with @xmath196 , @xmath180 , with a uniform mesh size @xmath195 . ] in this case @xmath197 , and @xmath198 is equal to @xmath189 on the largest cell containing @xmath103 . for reference , @xmath199 is about @xmath42 , so that the low resolution case is not even resolving the equilibrium film . both methods perform comparably well at @xmath200 , with method i converging slightly faster . however , at low resolutions , @xmath199 , method ii performs significantly better . . initially , @xmath181 , and the drop spreads to its equilibrium configuration , defined by @xmath201 . the initial profile is shown by the dashed curve . as in fig . [ fig : eq_drop_profile ] , the pressure is normalized by the capillary pressure of the equilibrium drop ; @xmath184 . ] . ] . ] we now move onto the question of how a drop behaves when @xmath166 is far from @xmath173 . the initial shape of the drop is as above , with @xmath180 , @xmath202 , so that @xmath181 . at @xmath153 , we use free - slip since it allows the drop to reach its steady state more quickly , reducing computational time . the simulation domain is @xmath203 \times [ h_c,1+h_c]$ ] . for these simulations , we will set @xmath204 . we note that even with this choice of parameters , the dimensionless contact point speed is sufficiently small so that surface tension effects still dominate over viscous effects ( or precisely speaking , the capillary number defined based on the speed of the contact point is still small ) . as before , we use the same material parameters for both phases , and note that , provided that the capillary number is small , varying the ratios between the phases will only affect the relaxation time ; we will expand on this point below . here , we impose a small contact angle of @xmath205 . unlike the above , we use an adaptive mesh which refines regions to a level of @xmath206 near the liquid / vapor interface , or if there is a high gradient in @xmath207 . we will vary @xmath206 in what follows to study the convergence with respect to the maximum resolution . figure [ fig : spread_profile ] shows the initial and equilibrium profiles and pressure distribution when the drop reaches equilibrium , computed using method ii . the pressure inside the liquid phase is near unity , while the pressure in the vapor phase just above the equilibrium film is five times the capillary pressure . the equilibrium contact angle in fig . [ fig : spread_profile ] is @xmath208 . to equilibrium , with an imposed contact angle of @xmath205 using method ii . the black ( dashed ) line is proportional to @xmath209 , and the agreement with the blue ( solid ) curve shows that the drop spreading approximately satisfies the cox - voinov law . ] figure [ fig : front_plots ] shows the front locations for the spreading drops as a function of time for method i and method ii . both methods show broadly similar results , with method ii appearing to slightly outperform method i in terms of convergence . in order to compute the convergence , we calculate the error as : @xmath210 here @xmath211 is the front location computed with @xmath212 , and @xmath213 is the front location computed with @xmath214 . figure [ fig : spread_error ] shows the errors . method ii displays significantly better convergence for the front location as a function of time . for various values of @xmath42 . the black ( dotted ) profile shows the initial condition for @xmath215 . we plot @xmath216 so that for each curve the @xmath67 range is @xmath217 . ] we briefly compare the qualitative behavior of the spreading drop to the well known cox - voinov law @xcite . for a drop displacing another immiscible fluid on a solid surface , the speed of the contact point , @xmath209 , is related to @xmath166 and @xmath173 to leading order by @xcite : @xmath218 note that eq . is derived under the assumption that @xmath219 , i.e. that the capillary number defined using the front velocity is small . provided that one is in this regime , the choice of material parameters only impacts the constant of proportionality in eq . . figure [ fig : cox_voinov ] shows @xmath209 versus @xmath220 using method ii , for @xmath221 . the blue ( solid ) line shows the numerical results , and the black ( dashed ) line shows the slope expected if the cox - voinov law is obeyed . we see that after initial transients @xmath209 decreases with @xmath220 , and the drop spreading approximately satisfies the cox - voinov law for @xmath222 . for ( a ) @xmath223 and ( b ) @xmath224 . the black ( dotted ) profile shows the initial condition for @xmath215 . as in fig . [ fig : conv_hstar_profiles ] , we plot @xmath216 so that for each curve the @xmath67 range is @xmath217 . ] for ( a ) @xmath223 and ( b ) @xmath224 . the black ( dotted ) profile shows the initial condition for @xmath215 . as in fig . [ fig : conv_hstar_profiles ] , we plot @xmath216 so that for each curve the @xmath67 range is @xmath217 . ] next , we turn our attention to the precise value of the contact angle . as shown in figs . [ fig : eq_drop_profile ] and [ fig : spread_profile ] , the actual equilibrium contact angle , @xmath173 , is generally smaller than the imposed angle @xmath6 . this is due to the fact that eq . is derived under the assumption of small @xmath42 , while in our simulations the ratio of drop radius and @xmath42 is about 25 . to confirm the above statement , we analyze in more detail the resulting contact angles as @xmath225 is varied . we set @xmath196 using method ii , and the resolution is fixed at a uniform @xmath184 . the value of @xmath42 varies over @xmath226 , @xmath227 , and @xmath228 . the initial condition is imposed with @xmath181 as in eq . . the drop is again permitted to relax to its equilibrium with a fixed time step for 1.75 units of time . figure [ fig : conv_hstar_profiles ] shows the equilibrium profiles for various values of @xmath42 . the black ( dotted ) profile is the initial condition for @xmath215 , and is included as a reference . as @xmath42 is decreased , the equilibrium profiles are characterized by contact angles closer to @xmath6 . .dependence of @xmath173 on @xmath42 . we calculate @xmath173 using a circle fit . the third column gives the relative difference between @xmath173 and @xmath6 . the fourth column is the @xmath193 norm of the difference between the initial volume fractions @xmath229 and the equilibrium @xmath230 . [ cols="<,<,<,<",options="header " , ] we quantify the dependence of @xmath173 on @xmath42 in table [ tab : theta ] . as before , we compute the contact angle @xmath173 using a circle fit of the drop after @xmath231 units of time , at a fixed time step for all simulations . as @xmath42 is reduced , the calculated @xmath173 approaches @xmath196 . the relative difference between @xmath173 and @xmath6 reduces with @xmath42 approximately linearly . these measures however depend on the accuracy of the estimation of @xmath173 ; to analyze the convergence more directly , we compare the volume fractions of the initial condition , @xmath229 , with the equilibrium state @xmath230 , using an @xmath193 norm computed as eq . . initially , @xmath232 , so this comparison provides a measure of the difference between @xmath173 and @xmath6 . this difference again decreases approximately linearly with @xmath225 . finally , we consider the effects of varying @xmath42 for @xmath6 other than @xmath233 . figure [ fig : dep_hstar_pi4 ] shows the equilibrium drop profiles for @xmath234 and varying @xmath42 , and fig . [ fig : dep_hstar_3pi4 ] shows the same for @xmath235 . for @xmath223 , as @xmath42 decreases , the profile quickly converges to the initial condition and hence the contact angle to @xmath6 , as seen by the comparison of the black ( dotted ) curve with that of @xmath236 in fig . [ fig : dep_hstar_pi4 ] . for @xmath224 , it can be seen that even for @xmath236 , the drop profile still shows some difference between @xmath173 and @xmath6 . nonetheless , this plot demonstrates a central advantage of our methods : @xmath173 larger than @xmath233 can be simulated with the van der waals force . in this paper , we have described a novel approach for including the fluid / solid interaction forces , into a direct solver of the navier - stokes equations with a volume of fluid interface tracking method . the model does not restrict the contact angles to be small , and therefore , can be used to accurately model wetting and dewetting of fluids on substrates characterized by arbitrary contact angles . we study the problem of a two - dimensional drop on a substrate and compare the results with the cox - voinov law for drop spreading . these validations and results illustrate the applicability of our proposed method to model flow problems involving contact lines . furthermore , our approach has the desirable property of regularizing the viscous stress singularity at a moving contact line since it naturally introduces an equilibrium fluid film . we have considered two alternative finite - volume discretizations of the van der waals force term that enters the governing equations to include the fluid / solid interaction forces . these two methods are complementary in terms of their accuracy and the ease of use , and therefore the choice of the method could be governed by the desired features of the results . in particular , method i discretizations can be straightforwardly extended to the third spatial dimension . however , we have shown that , when implementing method i , to guarantee accurate results , sufficient spatial resolution of the computational mesh must be used . we also show that method ii does not suffer accuracy deterioration at low mesh resolutions , and is therefore , superior to method i , and furthermore outperforms method i for spreading drops at all resolutions , albeit being more involved to implement . the presented approach opens the door for modeling problems that could not be described so far , in particular dewetting and associated film breakup for fluids characterized by large contact angles . furthermore , the model allows to study the effects of additional mechanisms , such as inertia . we will consider these problems in future work . this research was supported by the the national science foundation under grants nsf - dms-1320037 and cbet-1235710 . the authors acknowledge many useful discussions with javier diez of universidad nacional del centro de la provincia de buenos aires , argentina . 10 url # 1`#1`urlprefixhref # 1#2#2 # 1#1 b. derjaguin , l. landau , theory of the stability of strongly charged lyophobic sols and of the adhesion of strongly charged particles in solutions of electrolytes , acta physico chemica ussr 14 ( 1941 ) 633 . t. baer , r. cairncross , p. schunk , r. rao , p. sackinger , a finite element method for free surface flows of incompressible fluids in three dimensions . dynamic wetting lines , int . j. for num . meth . in fluids 33 ( 3 ) . m. m. francois , s. j. cummins , e. d. dendy , d. b. kothe , j. m. sicilian , m. w. williams , a balanced - force algorithm for continuous and sharp interfacial surface tension models within a volume tracking framework , j. comput . 213 ( 2006 ) 141 . t. nguyen , m. fuentes - cabrera , j. fowlkes , j. diez , a. gonzlez , l. kondic , p. rack , competition between collapse and breakup in nanometer - sized thin rings using molecular dynamics and continuum modeling , langmuir 28 . m. fuentes - cabrera , b. rhodes , j. fowlkes , a. lpez - benzanilla , h. terrones , m. simpson , p. rack , molecular dynamics study of the dewetting of copper on graphite and graphene : implications for nanoscale self - assembly , phys . e 83 ( 2011 ) 041603 .

350 has been the subject of many studies in the optical / near infrared and radio regimes and most wavelengths in between . it is highly obscured at short wavelengths and bright at long wavelengths . the object was originally part of the planetary nebulae survey of @xcite , but was later reclassified as an region . @xcite observed 350 at mid infrared wavelengths and showed the optical nebula was offset from the radio source due to a strong gradient in the foreground extinction . @xcite determined precise offsets between the radio , infrared and optical sources which validate the @xcite model . 350 , also known as the galactic radio source g70.31.6 , has been extensively observed at long wavelengths . @xcite provided a high angular resolution map at 2 cm which resolved the ionized gas into a clumpy arc like or shell like structure on scales smaller than 1@xmath1 . @xcite included this object in their catalog of regions , describing it as a core halo type while @xcite observed the larger nebula in the continuum and radio recombination line h76@xmath4 . the radio morphology is extended in two lobes to the nw and se and @xcite found a velocity gradient along the lobes which led them to suggest the structure was an ionized outflow extending 10 s of arcseconds from the center of the region . @xcite presented br@xmath3 and br@xmath4 emission line maps at greater than arcsecond resolution and followed this work @xcite with molecular line maps which they modeled as arising from a large scale molecular torus ( roughly perpendicular to the ionized outflow of * ? ? ? * ) . more recent work has centered on near infrared and mid infrared wavelengths and higher angular resolution data . @xcite observed the nebula in the @xmath0band using speckle techniques to provide an order of magnitude higher angular resolution than previous near infrared work , and @xcite used the subaru 8 m telescpe and mid infrared imager comics to obtain 0.5@xmath1 resolution images and spectra near 10 . in this paper , we present @xmath0band integral field spectroscopy of 350 obtained on the gemini north 8 m telescope with the near infrared integral field spectrograph . these data were taken in conjunction with the facility adaptive optics system and result in the most detailed data cube yet for the core of 350 . in an earlier paper @xcite , we presented similar data for the region g45.45@xmath50.6 . in that case , several massive stars were identified by photospheric lines and other objects showed signatures of buried massive young stellar objects . g45.45@xmath50.6 clearly harbors a rich cluster . @xcite make similar conclusions based on their eso vlt sinfoni integral field spectrograph observations for the region in the cluster [ dbs2003]8 . previous investigators have suggested the same may be true of 350 : @xcite , @xcite , and @xcite . our data identify multiple continuum sources in 350 and support the multiple star model , but we also show that the excitation in the central few arcseconds of 350 is likely dominated by a single object ( the unresolved continuum peak ) . in the following , the distance to the nebula is based on the work of @xcite . the @xcite distance is derived from a compilation of velocities and a rotation model ; harris distance needs to be reduced due to a revised distance to the galactic center ( * ? ? ? * used a distance of 10 kpc ) . adopting a value of 8 kpc gives a reduction by a factor of 0.8 resulting in a distance to the region of about 7 kpc . data were obtained with the near infrared integral field spectrograph ( nifs ) at the cassegrain focus of the gemini north fredrick c. gillett 8m telescope on mauna kea , hawaii on the night of 20 july , 2006 ( hst ) . nifs was used with the facility adaptive optics ( ao ) module altair in natural guide star ( ngs ) mode . nifs is fully described by @xcite ; see also @xcite for a more detailed description of similar observations with nifs as described here . briefly , nifs slices an approximately three arc second by three arc second field into 29 spectral segments of 0.1@xmath1 width ( the `` slit '' width ) and @xmath6 3@xmath1 in length . the scale along the spatial dimension is 0.043@xmath1 pix@xmath7 . the resulting `` spaxels , '' or spatial pixels , are thus 0.043@xmath1 @xmath8 0.1@xmath1 in size , and each contains a full spectrum covering one of the near infrared bands . in the present paper , @xmath0band spectra are presented for the galactic region 350 . the ao guide star used for altair is located 10.2@xmath1 to the south and west of 350 and has an @xmath9 magnitude of 12.7 according to the usno catalog ( source i d : u1200_14213688 ) . there was thin cirrus at the time of the observations , and the seeing ( at 5000 , corrected to zenith ) reported during the observations by altair was approximately 0.45@xmath1 . the observations were obtained at relatively high airmass ( 1.5 ) , and altair was run at 500 hz . the observations consisted of a single coadded frame taken on source and a second frame obtained on a nearby ( @xmath6 50@xmath1 west ) blank field . each frame had a total exposure time of 600 seconds ( 15@xmath840s ) on source and 600 seconds on sky . nifs was oriented with a position ( pa ) of 70 degrees east of north to correspond approximately to the orientation of the large scale ionized flow from 350 @xcite . this pa is aligned with the outflow direction . the spectral resolving power of nifs in the @xmath0band is @xmath10 5160 which results in a linear dispersion of 2.13 / pixel . this dispersion , combined with the large format array gives a full wavelength coverage at @xmath11 of about 4200 accounting for some minor truncation in the final data cube due to the systematic shift of wavelength in each slitlet from the staircase design of the image slicer . the line maps and spectra presented here were obtained using the gemini nifs iraf data reduction package ( version 1.9 ) . the nifs iraf package allows for full reduction to the `` image cube '' stage where a final cube has a roughly 60@xmath862 pixel image plane and a 2040 pixel spectral depth . first , raw images are prepared for reduction by standard gemini procedures that create the fits image variance and data quality extensions . next , the data are sky subtracted , flatfielded , rectified spatially , and wavelength calibrated ( see * ? ? ? * for a full description of these steps ) . the last two steps result in a uniform spatial spectral trace for each image slice row and a linear wavelength scale . it is important to remember that the final iraf data cube resamples each slitlet to two 0.05@xmath1 pixels for convenience ; the angular resolution in this dimension is still 0.1@xmath1 . a lamp image of ar and xe lines was used to determine the dispersion along each slice and as a function of the spatial dimension of each slice . a quadratic polynomial was used in this case which produced typical uncertainties in the position of a spectral line of approximately @xmath2 0.2 ( @xmath6 1/10 pixel ) . final spectra are interpolated to a linear wavelength solution . the spectra were next corrected for telluric absorption by division by the spectrum of an a0 v star ( hip 102568 ) . the spectrum of hip 102568 was corrected for intrinsic br@xmath3 absorption by fitting a voigt profile to the telluric spectrum . particular care was taken in the fit in order to remove this feature from the telluric standard as accurately as possible . the final wavelength calibrated , spatially rectified , and telluric corrected images were transformed into data cubes . the spatial pixel scale was resampled to 0.05@xmath1 in the fine dimension and block replicated to 0.05@xmath1 in the course dimension providing for a uniform scale as described in the previous section . line maps presented below were extracted from these cubes using nfmap . the zero point of the wavelength calibration is confirmed by the positions of oh airglow lines in the sky image . analysis of these lines shows the zero point is accurate to @xmath2 2.5 . the calibration results in an observed blueshift of the nebular lines of approximately 3.8 ( 1.8 pixels or @xmath1253 ) . lines are identified on a spectrum extracted from a 2.5@xmath1 diameter aperture centered on the nifs fov ; see figure [ spec ] . the spectrum is flux calibrated using the results of ( * ? ? ? * see below ) . the line positions and relative strengths compared to br@xmath3 for this integrated spectrum are given in table 1 . an image of the continuum near 21700 is shown in figure [ c217 ] along with the radio continuum image from @xcite . the continuum passband extracted was 9 pixels ( 19.2 ) wide . the appearance of sources in our continuum image is similar to the speckle @xmath0band image presented by @xcite who report an angular resolution of 0.11@xmath1 in their image derived from observations obtaddiained under 1.1@xmath1 seeing . @xcite identify 10 point sources in their image , though some of these appear more extended than others , and this is consistent with the continuum image shown in figure [ c217 ] . the brightest source in figure [ c217 ] is source # 1 of @xcite . the size of the most compact source in our image , source # 8 of @xcite , is about 3.7 pixels in the fine sampling direction suggesting a fwhm of 0.19@xmath1 . source # 1 is more extended than this in the present image ( @xmath6 4.9 pixels ) , but this is due to crowding with nearby blended sources and the intense nebular emission surrounding it . @xcite report source # 1 to be `` point like . '' the 10 sources identified by @xcite are clearly seen in figure [ c217 ] with the possible exception of their source # 2 which is close to source # 1 and not obvious in our image . the @xcite sources are indicated by filled circles in figure [ c217 ] . an additional point source ( named `` 8n '' ) appears near @xmath13 which is indicated by a single contour in figure 2 of @xcite . in addition to these objects , there are other knots or clumps which might contain embedded sources , or may be density enhancements in the region . these clumps appear to match well the morphology in figure 1 of @xcite . an exception is the `` x - like '' structure centered on source # 1 described by @xcite . we see no evidence of this feature in the lower contours of the nebular emission even though similarly bright clumps from the @xcite image do show up in our image . the mid infrared images of @xcite also do not show this feature , though diffraction spikes from the telescope secondary are present in their images . @xcite identified four bright sources in the core region of 350 covered by figure [ c217 ] in 913 images with somewhat lower angular resolution . the three brightest , okym13 , were unresolved as individual point sources . okym3 was associated with source # 1 by @xcite . the source labeled stho1 in figure [ c217 ] is the peak of [ ] emission in the maps of @xcite . these authors suggest stho1 is the location of an embedded massive star . there is no obvious point source in our continuum image or the @xmath0band image of @xcite at this position . the diamonds in figure [ c217 ] indicate the mid infrared sources identified by @xcite . the offsets between these sources ( okym 14 and stho1 ) are taken from @xcite and the postion of okym 3 is assumed to coincide with source # 1 ( also assumed to be the case by * ? ? ? @xcite identified a @xmath0band source with each of these @xcite sources within the positional uncertainties of the mid infrared image . these sources are the obvious ones in figure [ c217 ] near each of the okym sources . the continuum subtracted br@xmath3 emission is shown in figure [ brgmap ] . @xcite present narrow band brackett imaging . their data are seeing limited and do not resolve any of the structure shown in figure [ brgmap ] . @xcite report a br@xmath3 flux of 3.7@xmath810@xmath14 erg @xmath15 s@xmath7 in a 3.4@xmath1 aperture centered on the br@xmath3 peak in their images . this covers approximately the nifs field of view . equating this flux to the total counts in an approximately 2.5@xmath1 aperture results in a peak intensity ( one pixel ) of 5.5@xmath810@xmath16 erg @xmath15 s@xmath7 pix@xmath7 or 2.2@xmath810@xmath14 erg @xmath15 s@xmath7 arcsec@xmath17 . the peak br@xmath3 emission is to the sw of the continuum peak , although there is a second peak nearly as strong and very compact ( point like , superposed on extended emission ) centered on the continuum peak ( source # 1 ) . the @xcite sources # 4#7 are also centered on local br@xmath3 peaks . the rest of the @xcite continuum sources are offset from nearby local br@xmath3 emission peaks . 1 emission follows the br@xmath3 emission closely . the left panel of figure [ ratio ] shows the ratio of to br@xmath3 for the 1 @xmath18p@xmath19@xmath18s line at 21127 . the ratio is remarkably uniform over most of the fov and has a value of approximately 0.04 ( uncorrected for extinction ) . this value , which would be larger in the absence of extinction , already suggests a hot star ( s ) with effective temperature 40000 k @xcite . the stronger 1 line , @xmath20s@xmath20p@xmath19 at 20587 , is not uniform , however , in comparison to br@xmath3 ( see the right panel of figure [ ratio ] and discussion in 5.2.3 ) . figure [ spec ] shows that strong emission from multiply ionized species exist in the region . these include [ ] , [ ] , and [ ] . the first and last are the strongest lines apart from 1 and br@xmath3 . figure [ feiii ] shows the line maps for two of these lines . the [ ] emission is clumpy and peaks on the source okym4 while the [ ] emission is more intense and somewhat more concentrated to the central `` rectangle '' or `` diamond '' . the [ ] emission has a maximum at the continuum peak ( @xmath120.2 , 0.6 ) which is unresolved ( i.e. compact ) . the [ ] emission also has a local maximum at the continuum peak , but it appears to be affected by one or more bad pixels at the very center . these lines ( including [ ] ) appear to be common in ; see @xcite and @xcite . the ionization potentials of , , , , , and are 30.7 , 54.8 , 37.0 , 52.5 , 30.8 , and 42.9 ev , respectively . this suggests that the region with strong [ ] emission corresponds with ( for which there are no identified lines ) . the range of species seen is consistent with the presence of recombination lines ( ionization potential @xmath21 24.4 ev ) and the lack of recombination lines ( ionization potential @xmath21 54.4 ev ) . molecular hydrogen emission is also detected ( see figure [ spec ] and table 1 ) . the strongest line is the v@xmath2110 s(1 ) line at 21218 . a map of the line is shown in figure [ h2 ] with the contours of the br@xmath3 emission overlaid . the projection of the h@xmath22 emission seems to follow the ionized emission with the peak emission coincident with the br@xmath3 peak . there is a local maximum at the position of the continuum peak as well . the location and excitation of the h@xmath22 emission are discussed further in 5.4 . figure [ velmap ] shows the measured br@xmath3 line center velocity ( in ) for the entire fov . the mean observed br@xmath3 velocity of @xmath1250 has been subtracted to show relative velocity differences in this map . this mean corresponds to a velocity with respect to the local standard of rest ( lsr ) @xmath23 @xmath21 @xmath1262.5 @xmath2 5 . the velocities are overplotted with the br@xmath3 emission contours . each spatial pixel in the br@xmath3 map was fit with a gaussian profile over a restricted velocity range of @xmath12250 to 50 . this choice was made so that the 1 line at 21642 ( see table 2 ) did not interfere with the continuum fit . the velocity map shown in figure [ velmap ] was made with a modified version of the euro3d program @xcite produced by one of us ( pjm ) and called nf3d . the br@xmath3 velocity map shows a @xmath6 @xmath2 25 range in velocity centroids . this should be compared with our fwhm velocity resolution of @xmath6 60 . two br@xmath3 emission clumps are apparent in figure [ velmap ] . these are separated by 1@xmath1 @xmath12 2@xmath1 and oriented approximately at a position angle of 45 degrees . a single gaussian provides an adequate fit to the br@xmath3 emission over the entire fov . however , as we show in the following section , there is a broad blue wing to the emission near the continuum peak . in what follows , we call the single gaussian fit to the bright component of the line `` the principal component . '' when two gaussians are fit to spaxels near the continuum peak , the effect is to redshift the principal component by a further @xmath6 5 in this vicinity . the continuum peak is not located near any point of symmetry with respect to the velocity distribution . the nearest possible point source to the apparent center of the velocity distribution ( figure [ c217 ] ) is stho1 which was not detected in our continuum map . this source is only detected in the mid infrared [ ] map of @xcite and not in any continuum image presented here or by @xcite or @xcite . the region k350a exhibits a rich 1 spectrum in the @xmath0band . table 2 lists the 1 lines detected and their properties . using the on line data base compiled by peter van hoofpeter / atomic/ ; the on line database uses data from @xcite . ] ; 13 lines are identified between 20000 and 22300 . other lines redward of 22300 may be blended with the pfund series ( figure [ spec ] ) . multiple lines are present in the region around b@xmath3 . figure [ detail ] shows spectra extracted from individual spaxels at a typical location @xmath6 1.3@xmath1 se of the continuum peak ( figure [ spec ] , left panel ) and a location @xmath6 0.2@xmath1 se of the continuum peak ( figure [ spec ] , right panel ) . five gaussian functions have been fit simultaneously to these features using the spectral analysis program liner . good fits with rms residuals of @xmath24 1@xmath25 have typically been achieved by assuming that all five lines have the same fwhm as br@xmath3 ( figure [ spec ] , left panel ) . poorer fits with rms residuals of @xmath6 3@xmath25 are obtained in the region within @xmath6 0.5@xmath1 of the continuum peak where the br@xmath3 line has a strong blue wing ( e.g. , figure [ spec ] , right panel ) . the gaussian component fit to this wing has a central wavelength ( which is a fit parameter ) that is close to that of the @xmath26 @xmath20f@xmath19@xmath20d , @xmath18f@xmath19@xmath18d 21655 line ( see table 2 ) . however , this line is expected to be too weak to account for the blue wing near the continuum peak . in fact , based on atomic data used in the cloudy photoionization code @xcite , the 21655 line is expected to be weaker than the @xmath27 @xmath20f@xmath19@xmath20 g , @xmath18f@xmath19@xmath18 g 21647 line that is apparent in figure [ spec ] . better fits are obtained for spaxels near the continuum peak if the fwhm of the blue wing component is allowed to vary . rms residuals of @xmath24 1@xmath25 are then obtained with a gaussian fwhm that is about twice as broad as the principal br@xmath3 line . there is no apparent reason why the 21655 line should be anomalously strong and have a width twice that of other lines . we therefore conclude that the blue wing on br@xmath3 is due to a mix of a second kinematic component of br@xmath3 that dominates in a small region near the continuum peak and weaker emission that occurs over most of the fov . it is interesting that neither the peak emission nor the width of this second kinematic component is centered on the continuum peak . instead , this emission forms an arc that is offset se of the continuum peak by @xmath24 0.5@xmath1 . whatever the source of the broad component , the contribution of the lines near 21600 make a significant contribution to the integrated `` br@xmath3 '' flux . for the spaxel shown in figure [ detail ] , the broad line produces 50@xmath25 of the corresponding br@xmath3 flux . @xcite measured the optical , near infrared , and mid infrared positions of 350 with respect to the radio continuum @xcite using background stars as a reference . the @xcite position is consistent with later radio positions including the high angular resolution study of @xcite , and thus , to our knowledge , the @xcite positions are still the best available . more recent studies have assumed the near and mid infrared peaks are coincident with the radio continuum peak @xcite . @xcite find that the 2.2 and 10 peaks are coincident , within the @xmath6 1@xmath1 uncertainties , with the radio continuum peak . the optical and 1.6 peaks are located to the south west of the radio peak , and this is consistent with an extinction gradient across 350 as suggested by @xcite and @xcite and found also by @xcite . @xcite observed 350 in the radio continuum ( 15 ghz ) and in the h76@xmath4 recombination line . the angular resolution of these data is about 1.5@xmath1 . the continuum appears to be extended along an axis with pa ( e from n ) of 160 degrees and there is a strong velocity gradient from nw to se along the extended continuum source ( see also * ? ? ? this has led to the accepted model of an ionized outflow along the 160 degree axis . the nifs continuum is shown superposed with the @xcite radio continuum image in figure [ c217 ] . the two images have been aligned assuming the radio and @xmath0band continuum peaks are coincident . the large difference in scale and angular resolution are apparent . the overlay is instructive for the following discussion , but there are no obvious connections which can be drawn between features in the two maps . molecular line studies have been used to argue that there is a rotating torus of material with major axis roughly coincident with and perpendicular to the radio source . low angular resolution co data and modeling of hco@xmath28 line emission suggests this material may be rotating about an axis parallel to the outflow axis . @xcite first drew attention to a possible link between their ionized outflow and the molecular gas traced by co line emission @xcite , though they cautioned that the low angular resolution of the co data ( 30@xmath1 ) and the marginal evidence for a velocity gradient along its major axis made any link between the two structures `` premature . '' later , @xcite used observations of hco@xmath28 to argue for a rotating torus roughly perpendicular to the ionized outflow . the peak lobes of hco@xmath28 emission are separated by 8@xmath1 , so the scale of the rotating molecular torus is larger than the fov covered by nifs . the velocity difference measured between the peaks of the hco@xmath28 is @xmath6 @xmath2 2 , and the best fit to the velocity profile was found to be proportional to the square root of the radius . higher angular resolution observations were made by @xcite in the mid infrared ( near 10 ) and by @xcite in the @xmath0band ( 2.2 ) . the former observations were diffraction limited based on the 0.4@xmath1 fwhm obtained on standard stars , but the images of 350 do not clearly separate individual objects owing to intense resolved structure / emission and blended point sources . @xcite fit various point sources ( figure [ c217 ] ) to their data . the @xcite images , on the other hand , were obtained with a speckle imaging technique and are of similar angular resolution as the data presented here . the higher angular resolution and lower background emission result in more identifiable point sources within the central few arcseconds of 350 . @xcite and @xcite conclude that 350 contains multiple massive stars ; the former through excitation measurements of mid - infrared ionized lines and the latter based on the detection of multiple point sources in the center of the region . the nifs data confirm various aspects of the earlier work , principally the @xmath0band morphology of the point sources and brighter clumps of emission which are dominated by br@xmath3 . the @xmath0band image of @xcite , the @xcite mid infrared images , and our nifs images all indicate an abrupt cutoff in emission to the north of the bright continuum peak . the gradient is extremely sharp and the emission drops precipitously in 0.5@xmath1 towards the north of the region . the gradient in extinction was noted in earlier work ( e.g. , * ? ? ? * ) but is seen very clearly in the @xmath29 and @xmath11 images of @xcite . the center of the emission shifts systematically to the south . this geometry has been described as an opening cone with the continuum peak at the cone s apex @xcite and the opening toward the south south east . the dense molecular torus @xcite and other overlying material form the cone structure by obscuring ionized material to the north ; see figure 7 of @xcite . the constraining effect of the torus may be responsible for helping to shape the ionized outflow @xcite . the very strong gradient in emission evident in figure [ c217 ] and the @xmath0band image of @xcite suggest a possible obscuring disk on small scales , but no kinematic or morphological evidence exists to support this picture , and the radio and near infrared peaks are offset from the center of the torus ( see figure 14 of * ? ? ? if the embedded massive star associated with the continuum peak formed from the molecular torus , then it may have separated itself from this `` parental cloud . '' @xcite presented ionic line measurements of collisionally excited , , and at approximately 9 , 11 , and 13 , respectively . ratios of these lines suggest a relatively low excitation , typical of late o type or early b - type stars . the radio continuum @xcite , however , requires more ionizing photons than a single late o type star produces . @xcite thus concluded that multiple stars ionize 350 ( as discussed above for the nifs images , we detect continuum sources , # 1 , # 4#7 at least , which are centered on local peaks of the br@xmath3 emission ) . @xcite made a similar analysis and conclusion for w51 irs2 , also a massive star forming region . but recently , @xcite identified the exciting star of w51 irs2 directly through its near infrared spectrum , and it is much hotter ( approximately o3o4 ) . for g29.26@xmath120.02 , @xcite find a best fit nebular model which also has a cooler central source than the observed spectral type @xcite , though their acceptable models do cover a range of which includes the observed spectral type . the nebular emission in g45.45@xmath50.06 was consistent with the observed spectral types found by @xcite , but detailed modeling was not done . a grid of ionization models has been run to analyze the near infrared lines presented here in order to further explore the excitation of the region . cloudy was run on a grid of input parameters which sample a range of ionizing flux and of the central source as described by @xcite . a base line model grid was run with density 10@xmath30 @xmath31 and the @xcite atmospheres ( log ( g ) @xmath21 5.0 and z @xmath21 0 ) . the simulations were run until the temperature into the cloud reached 30 k. a model he atom was used with 10 energy levels to better resolve the level populations for the near infrared lines . the `` _ ism '' abundance set available in cloudy was used which has a he abundance by number relative to h of 0.10 . in the following sections , we discuss how the nifs observations constrain inputs to the models and also what can be inferred from a comparison of the models and observations . the density used in the cloudy runs was constrained by the observed ratio of [ ] lines . the ratio of [ ] 22184 to 21457 is sensitive to density @xcite . the observed ratio ( see table 1 ) is 3.12 and corrected for a@xmath32 of 2.17 @xcite leads to a ratio of 2.8 and a density of approximately 10@xmath33 @xmath31 according to the 7000 k model of @xcite or the 10000 k model of @xcite . @xcite give densities for the ratio of [ ] 22427 to 22184 as well , but the observed ratio of 0.5 is on the flat part of their curve and gives a low density of 10@xmath34 @xmath31 . the small reddening correction would push toward even lower density . all three lines are stronger toward the south in the fov and weak near the continuum peak , thus the ratios do not constrain the density near the continuum peak . @xcite derived a density of 10@xmath35 @xmath31 in the central few arcseconds from their 2 cm data ; similar radio data were presented by @xcite who find a density of 4@xmath810@xmath30 @xmath31 . we discuss models for a range of densities below ( see 5.2.2 and 5.2.3 and figure [ density ] ) . the [ ] line ratios to br@xmath3 reported in table 1 are similar to those reported by @xcite in orion . though the emission is relatively strong in the 350 spectrum , it may be consistent with most of the fe being condensed onto dust grains as is the case for orion ; similar ratios and physical conditions in orion are due to a gas phase abundance which is only a percent of the solar iron abundance @xcite . the distribution of and discussed above suggests some of the gas phase iron in 350 must be in , but there are no lines to estimate the abundance in that species ( the upper levels of observable lines are generally too high to be excited in regions @xcite ) . for the models described below which fit the data best for the excitation of the nebula , most of the fe is in and . in some cases , the fraction is @xmath6 90 @xmath25 , but this still requires the bulk of the iron to be condensed on grains if the total abundance is near or above the solar value . the map of the observed ratio of 1 21127 to br@xmath3 is shown in the left panel of figure [ ratio ] . the ratio is approximately uniform across the main emission in the center of the image and has a value of between 0.04 and 0.05 . the highest value is 0.055 near the continuum peak , and regions around the periphery have values in the range 0.0250.03 . the value in the average spectrum of figure [ spec ] is 0.046 ( table 2 ) which includes the effect of the blend of lines near br@xmath3 . these other lines are not included in the model prediction shown in figure [ he1brg ] , although their effect is small . the average ratio of 0.046 is consistent with an o star which has 40000 k @xcite . a small correction for extinction gives an intrinsic ratio of 0.051 if we adopt the observed ratio in table 2 . cloudy models were run for a grid of ionizing flux ( phi , s@xmath7 cm@xmath34 ) versus exciting source . the former parameter was varied from 12.0 to 16.0 dex and the latter varied from 25000 k to 45000 k. the results for our baseline case ( density @xmath21 10@xmath30 @xmath31 and blister geometry ) are given in figure [ he1brg ] . the ratio along a given contour becomes approximately constant above @xmath21 40000 k , but the detailed ratio depends on the ionizing flux as well . in figure [ he1brg ] , the overall emission due to dust compared to the br@xmath3 flux as predicted by cloudy is also plotted ( as _ dashed _ contours ) . the intersection of this contour with the 1 to br@xmath3 contour can give an idea of the true excitation of the nebula under the assumption of a single dominant source ( which might not be the case ) ; see @xcite and @xcite . the dust emission was calculated approximately from the spectral energy distribution ( sed ) shown in figure 2 of @xcite . no attempt was made to correct for the different beam sizes associated with the many distinct data sets which make up the @xcite figure . the most important is the @xmath6 100 point originally reported by @xcite since this is the peak in the sed . the @xcite beam was 50@xmath1 in diameter and included flux from the nearby k350b region . on the other hand , higher angular resolution data at 450 @xcite show a very compact source ( unresolved at the 8@xmath1 beam size ) with a similar flux to the sed presented by @xcite . @xcite report a br@xmath3 flux of 3.2@xmath810@xmath37 ergs @xmath15 s@xmath7 corrected for @xmath38 @xmath21 2.17 mag . we find a integrated dust emission of approximately 1.8@xmath810@xmath39 ergs @xmath15 s@xmath7 from the @xcite sed . thus the lower limit to the ratio is 1.78@xmath810@xmath40 . the intersection of the contours for 1 and dust compared to br@xmath3 suggest a of about 38000 k and log10(phi ) @xmath21 13.8 . since the br@xmath3 to dust ratio is a lower limit according the discussion above , the true would be hotter . the ionization model grid was repeated for a density of 10@xmath35 @xmath31 . the results are similar to those in figure [ he1brg ] with the intersection of the contours for the observed ratios moving to slightly lower and higher ionizing flux , i.e up and to the left in the figure . the formal intersection is at 37000 k and log10(phi)=14.5 . changing the geometry from plane parallel ( open ) to spherical ( closed ) in cloudy requires higher incident ionizing flux due to the radiation field dilution by @xmath41 . in the model grid , the inner radius of the cloud was set to 0.5@xmath1 or 5@xmath810@xmath42 cm . the observed ratios for the spherical case intersect at somewhat higher , 40000 k , and the corresponding ionizing flux is log10(phi ) @xmath21 14.7 to 14.4 . if the dust emission is too high by a factor of two due to beam size mismatches , then the br@xmath3 to dust ratio is too low by a factor of two . in this case , the model could be as high as 45000 k for a spherical geometry . we conclude that the excitation of 350 is consistent with a central star whose is greater than 37000 k and less than 45000 k. this is higher than @xcite derive based on line ratios of collisionally excited ions . a similar result is obtained in w51 irs2 when comparing the results of @xcite for line ratios to the recent spectral type obtained for the star located at the position of irs2 ( whose region is known as w51d ) ; see @xcite . evidence is presented below that the geometry may be more open like , see also][]col91 , depree94 which would lead to values near 40000 k. @xcite report a lyman continuum luminosity , lyc @xmath21 2.1@xmath810@xmath43 s@xmath7 for a distance of 8.7 kpc . adopting 7 kpc ( see 1 ) , results in a reduced number of 1.3@xmath810@xmath43 s@xmath7 . @xcite find lyc @xmath21 1.95@xmath810@xmath43 s@xmath7 for a distance of 8.3 kpc , which scaled to 7 kpc gives a similar number . this lyc luminosity is consistent with a single o type star with effective temperature of approximately 40000 k based on the properties for o stars given by @xcite or 41000 k if the data from @xcite are used . figure [ c217 ] shows there are multiple continuum sources ( especially in the vicinity of the peak emission ) , but the excitation of the nebula based on the nifs spectrum is consistent with a single , dominant star . the strongest 1 line in 350 is the 20587 line ( figure [ spec ] ) . the right panel of figure [ ratio ] shows the line ratio relative to br@xmath3 . comparison to the left panel of figure [ ratio ] indicates the 20587 line emission is more variable across the nebula than is the case for the 21127 emission . the strong degree of uniformity for the latter suggests the 20587 line variations are not due to differential extinction , but instead are due to structure in the nebula . if the variations seen in the right panel of figure [ ratio ] were due only to variations in the line of sight extinction , then this would require a change in a@xmath32 of more than 10 magnitudes . the continuum map morphology is not consistent with this picture , nor is the average extinction found by @xcite in the central few arcseconds ( a@xmath32 @xmath6 2 ) . finally , the br@xmath3 and 1 21127 intensities are near maxima at the position of the continuum peak while the 20587 line itself is weak there . the formation of the 1 20587 line is complicated and depends on the physical conditions in the ionized gas @xcite . a large ratio to br@xmath3 is due primarily to high density @xcite which allows for enhanced collisional population of the 2 @xmath44p@xmath19 level . however , this level population can also depend strongly on l@xmath4 584 resonance fluorescence which is affected by dust absorption of the he l@xmath4 line and ionization of neutral h by the he l@xmath4 line since both processes destroy the photons which would otherwise pump the line @xcite . the biggest changes in 20587 line formation are expected for a range of central star . the @xcite models were for an average over the nebula ; in the present case , the nebula is spatially resolved and clearly is dominated by a point source ( or unresolved sources ) at one end of the fov . in figure [ depth ] , the ratio of 1 to br@xmath3 emissivity with depth into the cloud ( for a blister model ) is plotted for the 21127 and 20587 lines and a central source with @xmath21 38000 k. the 20587 line strength changes sharply from inner ( close to central source ) to outer ( far from the central source ) positions in the cloud . thus , at any given point in the nebula , we would expect a relatively high ratio in the right panel of figure [ ratio ] since the effects shown in figure [ depth ] would be integrated along the line of sight . the range of observed values for the ratio in the right panel of figure [ ratio ] then suggests that real density variations may exist across the field of view ( lower density near the continuum peak ) . a possible exception to this would be if we were looking along a line of sight such that a slab or blister was illuminated with a geometry whose normal to the slab was in the plane of the sky . but this geometry is inconsistent with the molecular hydrogen morphology ( see below , 5.4 ) which suggests a face on blister geometry . in figure [ density ] , the cloudy predictions for the ratios of the 20587 and 21127 lines to br@xmath3 versus density from cloudy are plotted . the latter line has an observed ratio of approximately 0.051 ( corrected for extinction ) and is nearly uniform over the nebula . the former line has an observed ratio ( corrected for extinction ) that varies from about 0.33 to 0.84 . figure [ density ] shows the cloudy prediction for the case of @xmath21 38000 k and log10(phi ) @xmath21 14.0 . these results indicate that the 21127 line ratio is nearly constant over a range in density of 100 to 100000 @xmath31 while the 20587 line changes from 0.38 to nearly 1.0 over the same range . the ratios plotted in figure [ density ] are integrated along the line of sight through the nebula in contrast to the ratios plotted in figure [ depth ] . the kinematics evident in the bright br@xmath3 line ( figure [ velmap ] ) do not fit neatly with any of the previous kinematic data that were obtained at lower angular resolution . figure [ velmap ] shows a double lobed structure , possibly a flow , with a major axis distinct from any of the larger scale components discussed above . furthermore , the ordered red and blue shifted material is not well correlated with much of the strong br@xmath3 emission and the point of symmetry of the indicated flow is not coincident with the continuum peak . a possible explanation for the small scale flow is that it is created by a lower mass protostar in the nascent cluster . this molecular outflow is then ionized by the massive star at the continuum peak . a similar scenario was recently described for the compact high mass star forming region irs2 in w51 @xcite . another possibility is that the large scale ionized flow is related to the inner flow . this would require that the outflow is precessing or that details of the interaction of the out - flowing gas on larger scales than the nifs fov reorient the flow since the axis of the inner flow does not line up with the axis of the large scale ionized flow . the contours of the larger flow do , however , appear to twist @xcite and figure [ c217 ] clearly shows more detailed radio ( kinematic ) observations would be helpful in understanding the transition between the inner scale and larger scale gas kinematics . @xcite argued that the large scale ionized flow might result from the expansion of material coming from within a small ( few arcsecond ) shell . they took as evidence for a shell the high angular resolution 2 cm map of @xcite . the nifs br@xmath3 morphology appears to match well with the 2 cm emission , but does not appear shell like ( * ? ? ? * noted the classification of 350 as shell like was problematic ) . still , the nifs data suggest an outflow originating within this small region which is likely confined or directed by the material around it . the large scale flow has a velocity gradient of about 6 arcsec@xmath7 while the small scale flow gradient evident in figure [ velmap ] is 25 arcsec@xmath7 . the velocity reported for the core of the radio emission is @xmath1233 with respect to the lsr @xcite . the integrated br@xmath3 emission has @xmath23 @xmath21 @xmath1262.5 ( see table 1 ) . the red shifted emission in figure [ velmap ] is at @xmath23 @xmath6 @xmath1230 . this suggests that we may be only seeing part of radio emitting gas in the br@xmath3 map and that the red lobe of the br@xmath3 flow is really tracing the bulk of the radio flow ; that velocity is presumably the systemic velocity of the radio region . in this model , all the rest of the br@xmath3 emission is tracing blue shifted gas outflowing from the central source . the spatially unresolved br@xmath3 emission at the continuum peak ( see figure [ brgmap ] and 5.6 ) also has a @xmath23 @xmath6 @xmath1232 that is close to the systemic velocity defined by the h76@xmath4 emission . in 4.3 , a broad component of emission was identified to the blue of the br@xmath3 line ( see figure [ detail ] ) . the position of the line is consistent with one of the 1 lines from the n@xmath21 74 complex , but the emission is much stronger than expected based on transition probabilities and broader than any other 1 line by about a factor of two . thus , it is more likely a second br@xmath3 component . the line velocity is about 40 blue shifted compared to the background or cloud velocity ( the green regions in figure [ velmap ] ) . this is larger than the blueshifted velocity structure of the principal component of br@xmath3 shown in that figure . the width of the broad component is about twice the width of the principal br@xmath3 component ( 120 compared to 70 ) . the broad component is distributed in an arc to the sse of the continuum source ( @xmath6 0.5@xmath1 in length and offset from the peak by @xmath6 0.2@xmath10.3@xmath1 ) and is confined to this vicinity of the continuum source . it is not clear how the high velocity , broad component is produced . it may be related to hot gas escaping a tightly constrained volume near the central source . the scale associated with the broad component is small , only 0.0070.010 pc ( for a distance of 7 kpc ) . in champagne flows that are produced as the ionizing radiation from the hot star penetrates into a uniform medium ( e.g. , * ? ? ? * ) , the hot star quickly carves out a larger diameter cavity ( 0.1 pc ) . perhaps shocked gas initially flowing parallel to the plane of the sky and then being redirected along our line of sight could explain the broad component . this putative flow might be related to the central source stellar wind . in any case , gas must escape to the north and eventually form the larger ionized flow seen in the @xcite radio images . the immediate vicinity of the continuum peak may be filled with low density gas , and this gas may surround ( or overlay ) higher density material within the projected `` cone '' which extends to the south of the continuum source . in order to explain the low ratio of 1 20587 to br@xmath3 near the continuum peak , the particular line of sight must contain very little dense ionized gas indicating this area has been cleared away as the hot star emerges from its parent cloud ( which remains mostly to the north ) . the high angular resolution 2 cm image of @xcite and the molecular maps of @xcite are consistent with clearing of high density material in the vicinity of the continuum peak ; both show voids or gaps in the emission . none of the current absolute positions is good enough to place the different wavelength data to better than about 0.5@xmath1 which is the scale of the structure in our data and the 2 cm map . the nebula is seen in projection against h@xmath22 emission , primarily in the lower left quadrant where the br@xmath3 emission is strongest ( figure [ h2 ] ) but also ( unresolved ) at the position of the continuum peak . three prominent lines of molecular hydrogen are listed in table 1 . the average ratios of the two longer wavelength lines to the v@xmath2110 s(1 ) 21218 line ( 0.42 @xmath2 0.1 , 0.23 @xmath2 0.1 for h@xmath22 v@xmath21 10 s(0 ) and h@xmath22 v@xmath21 21 s(1 ) , respectively ) can be compared to the diagnostic diagram given by @xcite which shows these same ratios for a large sample of regions along with several model predictions for shock excited and fluorescent emission . the bulk of the @xcite sample is consistent with dense ( 10@xmath45 @xmath31 ) photo dissociation regions ( pdr ) . the results for 350 place it in the same region of the diagram showing it too is consistent with a dense pdr on the boundary of its ionized nebula . since we see both the ionized gas and molecular emission , it is likely the h@xmath22 emission is behind the ionized gas ( a large column of molecular gas in front would obscure our view ) . if this is the case , then the region may be best described as blister like ; the hot star has emerged from the parent cloud and is ionizing the face of the molecular material to the south . in this case , the `` cone '' is really a flattened structure seen face on and lying on the surface of a dense cloud of molecular gas . as mentioned above , @xcite analyzed [ ] / [ ] and [ ] / [ ] ratios in 350 . the observed ratios are consistent with other regions , but inconsistent with simple ionization models . the models predict stronger [ ] / [ ] than [ ] / [ ] for a given , but the observations show the average ratios are of similar value and the peak ratios have substantially larger [ ] / [ ] than [ ] / [ ] . our grid of models agrees with @xcite in the sense that for a given , the [ ] line ratio is predicted to be stronger . @xcite computed ar , s , and ne abundances for 350 from iso mid infrared spectra and find abundances which differ from the solar values ( and our cloudy models ) . in particular , the ne abundance compared to h is approximately solar , while the ar abundance is a factor of 2 lower than the solar value , and the s abundance is a factor of 10 lower . however , this large reduction in s abundance may be ( in part ) due to the high density in 350 depopulating the upper levels producing the s emission as @xcite point out . detailed models used to derive the s abundance in 350 by @xcite give a value only a factor of two lower than solar . the iso spectra were obtained through large apertures ( @xmath6 20@xmath1 ) , and the line ratios are smaller than the peak ratios in the higher angular resolution @xcite data . the iso line ratios are nearly equal and have a value of approximately 0.25 . this is comparable to the low end of the range of values obtained by @xcite . the observed high peak line ratios of @xcite ( [ ] / [ ] @xmath21 3 , [ ] / [ ] @xmath21 4 ) would appear to require lower s abundance and higher to be consistent with the cloudy ionization models , but the precise line ratios do not converge in our grid of models even when adopting the detailed abundances of @xcite . in particular , reproducing the high line ratio of [ ] / [ ] requires to be @xmath46 45000 k , the limit of the range based on the analysis of the 1 21127 line and the dust emission , and such a model would produce a much too high [ ] / [ ] ratio . earlier work based on kuiper airborne observatory observations @xcite also did not succeed in fitting the ratios of these infrared lines . the @xcite data were obtained through apertures approximately 5@xmath1 in diameter and the line ratios are larger than the iso values ( about unity for both ) and somewhat smaller than the @xcite peak values . @xcite and @xcite have argued that multiple ionizing sources might explain the observations . the nifs data generally ( continuum , 1 lines , resolved nebular structure ) seem to indicate the central few arcseconds are dominated by the continuum peak . figure [ feiii ] shows that the excitation in the nebula is higher toward the continuum peak . the [ ] emission is weak in the vicinity of the continuum peak compared to the [ ] line . the [ ] line is weak as well and this is consistent with its lower ionization potential than the other species present . in the case of okym4 ( figure [ c217 ] ) , there may be some evidence for lower excitation ; @xcite preferred a late o or b star for this source . the left panel of figure [ ratio ] shows a dip in the 21127 to br@xmath3 ratio at the position of okym4 ( the ratio is about 0.025 ) . this source is about 2@xmath1 away from the continuum peak and this is roughly the scale of an associated stromgren sphere for the central star ( figure [ depth ] ) . further , the [ ] emission peaks at the location of this source . indeed , the emission morphology of the line maps suggest the boundary of the region which is principally due to the continuum peak is on this same scale . if this is correct the `` cone '' morphology arises from dust obscuration to the north and the ionization boundary of the nebula to the south . the main nebula is largely excited by the continuum peak , and okym4 may be locally internally excited . the compactness of okym4 then suggests the source is very young and/or highly constrained . lower mass ob stars should not greatly affect the structure within the `` cone , '' however . @xcite and @xcite argued that the low _ average _ [ ] / [ ] and [ ] / [ ] ratios might be due to multiple lower sources ( as may be the case for okym4 ) , but a successful model must also explain the detailed ratios including their very high peak ratios near the continuum peak . while we have not been able to produce such a model , higher ratios are more consistent with hotter stars . in the preceding sections , the structure of the nebula was described , and it was concluded that the central source is a hot star emerging from its natal cocoon . the star is ionizing the cloud to the south in our fov and produces an intense region and associated pdr . it was argued that the density in the nebula may be lower in the vicinity of the continuum peak . the continuum peak ( by definition ) dominates the @xmath0band emission . in addition , br@xmath3 , 1 21127 , [ ] ( the highest ionization line observed ) , and h@xmath22 show unresolved emission peaks ( or near peaks ) at this location . there is clearly a very compact source of emission here which includes stellar continuum ( from multiple sources ) re - radiated by hot dust , and possibly line emission from the circumstellar environment . an ionized circumstellar disc is a possible source of the compact emission . though no kinematic evidence is available to support this possibility . @xcite and @xcite presented kinematic evidence of discs around ( somewhat less ) massive young stars based on the emission profiles of the co v@xmath2120 rotational vibrational bandhead at 22935 . this line is not detected in 350 . whatever its geometry , there must be compact circumstellar material ( partially ) surrounding the central source which is not directly linked ( at least now ) to the larger resolved region covered by the nifs fov . high angular resolution ( @xmath6 0.2@xmath1 ) @xmath0band spectral imaging has been presented for the galactic region 350 . the data were obtained with adaptive optics and the image slicing integral field spectrograph , nifs , at gemini north . the combination of excellent wavelength coverage , moderate spectral resolution ( 5160 ) , and high angular resolution provide the most detailed look to date at this massive young stellar object and its immediate environment . the emission morphology matches well with earlier speckle observations . spatial and ionization structure are resolved in the 3@xmath1 nebula . the 1 emission morphology suggests changes in density across the field which are attributed to the break out and clearing phase of massive star birth . the ratio of 1 20587 to br@xmath3 varies from about 0.3 to 0.8 and requires that the density near the central source be more than a factor 10 less than farther away . the ionization in the nebula can be traced across the nebula by the varying intensity of [ ] and [ ] lines . a grid of ionization models has been run with the code cloudy @xcite , and the results of these calculations were compared to the observed 1 21127 line to br@xmath3 line ratio and dust emission properties . the combination of observed values constrain the nebular excitation to a source with 37000 k. this is somewhat hotter than expected from collisionally excited emission line ratios in the mid infrared @xcite , but those line ratios , particularly their peak values on scales of @xmath6 1@xmath1 , are not explained by our models ( or those of * ? ? ? despite the high angular resolution of these data which enhances the contrast between background and point source emission , no photospheric lines are detected in the spectrum of the bright continuum source ; it remains buried within intense continuum emission . the central source is likely also blended with other point sources as pointed out by other investigators . there is spatially unresolved , bright emission in the lines of br@xmath3 , 1 , and [ ] ( and possibly [ ] ) at the continuum peak position and these are attributed to the circumstellar environment of the massive young stellar object . the nifs data cube reveals a striking kinematic signature in 350 . however , this `` flow '' is not aligned with the large scale ionized out flow reported in earlier radio continuum and recombination line studies ( 10 s of arcseconds in extent ) . the nifs kinematic signature is not symmetric about any detected point source in the nifs image either . it is possible the bipolar kinematics arise from a lower mass protostar associated with the central star of 350 , and the associated material is ionized by the hot central source . a second broad component of br@xmath3 emission is seen within a few tenths of an arcsecond of the continuum peak . this may be hot gas escaping from the cavity resulting from the action of the ionizing radiation on the natal molecular cloud or large scale torus . the authors would like to thank gary ferland , ryan porter , and peter van hoof for useful input regarding cloudy and the 1 model . the authors also thank manuel bautista for useful conversations regarding the line ratios and chris depree for kindly providing the 14.7 ghz fits image used in figure [ c217 ] . the plots and analysis in this article made use of the yorick programing language . this article made use of the simbad database at the cds . based on observations obtained at the gemini observatory ( proposal i d gn-2006a - c-11 ) , which is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the science and technology facilities council ( united kingdom ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , cnpq ( brazil ) and secyt ( argentina ) . ralchenko , yu . , jou , f .- c . , kelleher , d.e . , kramida , a.e . , musgrove , a. , reader , j. , wiese , w.l . , & olsen , k. ( 2007 ) . nist atomic spectra database ( version 3.1.2 ) , [ online ] . available : http://physics.nist.gov/asd3 [ 2007 , august 24 ] . national institute of standards and technology , gaithersburg , md . ( beam pa@xmath2179@xmath19 ) . the contours correspond to the nifs continuum ( _ right _ panel ) where the peak emission in both images is made to match . _ right _ panel : @xmath0band continuum at 21700 ( 19.2 , 9 pixels wide ) shown in grayscale ( log intensity ) and with the same contours plotted in the _ left _ panel . the field center is located at approximately ra(2000)@xmath2120@xmath47 01@xmath48 45.71@xmath49 , dec(2000)@xmath5033@xmath51 32@xmath52 42.1@xmath1 ( see text for a discussion of source position versus wavelength ) . the position angle on the sky was 70(east of north ) ; this image has been rotated so that north is up , and east is left . the _ red _ filled circles are sources from @xcite and the _ yellow _ diamonds are the mid infrared sources of @xcite ; see text . the continuum peak appears as an unresolved point source near 0.2 , 0.6 and is assumed to be source # 1 of @xcite and okym3 of @xcite . the long line starting at source # 1 indicates the approximate direction of the ionizied outflow described by @xcite . the second line is our estimate of the rotation axis of the hco@xmath28 torus described by @xcite ; see text . the torus is centered a few arcseconds n - w of the continuum peak . [ c217],width=6 ] . [ ] 22184 is shown on the left , and [ ] 22867 is shown on the right . the ionization potential of is lower than that for showing that the excitation is higher closer to the continuum peak . the peak surface brightnesses are 6.6@xmath810@xmath53 erg @xmath15 s@xmath7 arcsec@xmath17 and 1.1@xmath810@xmath54 erg @xmath15 s@xmath7 arcsec@xmath17 for [ ] and [ ] , respectively , and the contour intervals are evenly spaced at an interval 10 times less than the peak value . [ feiii],width=6 ] lcccc [ lines ] h@xmath22 v@xmath2110 s(1 ) & 21218 & 21215 & 59 & 0.0076 @xmath2 0.0001 + @xmath55@xmath56 $ ] @xmath18h@xmath18 g & 21457 & 21454 & 71 & 0.0052 @xmath2 0.0001 + br@xmath3 ( 74 ) & 21661 & 21657 & 71 & 1.0000 @xmath2 0.0002 + @xmath55@xmath56 $ ] @xmath18p@xmath57@xmath18p@xmath22 & 21987 & 21983 & 57 & 0.0041 @xmath2 0.0001 + @xmath55@xmath56 $ ] @xmath18h@xmath18 g & 22184 & 22183 & 62 & 0.0164 @xmath2 0.0001 + h@xmath22 v@xmath2110 s(0 ) & 22233 & 22230 & 70 & 0.0034 @xmath2 0.0001 + @xmath55@xmath56 $ ] @xmath18h@xmath18 g & 22427 & 22422 & 54 & 0.0086 @xmath2 0.0001 + h@xmath22 v@xmath2121 s(1 ) & 22477 & 22473 & 55 & 0.0020 @xmath2 0.0001 + @xmath55@xmath56 $ ] @xmath34p@xmath58@xmath34p@xmath59 & 22867 & 22863 & 71 & 0.0291 @xmath2 0.0003 lcccc [ helines ] 1 @xmath60 @xmath18s@xmath18p@xmath19 & 20430 & 20427 & 79 & 0.0055 @xmath2 0.0001 + 1 @xmath61 @xmath20s@xmath20p@xmath19 & 20587 & 20583 & 67 & 0.6778 @xmath2 0.0002 + 1 @xmath62 @xmath18p@xmath19@xmath18d & 20607 & 20603 & 89 & 0.0067 @xmath63 0.0002 + 1 @xmath64 @xmath18p@xmath19@xmath18s & 21127 & 21122 & 70 & 0.0462 @xmath2 0.0001 + 1 @xmath64 @xmath20p@xmath19@xmath20s & 21138 & 21134 & 68 & 0.0110 @xmath2 0.0001 + 1 @xmath65 @xmath18p@xmath19@xmath18s & 21500 & 21497 & 70 & 0.0012 @xmath2 0.0001 + 1 @xmath66 @xmath18d@xmath18f@xmath19 & 21614 & 21609 & 71 & 0.0250 @xmath2 0.0002 + 1 @xmath66 @xmath20d@xmath20f@xmath19 & 21623 & 21620 & 70 & 0.0100 @xmath2 0.0002 + 1 @xmath27 @xmath20f@xmath19@xmath20 g , @xmath18f@xmath19@xmath18 g & 21647 & 21642 & 70 & 0.0508 @xmath2 0.0002 + 1 @xmath26 @xmath20f@xmath19@xmath20d , @xmath18f@xmath19@xmath18d , br@xmath3 ? & 21655 & 21650 & 70 & 0.0670 @xmath2 0.0002 + 1 @xmath67 @xmath18d@xmath18p@xmath19 & 21821 & 21817 & 62 & 0.0044 @xmath2 0.0001 + 1 @xmath62 @xmath20p@xmath19@xmath20d & 21846 & 21842 & 62 & 0.0023 @xmath2 0.0001 + 1 @xmath65 @xmath20p@xmath19@xmath20s & 22291 & 22287 & 78 & 0.0012 @xmath2 0.0001

topological transport phenomena have been attracting a great deal of interest , because they exhibit universal properties that are insensitive to perturbations and independent of material details . a classical example of such a transport phenomenon is the integer quantum hall ( iqh ) effect in two - dimensional ( 2d ) electron systems , first discovered in 1980 , @xcite which is characterized by an integer quantization of the hall conductivity in unit of @xmath2 . the iqh effect has been observed in a large variety of materials , ranging from traditional semiconductors , to oxides , @xcite graphene , @xcite and topological insulators ( tis ) . @xcite laughlin @xcite interpreted the iqh effect in terms of an adiabatic charge pump . thouless , kohmoto , nightingale , and nijs @xcite established a relation between the quantized hall conductivity of the iqh system and a topological invariant , the first chern number . thouless and niu @xcite also related the amount of charge pumped in a 1d charge pump to the chern number . a variant of the iqh effect , the quantum spin hall ( qsh ) effect , was proposed recently , @xcite which has been experimentally realized in hgte quantum wells @xcite and inas / gasb bilayers . @xcite extension of the idea of the qsh effect has led to the discovery of 3d tis . @xcite a qsh system , which is also called a 2d ti , has an insulating band gap in the bulk and a pair of gapless helical edge states at the sample boundary . when the electron spin is conserved , a qsh system can be viewed as two independent iqh systems without landau levels . @xcite different from the charge , the spin does not obey a fundamental conservation law . in general , when the spin conservation is absent , unconventional topological invariants , either the @xmath3 index @xcite or the spin chern numbers , @xcite are needed to describe the qsh systems . the time - reversal ( tr ) symmetry is considered to be a prerequisite for the qsh effect , which protects both the @xmath3 index and gapless nature of the edge states . however , based upon the spin chern numbers , it was shown that the bulk topological properties remain intact even when the tr symmetry is broken . @xcite this finding evokes interest to pursue direct investigation and possibly utilization of the robust topological properties of the tis , besides using their symmetry - protected gapless edge states which are more fragile in realistic environments . unlike the first chern number underlying the iqh systems , which is embedded into the hall conductivity , up to now the topological invariants in the tis have not been directly observable . several experimental methods were proposed , but have not been realized . one was to measure the topological magnetoelectric effect , @xcite for which experimental complexities exist . @xcite fu and kane @xcite put forward an abstract 1d model , in which the spin pumping was related to the @xmath4 index in the limit of weak coupling . however , how this fictitious model could be implemented is still unknown . furthermore , from the viewpoint of application , generalization of the idea of the @xmath4 pump to higher dimension is meaningless , because according to the @xmath4 theory , @xcite only the states at the tr - invariant point of the brillouin zone can contribute to the spin pumping , and so the pumping rate can not be enhanced by increment of dimension . in a recent work , @xcite the more general case of finite coupling between the pump and electrode is investigated by using the scattering matrix method . it was found that the spin pumping in the model of fu and kane can survive finite scattering of magnetic impurities , and so may be attributed to the spin chern numbers rather than the @xmath3 index . some other authors @xcite proposed to pump quantized charge through the helical edge states by precessing a magnet covering the edge of a 2d ti , so that the number of gapless edge channels can be counted through electrical measurement . this method is indirect , in the sense that the topological invariants are intrinsic properties of the bulk electron wavefunctions , which do not immediately determine the charge pumping in the edge channels . here we predict an intriguing bulk topological pumping effect , directly driven by nonzero spin chern numbers , in a qsh system electrically modulated in adiabatic cycles . as a consequence of the topological spectral flows of the spin - polarized wannier functions ( spwfs ) in the bulk of the system , spin can be pumped into a nonmagnetic electrode continuously without net charge transfer . the total amount of spin pumped per cycle is proportional to the ( cross - section ) width of the sample , and insensitive to the material parameters and spin - mixing effect due to the rashba spin - orbit coupling . this electrical spin pump establishes a basis , on which spintronic applications taking advantage of the robust topological properties of the tis can be developed . especially , if a half - metallic electrode with spin polarization parallel ( or antiparallel ) to the @xmath5-axis is used , a quantized charge pumping conductivity , @xmath6 ( or @xmath7 ) , can be measured by electrical means , demonstrating a way to observe the spin chern numbers @xmath0 directly . let us consider a 2d model hamiltonian @xmath8 with @xmath9-m(t)\hat{\sigma}_{z}\ . \label{hamil}\ ] ] here @xmath10 is the electron charge , @xmath11 is the 2d momentum , @xmath12 is the vector potential of an @xmath13 electric field @xmath14 applied along the @xmath15 direction with @xmath16 and frequency @xmath17 being designated , and @xmath18 . this model can describe both the qsh materials , the hgte quantum wells , @xcite and inas / gasb bilayers , @xcite in the linear order in momentum . for definity of discussion , we confine ourselves to the hgte quantum wells , for which @xmath19 with @xmath20 are the pauli matrices for spin , and @xmath21 for the electron and hole bands . as will be discussed below , the time - dependent mass term @xmath22 can be induced by varying the voltages of the dual gates . @xmath23 represents the rashba spin - orbit coupling @xcite @xmath24\ . \label{rashba}\ ] ] to the linear order in momentum , the rashba spin - orbit coupling is nonvanishing only in the electron band . @xcite within the adiabatic approximation , for a bulk sample there exists a finite energy gap between the conduction and valence bands for @xmath25 or @xmath26 . at @xmath27 and @xmath26 , the conduction and valence bands touch at @xmath28 and @xmath29 or @xmath30 with @xmath31 . to clarify the topological properties underlying the spin / charge pumping , we consider @xmath32 as a parameter , and calculate the spin chern numbers @xmath0 in the standard way , @xcite on the torus of the two variables @xmath33 and @xmath34 with @xmath35 as the period . the spin chern numbers are obtained as @xmath36 for @xmath37 , and vanish elsewhere . not surprisingly , the band touching points @xmath38 serve as the critical points . ( vertical axis ) . the parameters are taken to be @xmath39 , @xmath40 , @xmath41 , and @xmath42 , with @xmath43 as the lattice constant and @xmath44 ( @xmath45 ) as the hopping integral of the tight - binding hamiltonian . , width=288 ] we now consider a system consisting of a pump for @xmath46 , an electrode for @xmath47 , and a potential barrier in between . the total hamiltonian of the system reads @xmath48 where @xmath49 has been given above , and @xmath50 is the hamiltonian of the electrode . a possible experimental setup for realizing this hamiltonian is explained in appendix a in more details . in the barrier region , the term @xmath51 opens an insulating gap of size @xmath52 , which accounts for contact deficiencies between the pump and electrode . the crucial role of the nonzero spin chern numbers in the spin / charge pumping process can be visualized by using the spwfs , which were first introduced in ref . we construct a tight - binding hamiltonian for the effective 1d system at any given @xmath53 according to eq . ( [ htot ] ) , and diagonalize the total hamiltonian of the pump and electrode numerically . following the same procedure as calculating the spin chern numbers , @xcite the space occupied by electrons is partitioned into two spin sectors after diagonalizing the spin operator @xmath54 in the occupied space . by definition , the states in the two spin sectors are essentially the maximally spin - polarized states . then we construct the wannier functions @xcite for the spin - up and spin - down sectors , respectively , which are called the spwfs . the evolution of the centers of mass of the spwfs for @xmath55 and @xmath56 is shown in fig . [ fig_1 ] . we see that the wannier centers for the spin - up sector move right and those for the spin - down sector move left , each center shifting on average a lattice constant per cycle . within the adiabatic approximation , time @xmath57 plays the same role as the momentum of an additional dimension , @xcite namely , @xmath58 . therefore , when @xmath53 is considered as a parameter , the evolution of the wannier functions of the effectively 1d system related to various @xmath59 with time @xmath60 can be understood from the static properties of a 2d system associated with various @xmath59 and @xmath61 . in the general theory , @xcite the relationship between the chern number and the spectral flows of the wannier functions in a 2d system has been established . according to this theory , the average displacement of each of the centers of the spwfs in the spin - up ( spin - down ) sector with changing @xmath61 ( or @xmath60 ) from @xmath62 to @xmath63 , in units of the lattice constant , must equal to the spin chern number @xmath64 ( @xmath65 ) . therefore , the nontrivial transfer of the spwfs observed in fig . 1 is a direct manifestation of the nonzero spin chern numbers @xmath66 in the pump ( for @xmath67 ) . more interestingly , we see that such spectral flows can go across the finite barrier ( @xmath68 ) , and extend into the electrode , even though the barrier and electrode are topologically trivial . physically , because the system needs to recover its original eigenstates when each cycle ends , the nontrivial spectral flows of the spwfs in the ti need to constitute closed loops through formation of edge states at the boundary , @xcite or extend into the electrode . however , localized edge states can not exist at the finite barrier due to quantum tunneling effect , so the transfer of the spectral flows of the spwfs into the electrode occurs . this result will be further confirmed by direct calculation based upon the scattering matrix theory in the next section . the spwfs are just another equivalent representation of the occupied space , and so the counter spectral flows of the wannier centers in the two spin sectors represent the true movements of the electrons . if the rashba spin - orbit coupling were neglected , the wannier functions would be the eigenstates of @xmath69 . the nontrivial spectral flows indicate that at the given @xmath53 , in each cycle a spin - up electron goes from the pump into the electrode , and a spin - down electron moves oppositely . therefore , no net charge transfer occurs but a quantized spin of @xmath70 is pumped into the electrode . when the small rashba spin - orbit coupling is turned on , while the topological spectral flows remain intact , as seen from fig . [ fig_1 ] , the spin polarizations of the wannier functions are no longer fully parallel to the @xmath5-axis , and may also vary with time . as a consequence , the amount of spin pumped per cycle will deviate from the quantized value . in general , the amount of the spin pumped can be conveniently calculated by using the scattering matrix formula . @xcite the @xmath5-component of the spin pumped per cycle is given by @xcite @xmath71 where @xmath72 ( @xmath73 ) is the reflection amplitude for an electron at the fermi energy incident from the spin-@xmath74 channel of the electrode and reflecting back into the spin-@xmath75 channel . in the following calculations , the fermi energy is set to be zero ( @xmath76 ) , and the rashba spin - orbit coupling is treated as a perturbation . as shown in appendix b , to the linear order in @xmath77 , we obtain @xmath78}{\mbox{ch}(2\gamma_{0}d)-\sin(2\theta ) \mbox{sh}(2\gamma_{0}d)}+{\cal o}(\epsilon^2)\ , \label{rupup0}\ ] ] @xmath79 } { \mbox{ch}(2\gamma_{0}d)-\sin(2\theta)\mbox{sh}(2\gamma_{0}d ) } + { \cal o}(\epsilon^2)\ , \label{rdnup0}\ ] ] and @xmath80 and @xmath81 , where @xmath82 and @xmath83 $ ] . we note that the dimensionless quantity @xmath84 appears as the small expansion parameter . since @xmath85 and @xmath86 are always real , the contributions from the third and fourth terms in eq . ( [ ds1 ] ) vanish . consequently , @xmath87 this expression has a geometric explanation : the amount of spin pumped per cycle equals to the difference between the areas enclosed by the directional trajectories of @xmath88 and @xmath89 on the complex plane , multiplied by @xmath90 . due to the relation @xmath91 , yielding @xmath92 , the two terms in eq . ( [ ds2 ] ) make an equal contribution , so that we can focus on the first term . while the expression ( [ rupup0 ] ) for @xmath88 is independent of the rashba spin - orbit coupling , as will be shown soon , a combination of eqs . ( [ rupup0 ] ) and ( [ rdnup0 ] ) allows us to evaluate the amount of spin pumped up to the second order in @xmath93 . , as a function of @xmath94 for four sets of @xmath95 . the other parameters are taken to be @xmath96 and @xmath97 with @xmath98 . inset : trajectories of @xmath88 in a cycle on the complex plane.,width=230 ] we first consider the case of @xmath96 . from eq . ( [ rupup0 ] ) , it is easy to show @xmath99 . in fig . [ fig_2 ] , we plot the argument @xmath100 of @xmath101 as a function of @xmath94 for several parameter sets . for either @xmath102 ( ideal contact ) or @xmath103 ( strong potential barrier ) , @xmath100 always increments @xmath104 in a cycle as long as @xmath105 . in this case , the trajectories of @xmath101 always form a unit circle on the complex plane , oriented counterclockwise , as shown in the inset of fig . [ fig_2 ] , suggesting @xmath106 ( for @xmath67 ) . for @xmath107 , however , the situation is quite different . @xmath100 does not change after going through a cycle , and the trajectory of @xmath101 does not enclose a finite area , so that @xmath108 . apparently , the present result conforms to the spin chern numbers given by eq . ( [ schern ] ) and the spectral flows of the spwfs . ( in unit of @xmath109 ) as a function of @xmath110 for @xmath39 and three different values of @xmath77 . the other parameters are taken to be the same as in fig . [ fig_2 ] . inset : the trajectory of @xmath88 in a cycle for @xmath56 and @xmath111 , with the unit circle indicated by the dotted line.,width=249 ] next we study the correction to @xmath112 due to nonzero rashba spin - orbit coupling . by expressing @xmath113 in the polar coordinate system , where @xmath114 is the argument at @xmath115 , and @xmath116 and @xmath117 stand for the second - order corrections to @xmath118 and @xmath119 , respectively , due to the rashba spin - orbit coupling , eq . ( [ ds2 ] ) becomes @xmath120 + { \cal o}(\epsilon^3)$ ] . we notice that @xmath116 is a small quantity fluctuating around @xmath62 and periodic in time , @xmath121 , so that @xmath122 . using the identity @xmath123 , we then obtain @xmath124 , where @xmath125 can be calculated from eq . ( [ rupup0 ] ) and @xmath85 has been given by eq . ( [ rdnup ] ) . this is an expression for @xmath112 accurate to the second order in @xmath93 . at @xmath102 , we obtain @xmath112 as @xmath126(c_{+}-c_{-})\ , \label{ds_r}\ ] ] for @xmath37 , and @xmath108 elsewhere . as expected , nonzero rashba spin - orbit coupling causes @xmath112 to deviate from its quantized value , i.e. , @xmath127 . in real materials , @xmath77 is usually much smaller ( by an order of magnitude or more ) than @xmath128 , @xcite so that the deviation is less than a percent for an ideal connection between the pump and electrode . for @xmath129 , @xmath112 can be evaluated numerically and its calculated result is plotted in fig . [ fig_3 ] as a function of @xmath110 for three different strengths of the rashba spin - orbit coupling . for @xmath96 , @xmath112 is quantized to @xmath130 , independent of @xmath110 . for @xmath131 and @xmath132 , weak potential barrier ( @xmath133 ) has little effect on @xmath112 . this is reasonable as the leading - order correction of small @xmath110 must be @xmath134 . appreciable deviations from the quantized value occur for strong potential barrier ( e.g. , @xmath135 ) . we note that @xmath116 does not affect the orbit of @xmath88 , as it represents a variation in the tangent direction of the orbit , and the orbit can be determined by @xmath136 . inset shows the trajectory of @xmath88 on the complex plane for @xmath56 and @xmath111 . for such a strong potential barrier , the orbit of @xmath88 deviates from the unit circle visibly . the above result suggests that improving the contact quality between the pump and electrode is helpful for obtaining a nearly integer - quantized value of the pumped spin . by summing over @xmath53 between @xmath30 and @xmath137 , we obtain for the total spin pumped per cycle @xmath138 with @xmath139 ( c_{+}-c_{-})\ , \ ] ] for a good contact ( @xmath133 ) . @xmath140 is in scale with width @xmath141 of the pump . by noting that @xmath142 is proportional to the number of cycles per unit time , @xmath143 can be considered as the spin pumping conductivity . now we discuss a possible way to experimentally observe the spin chern numbers , by using a half - metallic electrode , in which conducting channels for electron spin antiparallel to the spin polarization are absent . we first consider the case , where the spin polarization of the electrode is parallel to the @xmath5 axis . the hamiltonian of the electrode is taken as @xmath144 . in this case , as shown in appendix c , @xmath88 is still given by eq . ( [ rupup0 ] ) but @xmath145 . it follows that for any @xmath32 between @xmath146 and @xmath147 , the charge pumped per cycle is integer - quantized and equal to @xmath148 . similarly , for the spin polarization of the electrode antiparallel to the @xmath5 axis , the charge pumped is equal to @xmath149 . therefore , the total charge pumped per cycle is given by @xmath150 with @xmath151 where the spin chern number @xmath152 ( @xmath153 ) is taken for the spin polarization of the electrode parallel ( antiparallel ) to the @xmath5 axis . we emphasize that eqs . ( [ totcharge ] ) and ( [ ccond ] ) obtained above are valid for finite rashba spin - orbit coupling and finite potential barrier between the pump and electrode , indicating that the quantized charge pumping is robust against small perturbations . experimentally , @xmath154 can be obtained by measuring the electrical current in the electrode , and from eqs . ( [ totcharge ] ) and ( 13 ) , @xmath0 can be evaluated , yielding an experimental method to measure the spin chern numbers directly . the sign inversion of @xmath154 with reversing the spin polarization of the half - metallic electrode , as indicated by eqs . ( [ totcharge ] ) and ( [ ccond ] ) , is a hallmark of the present spin chern charge pump , which can be used to distinguish it from the conventional thouless charge pump @xcite . we have used the single - electron approximation , where the electron interaction is not taken into account . in particular , in the half - metallic electrode case , one of the spin channel is blocked at the boundary , which naturally induces some charge and spin accumulations , and consequently changes the potential profile . however , owing to the screening effect , the change in the potential profile is expected to be localized at the boundary , which in effect modifies the potential barrier between the pump body and the electrode . as has been shown above , the charge pumping effect is independent of the existence and details of the potential barrier , and so we believe that the pumping effect will survive the charge and spin accumulations . up to now , all the results obtained from the scattering matrix formula are apparently in complete agreement with the spin chern numbers given by eq . ( [ schern ] ) . these results can not be explained within the framework of the @xmath3 theory . @xcite while one can define a @xmath3 index at the tr - invariant point @xmath155 , the effective 1d hamiltonian given by eqs . ( [ hamil ] ) and ( [ rashba ] ) for any given nonzero @xmath32 does not preserve the tr symmetry , as its tr partner is at @xmath156 , making the @xmath4 index invalid . the @xmath4 theory predicted that the tr symmetry is crucial for the topological spin pumping , @xcite suggesting that only the states at the tr - invariant point @xmath155 can contribute to the spin pumping . this clearly contradicts the present result that all the states with @xmath157 contribute equally , which is obtained directly from the scattering matrix formula . this point is also evidenced by the fact that the total amount of charge or spin pumped per cycle is in proportion to the sample width @xmath141 . for the same reason , the pumping effect found in this work is also essentially different from that via edge states in refs . @xcite , where the amount of spin or charge pumped per cycle is proportional to the number of the gapless edge channels . in conclusion , our work uncovers a bulk topological pumping effect due to direct transfer of the spwfs between the pump and electrode , without the participation of edge states . this measurable effect reveals the bulk topological properties of the system that are neither captured by the @xmath3 index nor reflected by the number of gapless edge channels . it can be accurately described by the spin chern numbers . this spin chern pump may lay the foundation for direct experimental study and possibly utilization of the robust topological properties of the tis . the previous experimental work @xcite evidenced the difficulty of modulating in time the properties of an open quantum dot without generating undesired bias voltages due to stray capacitances . this problem might not be significant in our pumping setup , where a much larger bulk sample of the ti can be used and the stray capacitances can be greatly reduced . moreover , a possible way around the obstacle is to use the @xmath13 josephson effect to induce periodically time - dependent andreev reflection amplitudes in a hybrid normal - superconducting system . @xcite concrete design of a spin chern pump based upon the josephson effect will await future work . while the proposed spin pumping scheme may have the advantage of low noises , its practical application in spintronic devices still relies on the discovery of new tis with bulk band gaps much greater than room temperature , which determine the temperature range where the spin chern pumping effect can survive . currently , precessing magnetization is a feasible method to generate robust spin currents in spintronic devices at room temperature . @xcite this work was supported by the state key program for basic researches of china under grants numbers 2015cb921202 , 2014cb921103 ( ls ) , 2011cb922103 and 2010cb923400 ( dyx ) , the national natural science foundation of china under grant numbers 11225420 ( ls ) , 11174125 , 91021003 ( dyx ) and a project funded by the papd of jiangsu higher education institutions . we also thank the us nsf grants numbers dmr-0906816 and dmr-1205734 ( dns ) . in what follows we expand on the model setup and possible experimental realization of the spin chern pump in more details . a possible experimental realization of the hamiltonian eq . ( [ hamil ] ) for the pump is illustrated in fig . a hgte / cdte quantum - well heterostructure with dual gates ( top and bottom ) is placed between two conductive plates . it is known that when the width of the quantum well ( thickness of the hgte film ) is above a critical size @xmath158 , @xcite the band structure is inverted , characterized a negative mass term @xmath159 in the hamiltonian , corresponding to the qsh state . if the width of the quantum well falls below @xmath160 , the band structure will be aligned in a `` normal '' way with a positive mass term @xmath161 , corresponding to a normal insulator . as has been discussed in refs . @xcite and @xcite , the topological phase transition between the qsh phase and normal insulator can also be tuned by applying a gate voltage , which effectively reduces the width of the quantum well . it is assumed that the quantum well under consideration has a width somewhat greater than @xmath160 , and so has a negative mass term @xmath159 initially . with increasing the gate voltage , the electron mass increases , and can invert its sign . usually , increasing the gate voltage may also adjust the carrier density . nevertheless , it has been shown @xcite that , by using dual gates and properly tuning their voltages @xmath162 and @xmath163 , it is generally possible to change the electron mass in the desired manner to be @xmath164 , while keeping the electron fermi energy still in the band gap . the effect of the conductive plates is easily understood . when a voltage drop @xmath165 is applied across the plates , a uniform electric field @xmath166 will be generated in the space between the two plates . the electrons in the quantum well experience a vector potential @xmath167 with @xmath168 defined as @xmath169 . if the electric field is chosen to be @xmath170 , one gets @xmath12 with @xmath16 , as desired . we point out that the exact time dependencies of @xmath22 and @xmath168 are not essential for realizing the spin chern pump , provided that they have the same periodicity and a constant relative phase shift . + direction.,width=163 ] the pumping effect is insensitive to material details of the electrode . the electrode is taken to be a normal metal with a 2d parabolic hamiltonian @xmath171 . when @xmath172 is sufficiently large , for a given @xmath173 , we can linearize the effective 1d hamiltonian @xmath174 at the right and left fermi points @xmath175 with @xmath176 . a pauli matrix @xmath177 is introduced to describe the two branches . to be consistent with the form of the hamiltonian in the pump , we use @xmath178 and @xmath179 , respectively , to represent the right - moving and left - moving branches for @xmath180 and oppositely for @xmath181 . as a result , the hamiltonian of the electrode becomes @xmath182 at @xmath183 , where @xmath184 and @xmath185 . the spin pumping effect is usually dominated by small @xmath53 , so that we can further approximate @xmath186 , with purpose to minimize the number of adjustable parameters in the model . for the present dirac - like hamiltonian , an ordinary potential barrier has a very weak effect on the electron transmission due to the klein paradox . therefore , we take the hamiltonian for the barrier to be @xmath187 . the inclusion of potential @xmath188 opens up an insulating energy gap of size @xmath52 around the fermi level , which presumably is more efficient for describing the contact deficiencies and structural mismatch between the pump and electrode . the half metal , e.g. , cro@xmath189 , la@xmath190sr@xmath191mno@xmath192 , etc . , is a substance that acts as a conductor to electrons of one spin orientation , but as an insulator to those of the other spin orientation . from the viewpoint of the electronic structure , one of the spin subbands is metallic , whereas the fermi level falls into an energy gap of the other spin subband . to simulate the half - metallic electrode , @xmath174 is taken to be @xmath193 , where @xmath194 stands for the spin polarization of the electrode parallel and antiparallel to the @xmath5-axis , respectively . the second term opens an energy gap of size @xmath195 around the fermi level for electron spin antiparallel to the spin polarization of the electrode , without affecting the other spin subband . as a result , the electron density of states is fully spin - polarized at the fermi energy . @xmath196 is set to be infinity in the final result . we now solve the scattering problem for an electron at the fermi energy incident from the electrode . the fermi energy will be taken to be @xmath183 , which is in the band gap of the pump . therefore , the incident electron will be fully reflected back into the electrode . the rashba spin - orbit coupling is treated as a perturbation , and the result will be calculated to the linear order in the small quantity @xmath197 . the wavefunctions of the pump ( @xmath198 ) , barrier ( @xmath199 ) and electrode ( @xmath200 ) are denoted by @xmath201 , @xmath202 , and @xmath203 , respectively . we have two boundary conditions : @xmath204 and @xmath205 . we use @xmath206 and @xmath207 to represent the eigenstates of @xmath69 , and @xmath208 and @xmath179 to represent those of @xmath209 . on the basis @xmath210 , @xmath211 , @xmath212 , and @xmath213 , the hamiltonian of pump ( the eqs . ( 1 ) and ( 2 ) in the manuscript ) can be expanded as a @xmath214 matrix @xmath215 where @xmath216 . for energy @xmath217 , the eigen - equation is obtained from eq . ( [ hp_mat ] ) @xmath218 ^ 2-m^2(t)r^2_{0}\tilde{k}^2=0\ , \label{eigen_mat}\ ] ] with @xmath219 , and up to a normalization factor , the eigenfunctions are @xmath220 where @xmath221 and @xmath222/r_{0}$ ] . we need to solve @xmath59 from the eigen - equation eq . ( [ eigen_mat ] ) . we notice that the equation is a 4th - degree polynomial of @xmath59 with real coefficients , so complex conjugate roots must appear in pairs . moreover , eq . ( [ eigen_mat ] ) is even in @xmath59 , so positive and negative roots appear in pairs . in combination , ( [ eigen_mat ] ) must have four roots of the form @xmath223 , @xmath224 , @xmath225 , and @xmath226 . by substitution of the four roots into eq . ( [ eigenfunc ] ) , we can in principle obtain four different eigenfunctions . for the present scattering problem , we only need the two eigenfunctions that are decaying into the pump , which correspond to the two roots with negative imaginary parts . for @xmath96 , it is easy to obtain for the roots for the two decaying modes : @xmath227 with @xmath228 , which are two - fold degenerate . the corresponding two decaying eigenfunctions are given by @xmath229 @xmath230 where @xmath231 $ ] . for @xmath232 , we write the roots of @xmath233 as @xmath234 , and also write @xmath235 as @xmath236 . to the second order in @xmath237 , we can solve for @xmath238 from the eigen - equation eq . ( [ eigen_mat ] ) @xmath239 noticing that the expression for @xmath240 given below eq . ( [ eigenfunc ] ) has a factor @xmath77 in the denominator , we keep @xmath238 to the second order , for the purpose to calculate @xmath240 to the linear order . by using the relation @xmath241 , we derive from eq . ( [ dkx2 ] ) @xmath242 with these relations , we obtain for @xmath243 and @xmath240 @xmath244\\ & \pm&i\frac{r_{0}}{2}(\hbar\eta)^2\sin^3(2\theta)+{\cal o}(\epsilon^2)\ , \end{aligned}\ ] ] and @xmath245 we can always eliminate any common factor that appears in all the four components of eq . ( [ eigenfunc ] ) , whenever possible . by eliminating a common factor @xmath246 , we rewrite @xmath243 and @xmath240 as @xmath247 and @xmath248 then the two decaying wavefunctions can be derived to be @xmath249 \end{array } \right)e^{\eta_{\mp}x}\ , \label{eigenfuncr0}\ ] ] where @xmath250 $ ] . some remarks are in order . with respect to the wavefunctions at @xmath96 , namely , eqs . ( [ eigenfuncwithoutr00 ] ) and ( [ eigenfuncwithoutr01 ] ) , nonzero @xmath77 leads to nonperturbative change in the wavefunctions eq . ( [ eigenfuncr0 ] ) , in the sense that eq . ( [ eigenfuncr0 ] ) will not recover eqs . ( [ eigenfuncwithoutr00 ] ) and ( [ eigenfuncwithoutr01 ] ) in the limit @xmath251 . this is reasonable , just as what always happens in the degenerate perturbation theory of the quantum mechanics . for the present problem , the wavefunction @xmath201 in the pump is always expressed as an arbitrary linear superposition of the two decaying modes : @xmath252 . by defining @xmath253 $ ] and @xmath254 $ ] , we can rewrite @xmath201 as @xmath255 . the final result for the reflection amplitudes depends only on @xmath256 , so we explicitly write out the expression for @xmath256 as follows @xmath257 where @xmath258 @xmath259 now we see that in the limit @xmath251 , the total wavefunction @xmath260 will go back to the form of a superposition of the two decaying wavefunctions given in eqs . ( [ eigenfuncwithoutr00 ] ) and ( [ eigenfuncwithoutr01 ] ) . in conclusion , while small rashba spin - orbit coupling may cause a nonperturbative change of the individual decaying wavefunctions , it modifies the `` space '' spanned by the two decaying modes in a perturbative manner . it is this `` space '' which determines the final result of the reflection amplitudes . this is the physical reason why in the final result , the rashba spin - orbit coupling modifies the reflection amplitudes in a perturbative manner . the wavefunction in the potential barrier can be written as @xmath261 where @xmath82 . for an electron incident from the spin - up channel , the wavefunction in the electrode is given by @xmath262 first , matching the wavefunctions eqs . ( [ wavebarrier ] ) and ( [ waveelectrode ] ) at @xmath263 , one obtain @xmath264e^{-\gamma_{0}d}\ , \label{c_1}\\ c_{2}&=&\frac{1}{2}[(1+i)+r_{\uparrow\uparrow}(1-i)]e^{\gamma_{0}d}\ , \label{c_2}\\ c_{3}&=&\frac{r_{\downarrow\uparrow}}{2}(1+i)e^{-\gamma_{0}d}\ , \label{c_3}\\ c_{4}&=&\frac{r_{\downarrow\uparrow}}{2}(1-i)e^{\gamma_{0}d}\ . \label{c_4}\end{aligned}\ ] ] in the next step , we will match wavefunctions at @xmath265 . substituting eqs . ( [ c_1]-[c_4 ] ) into eq . ( [ wavebarrier ] ) , we can write eq . ( [ wavebarrier ] ) at @xmath266 as @xmath267 where @xmath268 . now equating eq . ( [ psim ] ) with eq . ( [ psip0 ] ) , we obtain @xmath269 it follows from eq . ( [ abcd0 ] ) @xmath78}{\mbox{ch}(2\gamma_{0}d)-\sin(2\theta ) \mbox{sh}(2\gamma_{0}d)}+{\cal o}(\epsilon^2)\ , \label{rupup}\ ] ] and @xmath79 } { \mbox{ch}(2\gamma_{0}d)-\sin(2\theta)\mbox{sh}(2\gamma_{0}d ) } + { \cal o}(\epsilon^2)\ . \label{rdnup}\ ] ] the reflection amplitudes for an electron incident from the spin - down channel can be solved similarly . now the wavefunction in the electrode is given by @xmath270 the forms of the wavefunctions in the pump and barrier remain to be the same . by some algebra , we arrive at @xmath271 equating eq . ( [ psim ] ) with eq . ( [ psip1 ] ) , we obtain @xmath272 it follows from eq . ( [ abcd1 ] ) @xmath273}{\mbox{ch}(2\gamma_{0}d)-\sin(2\theta ) \mbox{sh}(2\gamma_{0}d)}+{\cal o}(\epsilon^2)\ , \label{rdndn}\ ] ] and @xmath274 } { \mbox{ch}(2\gamma_{0}d)-\sin(2\theta)\mbox{sh}(2\gamma_{0}d ) } \left(\frac{r_{0}}{v_{\mbox{\tiny f}}}\right)+{\cal o}(\epsilon^2)\ . \label{rupdn}\ ] ] the total hamiltonian of the system is invariant under the transformation @xmath275 so the corresponding transformation of eq . ( [ waveelectrode ] ) @xmath276 must also be an eigenstate of @xmath277 . this result tells that an electron incident from the spin - down channel will be reflected into the spin - down channel with amplitude @xmath278 , and also into the spin - up channel with amplitude @xmath279 . comparing it with eq . ( [ waveelectrode1 ] ) and noticing that @xmath280 is equivalent to @xmath281 , we find immediately the following relations @xmath282 and @xmath283 the reflection amplitudes given in eqs . ( [ rupup ] ) , ( [ rdnup ] ) , ( [ rdndn ] ) , and ( [ rupdn ] ) apparently satisfy these relations . to simulate the half - metallic electrode , the hamiltonian of the electrode is taken to be @xmath193 . consider first the case , where the spin polarization of the electrode is parallel to the @xmath5-axis . now the hamiltonian of the potential barrier becomes @xmath284 for simplicity , we will take @xmath285 limit in the final result . in this case , an electron incident from the spin - up channel has some probability to be reflected into the spin - down channel from the pump , but the reflected wave will decay quickly to @xmath62 within the barrier , and has no chance to reach the electrode . therefore , @xmath145 and the wavefunction in the electrode becomes @xmath286 the wavefunction in the potential barrier can be written as where @xmath82 and @xmath288 . the wavefunction in the pump remains to be same . repeating the same calculation as in sec . ii , it is straightforward to obtain @xmath78}{\mbox{ch}(2\gamma_{0}d)-\sin(2\theta ) \mbox{sh}(2\gamma_{0}d)}+{\cal o}(\epsilon^2)\ . \label{rupup_down}\ ] ] this expression is identical in form to eq . ( [ rupup ] ) obtained in sec . ii for a nonmagnetic electrode . however , an important difference is @xmath145 . as a result , @xmath289 up to any order in @xmath237 . this means that @xmath290 in eq . ( [ rupup_down ] ) must be a correction only to the argument of @xmath88 , which as discussed in the manuscript , will not modify the orbit of @xmath88 on the complex plane . the orbit is always a unit circle for @xmath291 , and the amount of charge pumped per cycle by the @xmath53 state is quantized to @xmath292 . similarly , for the spin polarization of the electrode antiparallel to the @xmath5-axis , the charge pumped per cycle is @xmath293 . 99 k. klitzing , g. dorda , and m. pepper , phys . * 45 * , 494 ( 1980 ) . a. tsukazaki , a. ohtomo , t. kita , y. ohno , h. ohno , and m. kawasaki , science * 315 * , 1388 ( 2007 ) . y. zhang , y. w. tan , h. l. stormer , and p. kim , nature * 438 * , 201 - 204 ( 2005 ) . c. brne , c. x. liu , e. g. novik , e. m. hankiewicz , h. buhmann , y. l. chen , x. l. qi , z. x. shen , s. c. zhang , and l. w. molenkamp , phys . . lett . * 106 * 126803 ( 2011 ) . r. b. laughlin , phys . b * 23 * , 5632 ( 1981 ) . d. j. thouless , m. kohmoto , m. p. nightingale , and m. den nijs , phys . rev . lett . * 49 * , 405 ( 1982 ) . q. niu , and d. j. thouless , j. phys . a : math * 17 * , 2453 ( 1984 ) . d. j. thouless , phys . rev . b * 27 * , 6083 ( 1983 ) . c. l. kane , and e. j. mele , phys . lett . * 95 * , 226801 ( 2005 ) . b. a. bernevig , and s. c. zhang , phys . lett . * 96 * , 106802 ( 2006 ) . m. knig , s. wiedmann , c. brne , a. roth , h. buhmann , l. w. molenkamp , x. l. qi , and s. c. zhang , science * 318 * , 766 ( 2007 ) . i. knez and r. r. du , frontiers of phys . * 7 * , 200 ( 2012 ) . j. e. moore and l. balents , phys . b * 75 * , 121306 ( r ) ( 2007 ) . l. fu and c. l. kane , phys . b * 76 * , 045302 ( 2007 ) . m. z. hasan , and c. l. kane , rev . * 82 * , 3045 ( 2010 ) . x. l. qi and s. c. zhang , physics today * 63 * , 33 ( 2010 ) . f. d. m. haldane , phys . * 61 * , 2015 ( 1988 ) . c. l. kane and e. j. mele , phys . lett . * 95 * , 146802 ( 2005 ) . d. n. sheng , z. y. weng , l. sheng , and f. d. m. haldane , phys . lett . * 97 * , 036808 ( 2006 ) . e. prodan , phys . b * 80 * , 125327 ( 2009 ) . y. yang , z. xu , l. sheng , b. g. wang , d. y. xing , and d. n. sheng , phys . * 107 * , 066602 ( 2011 ) . x. l. qi , t. l. hughes , and s. c. zhang , phys b * 78 * , 195424 ( 2008 ) . a. m. essin , j. e. moore , and d. vanderbilt , phys . * 102 * , 146805 ( 2009 ) . l. fu , and c. l. kane , phys . b * 74 * , 195312 ( 2006 ) . c. q. zhou , y. f. zhang , l. sheng , r. shen , d. n. sheng , and d. y. xing , phys . rev . b * 90 * , 085133 ( 2014 ) . x. l. qi , t. l. hughes , and s. c. zhang , nature physics * 4 * , 273 ( 2008 ) . f. mahfouzi , b. k. nikoli , s. h. chen , and c. r. chang , phys . b * 82 * , 195440 ( 2010 ) . b. a. bernevig , t. l. hughes , and s. c. zhang , science * 314 * , 1757 - 1761 ( 2006 ) . m. knig , h. buhmann , l. w. molenkamp , t. l. hughes , c. x. liu , x. l. qi , and s. c. zhang , j. phys . * 77 * , 031007 ( 2008 ) . w. yang , k. chang , and s. c. zhang , phys . * 100 * , 056602 ( 2008 ) . c. x. liu , t. l. hughes , x. l. qi , k. wang , and s. c. zhang , phys . lett . * 100 * , 236601 ( 2008 ) . d. g. rothe , r. w. reinthaler , c. x. liu , l. w. molenkamp , s. c. zhang , and e. m. hankiewicz , new j. phys . * 12 * , 065012 ( 2010 ) . g. h. wannier , rev . mod . phys . * 34 * , 645 ( 1962 ) . x. l. qi , phys . lett . * 107 * , 126803 ( 2011 ) . h. li , l. sheng , and d. y. xing , phys . rev . lett . * 108 * , 196806 ( 2013 ) . m. bttiker , h. thomas , and a. prtre , z. phys . b * 94 * , 133 ( 1994 ) . p. w. brouwer , phys . b * 58 * , r10135 ( 1998 ) . m. switkes , c. m. marcus , k. campman , and a. c. gossard , science * 283 * , 1905 ( 1999 ) . f. giazotto , p. spathis , s. roddaro , s. biswas , f. taddei , m. governale , and l. sorba , nature physics * 7 * , 857 ( 2011 ) . g. e. w. bauer , and y. tserkovnyak , physics * 4 * , 40 ( 2011 ) .

entanglement plays a central role in quantum information theory . it is considered a resource for quantum communication and teleportation , as well as for various computational tasks @xcite . the importance of understanding entanglement in a relativistic setting has received considerable attention recently @xcite . such an understanding is certainly relevant from a fundamental point of view , since relativity is an indispensable component of any complete theoretical model . however , it is also important in a number of practical situations , for example , when considering the implementation of quantum information processing tasks performed by observers in arbitrary relative motion . entanglement was shown to be an invariant quantity for observers in uniform relative motion in the sense that , although different inertial observers may see these correlations distributed among several degrees of freedom in different ways , the total amount of entanglement is the same in all inertial frames @xcite . in non - inertial frames , entanglement was first studied indirectly by investigating the fidelity of teleportation between two parties in relative uniform acceleration @xcite . more recently , the observer - dependent character of entanglement was explicitly demonstrated by studying the entanglement between two modes of a free scalar field as viewed by two relatively accelerated observers @xcite . a uniformly accelerated observer is unable to access information about the whole of spacetime since , from his perspective , a communication horizon appears . this can result in a loss of information and a corresponding degradation of entanglement . in essence , the acceleration of the observer effects a kind of `` enviromental decoherence '' , limiting the fidelity of certain quantum information - theoretic processes . a quantitative understanding of such degradation in non - inertial frames is therefore required if one wants to discuss the implementation of certain quantum information processing tasks between accelerated partners . in curved spacetime two nearby inertial observers are relatively accelerated due to the geodesic deviation equation . accordingly , the results of @xcite indicate that in curved spacetime even two inertial observers will disagree on the degree of entanglement in a given bipartite quantum state of some quantum field . indeed , a thorough investigation into entanglement in an expanding curved spacetime shows that entanglement can encode information concerning the underlying spacetime structure @xcite . in this paper we analyze the entanglement between two modes of a dirac field described by relatively accelerated parties in a flat spacetime . we are interested in understanding how both the crucial sign change in fermi - dirac versus bose - einstein distributions , and the finite number of allowed states in fermionic systems due to the pauli exclusion principle ( in contrast to the unbounded excitations that can occur in bosonic systems ) , affect the degradation of entanglement produced by the unruh effect . we find that unlike the bosonic case , where the entanglement degrades completely in the infinite acceleration limit , in the fermionic case the entanglement is never completely destroyed . we analyze the degradation of entanglement in the system by applying the constraints of entanglement sharing @xcite and track the information originally encoded in these quantum correlations by employing a set of multi - qubit complementarity relations @xcite . as in @xcite , our results can be applied to the case that alice falls into a black hole whilst rob barely escapes through eternal uniform acceleration . the remainder of this paper is organized as follows . in sec . [ sec : setting ] we consider two modes of a free dirac field that are maximally entangled from an inertial perspective . two parties , an inertial observer named alice and a uniformly accelerating observer named rob , are each assumed to possess a detector sensitive only to one of the two modes . each measures the field with his / her detector and the results are compared in order to estimate the entanglement between the modes . section [ sec : unruh ] discusses the unruh effect for dirac particles as experienced by rob . if a given dirac mode is in the vacuum state from an inertial perspective , then rob s detector perceives a fermi - dirac distribution of particles . this has a strong effect on the entanglement that exists between alice and rob , and therefore plays an important role in any quantum information task they might perform that uses this entanglement as a resource . in sec . [ sec : ent ] we calculate the entanglement between the modes from the perspectives of both alice and rob . due to the presence of a rindler horizon , rob is forced to trace over a causally disconnected region of spacetime that he can not access . accordingly , his description of the system takes the form of a two - qubit mixed state . we calculate the entanglement using mixed state entanglement measures such as the entanglement of formation @xcite and the logarithmic negativity @xcite . we also estimate the total correlations ( classical plus quantum ) via the mutual information @xcite . our results show that the entanglement of formation does not vanish as it does in the bosonic case , but rather reaches a minimum of @xmath0 in the limit that rob moves with infinite acceleration . since the fermionic system we are considering is accurately described by a pure state of three qubits , we study the constraints placed on the system by the phenomenon of entanglement sharing in sec . [ sec : ent_sharing ] . our analysis shows that no inherently three - body correlations are generated in the quantum state . that is , all of the entanglement in the system is in the form of bipartite correlations , regardless of rob s rate of acceleration . using complementarity relations applicable to an overall pure state of three qubits , as well as to the various two - qubit marginals , we identify the different types of information encoded in the quantum state of our system in sec . [ sec : complementarity ] . this enables us to study how specific subsystem properties depend on rob s rate of acceleration and to explain how some of the entanglement from the inertial frame is able to survive in the fermionic system , even at infinite acceleration . finally , we summarize our results and suggest possible directions for further research in sec . [ sec : conclusions ] . consider a free minkowski dirac field in @xmath1 dimensions @xmath2where @xmath3 is the particle mass , @xmath4 the dirac gamma matrices , and @xmath5 is a spinor wavefunction . minkowski coordinates @xmath6 with @xmath7 are the most suitable to describe the field from an inertial perspective . the field can be expanded in terms of the positive ( fermions ) and negative ( antifermions ) energy solutions of the dirac equation @xmath8 and @xmath9 respectively , since they form a complete orthonormal set of modes , @xmath10 in the above , @xmath11 is a notational shorthand for the wavevector @xmath12 which labels the modes . the positive and negative energy minkowski modes have the form @xmath13 where @xmath14 , and @xmath15 is a constant spinor with @xmath16 indicating spin - up or spin - down along the quantization axis , satisfying the normalization relations @xcite @xmath17 , @xmath18 , with the adjoint spinor given by @xmath19 . the above positive and negative energy solutions satisfy the orthogonormality relations @xmath20 where the dirac inner product for two mode functions is given by @xmath21 the modes @xmath22 are classified as positive and negative frequency with respect to ( the future - directed minkowski killing vector ) @xmath23 for @xmath24 , i.e. @xmath25 the operators @xmath26 and @xmath27 are the creation and annihilation operators for the positive and negative energy solutions of momentum @xmath11 which satisfy the anticonmutation relations @xmath28with all other anticommutators vanishing . the minkowski vacuum state is defined by the absence of any mode excitations in an inertial frame , @xmath29where the @xmath30 superscript on the kets is used to indicate the particle and anti - particle vacua , respectively so that @xmath31 . we will use the notation here , and throughout the rest of the work , that the mode index ( @xmath32 ) will be a subscript affixed to the occupation number inside the ket , and that the absence of a subscript on the outside of the ket indicates a minkowski fock state . since @xmath33 , there are only two allowed states for each mode , @xmath34 and @xmath35 for particles , and similarly for anti - particles . consider two maximally entangled fermionic modes in an inertial frame , @xmath36 and @xmath37 indicate the modes associated with the observers alice and rob , respectively . all other modes of the field are in the vacuum state , and therefore the state can be written as @xmath38 . now assume that alice is stationary and has a detector sensitive only to mode @xmath39 . rob moves with uniform acceleration and takes with him a detector that only detects particles corresponding to mode @xmath40 . we ask the question of what is the entanglement between modes @xmath39 and @xmath40 observed by alice and rob , given that rob undergoes uniform acceleration . note that in order to determine the amount of entanglement , alice and rob perform measurements which are then compared by either party in order to estimate the correlations in the results . due to rob s acceleration , at some point alice s signals will no longer reach rob , but rob s signals will always be available to alice , ( see ) . at this point only alice can compare the measurement results and estimate the entanglement of the state . let us now consider the state observed by rob . consider rob to be uniformly accelerated in the @xmath41 plane ( @xmath42 ) . rindler coordinates @xmath43 are appropriate for describing the viewpoint of an observer moving with uniform acceleration . two different sets of rindler coordinates , which differ from each other by an overall change in sign , are necessary for covering minkowski space @xcite . these sets of coordinates define two rindler regions that are causally disconnected from each other . @xmath44 where @xmath45 denotes rob s proper acceleration . the above set of coordinates both give rise to the same metric @xmath46 where @xmath47 are the same in both minkowski and rindler spacetimes . are hyperbolas and lines of constant proper time @xmath48 for the accelerated observer run through the origin . note that while @xmath48 flows in the direction of @xmath49 in region i , it flows in the direction of @xmath50 in region ii , ( i.e the dashed line rotates counter - clockwise for increasing values of @xmath48 ) . a uniformly accelerated observer rob ( @xmath37 ) with acceleration @xmath45 travels on a hyperbola constrained to region i , while a fictitious observer anti - rob @xmath51 travels on a corresponding hyperbola in region ii given by the negative of rob s coordinates . the horizons @xmath52 are lines of @xmath53 which alice ( @xmath54 ) will cross at finite minkowski times.,width=336,height=240 ] a particle undergoing eternal uniform acceleration remains constrained to either rindler region i or ii and has no access to the opposite region , since these two regions are causally disconnected . figure [ fig1 ] serves to illustrate these ideas , as well as to introduce our labeling scheme , where we refer to the accelerated observer in region i as rob ( r ) and to the corresponding fictitious observer confined to region ii ( whose coordinates are the negative of rob s ) as anti - rob @xmath55 . the coordinates @xmath43 have the ranges @xmath56 separately in region i and in region ii . this implies that region i and ii each admit a separate quantization procedure with corresponding positive and negative energy solutions @xmath57 and @xmath58 . since the rindler metric is static ( independent of @xmath48 ) it will admit solutions of the form @xmath59 , with @xmath60 a spatially dependent spinor @xcite . particles and anti - particles will be classified with respect to the future - directed timelike killing vector in each region . in region i this is given by @xmath61 where @xmath62 which is a boost into the instantaneous comoving frame of rob . thus , mode solutions in region i having time dependence @xmath63 with @xmath64 represent positive frequency solutions since @xmath65 . however , in region ii @xmath61 points in the opposite direction of @xmath66 ( increasing @xmath48 flows in the direction of @xmath50 , see ) . hence in region ii the future - directed timelike killing vector is given by @xmath67 @xcite . thus , a solution in region ii with time dependence @xmath68 with @xmath64 is actually a negative frequency mode since @xmath69 . hence , the positive frequency mode in region ii is given by @xmath70 with @xmath64 satisfying @xmath71 . due to the causally disjoint nature of region i and ii , the modes @xmath72 have support only in region i and vanish in region ii , while the opposite is true for the modes @xmath73 in region ii . the rindler modes satisfy orthonormality relations analogous to the minkowski modes @xcite @xmath74 , and @xmath75 where @xmath76 . in region i , let us denote @xmath77 as the annihilation and creation operators for fermions ( particles ) and @xmath78 as the annihilation and creation operators for anti - fermions ( anti - particles ) . the corresponding particle and anti - particle operators in region ii are denoted as @xmath79 and @xmath80 . these obey the usual dirac anti - commutation relations @xmath81 @xmath82 @xmath83 , with all other anti - commutators , including those between operators in region i and the causally disconnected region ii , equaling zero . taking into account the two sets of modes in each rindler region , the dirac field can be expanded , in analogy to , as @xmath84 and represent the decomposition of the dirac field in minkowski and rindler modes , respectively . we can therefore relate the minkowski and rindler creation and annihilation operators by taking appropriate inner products . using the rindler orthogonality relations and we have @xmath85 . substituting for @xmath86 in this last expression yields c^_k = dk ( ^_kk a_k + ^_kk b^_k ) , \{i , ii } where the bogoliubov coefficients are given by the inner product of the rindler modes wavefunctions with the minkowski positive and negative frequency modes ^_kk = ( ^+_k , ^+_k ) , ^_kk = ( ^+_k , ^-_k ) . a similar calculation for @xmath87 yields the corresponding expression d^_k = dk ( ^_kk b_k + ^_kk a^_k ) , \{i , ii } with the same bogoliubov coefficient as in . in deriving use has been made of the following properties of the dirac inner product : @xmath88 . the calculation of the bogoliubov coefficients is straightforward , though lengthy and an exercise in special functions . details can be found elsewhere @xcite . for our purposes , the end result of such calculations yields a relationship between the minkowski and rindler creation and annihilation operators given by the bogoliubov transformation = , where @xmath89 with @xmath90 , the ratio of the frequency @xmath91 to the only naturally occurring frequency in the problem @xmath92 , and @xmath93 is an unimportant phase that can always be absorbed into the definition of the operators . it is easy to see from and its adjoint that given the anti - commutation relations of the rindler operators , the minkowski anti - commutation relations are preserved . in we have made the @xcite , which is valid if we consider rob s detector as sensitive to a single particle mode in region i such that we can approximate the frequency @xmath94 observed by alice to be the same as the frequency @xmath95 as observed by rob , i.e @xmath96 @xcite . note that this bogoliubov transformation mixes a particle in region i and an anti - particle in region ii . correspondingly , the bogoliubov transformation that mixes an anti - particle mode in region i and a particle in region ii is given by = . since the anti - commutators between particle and anti - particle operators and between region i and region ii rindler operators are zero , it is easy to see that the minkowski operators in anti - commute with the minkowski operators in , as they should ( since @xmath11 and @xmath97 represent two separate modes ) . as stated above , reveals that the minkowski particle annihilation operator is a bogoliubov transformation between a particle in region i of momentum @xmath11 and an anti - particle in region ii of momentum @xmath97 , i.e. a transformation that mixes creation and annihilation operators . we can understand this in terms of our previous discussion of the time dependence of positive frequency rindler modes in regions i and ii . for a massless dirac field a positive frequency rindler mode has the form @xcite @xmath98 ( for the scalar case see @xcite ) and a positive frequency rindler mode in region ii has the form @xmath99 . thus , in order to construct a positive frequency minkowski mode @xmath100 that extends @xmath101 analytically from region i to region ii , we need a linear combination of @xmath101 and @xmath102 so that both have the space and time dependence @xmath103 . this means that the minkowski operator @xmath104 must mix @xmath105 and @xmath106 as in . similarly , to construct a minkowski mode that analytically extends @xmath107 from region ii to region i we must form a linear combination of @xmath107 and @xmath108 so that both have the space and time dependence @xmath109 this corresponds to @xmath110 mixing @xmath111 and @xmath112 as in . for a massive dirac field the time dependence of the rindler modes remains the same , but the spatial dependence is more complicated @xcite . the above argument for the mixing of rindler particles ( anti - particles ) in region i ( ii ) and anti - particles ( particles ) in region ii ( i ) goes through unchanged . having related minkowski and rindler creation and annihilation operators , we now wish to relate the minkowski vacuum to the corresponding rindler vacuum . it is useful to note that can be written as a two - mode squeezing transformation @xcite @xmath113 = s\ , \left [ \begin{array}{c } c^{i}_k \\ d^{ii\dagger}_{-k } \\ \end{array } \right]\ , s^\dagger\ ] ] for the single mode @xmath11 with @xmath114 given by @xmath115.\ ] ] using the relation @xmath116 + [ b,[b , a]]/2 ! + [ b,[b,[b , a]]]/3 ! + \cdots$ ] and the identity @xmath117 = a\,\{b , c\ } - \{a , c\}\,b$ ] we find a_k & = & s c^i_k s^= c^i_k - r e^-id^ii_-k - c^i_k + e^-id^ii_-k + & = & r c^i_k - e^-i r d^ii_-k . and b^_-k & = & s d^ii_-k s^= d^ii_-k + r e^ic^i_k - d^ii_-k - e^ic^i_k + & = & r d^ii_-k + e^-i r c^i_k . now @xmath104 and @xmath118 , respectively , annihilate the single mode particle and anti - particle minkowski vacua @xmath119 and @xmath120 . since @xmath104 mixes particles in region i and anti - particles in region ii , we postulate that the minkowski particle vacuum for mode @xmath11 in terms of rindler fock states is given by |0_k^+ = _ n=0 ^ 1 a_n |n_k^+_i|n_-k^-_ii , where @xcite lll c^i_k |0_k^+_i = 0 & & d^ii_-k |0_-k^-_ii = 0 , + & & + c^i_k |0_k^+_i = |1_k^+_i & & d^ii_-k |0_-k^-_ii = |1_-k^-_ii . + as a comment on notation , the rindler region i and ii fock states carry a subscript i or ii , respectively , on the kets while the minkowski fock states are indicated by the absence of an outside subscript on the kets . momentum mode labels are attached to the fock occupation number , and the @xmath30 ket superscript indicates a particle or anti - particle state , respectively . applying @xmath104 from to , we have 0 & = & a_k |0_k^+ & = & ( r c^i_k - e^-i r d^ii_-k ) _ n=0 ^ 1 a_n e^-ir . @xmath122 yields @xmath123 so that we finally arrive at |0_k^+ & = & r where the second equality in is very useful for keeping track of any possible transposition signs arising from the anti - commuting operators when applying minkowski operators to @xmath124 . as an important example that we will use subsequently , a short calculation of @xmath125 using the adjoint of a_k^= r c^i_k - e^+i r d^ii_-k , acting on the lower expression for @xmath124 in , yields a_k^|0_k^+ & = & ( ^2 r c^i_k - ^2 r d^ii_-kc^i_kd^ii_-k ) |0_k^+_i|0_-k^-_ii , & = & ( ^2 r c^i_k + ^2 r c^i_kd^ii_-kd^ii_-k , ) |0_k^+_i|0_-k^-_ii & = & c^i_k|0_k^+_i|0_-k^-_ii , the anti - commutation relations between @xmath126 and @xmath127 to obtain a minus sign upon transposition , and in the third equality we have used the anti - commutation relations to write @xmath128 , noting that the latter term annihilates the anti - particle vacuum . lastly , one can easily verify that @xmath129 acting on @xmath130 yields zero , a_k^|1_k^+ = ( r c^i_k - e^+i r d^ii_-k ) |1_k^+_i|0_-k^-_ii = 0 , ensuring that @xmath131 . we note that the form of the dirac particle vacuum for mode @xmath11 in , which can be written as @xmath132 , is complementary to the form of the minkowski charged scalar vacuum for mode @xmath11 , which is given by @xcite @xmath133 where in the latter case @xmath134 is defined by @xmath135 . qualitatively , in going from the scalar field to the dirac field , scalar mode functions are replaced by spinors , the infinite number of equally spaced bosonic levels are replaced by two fermionic levels , and the hyperbolic functions in the bogoluibov transformation between minkowski and rindler modes / operators are replaced by the corresponding trigonometric functions ( in essence @xmath136 in going from the scalar to the dirac field ) . the two minkowski states , @xmath124 and @xmath130 , correspond to the particle field of mode @xmath11 observed by alice . on the other hand , an observer moving with uniform acceleration @xmath45 in one of the regions has no access to field modes in the causally disconnected region . therefore , the observer must trace over the inaccessible region , constituting an unavoidable loss of information about the state , which essentially results in the detection of a mixed state . thus , when a minkowski observer detects a vacuum state @xmath137 for mode @xmath11 , an accelerated observer in region i sees a distribution of particles according to the marginal state describing region i , _ k^i & = & tr_ii&= & ^2 r|0_k^+_i^+_i0_k| + ^2 r|1_k^+_i^+_i1_k| . as the region i observer accelerates through the minkowski particle vacuum @xmath124 of mode @xmath11 his detector registers a number of particles given by & = & tr_i= ^2 r ^+_i1_k|c^i_k c^i_k |1_k^+_i & = & ^2 r = where use has been made of @xmath89 with @xmath138 , and we have defined the unruh temperature ( where @xmath139 is boltzmann s constant ) as t = . equation ( [ thermalization ] ) is known as the unruh effect @xcite which shows that the uniformly accelerated observer in region i detects a thermal fermi - dirac ( fd ) distribution of particles as he traverses the minkowski vacuum . qualitatively , we can understand the unruh effect as follows . the constant proper force @xmath140 that acts on a mass @xmath3 ( a detector ) in rob s instantaneous co - moving frame , to keep it under uniform acceleration , can be written as @xmath141 ( which can be integrated as @xmath142 where @xmath143 , @xmath144 , to yield the hyperbolic orbits in ) . the work @xmath145 that is performed on the particle to keep it in uniform acceleration is the product of @xmath140 times a characteristic distance through which the force acts , which we take to be the compton wavelength of the particle @xmath146 . this yields @xmath147 , which is proportional to @xmath148 in . thus , the energy that is supplied to keep @xmath3 under constant acceleration goes into exciting rob s detector , and curiously has a thermal spectrum ( see @xcite ) . in the general relativistic case this energy is supplied by the gravitational field acting on a stationary observer ( rob ) situated outside a black hole , who experiences constant acceleration by nature of his stationarity . a freely falling observer ( alice ) who eventually crosses the event horizon would experience the kruskal vacuum ( the minkowski - like vacuum of ) and thus detect no unruh radiation with a co - moving detector . note that in the case of a scalar field , has a crucial minus sign in the denominator leading to a bose - einstein ( be ) distribution of particles appropriate for bosons . in both the scalar and dirac case , the state of the minkowski vacuum for mode @xmath11 is a two - mode squeezed state ( bosonic and fermionic , respectively ) , which to a uniformly accelerated observer confined entirely to region i , is detected as a thermal state ( be and fd , respectively ) . in the following we investigate how the unruh effect for dirac particles affects the entanglement between various dirac modes . specifically , we study the entangled bell state given by in the case that rob is uniformly accelerated in region i. assuming a detector for rob that is sensitive to a single rindler particle mode ( @xmath149 ) , we decompose rob s single particle minkowski states in into the appropriate rindler particle and anti - particle states utilizing eqs . ( [ vacm ] ) and ( [ 1 m ] ) . we then proceed to evaluate various measures of entanglement . as stated in the introduction , the advantage of utilizing a dirac field over a bosonic scalar field is that due to the finite occupation of the fermionic states , we obtain finite dimensional density matrices that lead to closed form expressions for the entanglement measures that are more easily interpreted than their infinite dimensional bosonic counterparts . in this and subsequent sections we will use the following notation : @xmath54 will indicate the inertial observer alice , i will indicate the uniformly accelerated observer rob ( r ) confined to region i , and ii will indicate the fictitious complementary observer anti - rob ( @xmath150 ) in region ii ( see ) , that arises from the second ( negative ) set of rindler coordinates in . furthermore , since we are in the single mode approximation , we will drop all labels ( @xmath151 ) on states and density matrices indicating the specific mode . thus , the minkowski particle mode @xmath152 for alice will be written @xmath153 , the rindler region i particle mode @xmath154 ( for @xmath155 ) . likewise , we will refer to the minkowski mode for alice " simply as mode @xmath54 " , the rindler particle mode in region i as mode i , " and the rindler anti - particle mode in region ii as mode ii . " the density matrix for the minkowski entangled state in eq . ( [ eq : ent ] ) is , from an inertial perspective @xmath156 in the basis @xmath157 with @xmath158 . here , and only here , we have used the subscript @xmath37 to indicate rob s minkowski fock states in . to describe the entanglement of the state as seen by an inertial alice and a uniformly accelerated rob , we expand the minkowski particle states @xmath159 and @xmath160 into rindler region i and ii particle and anti - particle states using eqs . ( [ vacm ] ) and ( [ 1 m ] ) to obtain @xmath161 which we write in matrix form as @xmath162 in the basis @xmath163 where for notational convenience we have defined @xmath164 . note that the unimportant phase factor @xmath93 discussed in section [ sec : unruh ] has been absorbed into the definition of the creation and annihilation operators . since rob is causally disconnected from region ii , we take the trace over the mode in this region , which results in a mixed density matrix between alice and rob @xmath165 in the basis @xmath157 where @xmath166 . to determine whether or not this state is entangled we use the partial transpose criterion @xcite . this criterion specifies a necessary and sufficient condition for the existence of entanglement in a mixed state of two qubits . if at least one eigenvalue of the partial transpose of the density matrix is negative , then the density matrix is entangled . the partial transpose is obtained by interchanging alice s qubits @xmath167 @xmath168@xmath169 has eigenvalues @xmath170 , the last of which @xmath171 , is always negative for @xmath172 i.e. for @xmath173 . this means that the state is always entangled . quantifying entanglement for mixed states is fairly involved entanglement . a pure state of a bipartite system @xmath45 and @xmath174 can always be written in the schmidt basis , @xmath175 , where the quantum correlations between the states are evident . entanglement between the systems is given by the von - neumann entropy of the reduced density matrix @xmath176 defined as @xmath177 , which is a function of the schmidt coefficients @xmath178 . unfortunately , there is no analog of the schmidt decomposition for mixed states , and the von - neumann entropy is no longer a good measure of mixed state entanglement . a set of conditions that mixed state entanglement measures should satisfy is well known @xcite . there is no unique measure , and several different mixed state entanglement measures have been proposed . among the most popular are those related to the formation and distillation of entangled states . consider the number @xmath3 of maximally entangled pairs needed to create @xmath179 arbitrarily good copies of an arbitrary pure state using only local operations and classical communication . the entanglement of formation is defined as the asymptotic conversion ratio , @xmath180 in the limit of infinitely many copies @xcite , @xmath181where the minimum is taken over all the possible realizations of the state @xmath182 . the opposite process gives rise to the definition of the entanglement of distillation . this is the asymptotic rate of converting non - maximally entangled states into maximally entangled states by means of a purification procedure . the entanglement of distillation is in general smaller than that of formation . this shows that the entanglement conversion is irreversible and is due to a loss of classical information about the decomposition of the state . bound entangled states are a consequence of this : no entanglement can be distilled from them even though they are inseparable . the entanglement of formation can be explicitly calculated for two qubits . it is given by @xmath183 where @xmath184 is the concurrence . the concurrence of a pure state @xmath185 of two qubits is given by @xmath186 where @xmath187 represents the ` spin - flip ' of @xmath185 and * " denotes complex conjugation in the standard basis . the generalization of the concurrence to a mixed state @xmath188 of two qubits follows by minimizing the average concurrence over all possible pure state ensemble decompositions of @xmath188 , defined by a convex combination of pure states @xmath189 , such that @xmath190 . in this way , @xmath191 wootters @xcite succeeded in deriving an analytic solution to this difficult minimization procedure in terms of the quantities @xmath192 defined as the square roots of the eigenvalues ( which are all positive ) of the non - hermitian operator @xmath193 , ordered in decreasing order . the matrix @xmath194 is the spin - flip of the quantum state @xmath195 which not only exchanges the states @xmath196 and @xmath197 , but , in general , also introduces a relative phase . the closed form solution for the concurrence of a mixed state of two qubits is given by @xmath198 unfortunately , the entanglement of distillation can not be explicitly calculated in this case . therefore we will use the logarithmic negativity @xcite which serves as an upper bound on the entanglement of distillation . it is not an entanglement measure because it does not satisfy the requirement of being equal to the von - neuman entropy for pure states . however , it is an entanglement monotone since it satisfies all other criteria to quantify entanglement . it is defined as @xmath199 where @xmath200 is the trace - norm of the partial transpose density matrix @xmath201 i.e. , the sum of the eigenvalues of @xmath202 . since the matrices @xmath203 in this work are symmetric , @xmath204 is simply given by @xmath205 of the sum of the absolute values of the eigenvalues of @xmath201 . to quantify the entanglement of @xmath206 in eq . ( [ eq : state ] ) we compute the spin - flip matrix @xmath207 @xmath208 and find that @xmath209 has eigenvalues @xmath210 . the concurrence is then given by @xmath211 , which is unity at zero acceleration , as expected , and approaches the value @xmath0 for infinite acceleration @xmath212 . the entanglement of formation is @xmath213 and the logarithmic negativity is @xmath214 in an inertial frame , @xmath215 , and the state of the system defined by is maximally entangled . in the limit of infinite acceleration , the logarithmic negativity is @xmath216 , implying that the entanglement in the infinite acceleration limit is finite . this means that the state is always entangled and can be used as a resource for performing certain quantum information processing tasks . this is in strong contrast to the bosonic case @xcite , where it is found that the entanglement goes to zero in this limit . as in @xcite the infinite acceleration limit can be interpreted as alice falling into a black hole while rob hovers outside the black hole , barely escaping the fall . since close to a black hole horizon spacetime is flat , rob must be uniformly accelerated in order to escape the black hole . however , alice falls in as she is an inertial observer . the connection to the present situation is made by recognizing that , according to rob , a communication horizon appears causing him to lose information about the state in the whole of spacetime . as a result , the inertial entanglement is degraded , and there will be a reduction in the fidelity of any information processing task performed by alice and rob using this state . the analogy with the black hole scenario is further strengthened by observing that classical information can only flow from rob to alice after alice has crossed the horizon . we may also calculate the total correlations between any two subsytems of the overall system by using the mutual information @xcite , @xmath217this measure quantifies how much information two correlated observers ( one with access to subsystem @xmath45 and the other with access to subsystem @xmath174 ) possess about one another s state . equivalently , it represents the distance between the actual joint distribution and the product state obtained when all correlations are neglected . calculating the relevant marginal density operators , the mutual information of the state ( [ eq : state ] ) is found to be @xmath218where we have used the fact that the density matrix in region i is @xmath219 $ ] , so that @xmath220similarly , the density matrix of mode @xmath54 is @xmath221 , yielding @xmath222 . finally , using we find that @xmath223 the mutual information is equal to @xmath224 at @xmath215 and goes to unity in the infinite acceleration limit this behavior is reminiscent of that seen in the bosonic case @xcite . to explore entanglement in this system in more detail we consider the tripartite system consisting of the modes @xmath54 , i , and ii . in an inertial frame the system is bipartite , but from a non - inertial perspective an extra set of modes in region ii becomes relevant . we therefore calculate the entanglement in all possible bipartite divisions of the system as well as any tripartite correlations that may exist . we investigate pure state entanglement in three different bipartite divisions of the system . since the overall state of the system is pure , these entanglements are uniquely quantified by the von - neumann entropy . in what follows we make repeated use of the fact that , for a bipartite pure state @xmath225 with subsystems @xmath45 and @xmath174 , @xmath226 and @xmath227 . there are three separate cases to consider . * the entanglement between mode @xmath54 and ( modes i and ii ) is given by @xmath228 . thus , these two subsystems are always maximally entangled , regardless of the value of the parameter @xmath134 . * the entanglement between mode i and ( modes @xmath54 and ii ) is given by eq . ( [ sr1 ] ) since @xmath229 . * finally , the entanglement between mode ii and ( modes @xmath54 and i ) is given by eq . ( [ skr1 ] ) since @xmath230 . figure ( [ fig : entropy ] ) shows the entanglement in each of these bipartite partitions as functions of @xmath134 . [ entropy ] for mixed state entanglement we can consider , besides the entanglement between the minkowski mode @xmath54 and rindler mode i calculated in sec . [ sec : ent ] , two additional bipartite divisions of the subsystems . * the entanglement between mode @xmath54 and mode ii . tracing over the mode in region i , we obtain the density matrix @xmath231the partial transpose of @xmath232 is given by @xmath233which has eigenvalues @xmath234 , the last of which @xmath235 , is less than or equal to zero . at @xmath236 the eigenvalue is zero , which means that there is no entanglement at this point . however , for any @xmath237 entanglement does exist between these two modes according to the partial transpose criterion . the logarithmic negativity in this case is given by @xmath238 . + calculating the spin - flip of @xmath232 @xmath239we find that @xmath240has eigenvalues @xmath241 . thus , the concurrence is given by @xmath242 , which is zero at zero acceleration as expected , and approaches the value @xmath0 for infinite acceleration @xmath212 . the entanglement of formation is @xmath243and the mutual information is@xmath244at @xmath215 the mutual information is zero and approaches unity as the rate of acceleration goes to infinity . * the entanglement between mode i and mode ii . tracing over the modes in @xmath54 , we obtain the density matrix @xmath245the partial transpose of @xmath246 ( obtained by interchanging @xmath247 s qubits ) is given by @xmath248which has eigenvalues @xmath249 , the last of which @xmath250 $ ] is less than or equal to zero . again , at @xmath215 there is no entanglement between these two subsystems . yet , similar to the last case , entanglement does exist between these two modes in non - inertial frames according to the partial transpose criterion . further , the logarithmic negativity @xmath251 is nonzero for all @xmath237 . the spin - flip of @xmath252 is given by @xmath253and the matrix @xmath254has eigenvalues @xmath255 . thus , the concurrence is given by @xmath256 which is zero at zero acceleration , and approaches the value @xmath257 for infinite acceleration @xmath212 . the entanglement of formation in this case is + @xmath258 + and the mutual information is@xmath259 again , we find that the mutual information is zero at @xmath215 , and increases to a finite value ( in this case @xmath260 ) , as the rate of acceleration goes to infinity . the results of the above calculations are shown in figs . ( [ fig : nega ] ) - ( [ fig : mutual ] ) . [ nega ] [ formation ] [ mutual ] entanglement in triparite systems has been studied by coffman , et . al . , @xcite for the case of three qubits . they found that such quantum correlations can not be arbitrarily distributed amongst the subsystems ; the existence of three - body correlations constrains the distribution of the bipartite entanglement which remains after tracing over any one of the qubits . for example , in a ghz - state , @xmath261 , tracing over one qubit results in maximally mixed state containing no entanglement between the remaining two qubits . in constrast , for a w - state , @xmath262 , the average remaining bipartite entanglement is maximal . coffman , et . al . , analyzed this phenomenon of entanglement sharing @xcite , using an entanglement monotone known as the tangle , defined as the square of the concurrence @xmath263 . they also introduced a new quantity , known as the residual tangle , in order to quantify the irreducible tripartite correlations in a system of three qubits ( @xmath45 , @xmath174 , and @xmath264 ) @xcite . the definition is motivated by the observation that the tangle of @xmath45 with @xmath174 plus the tangle of @xmath45 with @xmath264 can not exceed the tangle of @xmath45 with the joint subsystem @xmath265 , i.e. , @xmath266 subtracting the terms on the left hand side of eq . ( [ eq tangle ineq ] ) from that on the right hand side yields a nonnegative quantity referred to as the residual tangle @xmath267 , i.e. , @xmath268 the residual tangle or three tangle is interpreted as quantifying the inherent tripartite entanglement present in a system of three qubits , i.e. , the entanglement that can not be accounted for in terms of the various bipartite tangles . this interpretation is given further support by the observation that the residual tangle is invariant under all possible permutations of the subsystem labels @xcite . to quantify tripartite entanglement in our system we use the residual tangle @xmath269 . this quantity is zero for the situation we are considering , since @xmath270 , @xmath271 and @xmath272)=1 $ ] . thus , the state that we are studying has no tripartite correlations for any value of the acceleration rate . instead , any entanglement existing in the system is necessarily bipartite in nature . the marginal bipartite tangles , plotted in fig . [ fig : tangles ] as functions of @xmath134 , are found to be strongly constrained . for low rates of acceleration , modes @xmath54 and i remain almost maximally entangled while there is very little entanglement between modes i and ii and between modes @xmath54 and ii . as the acceleration grows , the entanglement between modes i and ii and between modes @xmath54 and ii increases , while the entanglement between modes @xmath54 and i is degraded . the main system of interest ( mode @xmath54 plus mode i ) becomes increasingly entangled to mode ii and therefore , after tracing over mode ii , we observe an effect analogous to environmental decoherence in which the modes in region ii play the role of the environment . the distribution of entanglement in the system is non - trivial since entanglement between modes @xmath54 and i is not conserved as it is in the inertial case . we next analyze our system in terms of several complementarity relations designed to identify the different types of information encoded in a quantum state in an attempt to better understand how various subsystem properties depend on rob s rate of acceleration . specifically , we are interested in explaining ( i ) the nonconservative nature of the entanglement discussed in the last section and ( ii ) the fact that not all of the initial entanglement between alice and rob is destroyed , even at infinite acceleration . the latter of these two results corresponds to the most obvious difference between the fermionic case studied here and the bosonic case investigated in @xcite . we begin by making use of the relationship @xcite @xmath273 which shows that an arbitrary state of two qubits ( @xmath45 and @xmath174 ) exhibits a complementary tradeoff between the amounts of separable uncertainty @xmath274 , bipartite entanglement ( as quantified by the tangle ) @xmath48 , and a unitarily invariant measure of information about the single particle properties @xmath275 that it encodes . the separable uncertainty @xmath276 @xmath277 is a measure of the uncertainty or ignorance encoded in the two - qubit mixed state @xmath195 regarding individual subsystem properties that is unrelated to the presence of entanglement between the qubits . in eq . ( [ eq : localcorrelations ] ) , @xmath278 represents the spin - flip of @xmath195 , and @xmath279 is the marginal mixedness of @xmath195 . similarly , @xmath280 is a measure of the information pertaining to a single qubit encoded in the marginal density operator @xmath281 . specifically , @xmath282 $ ] is the average of the squares of the single qubit properties associated with qubit @xmath283 . the first of these properties , the coherence @xmath284 of qubit @xmath11 , quantifies , e.g. , the fringe visibility in the context of a two - state system incident on an interferometer , and is given by @xmath285 where @xmath286 is the raising operator acting on qubit @xmath11 . similarly , the predictability @xmath287 which quantifies the _ a priori _ information regarding whether qubit @xmath11 is in the state @xmath288 or the state @xmath289 , e.g. , whether it is more likely to take the upper or lower path in an interferometer , is given by@xmath290 where @xmath291 and @xmath292 is the plus ( minus ) one eigenvector of @xmath293 . figures [ fig : tangles ] and [ fig : eta ] plot the bipartite components of eq . ( [ eq : complementarity ] ) for the various two - qubit marginals obtained after tracing over any one of the three subsystems ( @xmath54 , i , or ii ) as functions of the parameter @xmath134 , while fig . [ fig : single_part ] does the same for the relevant single particle information measures . the complementary nature of these quantities is illustrated by the pattern / color scheme chosen for the curves , where each pattern ( color ) corresponds to a unique set of properties satisfying eq . ( [ eq : complementarity ] ) . adding together all of the curves that contain a common attribute in figs . [ fig : tangles ] - [ fig : single_part ] yields the constant value one , independent of @xmath134 . [ tangles ] [ eta ] [ single_part ] for example , taking the sum of the two dashed ( red ) curves in figs . [ fig : tangles ] and [ fig : eta ] and the two curves containing dashes ( red ) in fig . [ fig : single_part ] , thick - thin dashed ( blue - red ) and dot - dashed ( green - red ) , corresponds to the equality @xmath294 . note that the curves in fig . [ fig : single_part ] , representing information encoded in the individual qubits , each contribute to exactly two distinct complementarity relations . the tradeoffs expressed by eq . ( [ eq : complementarity ] ) , and illustrated in figs . [ fig : tangles ] - [ fig : single_part ] , show how rob s acceleration determines the distribution of individual and bipartite properties of the system . for example , adding together the thick solid ( blue ) and dotted ( green ) curves in fig . [ fig : tangles ] shows that , although the entanglement in any bipartite subsystem changes with acceleration , the total entanglement between alice and ( the modes in regions i and ii ) is always maximal and constant . further , since rob is forced to trace over the causally disconnected modes in region ii , the entanglement arising between the modes in region i and region ii is ultimately responsible for the unruh radiation that he sees . indeed alice , who has access to the modes in both regions , always sees an undisturbed vacuum . however , our analysis shows that the maximum amount of thermalization that rob can experience ( even at infinite acceleration ) is limited by the amount of entanglement that exists between alice and rob in an inertial frame . one straightforward way to see this is to consider the case in which alice and rob initially share no entanglement ( e.g. an arbitrary product state between alice and rob ) . then , as rob s acceleration approaches infinity , we find that @xmath295 approaches its maximum possible value of one ( see with @xmath296 @xcite ) . this is in contrast to the situation considered here in which the initial entanglement between alice and rob is maximal , constraining @xmath295 to a maximum value of 1/4 . since the overall state of our system is always pure , we may also apply the result @xcite @xmath297 which holds for an arbitrary pure state of three qubits ( @xmath45 , @xmath174 , and @xmath264 ) . the quantity @xmath298 is a measure of the total pairwise entanglement shared between qubit @xmath11 and all other qubits . equation ( [ eq : entsharingrel ] ) quantifies an explicit tradeoff between the inherent three - body correlations in a tripartite system , the total pairwise entanglement of a selected qubit , and the single particle properties of the qubit . because the three tangle is always zero for our system , this pure state complementarity relationship simplifies to @xmath299 it is then straightforward to see that this tradeoff is also captured by the pattern / color scheme used in the above figures . simply adding the sum of any two curves in fig . [ fig : tangles ] to the curve in fig . [ fig : single_part ] that is composed of the same two attributes ( or colors ) again yields the constant value one , regardless of rob s acceleration . in particular , this shows that the entanglement between alice and rob in region i can not vanish at infinite acceleration as it did in the bosonic case . this is because the entanglement @xmath295 generated between regions i and ii ( dashed / red curve in fig . [ fig : tangles ] ) and the single particle properties @xmath300 that manifest for rob ( dashed / red - blue curve in fig . [ fig : single_part ] ) are insufficient to satisfy eq . ( [ eq : entsharingrelsimple ] ) . instead , the contribution from the entanglement @xmath301 between alice and rob ( solid / blue curve in fig . [ fig : tangles ] ) must also be taken into account to ensure that @xmath302 . finally , we note that the various two - qubit marginal density operators for our system have a very specific form ; they are all examples of the maximally entangled states with fixed marginal mixednesses ( memms ) identified in @xcite . indeed , this fact is closely related to the absence of tripartite correlations in our system , which implies that any entanglement that exists in the system must necessarily be bipartite . as shown in @xcite , the memms are characterized by the relationship @xmath303 , i.e. , the marginal mixednesses are equal to the separable uncertainties . thus , our fermionic system encodes the maximum amount of bipartite entanglement consistent with the separable uncertainty in each two - qubit marginal for every possible value of the acceleration rate . we have studied the behavior of the entanglement between two modes of a free dirac field in a non - inertial frame in flat spacetime from the point of view of two observers , alice and rob , in relative uniform acceleration . our results show that entanglement existing between alice and rob in an inertial frame is progressively degraded by the unruh effect as rob s rate of acceleration increases . however , unlike the bosonic case in which the inertial entanglement vanishes in the limit of infinite acceleration , in this case the entanglement achieves a minimum value of @xmath0 for the entanglement of formation ( and of @xmath304 for the tangle ) . this fundamental difference , a consequence of the fact that fermions have access to only two quantum levels vs. the infinite ladder of excitations available to bosons , means that in this case alice and rob always share some entanglement which can in principle be used as a resource for performing certain quantum information processing tasks . further analysis shows that the total ( quantum plus classical ) correlations in the system , as quantified by the mutual information , behaves in a manner reminiscent of the bosonic case , decreasing from 2 for inertial observers to unity in the case of infinite acceleration . considering the ( causally inaccessible to rob ) modes in region ii to be a third subsystem allows us to analyze this system in terms of entanglement sharing . in doing so , we find that the overall tripartite pure state never encodes any inherently three - body correlations , regardless of the rate of acceleration . any entanglement existing in the fermionic system is therefore necessarily bipartite . such entanglement is known to be an invariant quantity for inertial observers , although different inertial observers may see these quantum correlations distributed among multiple degrees of freedom in different ways . however , in this analysis we find no indication that the entanglement in any bipartite subsystem is conserved when rob is allowed to accelerate . in fact , our results show that the presence of the communication horizon , which isolates the modes in region ii from an accelerated observer in region i , plays a key role in degrading the inertial entanglement between alice and rob . further insight into this behavior is gained by applying both pure and mixed state complementarity relations to different divisions of the system into subsystems . the ability of this formalism to identify the different types of information encoded in a quantum state facilitates the study of how various subsystem properties depend on rob s rate of acceleration . for instance , the constraints imposed by these relations illustrate how multi - qubit complementarity prevents the entanglement between alice and rob from vanishing at infinite acceleration as it does in the bosonic case . additionally , we find that the amount of vacuum thermalization ( due to unruh radiation ) that rob experiences at infinite acceleration is constrained by the amount of entanglement that he shares with alice in an inertial frame . one possible avenue for further research along these lines is to study the entanglement between alice and rob in the case that alice has acceleration @xmath305 and rob has acceleration @xmath306 . in this case the density matrix , after tracing over region ii is where @xmath308 and @xmath309 . in this case the logarithmic negativity is @xmath310 . thus , we see that the entanglement is further degraded by having two accelerated observers , but again remains finite , in this case taking on the value @xmath311 in the infinite acceleration limit . another aspect of this problem that deserves further consideration is the nature of the communication that is and is not allowed between observers in different regions . of course , no communication is possible between the left and right rindler wedges ( regions i and ii ) . however , even when two parties are not causally disconnected from one another ( e.g. alice and rob in ) , the presence of a horizon due to the acceleration of an observer imposes a unidirectionality on any classical communication occurring after the inertial observer crosses the horizon . given the connection between the horizons seen by accelerated observers , and the event horizon of a black hole , this in turn suggests the potential for gaining further insight into questions related to the black hole information paradox . czachor , phys . a 55 , 72 ( 1997 ) ; d.r . terno and a. peres , rev . phys . * 76 * , 93 ( 2004 ) ; a. peres , p.f . scudo and d.r . terno , phys . * 88 * , 230402 ( 2002 ) ; czachor , phys . 94 , 078901 ( 2005 ) ; p.m. alsing and g.j . milburn , quant . * 2 * , 487 ( 2002 ) ; r.m . gingrich and c. adami , phys . * 89 * , 270402 ( 2002 ) ; j. pachos and e. solano , _ qic vol . 2 , pp.115 _ ( 2003 ) ; w.t . kim and e.j . son , quant - ph/0408127 ; d. ahn , h.j . lee , y.h . moon , and s.w . hwang , phys . a 67 , 012103 ( 2003 ) . d. ahn , h.j . lee , and s.w . hwang , e - print quant - ph:/0207018 ( 2002 ) ; d. ahn , h.j . lee , s.w . hwang , and m.s . kim , e - print quant - ph:/0207018 ( 2003 ) . h. terashima and m. ueda , int . info . 1 , 93 ( 2003 ) . a.j . bergou , r.m . gingrich , and c. adami , phys . a 68 , 042102 ( 2003 ) . c. soo and c.c.y . lin , int . info . 2 , 183 ( 2003 ) . to cover the minkowski regions iii and iv in one needs to switch @xmath312 in , which changes the rindler metric in these regions to the form @xmath313 @xmath314 @xmath315 . this makes @xmath48 a spacelike coordinate and @xmath316 a timelike coordinate in region iii and iv . we will not require the solutions to the dirac equations in regions iii and iv in order to discuss the entanglment between alice and observers in region i and ii . in this work , we are implicitly assuming that the spin of each mode is along the direction of acceleration , the @xmath317-axis . since this motion is rectilinear , the succesive boosts into rob s instantaneous co - moving frame all commute . hence , the spin does not suffer any thomas precession ( wigner rotation ) and therefore points along the same direction in both the inertial and non - inertial frames . see also @xcite . in an arbitrary metric @xmath318 the dirac inner product is given by @xmath319 where where @xmath320 is a spacelike hypersurface with normal @xmath321 and hyperspatial volume element @xmath322 , and @xmath323 is the adjoint of the dirac spinor wave function @xmath324 . here @xmath325 are the flat spacetime dirac matrices satisfying @xmath326 with the minkowski metric @xmath327 , while the matrices @xmath328 sastify the corresponding anti - commutation relations with the arbitrary metric , i.e. @xmath329 . the inner product @xmath330 is invariant under the choice of the spacelike hypersurface @xmath331 . for the choice @xmath332 corresponding to @xmath333 with @xmath334 in minkowski coordinates , the inner product reduces to the usual flat spacetime form @xmath335 . in rindler coordinates one can also choose to compute the inner product on the hypersurface @xmath336 , although it is often more convenient to perform the computation on the future horizon @xmath337 , ( @xmath338 ) . the minkowski annihilation operator for frequency @xmath94 is related to the rindler region i and region ii operators of frequency @xmath95 more directly through an intermediate set of modes called unruh modes @xcite . the unruh modes analytically extend the rindler region i modes to region ii , and the region ii modes to region i as discussed in the next paragraph . the unruh modes are directly related to the rindler region i and ii modes through a bogoliubov transformations of the form and @xcite . since the unruh modes exist over all minkowski space , they share the same vacuum as the minkowski annihilation operators . a minkowski annihilation operator of frequency @xmath94 ( a frequency which alice detects ) can be written as an integral ( essentially a fourier transform ) over frequencies @xmath95 ( that a rindler observer detects ) involving only unruh annihilation operators with phase factor coefficients ( see@xcite ) . the single mode approximation entails assuming that rob s detector is sensitive to a single particle mode frequency @xmath95 ( an extremely narrow band detector ) and that the previously discussed frequency dependent phase factors are highly peaked about @xmath339 . we can then approximate the minkowski annihilation operator by a single unruh annihilation operator , which is then related to a rindler particle annihilation operator in region i and a rindler anti - particle creation operator in region ii through a bogoliubov transformation ( see the discussion in the last paper in @xcite ) . similar relations exist for the operators @xmath340 and @xmath111 namely , @xmath341 , @xmath342 , @xmath343 , and @xmath344 . however , these operators correspond to the mode @xmath97 , which we have ignored in the single mode approximation . let the intial unaccelerated state between alice and rob be the tensor product of minkowski states @xmath345 where @xmath346 is any state ( possibly mixed ) for alice and @xmath86 is any pure state for rob . since entanglement is invariant under local unitaries , we can always rotate @xmath347 to @xmath124 , the minkowski vacuum for mode @xmath11 given in , without changing the initial entanglement . then the reduced state for rob is just @xmath348 , which has maximal entanglement between i and ii at infinite acceleration .

accretion provides the energy that powers radiation from the galactic center black hole , sagittarius a * @xcite . the vlt nir discovery of an infalling object onto sgr a * , named g2 , offers an especially exciting possibility for probing the nature of that accretion flow via potential interactions as the object goes through periastron @xcite . if it is a pressure - confined gas cloud , the g2 object is estimated to have a gas mass of @xmath2 , which is comparable to the total accretion flow mass within its closest approach . best estimates indicate a highly eccentric orbit with a periastron to sgr a*of @xmath3 schwarzschild radii ( @xmath4 ) , a mere @xmath5 of the bondi accretion radius , in early 2014 @xcite . the origin , structure , and fate of the object remain uncertain . while the vlt observations of the br-@xmath6 transition indicate that g2 is being disrupted , through tidal processes and possibly through gas dynamical processes , keck nir observations have shown a persistently compact @xmath7 continuum source @xcite . these observational results have led to a range of models , from diffuse clouds to dense , dusty stellar winds surrounding stars @xcite . both the nature of the accretion process and the timescale on which it takes place depend sensitively on the density profile of the accretion region and the initial density structure of the object , among other factors @xcite . proposed models anticipate a steep increase in the accretion rate , potentially by several orders of magnitude , and heightened variability on timescales of months to decades @xcite . the g2 encounter could shed light on the historic and episodic galactic center flares seen through x - ray fluorescence . the processing of g2 s material through the accretion flow around sgr a * presents a unique opportunity to study the radial density and temperature structure of the flow , filling in crucial gaps in our understanding of ultra - low - eddington - luminosity accretion . the long - term monitoring at radio and millimeter wavelengths for sgr a * demonstrates short - term variability and long - term stability , consistent with damped random walk evolution @xcite . since the detection of sgr a * at shorter wavelengths , x - ray and nir light curves have also not shown any secular evolution @xcite . intensive campaigns to monitor sgr a * over a wide range of wavelengths have been launched since the discovery of g2 was announced , none of which have detected significant changes in the flux density or activity of sgr a*@xcite , although those searches did lead to the discovery of the galactic center pulsar psr j1745 - 2900 @xcite . among the predictions for g2 are strong increases in radio through millimeter spectrum driven by two processes . increasing the mass accretion rate at separations of a few schwarzschild radii would be expected to lead to enhanced emission from the accretion flow and/or jet . the timescale for this accretion event may be as short as the free - fall time ( @xmath8 yr ) or as long as the viscous time ( 0.1 100 yr ) . additionally , a strong precursor effect has been predicted for low radio frequencies due to the potential shock forming between the incoming object and the accretion flow @xcite . in this paper , we present karl g. jansky very large array ( vla ) , atacama large millimeter array ( alma ) , and submillimeter array ( sma ) observations that span the time since the discovery of g2 through periastron . in section [ sec : obs ] , we present the observations and analysis techniques . in section [ sec : results ] , we give the results , which show a marked lack of any change in level of activity relative to historical variability . in section [ sec : discussion ] , we discuss and summarize our conclusions . vla observations were carried out as part of a service observing program on the advice of an ad hoc committee ( project code tobs0006 ; * ? ? ? * ) between late 2012 and mid-2014 ( table [ tab : vlaobs ] ) at frequencies between 1 and 41 ghz ( table [ tab : frequencytable ] ) . observations were obtained in 8 separate observing bands , each with a total bandwidth of 2 ghz with the exception of l band observations , which had a total bandwidth of 1 ghz . for each band , the correlator was configured with 16 spectral windows , each with 64 channels , for a total of 1024 channels spanning the bandwidth of 2 ghz ( 1 ghz in the case of l band ) . observations were snapshots of a few minutes duration in each of the eight frequency bands . observations of 3c 286 were obtained for absolute gain calibration . standard pipeline calibration techniques in casa were carried out by nrao staff @xcite . the pipeline produced calibrated visibility files that we then analyzed with our own software . table [ tab : vlaobs ] summarizes beam sizes for sgr a * for l and q band ( 1.5 and 40 ghz , respectively ) from images obtained from all of the data in each epoch . unlike the reduction described below , the images included short baselines as a means of demonstrating the large beams that occurred . the ratio of q to l band beam sizes does not exactly follow the expected frequency scaling ratio , which is indicative of the poorly shaped beams from these snapshot observations and frequency - dependent weighting in the imaging stage . we averaged the data in time into 30s intervals and in frequency for each spectral window . we then exported the data for further analysis with our code . we performed two kinds of analyses . for the first , we extracted flux densities for sgr a * by averaging visibility amplitudes on baselines longer than 50 @xmath9 . past experience has shown that these long baselines are sufficient to exclude the effects of extended structure on the flux density of sgr a * ( e.g. , * ? ? ? * ) . at high frequencies and in extended configurations , we determine flux densities for sgr a * that are consistent with those provided by @xcite . for the second analysis , we used a visibility subtraction method to determine deviations from the average flux density . this method is important for low frequency observations in compact configurations , where the extended flux dominates that of sgr a*. these epochs were reported to have upper limits as high as 15 jy ( e.g. , * ? ? ? * ) . for each epoch , we gridded the visibilities for each spectral window and then found individual grid cells for which there was an overlap between the individual epoch and the remaining ensemble . we differenced each epoch s visibility grid against the grid derived from the ensemble of visibilities ( minus the particular epoch ) . the differential flux density was determined as a weighted sum of the residual visibilities . we applied a weighting scheme that scales as the inverse of the @xmath10 distance . that is , longer baselines have greater weight than shorter baselines in order to bias against contamination from extended structure . different weighting schemes do not qualitatively alter our conclusions . the scatter in the differences determines the error . we then averaged over each observing band and , again , calculated the error based on the scatter in the measurements . this visibility subtraction scheme works well given the large number of data sets and the fine frequency resolution of the observations . the frequency resolution permits us to make comparisons that are not limited by the steep spectral indices of some of the extended structures present in the galactic center . the large number of data sets gives good overlap between extended and compact structures . the method is sensitive to the effect of pairs of observations with significant overlap in @xmath10 coverage , i.e. , those conducted at the same sidereal time and in the same configuration . in this case , the method effectively differences those two epochs and does not provide a difference with respect to the average . this effect can be seen in the epochs 20131129 and 20131229 , in which the flux difference for one mirrors the other . the apparent minimum in rms variability of the flux density excess at 5 ghz is the result of increasing variability toward high frequencies and decreasing accuracy of the method toward low frequencies because of the smaller number of long baselines . we summarize the mean flux densities and differential flux densities in table [ tab : vlaresults ] . we show total intensity and differential spectra for sgr a * in figures [ fig : specvla ] and [ fig : diffspecvla ] . we plot the total intensity spectrum alongside historical average spectra @xcite . ] ] we carried out alma observations between mid-2013 and mid-2014 ( table [ tab : almaobs ] ) , over eight epochs , each with band 6 ( 230 ghz ) and band 7 ( 345 ghz ) receivers . details of the frequency setup are presented in table [ tab : frequencytable ] . one spectral window in each band was configured with higher spectral resolution in order to permit detection of hydrogen recombination lines . we will discuss those results in a separate paper . observations were snapshots with durations of a few minutes . the total flux density , corresponding to stokes i , was produced by summing the response of parallel handed cross - correlations of the orthogonal linearly polarized feeds . we set absolute amplitude calibration with observations of titan and neptune and set bandpass gains with observations of the compact source j1427 - 421 . all calibrator sources and sgr a * were phase - self - calibrated on timescales of a single integration ( 10 sec ) . for sgr a * , phase self - calibration solutions were obtained only for baselines longer than @xmath11 . flux densities for calibrators were determined by bootstrapping amplitude self - calibration solutions to the absolute flux density calibrator . time - dependent amplitude gain solutions were applied to sgr a*. unfortunately , alma observations did not include system temperature measurements on sgr a*itself . the pipeline software applied system temperatures obtained for j1733 - 130 ( nrao 530 ) to sgr a * , which introduces an error because of the different atmospheric optical depth toward these two sources . to correct for this effect , we fitted system temperatures for all sources in each epoch to determine the atmospheric optical depth and then applied the correct system temperature to sgr a*. typical changes to the amplitude gain were a few percent , with a maximum value of 7% . flux densities for sgr a * were obtained through fitting a point - source of unknown flux to visibilities on baselines longer than @xmath11 . we report the results in table [ tab : almaresults ] . we do not report statistical errors on the flux densities because these are typically @xmath12 mjy , much less than the likely uncertainty from gain calibration errors for these bright sources . we show alma spectra for sgr a * and all sources in figure [ fig : specalma ] . ] alma flux densities for the calibrators are consistent with recent sma measurements of flux densities . we plot alma and sma light curves for sgr a * and j1733 - 130 in figures [ fig : almasma ] and [ fig : almasmaj1733 ] . sma data are described in the following section . where the data are contemporaneous , we see very good agreement in both alma bands . the variability of the calibrators sets an upper bound on the calibration accuracy for sgr a*. j1733 - 130 , j1700 - 261 , and j1427 - 421 show 5% , 9% , and 14% variability , respectively , in the fitted intensity at 230 ghz assuming a power - law spectrum for these sources . thus , we take 5% as the systematic calibration error in the sgr a*flux density . alma and sma light curves at 230 ghz ( top ) and 345 ghz ( bottom ) for sgr a*. each alma spectral window is plotted , covering 218 to 235 ghz ( top ) and 341 to 355 ghz ( bottom ) . the two lower frequency spectral windows for each alma band are plotted with filled symbols . these match most closely the plotted sma flux densities , which are in the range 212 to 241 ghz and 331 to 356 ghz , respectively , and have mean frequencies of 221 and 338 ghz . ] . [ fig : almasmaj1733 ] ] observations of the flux density of sgr a * were obtained using the sma , an interferometric array of eight 6-m diameter radio dishes located near the summit of mauna kea , hawaii @xcite . in most cases the sma was operating in a single receiver mode , providing 4 ghz continuum bandwidth in a single polarization , in two sidebands , with the array tuned to one of the three main sma bands ( 1.3 mm , 1.1 mm and 870 micron ) . data were obtained during short ( @xmath1320 min ) , self - contained observation sets , in which observations of sgr a * alternated with those of the relatively nearby blazar j1733 - 130 , and were bookended with observations of a flux standard ( typically neptune ) . these observations were `` piggy backed '' on full - track scheduled observations - of - the - day for the sma . this explains the apparent randomness in sky frequencies . in addition , the flux densities were estimated from a measurements with a single linear polarization . the fractional linear polarization in sgr a * is typically in the range of 5 - 10 percent and changes rapidly . the single polarization estimate of the flux density for a polarization of 10 percent has an rms error of 7 percent . this error is comparable to the level of the statistical errors . to this data set were added more extensive observations over several hours in dual polarization mode on 5 july and 15 aug 2013 , which allowed short - term monitoring . data from each observing period were calibrated with the millimeter interferometer reduction ( mir ) suite of reduction routines developed by the sma , involving removal of visibility phase scatter due to atmospheric instability and scaling visibility amplitude via comparison with the flux standard . neptune was the primary flux standard used ( 85% of observations ) , with uranus ( 13% ) and titan ( 2% ) as the other standards . for uranus and neptune , we used the broadband spectral model from @xcite . however , due to the presence of broad co absorption in the spectrum of neptune which is not included in the griffin and orton spectral fit , we opted to use data that were at least 8 ghz separated from either the co(1 - 2 ) or co(3 - 2 ) rotational transitions . for titan , we used the spectral model developed by author gurwell , which is now included in the casa data reduction package as well ( see * ? ? ? in all cases the expected error on the flux density scale based upon using these models is 5% . on the other hand , since we are most interested in tracking changes in flux with time , the intrinsic bias in the flux density model is less important . the presence of gas along the line of sight toward sgr a * leads to deep absorption of the sgr a * continuum at the co transitions , as well as their isotopologues . for this reason , we masked spectral regions around these transitions in the sgr a * data . likewise , the structure of the visibility data from sgr a * indicated we were sensitive to broad scale continuum emission from dust in the vicinity of sgr a*. in order to limit our data to emission from sgr a * itself , we used only visibility data from baselines that exceeded 35 k@xmath14 , or roughly scales of 6or finer , which was sufficient to isolate sgr a * from the broader diffuse emission . flux densities from the sma for sgr a * and j1733 - 130 are tabulated in table [ tab : smaresults ] . the mean spectrum of sgr a * from vla , alma , and sma observations is presented in table [ tab : meanspec ] and figure [ fig : meanspec ] , along with variability , minimum flux , and maximum flux . variability is substantially stronger at millimeter wavelengths than at radio wavelengths . the ratio of the rms variability to the total flux density , known as the modulation index , is 8% at 40 ghz and below , while it is @xmath15 at millimeter wavelengths . this variability can not be attributed to differences in calibration accuracy . mean spectrum of sgr a * from vla and alma observations . the solid black line shows the mean and the shaded regions show the @xmath16 boundaries for the flux density based on the rms of the variations . we also plot a curve showing the 230 to 690 ghz spectrum as measured by @xcite . we plot several model curves representing synchrotron emission from relativistic thermal maxwell - juttner distributions . we also plot nir mean flux densities and spectral indices from @xcite , along with a curve showing the extension of the nir spectral index to shorter wavelengths ( thin solid line ) . see text for details . ] at vla frequencies between 1 and 40 ghz , the mean spectral index @xmath17 is @xmath18 ( @xmath19 ) and shows evidence for steepening at short wavelengths . a mean spectral index of 0.5 between 40 and 218 ghz is required to connect the vla and alma spectra , comparable to what has been seen previously @xcite . we also fit the alma data with a power - law spectrum ( figure [ fig : alphaalma ] ) . the mean spectral index for frequencies between 217 and 355 ghz is @xmath20 , which is flatter and more variable than those of the calibrators . the calibrators have mean spectral indices between 217 and 355 ghz of @xmath21 , @xmath22 , and @xmath23 for j1733 - 130 , j1700 - 261 , and j1427 - 421 , respectively . @xcite found that the 230 to 690 ghz spectral index ranged in four epochs from -0.4 to + 0.2 , with a mean of -0.13 . our results span the same range of @xmath17 and are statistically indistinguishably from those of @xcite ; that is , the spectrum is consistent with a flat or slightly declining power - law spectrum above 230 ghz . assuming stationary statistics for sgr a * between the early observations and these new observations , we conclude that there is no evidence for spectral curvature or greater steepening in the spectrum between 230 and 690 ghz . ] the millimeter and submillimeter spectrum of sgr a * indicates that the optical depth is @xmath12 at frequencies as high as 690 ghz . the result suggests that the source must be composed of stratified regions near the optically thick - to - thin transition in order to produce a near - flat synchrotron spectrum over this broad frequency range . a single optically thin synchrotron component can not reproduce the spectrum . this spectrum can be explained in the context of the classical @xcite jet model , but also by inflowing accretion flow models with radially evolving density and magnetic fields together with some non - thermal particles @xcite . despite this degeneracy in the emission geometry , it is clear that above 350 ghz , sgr a * has to become optically thin . this conclusion is supported by the transition to higher linear polarization at higher frequencies @xcite , and the increased variability and power - law spectrum seen in the infrared band . our simultaneous radio through submillimeter data can provide some interesting constraints on the underlying particle distribution , when put into context with the infrared measurements . in fig . [ fig : meanspec ] we include the average ir flux based on vlt and keck measurements @xcite . these results are consistent with a mean flux density of @xmath24 mjy in the @xmath25 band and a stable spectral index @xmath26 between @xmath25 and @xmath27 band . previous estimates of the nir spectral index appear to show steeper spectral indices but may have suffered from non - simultaneous observations of the variable flux density . to illustrate the new limits on the radiating particle distributions , we plot for reference a series of synchrotron spectra from mildly relativistic thermal electrons in equipartition with the magnetic field , ranging from @xmath28 k , assuming an emission region of size @xmath29 . the observed brightness temperature of sgr a * at 230 ghz is @xmath30 k @xcite . these are fiducial values spanning the range from the literature ( see , e.g. , * ? ? ? * and references therein ) , to illustrate how stratified , self - absorbed regions can add to a near - flat spectrum , but also to place some limits on the fraction and distribution of non - thermal particles present . the peak of a given temperature component can shift by a factor of a few in frequency for different emission size regions or energy partition between radiating particles and the magnetic energy density . however , it is clear that a peak temperature ( in the plasma closest to the black hole ) below a few @xmath31 k can not account for the extension of the self - absorbed spectrum through the alma range shown , and that a peak temperature much higher than a few @xmath32 k would violate the ir limits . it is also clear from these spectra that a purely thermal distribution can not account for the ir flux . the intersection of the ir power - law with the intermediate ranges indicates that a non - thermal `` tail '' of particles is present , with normalization on the order of @xmath33% of the thermal peak . this fraction is consistent with the results of fits to sgr a * in quiescence , using a self - consistent calculation of the particle distribution given a mechanism for injecting a power - law of nonthermal particles @xcite . recent work exploring second - order acceleration processes in turbulent , magnetized plasmas by @xcite shows the self - consistent production of a such a nonthermal tail ; however , its slope and relation to the thermal peak may be difficult to match to these newest constraints . therefore the combination of simultaneous alma ( eventually using even higher frequency bands ) and ir data offers the best constraints yet on the shape of the radiating particle distributions and nonthermal fraction . the results presented here can already be used to guide implementation of the inclusion of particle acceleration in semi - analytical models and in the so - called `` painting '' of grmhd simulations to produce images ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? simultaneous observations between alma , ir , and x - ray , particularly during flaring , would provide strong constraints on the upper extreme of the nonthermal population and accelerating mechanisms . bow shock emission from the interaction of g2 with the accretion flow was anticipated to appear and peak months before the center of mass reached periastron @xcite . the bow shock would excite nonthermal electrons that then produce synchrotron radiation in the magnetic field of the accretion flow . the synchrotron radiation is expected to peak at frequencies near 1 ghz and decline with an optically thin power - law index @xmath34 . the predicted peak flux density of the radio light curve scales with a number of factors , including the relative velocity of the cloud , the accretion flow density profile , the efficiency of nonthermal electron energy production , and the timescale for electrons to cool . perhaps the most important term is the cross - sectional area of the shock as it impacts the accretion flow . @xcite use the geometrical size of g2 as observed at large radii , @xmath35 . @xcite , on the other hand , constructs a magnetically arrested , tidally distorted cloud , which has an area @xmath36 . @xcite model g2 as a wind driven from a hidden star with a radius that shrinks as the external pressure in the accretion flow increases , leading to a minimum area at periastron of @xmath37 . the flux density in all models scales linearly with the area . thus , predictions of the 1 ghz excess flux density excess range from @xmath38 jy for some models to @xmath39 jy for others . synchrotron lifetimes for the radiating electrons in the accretion inflow are long , so the timescale for the radio emission is determined by the dynamics of the bow shock . @xcite predicts a characteristic timescale of four months , which is comparable to predictions in other models . the l band data alone in the first and last epochs of our experiment show that no significant secular change has occurred over the course of this experiment . between october 2012 and may 2014 , the 1.5 ghz flux density changed by at most 12 mjy ( 2% ) we can examine in more detail whether there is evidence for shorter time scale variations through an analysis of all the measurements in the three lowest vla frequency bands . we compute estimates of the flux density excess above the average at 1.0 ghz using low frequency measurements ( figure [ fig : bowshock ] ) . we extrapolate the flux density excess at 1.5 , 3.1 , and 5.4 ghz to 1 ghz , assuming a power - law spectrum with index @xmath40 , appropriate for optically thin synchrotron emission . at frequencies above 5 ghz , intrinsic variability rises rapidly and is likely to exceed estimates of any bow shock - related variability . 1.5 ghz results alone are sometimes only weakly sensitive to variations because of limited overlap in @xmath10 coverage for the most compact configurations . ] the average 1 ghz flux density excess computed in this way is only significantly non - zero for a single epoch , 2013 august 08 , with a value of @xmath41 jy . the excess is positive and comparable for all three bands and very high significance at 5 ghz . the subsequent epoch , 2013 september 18 , has a comparable value , @xmath42 jy , but much lower significance . the total flux density spectrum for 2013 august 08 is one of the most elevated across the band , with a peak flux density of 1.8 jy at 40 ghz . these observations were obtained in a compact configuration and no total flux density measurements were obtained at 1.4 and 3.1 ghz . for 11 of 15 epochs , the 95% confidence upper limit on 1 ghz flux density excess is 0.4 jy . for the remaining four epochs , the upper limit is 0.8 jy . a four - month moving average of the average 1 ghz flux has a peak at 0.3 jy , centered on the 2013 august 08 epoch , and typical values of 0.1 jy . these results clearly exclude an object with size @xmath43 . at 95% confidence , we can set an upper limit to the size of g2 of @xmath44 . this can be attributed to the magnetically arrested cloud derived by @xcite or due to a pressure - confined stellar wind @xcite . alternatively , the detailed model of the accretion flow structure could be in error . the new data appear to show higher radio flux densities at frequencies @xmath45 ghz over historical averages ( figure [ fig : specvla ] and table [ tab : meanspec ] ) . the 40.9 ghz flux density , for instance , has a mean of 1.5 jy and a range of 1.3 to 1.9 jy , whereas @xcite found a 43 ghz flux density mean of 1.0 jy and a range of 0.6 to 1.9 jy . similar results with smaller amplitudes differences are apparent at lower frequencies . the flux densities from @xcite and @xcite are higher than those of @xcite but still show a lower average flux density at most frequencies relative to the new vla data . @xcite presented many epochs of archival data obtained between 1979 and 1999 . @xcite reported over 100 epochs with regular monitoring from 2000 to 2003 , while @xcite is from a single - epoch simultaneous spectrum . over the whole radio spectrum , we infer an increase of @xmath46 relative to @xcite and @xmath47 relative to @xcite in the total flux density . these differences are comparable to the variability between @xcite and @xcite and consistent with variability seen in low - luminosity agn ( llagn ) in general . thus , these differences can not be attributed to any effects associated with the g2 periastron . the mean flux density that we measure at 21 ghz , @xmath48 jy , is comparable to the 22 ghz flux density measured over a similar time period with the japanese vlbi network , @xmath49 jy @xcite . the lower variability in our result may reflect some of the difficulties of providing accurate amplitude calibration for vlbi arrays . a comparison against historical millimeter and submillimeter flux densities also does not reveal any significant change . @xcite presents historical data for sgr a * at 230 , 345 , and 690 ghz with dates from 2001 through 2012 . the flux densities from this paper span timescales from minutes to years and were obtained with multiple telescopes with varying calibration accuracy . the mean alma band 6 ( 230 ghz ) flux density is @xmath50 jy while the mean historical flux density at 230 ghz is @xmath51 jy . similarly , for band 7 ( 345 ghz ) and historical 345 ghz data , we find @xmath52 jy and @xmath53 jy , respectively . results from the sma data are comparable . thus , increases in the millimeter and submillimeter mean flux density are no more than 20% relative to historical averages , again consistent with typical llagn variability levels . a kolmogorov - smirnov test shows that the alma and sma measurements are consistent ( @xmath54 ) with originating from the same distribution . the alma and sma measurements differ from the @xcite measurements with varying significance . sma 230 and 345 ghz data differ from the @xcite distributions with @xmath55 , while the band 6 ( 230 ghz ) and dexter 230 ghz and band 7 and 345 ghz distributions differ with significances of @xmath56 and @xmath57 . these variations may be partly the result of different calibration accuracies and as the result of evolution of the light curves over @xmath58 yr . similar to the case with the radio spectrum , there is no clear evidence that the millimeter and submillimeter changes are the result of the g2 periastron . we have analyzed the light curve data for our measurements near 220 ghz from both alma and the sma . there are 104 data points spanning an interval of 524 days with the shortest interval being 0.02 days . the data were binned in a sparsely filled array of 32,768 elements and fourier transformed . the resulting spectrum was averaged into logarithmically spaced frequency intervals . the spectrum has a simple power law form characterized as @xmath59 over a frequency interval of 0.04 to 2 cycle / day . @xmath17 was found to be @xmath60 . the error bar was determined by bootstrap resampling . this result is intuitively obvious from the light curve , i.e. the variations over a period of a few day are uncorrelated and have the same range as the variations over the entire data span . the spectral variations from historical data at millimeter / submillimeter wavelengths are known to approximate white noise at frequencies less than 3 cycles / day , with a steepening power law above that frequency @xcite . it is plausible that an enhancement of emission caused by g2 might have produced long term variations indicative of red noise . however , there is no evidence for this effect . the spectrum of sgr a * from radio to submillimeter wavelengths has remained remarkably stable over the past 30 years . our new vla , alma , and sma results demonstrate that this stability has continued throughout the apparent periastron passage of g2 . in the case of low - frequency emission , the absence of significant change permits us to constrain the size of the ensuing bow shock to be a factor of 30 smaller than the apparent g2 size observed with nir wavelengths at larger distances from the black hole . these observations do not resolve the nature of g2 . the smaller bow shock size is consistent with both cloud and stellar - wind models . @xcite propose a model that may reconcile the apparently discrepant nir continuum and spectroscopic observations and may also be consistent with our results . @xmath7 continuum observations reveal a compact source while br-@xmath6 observations reveal an extended source that appears to be tidally disrupted . @xcite argue that the continuum emission is the photosphere of a binary stellar merger , while the ionized emission represents a smaller tidally disrupted tail . the @xmath7 source has a cross - sectional area of @xmath61 @xmath62 , much less than our upper limit of @xmath0 @xmath62 . however , the size of the extended br-@xmath6 emission , which is consistent between vlt and keck observations , primarily determines the creation of a bow shock . some form of confinement for the diffuse gas appears to be important for attenuating the shock amplitude . the lack of any enhanced short wavelength radio and millimeter emission from the black hole is also consistent with a compact size for g2 . simulations have shown that more extended objects are more readily disrupted and would therefore be more likely to have some gas initiate enhanced accretion onto the black hole . at shorter wavelengths , the mean flux density has not increased by more than @xmath15 relative to historical averages . this is within the range of variations seen in the past and in other llagn . thus , we can not attribute any increase in the flux density to a g2-induced enhancement of the accretion rate . the amplitude and timescale for a change in the accretion rate , should g2 be fully disrupted , are uncertain . in order for the disrupted material to enhance the flux density , the gas must reach a radius of a few schwarzschild radii to contribute to the accretion flow and/or enhance jet power . the free - fall time from @xmath63 is @xmath64 yr . if a fraction @xmath65 of the material were on a plunging orbit due to tidal disruption or a spread in orbital characteristics , we would have expected an enhancement of the overall flux density during these observations . the flux density should scale as @xmath66 where @xmath67 is a model - dependent constant for jet and accretion disk models . for a g2 mass of @xmath68 appropriate for the gas - cloud model and a steady - state accretion rate onto sgr a * of @xmath69 , and @xmath70 , we find @xmath71 for @xmath72 . that is , very little of g2 s mass has yet accreted directly onto the black hole . longer term monitoring ( 10 yr ) will be sensitive to @xmath73 reaching the black hole through an increase of the average flux density . the bulk of the material is more likely to accrete on the viscous time scale , @xmath74 . @xmath75 , where @xmath76 1 is the viscous parameter and @xmath27 is the height at radius @xmath77 . for reasonable parameters , @xmath78 100 yr . continued monitoring of the radio through millimeter flux densities can constrain further the properties and origin of g2 as well as the properties of the accretion flow at large radii . the long - term stability of the radio / millimeter spectrum indicates that episodic events such as a fully disrupted cloud falling onto the black hole are rare . such cloud disruptions may take place but they must have timescales that are long compared to the free - fall time . we confirm a flat spectrum for sgr a * emission in the millimeter / submillimeter regime . the emission is likely to originate from a stratified region with optical depth in transition between optically thick and thin . optically thick emission at frequencies of @xmath79 ghz may influence images obtained with the event horizon telescope @xcite . accretion disk models with tilted disks and/or high ratios of ion to electron temperature produce larger regions with optical depth greater than 1 @xcite . higher optical depths may obscure or complicate the interpretation of gravitational lensing effects such as the black hole shadow . on the other hand , higher optical depth in regions that are offset from the black hole as expected from doppler boosting may enhance the contrast associated with geometric features . new alma measurements of the spectrum at frequencies from 100 to 950 ghz can resolve the uncertainty in the spectrum between the submillimeter and the nir regime and assist in optimization of eht observations . the national radio astronomy observatory is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . this paper makes use of the following alma data : ads / jao.alma # 2012.1.00635.s . alma is a partnership of eso ( representing its member states ) , nsf ( usa ) and nins ( japan ) , together with nrc ( canada ) and nsc and asiaa ( taiwan ) , in cooperation with the republic of chile . the joint alma observatory is operated by eso , aui / nrao and naoj . the submillimeter array is a joint project between the smithsonian astrophysical observatory and the academia sinica institute of astronomy and astrophysics and is funded by the smithsonian institution and the academia sinica . we thank michael johnson for helpful discussions . , s. s. , weintroub , j. , rogers , a. e. e. , plambeck , r. , freund , r. , tilanus , r. p. j. , friberg , p. , ziurys , l. m. , moran , j. m. , corey , b. , young , k. h. , smythe , d. l. , titus , m. , marrone , d. p. , cappallo , r. j. , bock , d. , bower , g. c. , chamberlin , r. , davis , g. r. , krichbaum , t. p. , lamb , j. , maness , h. , niell , a. e. , roy , a. , strittmatter , p. , werthimer , d. , whitney , a. r. , & woody , d. 2008 , , 455 , 78 , r. p. , falcke , h. , karuppusamy , r. , lee , k. j. , champion , d. j. , keane , e. f. , desvignes , g. , schnitzeler , d. h. f. m. , spitler , l. g. , kramer , m. , klein , b. , bassa , c. , bower , g. c. , brunthaler , a. , cognard , i. , deller , a. t. , demorest , p. b. , freire , p. c. c. , kraus , a. , lyne , a. g. , noutsos , a. , stappers , b. , & wex , n. 2013 , , 501 , 391 , a. , baganoff , f. k. , schdel , r. , morris , m. , genzel , r. , bower , g. c. , marrone , d. , moran , j. m. , viehmann , t. , bautz , m. w. , brandt , w. n. , garmire , g. p. , ott , t. , trippe , s. , ricker , g. r. , straubmeier , c. , roberts , d. a. , yusef - zadeh , f. , zhao , j. h. , & rao , r. 2006 , , 450 , 535 , v. , alef , w. , anderson , j. , asada , k. , baudry , a. , broderick , a. , carilli , c. , colomer , f. , conway , j. , dexter , j. , doeleman , s. , eatough , r. , falcke , h. , frey , s. , gabnyi , k. , glvan - madrid , r. , gammie , c. , giroletti , m. , goddi , c. , gmez , j. l. , hada , k. , hecht , m. , honma , m. , humphreys , e. , impellizzeri , v. , johannsen , t. , jorstad , s. , kino , m. , krding , e. , kramer , m. , krichbaum , t. , kudryavtseva , n. , laing , r. , lazio , j. , loeb , a. , lu , r .- s . , maccarone , t. , marscher , a. , marti - vidal , i. , martins , c. , matthews , l. , menten , k. , miller , j. , miller - jones , j. , mirabel , f. , muller , s. , nagai , h. , nagar , n. , nakamura , m. , paragi , z. , pradel , n. , psaltis , d. , ransom , s. , rodr@xmath80iguez , l. , rottmann , h. , rushton , a. , shen , z .- q . , smith , d. , stappers , b. , takahashi , r. , tarchi , a. , tilanus , r. , verbiest , j. , vlemmings , w. , walker , r. c. , wardle , j. , wiik , k. , zackrisson , e. , & zensus , j. a. 2013 , arxiv e - prints , k. , gotthelf , e. v. , zhang , s. , an , h. , baganoff , f. k. , barrire , n. m. , beloborodov , a. m. , boggs , s. e. , christensen , f. e. , craig , w. w. , dufour , f. , grefenstette , b. w. , hailey , c. j. , harrison , f. a. , hong , j. , kaspi , v. m. , kennea , j. a. , madsen , k. k. , markwardt , c. b. , nynka , m. , stern , d. , tomsick , j. a. , & zhang , w. w. 2013 , , 770 , l23 , j. , nowak , m. a. , gammie , c. , dexter , j. , markoff , s. , haggard , d. , nayakshin , s. , wang , q. d. , grosso , n. , porquet , d. , tomsick , j. a. , degenaar , n. , fragile , p. c. , houck , j. c. , wijnands , r. , miller , j. m. , & baganoff , f. k. 2013 , , 774 , 42 , n. , esposito , p. , pons , j. a. , turolla , r. , torres , d. f. , israel , g. l. , possenti , a. , burgay , m. , vigan , d. , papitto , a. , perna , r. , stella , l. , ponti , g. , baganoff , f. k. , haggard , d. , camero - arranz , a. , zane , s. , minter , a. , mereghetti , s. , tiengo , a. , schdel , r. , feroci , m. , mignani , r. , & gtz , d. 2013 , , 775 , l34 , g. , eckart , a. , bremer , m. , zamaninasab , m. , shahzamanian , b. , valencia - s . , m. , schdel , r. , karas , v. , lenzen , r. , marchili , n. , sabha , n. , garcia - marin , m. , buchholz , r. m. , kunneriath , d. , & straubmeier , c. 2012 , , 203 , 18 , g. , ghez , a. m. , morris , m. r. , sitarski , b. n. , boehle , a. , naoz , s. , campbell , r. , becklin , e. e. , canalizo , g. , chappell , s. , do , t. , lu , j. r. , matthews , k. , meyer , l. , stockton , a. , wizinowich , p. , & yelda , s. 2014 , , 796 , l8 lrrrr 20121014 & 23:16 - 00:25 & @xmath81 , 11.7 & @xmath82 , 45.1 & a + 20121222 & 17:30 - 18:39 & @xmath83 , -15.8 & @xmath84 , -12.5 & a + 20130218 & 13:42 - 14:51 & @xmath85 , -13.6 & @xmath86 , -72.9 & d + 20130323 & 12:02 - 13:11 & @xmath87 , 29.3 & @xmath88 , 36.9 & d + 20130426 & 10:56 - 12:05 & @xmath89 , 28.0 & @xmath90 , 50.2 & d + 20130609 & 06:25 - 07:34 & @xmath91 , 42.9 & @xmath92 , -26.2 & c + 20130808 & 03:30 - 04:39 & @xmath93 , 24.2 & @xmath94 , 29.1 & c + 20130918 & 01:33 - 02:42 & @xmath95 , 63.1 & @xmath96 , 32.4 & cnb + 20131026 & 23:14 - 00:24 & @xmath97 , 28.9 & @xmath98 , 23.4 & b + 20131129 & 19:46 - 20:55 & @xmath99 , 23.4 & @xmath100 , 26.8 & b + 20131229 & 17:48 - 18:57 & @xmath101 , 33.7 & @xmath102 , -1.3 & b + 20140215 & 14:11 - 15:21 & @xmath103 , 37.3 & @xmath104 , -101.8 & bna + 20140322 & 13:26 - 14:35 & @xmath105 , 22.0 & @xmath106 , 24.0 & a + 20140426 & 10:19 - 11:28 & @xmath107 , 27.0 & @xmath108 , -27.4 & a + 20140531 & 08:11 - 09:20 & @xmath109 , 43.6 & @xmath110 , -28.1 & a + llrrr vla & l & 994.0 & 2006.0 & 1024 + vla & s & 1988.0 & 3948.0 & 1024 + vla & c & 4488.0 & 6448.0 & 1024 + vla & x & 7988.0 & 9948.0 & 1024 + vla & u & 12988.0 & 14948.0 & 1024 + vla & k & 20188.0 & 22148.0 & 1024 + vla & a & 32008.0 & 31968.0 & 1024 + vla & q & 39988.0 & 41948.0 & 1024 + alma & b6 - 1 & 217.0 & 219.0 & 128 + alma & b6 - 2 & 219.0 & 221.0 & 128 + alma & b6 - 3 & 231.0 & 232.9 & 3840 + alma & b6 - 4 & 232.8 & 234.8 & 128 + alma & b7 - 1 & 340.6 & 342.6 & 128 + alma & b7 - 2 & 342.6 & 344.6 & 128 + alma & b7 - 3 & 350.7 & 352.7 & 128 + alma & b7 - 4 & 352.7 & 354.6 & 3840 + lrrr 20121014 & 1.5 & @xmath111 & @xmath112 + 20121014 & 3.0 & @xmath113 & @xmath114 + 20121014 & 5.4 & @xmath115 & @xmath116 + 20121014 & 8.9 & @xmath117 & @xmath118 + 20121014 & 13.9 & @xmath119 & @xmath120 + 20121014 & 21.1 & @xmath121 & @xmath122 + 20121014 & 32.0 & @xmath123 & @xmath124 + 20121014 & 40.9 & @xmath125 & @xmath126 + 20121222 & 1.5 & @xmath127 & @xmath128 + 20121222 & 3.0 & @xmath129 & @xmath130 + 20121222 & 5.4 & @xmath131 & @xmath132 + 20121222 & 8.9 & @xmath133 & @xmath134 + 20121222 & 13.9 & @xmath135 & @xmath136 + 20121222 & 21.1 & @xmath137 & @xmath138 + 20121222 & 32.0 & @xmath139 & @xmath140 + 20121222 & 40.9 & @xmath141 & @xmath142 + 20130218 & 1.5 & & @xmath143 + 20130218 & 3.0 & & @xmath144 + 20130218 & 5.4 & & @xmath145 + 20130218 & 8.9 & & @xmath146 + 20130218 & 13.9 & @xmath147 & @xmath148 + 20130218 & 21.1 & @xmath149 & @xmath150 + 20130218 & 32.0 & @xmath151 & @xmath152 + 20130218 & 40.9 & @xmath153 & @xmath154 + 20130323 & 1.5 & & @xmath155 + 20130323 & 3.0 & & @xmath156 + 20130323 & 5.4 & & @xmath157 + 20130323 & 8.9 & & @xmath158 + 20130323 & 13.9 & @xmath159 & @xmath160 + 20130323 & 21.1 & @xmath161 & @xmath162 + 20130323 & 32.0 & @xmath163 & @xmath164 + 20130323 & 40.9 & @xmath165 & @xmath166 + 20130426 & 1.5 & & @xmath167 + 20130426 & 3.0 & & @xmath168 + 20130426 & 5.4 & & @xmath169 + 20130426 & 8.9 & & @xmath170 + 20130426 & 13.9 & & @xmath171 + 20130426 & 21.1 & @xmath172 & @xmath173 + 20130426 & 32.0 & @xmath174 & @xmath175 + 20130426 & 40.9 & @xmath176 & @xmath177 + 20130609 & 1.5 & & @xmath178 + 20130609 & 3.1 & & @xmath179 + 20130609 & 5.4 & @xmath180 & @xmath181 + 20130609 & 8.9 & @xmath182 & @xmath183 + 20130609 & 13.9 & @xmath184 & @xmath185 + 20130609 & 21.1 & @xmath186 & @xmath187 + 20130609 & 32.0 & @xmath188 & @xmath189 + 20130609 & 40.9 & @xmath190 & @xmath191 + 20130808 & 1.5 & & @xmath192 + 20130808 & 3.0 & & @xmath193 + 20130808 & 5.4 & @xmath194 & @xmath195 + 20130808 & 9.4 & @xmath196 & @xmath197 + 20130808 & 13.9 & @xmath198 & @xmath199 + 20130808 & 21.1 & @xmath200 & @xmath201 + 20130808 & 32.0 & @xmath202 & @xmath203 + 20130808 & 40.9 & @xmath204 & @xmath205 + 20130918 & 1.5 & & @xmath206 + 20130918 & 3.0 & @xmath207 & @xmath208 + 20130918 & 5.4 & @xmath209 & @xmath210 + 20130918 & 8.9 & @xmath211 & @xmath212 + 20130918 & 13.9 & @xmath213 & @xmath214 + 20130918 & 21.1 & @xmath215 & @xmath216 + 20130918 & 32.0 & @xmath217 & @xmath218 + 20130918 & 40.9 & @xmath219 & @xmath220 + 20131026 & 1.5 & @xmath221 & @xmath222 + 20131026 & 3.0 & @xmath223 & @xmath224 + 20131026 & 5.4 & @xmath225 & @xmath226 + 20131026 & 8.9 & @xmath227 & @xmath228 + 20131026 & 13.9 & @xmath229 & @xmath230 + 20131026 & 21.1 & @xmath231 & @xmath232 + 20131026 & 32.0 & @xmath233 & @xmath234 + 20131026 & 40.9 & @xmath235 & @xmath236 + 20131129 & 1.5 & @xmath237 & @xmath238 + 20131129 & 3.0 & @xmath239 & @xmath240 + 20131129 & 5.4 & @xmath241 & @xmath242 + 20131129 & 8.9 & @xmath243 & @xmath244 + 20131129 & 13.9 & @xmath245 & @xmath246 + 20131129 & 21.1 & @xmath247 & @xmath248 + 20131129 & 32.0 & @xmath249 & @xmath250 + 20131129 & 40.9 & @xmath251 & @xmath252 + 20131229 & 1.5 & & @xmath253 + 20131229 & 3.0 & @xmath254 & @xmath255 + 20131229 & 5.4 & @xmath256 & @xmath257 + 20131229 & 8.9 & @xmath258 & @xmath259 + 20131229 & 13.9 & @xmath260 & @xmath261 + 20131229 & 21.1 & @xmath262 & @xmath263 + 20131229 & 32.0 & @xmath264 & @xmath265 + 20131229 & 40.9 & @xmath266 & @xmath267 + 20140215 & 1.5 & @xmath268 & @xmath269 + 20140215 & 3.0 & @xmath270 & @xmath271 + 20140215 & 5.4 & @xmath272 & @xmath273 + 20140215 & 8.9 & @xmath274 & @xmath275 + 20140215 & 13.9 & @xmath276 & @xmath277 + 20140215 & 21.1 & @xmath278 & @xmath279 + 20140215 & 32.0 & @xmath280 & @xmath281 + 20140215 & 40.9 & @xmath282 & @xmath283 + 20140322 & 1.5 & @xmath284 & @xmath285 + 20140322 & 3.0 & @xmath286 & @xmath287 + 20140322 & 5.4 & @xmath288 & @xmath289 + 20140322 & 8.9 & @xmath290 & @xmath291 + 20140322 & 13.9 & @xmath292 & @xmath293 + 20140322 & 21.1 & @xmath294 & @xmath295 + 20140322 & 32.0 & @xmath296 & @xmath297 + 20140322 & 40.9 & @xmath298 & @xmath299 + 20140426 & 1.5 & @xmath300 & @xmath301 + 20140426 & 3.0 & @xmath302 & @xmath303 + 20140426 & 5.4 & @xmath304 & @xmath305 + 20140426 & 8.9 & @xmath306 & @xmath307 + 20140426 & 13.9 & @xmath308 & @xmath309 + 20140426 & 21.1 & @xmath310 & @xmath311 + 20140426 & 32.0 & @xmath312 & @xmath313 + 20140426 & 40.9 & @xmath314 & @xmath315 + 20140531 & 1.5 & @xmath316 & @xmath317 + 20140531 & 3.0 & @xmath318 & @xmath319 + 20140531 & 5.4 & @xmath320 & @xmath321 + 20140531 & 8.9 & @xmath322 & @xmath323 + 20140531 & 13.9 & @xmath324 & @xmath325 + 20140531 & 21.1 & @xmath326 & @xmath327 + 20140531 & 32.0 & @xmath328 & @xmath329 + 20140531 & 40.9 & @xmath330 & @xmath331 + llrrl 20130708 & 02:14 - 02:44 & @xmath332 , 65.1 & @xmath333 , 73.8 & titan + 20131102 & 22:39 - 22:57 & @xmath334 , -84.2 & @xmath335 , -87.2 & neptune + 20131218 & 14:07 - 14:27 & @xmath336 , 88.3 & @xmath337 , 86.0 & titan + 20140128 & 10:26 - 10:44 & @xmath338 , 78.8 & @xmath339 , 77.6 & titan + 20140222 & 09:20 - 09:45 & @xmath340 , 69.3 & @xmath341 , 68.2 & titan + 20140321 & 09:17 - 09:35 & @xmath342 , 79.3 & @xmath343 , 77.1 & titan + 20140425 & 06:59 - 07:29 & @xmath344 , 78.7 & @xmath345 , 88.8 & titan + 20140527 & 05:47 - 06:06 & @xmath346 , -75.2 & @xmath347 , -75.8 & titan + lrrrrrrrrr 20130708 & sgrastar & 3.792 & 3.795 & 3.872 & 3.892 & 4.259 & 4.278 & 4.270 & 4.265 + 20131102 & sgrastar & 4.635 & 4.649 & 4.752 & 4.758 & 4.537 & 4.564 & 4.569 & 4.537 + 20131218 & sgrastar & 2.976 & 2.970 & 2.997 & 3.010 & 3.032 & 3.046 & 3.012 & 3.003 + 20140128 & sgrastar & 3.868 & 3.868 & 3.944 & 3.957 & 4.293 & 4.313 & 4.309 & 4.294 + 20140222 & sgrastar & 2.811 & 2.803 & 2.800 & 2.801 & 2.332 & 2.341 & 2.311 & 2.301 + 20140321 & sgrastar & 4.078 & 4.047 & 3.983 & 4.002 & 3.476 & 3.447 & 3.428 & 3.406 + 20140425 & sgrastar & 4.121 & 4.111 & 4.004 & 4.163 & 4.296 & 4.287 & 4.295 & 4.064 + 20140527 & sgrastar & 3.054 & 3.048 & 3.054 & 3.051 & 2.587 & 2.596 & 2.568 & 2.556 + 20130708 & j1733 - 130 & 1.359 & 1.337 & 1.256 & 1.264 & 0.952 & 0.951 & 0.917 & 0.857 + 20131102 & j1733 - 130 & 1.586 & 1.570 & 1.429 & 1.413 & 0.998 & 0.994 & 0.995 & 0.998 + 20131218 & j1733 - 130 & 1.771 & 1.754 & 1.668 & 1.657 & 1.311 & 1.273 & 1.261 & 1.263 + 20140128 & j1733 - 130 & 1.576 & 1.561 & 1.477 & 1.482 & 1.203 & 1.190 & 1.181 & 1.173 + 20140222 & j1733 - 130 & 1.649 & 1.638 & 1.557 & 1.560 & 1.214 & 1.201 & 1.192 & 1.187 + 20140321 & j1733 - 130 & 1.517 & 1.502 & 1.426 & 1.426 & 1.125 & 1.109 & 1.097 & 1.092 + 20140425 & j1733 - 130 & 1.545 & 1.532 & 1.406 & 1.449 & 1.140 & 1.126 & 1.114 & 1.053 + 20140527 & j1733 - 130 & 1.619 & 1.600 & 1.517 & 1.514 & 1.179 & 1.169 & 1.142 & 1.150 + 20131218 & j1427 - 421 & 2.899 & 2.876 & 2.739 & 2.750 & 2.131 & 2.079 & 2.065 & 2.058 + 20140128 & j1427 - 421 & 3.073 & 3.041 & 2.908 & 2.907 & 2.422 & 2.387 & 2.375 & 2.379 + 20140222 & j1427 - 421 & 3.370 & 3.353 & 3.201 & 3.206 & 2.665 & 2.658 & 2.639 & 2.645 + 20140321 & j1427 - 421 & 3.773 & 3.735 & 3.570 & 3.582 & 2.931 & 2.900 & 2.859 & 2.856 + 20140425 & j1427 - 421 & 4.015 & 3.984 & 3.669 & 3.799 & -0.019 & -0.019 & -0.016 & -0.015 + 20140527 & j1427 - 421 & 4.227 & 4.187 & 4.002 & 4.002 & 3.253 & 3.217 & 3.191 & 3.176 + 20130708 & j1700 - 261 & 0.967 & 0.962 & 0.914 & 0.916 & 0.701 & 0.685 & 0.666 & 0.651 + 20131102 & j1700 - 261 & 0.791 & 0.784 & 0.721 & 0.720 & 0.517 & 0.518 & 0.525 & 0.522 + 20131218 & j1700 - 261 & 0.774 & 0.775 & 0.732 & 0.734 & 0.592 & 0.583 & 0.574 & 0.585 + 20140128 & j1700 - 261 & 0.887 & 0.882 & 0.842 & 0.846 & 0.716 & 0.707 & 0.706 & 0.700 + 20140222 & j1700 - 261 & 0.969 & 0.961 & 0.921 & 0.924 & 0.749 & 0.744 & 0.738 & 0.738 + 20140321 & j1700 - 261 & 0.846 & 0.838 & 0.800 & 0.804 & 0.664 & 0.657 & 0.650 & 0.646 + 20140425 & j1700 - 261 & & & & & 0.773 & 0.763 & 0.757 & 0.723 + 20140527 & j1700 - 261 & 1.072 & 1.062 & 1.004 & 1.005 & 0.824 & 0.799 & 0.803 & 0.799 + lrrrr 20130601 & 13:18 & 218.60 & @xmath348 & @xmath349 + 20130605 & 13:23 & 336.40 & @xmath350 & @xmath351 + 20130609 & 13:21 & 271.80 & @xmath352 & @xmath353 + 20130610 & 13:32 & 265.99 & @xmath354 & @xmath355 + 20130612 & 12:15 & 271.79 & @xmath356 & @xmath357 + 20130615 & 12:08 & 218.48 & @xmath358 & @xmath359 + 20130616 & 12:53 & 218.95 & @xmath360 & @xmath361 + 20130617 & 12:15 & 219.71 & @xmath362 & @xmath363 + 20130618 & 12:03 & 218.47 & @xmath364 & @xmath365 + 20130619 & 12:41 & 218.95 & @xmath366 & @xmath367 + 20130620 & 12:13 & 214.60 & @xmath368 & @xmath369 + 20130621 & 12:37 & 218.77 & @xmath370 & @xmath371 + 20130622 & 12:41 & 218.93 & @xmath372 & @xmath373 + 20130623 & 11:39 & 218.68 & @xmath374 & @xmath375 + 20130624 & 12:10 & 218.75 & @xmath376 & @xmath377 + 20130626 & 11:43 & 218.68 & @xmath378 & @xmath379 + 20130630 & 12:13 & 333.40 & @xmath380 & @xmath381 + 20130701 & 10:35 & 342.96 & @xmath382 & @xmath383 + 20130705 & 07:18 & 226.86 & @xmath384 & @xmath385 + 20130705 & 08:55 & 226.86 & @xmath386 & @xmath387 + 20130705 & 11:41 & 226.86 & @xmath388 & @xmath389 + 20130708 & 11:07 & 218.87 & @xmath390 & @xmath391 + 20130713 & 10:55 & 336.38 & @xmath392 & @xmath393 + 20130714 & 10:44 & 333.40 & @xmath394 & @xmath395 + 20130716 & 10:36 & 218.84 & @xmath396 & @xmath397 + 20130717 & 11:05 & 265.99 & @xmath398 & @xmath399 + 20130719 & 10:32 & 265.97 & @xmath400 & @xmath401 + 20130720 & 10:42 & 271.78 & @xmath402 & @xmath403 + 20130721 & 10:30 & 218.69 & @xmath404 & @xmath405 + 20130723 & 10:43 & 218.84 & @xmath406 & @xmath407 + 20130724 & 10:45 & 218.90 & @xmath408 & @xmath409 + 20130728 & 10:12 & 336.07 & @xmath410 & @xmath411 + 20130731 & 10:07 & 218.79 & @xmath412 & @xmath413 + 20130801 & 09:58 & 218.79 & @xmath414 & @xmath415 + 20130802 & 09:38 & 218.79 & @xmath416 & @xmath417 + 20130803 & 09:01 & 218.82 & @xmath418 & @xmath419 + 20130804 & 08:46 & 334.73 & @xmath420 & @xmath421 + 20130804 & 09:39 & 215.44 & @xmath422 & @xmath423 + 20130805 & 08:51 & 336.98 & @xmath424 & @xmath425 + 20130806 & 09:49 & 218.96 & @xmath426 & @xmath427 + 20130808 & 09:09 & 333.98 & @xmath428 & @xmath429 + 20130809 & 08:34 & 218.85 & @xmath430 & @xmath431 + 20130810 & 09:13 & 218.77 & @xmath432 & @xmath433 + 20130811 & 08:55 & 218.77 & @xmath434 & @xmath435 + 20130812 & 09:10 & 218.99 & @xmath436 & @xmath437 + 20130813 & 09:10 & 218.66 & @xmath438 & @xmath439 + 20130814 & 08:19 & 334.00 & @xmath440 & @xmath441 + 20130815 & 03:27 & 226.85 & @xmath442 & @xmath443 + 20130815 & 03:56 & 226.85 & @xmath444 & @xmath445 + 20130815 & 04:27 & 226.85 & @xmath446 & @xmath447 + 20130815 & 04:56 & 226.85 & @xmath448 & @xmath449 + 20130815 & 05:27 & 226.85 & @xmath450 & @xmath451 + 20130815 & 05:56 & 226.85 & @xmath452 & @xmath453 + 20130815 & 06:25 & 226.85 & @xmath454 & @xmath455 + 20130815 & 06:57 & 226.85 & @xmath456 & @xmath457 + 20130815 & 07:25 & 226.85 & @xmath458 & @xmath459 + 20130815 & 08:01 & 226.85 & @xmath460 & @xmath461 + 20130815 & 08:26 & 226.85 & @xmath462 & @xmath463 + 20130815 & 08:53 & 226.85 & @xmath464 & @xmath465 + 20130815 & 09:24 & 226.85 & @xmath466 & @xmath467 + 20130816 & 08:12 & 220.59 & @xmath468 & @xmath469 + 20130914 & 06:24 & 218.93 & @xmath470 & @xmath471 + 20130916 & 06:28 & 331.32 & @xmath472 & @xmath473 + 20130917 & 06:43 & 219.19 & @xmath474 & @xmath475 + 20130918 & 06:15 & 331.31 & @xmath476 & @xmath477 + 20130920 & 06:12 & 331.31 & @xmath478 & @xmath479 + 20130921 & 05:01 & 219.87 & @xmath480 & @xmath481 + 20130921 & 05:01 & 334.79 & @xmath482 & @xmath483 + 20130922 & 05:49 & 220.64 & @xmath484 & @xmath485 + 20130922 & 05:49 & 355.92 & @xmath486 & @xmath487 + 20130924 & 05:38 & 331.37 & @xmath488 & @xmath489 + 20130925 & 06:30 & 219.21 & @xmath490 & @xmath491 + 20130926 & 06:28 & 218.93 & @xmath492 & @xmath493 + 20130928 & 05:24 & 220.71 & @xmath494 & @xmath495 + 20130928 & 05:24 & 355.92 & @xmath496 & @xmath497 + 20130930 & 05:54 & 220.28 & @xmath498 & @xmath499 + 20131002 & 05:32 & 220.27 & @xmath500 & @xmath501 + 20131003 & 05:30 & 220.27 & @xmath502 & @xmath503 + 20131004 & 04:22 & 219.85 & @xmath504 & @xmath505 + 20131006 & 05:06 & 334.92 & @xmath506 & @xmath507 + 20131007 & 05:00 & 220.27 & @xmath508 & @xmath509 + 20131008 & 04:37 & 239.61 & @xmath510 & @xmath511 + 20131012 & 04:26 & 218.83 & @xmath512 & @xmath513 + 20131013 & 04:27 & 218.83 & @xmath514 & @xmath515 + 20131022 & 04:08 & 239.61 & @xmath516 & @xmath517 + 20131101 & 03:25 & 218.90 & @xmath518 & @xmath519 + 20131102 & 03:16 & 219.37 & @xmath520 & @xmath521 + 20140116 & 19:58 & 225.02 & @xmath522 & @xmath523 + 20140408 & 13:59 & 266.10 & @xmath524 & @xmath525 + 20140409 & 16:42 & 334.01 & @xmath526 & @xmath527 + 20140410 & 17:01 & 219.74 & @xmath528 & @xmath529 + 20140412 & 16:41 & 218.92 & @xmath530 & @xmath531 + 20140415 & 16:22 & 356.05 & @xmath532 & @xmath533 + 20140416 & 16:40 & 218.78 & @xmath534 & @xmath535 + 20140417 & 16:37 & 218.43 & @xmath536 & @xmath537 + 20140422 & 16:10 & 220.56 & @xmath538 & @xmath539 + 20140423 & 16:47 & 218.97 & @xmath540 & @xmath541 + 20140425 & 15:19 & 215.27 & @xmath542 & @xmath543 + 20140505 & 15:01 & 219.02 & @xmath544 & @xmath545 + 20140506 & 14:49 & 218.74 & @xmath546 & @xmath547 + 20140510 & 13:43 & 336.95 & @xmath548 & @xmath549 + 20140511 & 14:15 & 336.95 & @xmath550 & @xmath551 + 20140512 & 15:43 & 218.93 & @xmath552 & @xmath553 + 20140521 & 14:31 & 219.00 & @xmath554 & @xmath555 + 20140522 & 14:12 & 219.00 & @xmath556 & @xmath557 + 20140523 & 14:15 & 332.14 & @xmath558 & @xmath559 + 20140826 & 08:13 & 335.55 & @xmath560 & @xmath561 + 20140829 & 07:09 & 353.75 & @xmath562 & @xmath563 + 20140902 & 07:39 & 241.36 & @xmath564 & @xmath565 + 20140903 & 07:45 & 271.52 & @xmath566 & @xmath567 + 20140904 & 07:07 & 220.77 & @xmath568 & @xmath569 + 20140905 & 07:20 & 220.77 & @xmath570 & @xmath571 + 20140909 & 07:18 & 241.36 & @xmath572 & @xmath573 + 20140912 & 06:06 & 335.59 & @xmath574 & @xmath575 + 20140915 & 05:45 & 220.20 & @xmath576 & @xmath577 + 20140917 & 06:19 & 333.39 & @xmath578 & @xmath579 + 20140925 & 05:40 & 214.17 & @xmath580 & @xmath581 + 20140930 & 05:15 & 219.42 & @xmath582 & @xmath583 + 20141001 & 05:50 & 213.23 & @xmath584 & @xmath585 + 20141002 & 05:05 & 213.44 & @xmath586 & @xmath587 + 20141003 & 05:39 & 219.52 & @xmath588 & @xmath589 + 20141005 & 05:25 & 218.68 & @xmath590 & @xmath591 + 20141006 & 05:36 & 218.67 & @xmath592 & @xmath593 + 20141009 & 05:18 & 218.62 & @xmath594 & @xmath595 + 20141015 & 05:24 & 219.71 & @xmath596 & @xmath597 + 20141016 & 05:16 & 218.90 & @xmath598 & @xmath599 + 20141021 & 05:03 & 220.26 & @xmath600 & @xmath601 + 20141025 & 04:09 & 219.42 & @xmath602 & @xmath603 + 20141029 & 04:10 & 335.58 & @xmath604 & @xmath605 + 20141030 & 04:03 & 335.58 & @xmath606 & @xmath607 + 20141031 & 04:18 & 356.16 & @xmath608 & @xmath609 + 20141103 & 04:10 & 220.27 & @xmath610 & @xmath611 + 20141104 & 04:09 & 218.63 & @xmath612 & @xmath613 + 20141107 & 03:49 & 220.26 & @xmath614 & @xmath615 + 20141109 & 03:55 & 333.33 & @xmath616 & @xmath617 + 20141110 & 03:46 & 333.36 & @xmath618 & @xmath619 + 20141112 & 03:51 & 333.36 & @xmath620 & @xmath621 + rrrrrrrr 1.6 & 0.592 & 0.028 & 0.512 & 0.683 & 8 & vla + 3.1 & 0.702 & 0.032 & 0.659 & 0.896 & 10 & vla + 5.4 & 0.870 & 0.118 & 0.647 & 1.134 & 12 & vla + 9.0 & 0.932 & 0.129 & 0.791 & 1.276 & 12 & vla + 14.0 & 1.075 & 0.135 & 0.887 & 1.749 & 14 & vla + 21.1 & 1.164 & 0.052 & 0.999 & 1.562 & 15 & vla + 32.0 & 1.382 & 0.087 & 1.146 & 1.648 & 15 & vla + 40.9 & 1.485 & 0.073 & 1.275 & 1.882 & 15 & vla + 218.0 & 3.667 & 0.650 & 2.811 & 4.635 & 8 & alma + 220.0 & 3.661 & 0.652 & 2.803 & 4.649 & 8 & alma + 231.9 & 3.676 & 0.664 & 2.800 & 4.752 & 8 & alma + 233.8 & 3.704 & 0.680 & 2.801 & 4.758 & 8 & alma + 341.6 & 3.602 & 0.866 & 2.332 & 4.537 & 8 & alma + 343.6 & 3.609 & 0.870 & 2.341 & 4.564 & 8 & alma + 351.7 & 3.595 & 0.884 & 2.311 & 4.569 & 8 & alma + 353.6 & 3.553 & 0.860 & 2.301 & 4.537 & 8 & alma + 216.8 & 3.677 & 0.762 & 2.154 & 5.420 & 68 & sma + 223.9 & 3.391 & 0.489 & 2.586 & 4.796 & 23 & sma + 238.2 & 3.310 & 0.424 & 2.955 & 4.035 & 4 & sma + 266.8 & 3.369 & 0.096 & 3.237 & 3.482 & 4 & sma + 274.0 & 3.526 & 0.697 & 2.799 & 4.743 & 4 & sma + 331.1 & 3.205 & 1.074 & 1.999 & 5.492 & 14 & sma + 338.3 & 3.436 & 0.863 & 2.421 & 5.751 & 13 & sma + 352.6 & 4.890 & 0.721 & 4.146 & 6.090 & 4 & sma +

the current understanding of the details of quantum hall effect relies crucially on the composite fermion picture @xcite . an important step in the development of this understanding was the introduction of statistical gauge fields and the rle they play in the landau - ginsparg effective action , @xcite . in fact the law of corresponding states of reference @xcite has an interpretation as a beautiful and extremely powerful generalisation of kramers - wannier duality , @xcite @xcite , though references @xcite and @xcite were completely independent of one another and appeared almost simultaneously . kramers - wannier duality is a discrete @xmath0 map of the partition function for the 2-dimensional ising model at one temperature to the same partition function at a different temperature . its power lies in the prediction of the critical temperature without having to solve the model . since its discovery however other techniques have proven more useful not only because the model has since been solved exactly in 2-dimensions but also because , being a discrete symmetry , duality is all or nothing affair . there is no algorithm for finding such symmetries it is hit and miss guesswork . this is one of the reasons why the renormalisation group approach has proven more powerful as a technique for understanding the physics at and near critical points associated with second order phase transitions not only for the 2 and 3-dimensional ising models in the presence of an external magnetic field but also for many other models . the renormalisation group ( rg ) approach can not really be called an algorithm ( it has many different manifestations all of which come under the same general heading of `` rg '' ) but it is nevertheless algorithmic in nature given a field theoretical model one can perturb around free field theory and hope to apply some version of the rg to extract physical information without actually having to solve the model . nevertheless interest in duality has never died perhaps because of the elegance of the idea but also because , if one _ can _ find a duality , one can obtain exact results whereas the perturbative rg is always an approximation . with duality the chances of success are smaller but the payoff is greater . a @xmath0 duality for massive maxwell - chern - simons theory was discovered in @xcite which necessitated the introduction of a complex coupling with positive imaginary part . a generalisation of kramers - wannier duality from @xmath0 to an infinite discrete non - abelian group ( called the modular group , which is defined in the next section ) was discovered for a coupled clock model in @xcite . being a larger group this leads to a much more complicated phase diagram and it has been suggested that a version of this duality applies to potts models with a complex temperature , @xcite . the importance of the modular group , and some of its sub - groups , for the quantum hall effect was realised in @xcite and @xcite and this led to the development of the phase diagram compatible with modular symmetry first presented in @xcite . interest in duality has also seen a major revival recently in the context of super - symmetric gauge theories and string theory ( for a review see @xcite ) . very strong results can be obtained by combining duality symmetry ( when one can be derived ) with rg techniques . this was first done for the quantum hall effect in @xcite . the basic idea is that the duality map should commute with the rg flow which puts very strong constraints on the @xmath1-functions of the theory . for example fixed ( i.e. self - dual ) points of duality symmetry correspond to fixed points of the rg and therefore the @xmath1-functions would be expected to vanish at these points even if they correspond to strongly coupled theories in the field theoretic sense . since @xcite appeared a number of authors has examined the constraints put on the rg flow by modular symmetry under various other assumptions , @xcite,@xcite,@xcite,@xcite,@xcite . indeed one can use this approach , together with particle - hole symmetry , to derive the semi - circle law for transitions for quantum hall transitions very generally without any recourse to a specific microscopic model it follows very generally from the duality symmetry , @xcite . the aim of this paper is to explore further the consequences of this infinite generalisation of kramers - wannier duality for the quantum hall effect . in particular electron spin is incorporated into the duality picture and it is argued that when the splitting of the landau levels due to electron spin is small compared to the cyclotron energy the relevant duality symmetry is smaller than when the landau levels are all well separated . technically speaking the symmetry group is @xmath2 in the former case and @xmath3 in the latter ( these groups are defined in the next section ) . in order to compare this prediction with experiment some extra assumptions about the form of the @xmath1-functions are necessary and here the analysis of @xcite is applied to the case of small spin splitting . this involves a very specific assumption concerning the form of the @xmath1-functions that they are meromorphic in the sense of a natural complex conductivity , @xmath4 ( throughout this paper units are used in which @xmath5 ) . unfortunately this assumption has no microscopic justification to date , rather it is made on the basis of analogy with other models with modular symmetry as a duality , @xcite . however it leads to such specific predictions that is easily falsifiable , though comparisons with experimental data show that its predictions agree very well with the currently available data lending some confidence that a justification may be found in the future . the available data to date , at strong fields and at low enough temperatures for the law of corresponding states to represent a symmetry , seem to be compatible with @xmath3 when the meromorphic ansatz is used , so it is not possible to test the specific predictions made here concerning the @xmath2 case . this must wait for future experiments . an important aspect of the comparison with experimental data made in section six is the normalisation of the longitudinal resistivity , @xmath6 . this is a problem for experiment , which in reality measures the longitudinal resistance @xmath7 , and it is usually assumed these are related by @xmath8 in a rectangular sample of length @xmath9 and width @xmath10 . the assumptions that go into this relation are examined in section five where it is argued that a better formula is @xmath11 where @xmath12 is an undetermined sample dependent function with the universal property that @xmath13 . the duality symmetry of the quantum hall effect however suggests an alternative method for extracting @xmath6 from the experimental @xmath7 , at least for hall - hall transitions . given that the duality group , together with particle - hole symmetry , predicts the semi - circle law one can use this to determine the normalisation of @xmath7 by _ assuming _ the semi - circle law at sufficiently low temperatures . this technique can not determine the normalisation by using a laughlin state - insulator transition because these correspond to vertical lines in the complex @xmath14-plane , which remain vertical under any re - scaling of @xmath6 . however this very fact dictates that laughlin state - insulator crossovers should be semi - circles in the @xmath15-plane for any normalisation of @xmath6 and this gives a quantitative method of determining when the temperature is ` low enough ' for the infinite duality symmetry to be valid the laughlin state - insulator transitions must be semi - circles in the complex @xmath15-plane . when this is the case the normalisation of @xmath6 for a hall - hall transition can be extracted from the experimental data by choosing the normalisation so as to get as close as possible to a semi - circle for the crossover in the complex @xmath14-plane . the layout of the paper is as follows : in section two the definition of the modular group and its various relevant sub - groups is reviewed together with its action in the quantum hall effect ; section three discusses the relation between the modular group and scaling ; section four introduces the specific ansatz of meromorphic @xmath1-functions ; section five is devoted to the relation between the longitudinal resistance and resistivity while section six compares the predictions of section four to existing experimental data . finally section seven contains a summary and conclusions . the law of corresponding states @xcite was originally proposed on the basis of an effective field theory maxwell - chern - simons theory . it provides a powerful method for classifying quantum hall states and transitions between them , both integral and fractional , as well as for hall - insulator transitions . the law is expressed by writing the conductivities ( both longitudinal @xmath16 and transverse @xmath17 ) as functions of the filling factor @xmath18 . then the following transformations were defined in @xcite : * landau level addition transformation ( * l * ) @xmath19 * flux attachment transformation ( * f * ) @xmath20 * particle - hole transformation ( * p * ) @xmath21 as discussed in @xcite under certain circumstances the arrows above can be replaced by equalities and then these transformations become _ symmetries_. this is not expected to be true in general but should hold , for example , as the temperature @xmath22 tends to zero . at almost exactly the same time , but from a completely different perspective , the modular group @xcite or some of its sub - groups @xcite were also implicated as being relevant to these phenomena . in this approach one defines a _ complex _ conductivity , @xmath23 , which by necessity lives on the upper - half complex plane since @xmath24 , and a group action @xmath25 where @xmath26 and @xmath27 are any four integers satisfying @xmath28 . this condition can re - written by defining a @xmath29 matrix @xmath30 and demanding that @xmath31 . it is easy to check from the definition ( [ gammadef ] ) that , for any three such matrices satisfying @xmath32 , we have @xmath33 . thus the group multiplication law is given by matrix multiplication . note that for any given matrix @xmath34 the matrix @xmath35 gives the same transformation since the minus signs cancel above and below in ( [ gammadef ] ) . the resulting group is called the modular group in the mathematical literature , sometimes denoted @xmath36 , and it is related to @xmath37 . matrices of real numbers with unit determinant ( @xmath38 ) obtained by restricting to integral entries . ] they are not quite the same group because @xmath37 distinguishes between @xmath34 and @xmath35 whereas @xmath36 does not , so @xmath36 is obtained from @xmath37 by identifying @xmath34 and @xmath35 as group elements , or in other words projecting out by a @xmath0 factor , so @xmath39 . any element of the modular group can be obtained by taking a string of products of the two generators @xmath40 and @xmath41 . in this matrix notation landau level addition and flux attachment are represented by @xmath42 and @xmath43 and these two transformations generate an infinite discrete group , which is a sub - group of @xmath36 , often denoted by @xmath44 in the mathematical literature ( see @xcite for example , another notation commonly used is @xmath45 , @xcite ) . any element @xmath46 can be represented by some string of products of @xmath47 and @xmath48 . this sub - group is most succinctly described by restricting the integer @xmath49 in the matrix @xmath34 to be even . the particle - hole transformation @xmath50 is an outer auto - morphism of the modular group and also of the sub - group , @xmath3 . the hypothesis that these transformations represent a symmetry can now be expressed as @xmath51 ( flux attachment is easily transcribed from resistivities to conductivities , in complex notation , by @xmath52 , with @xmath53 ) . other sub - groups of @xmath36 have been proposed in the literature as being relevant to the quantum hall effect , @xcite , @xcite , @xcite and it shall be argued here that the group proposed in @xcite is relevant when spin splitting of the electrons results in the landau levels that are not very well separated . obviously the zeeman effect would result in landau levels that are split but it is also argued in @xcite the exchange energy between electrons of opposite spin in electron - electron interactions can be larger than the zeeman energy . thus , despite the difference in philosophy , the two approaches in @xcite and @xcite give the same results at zero temperature , where the law of corresponding states becomes a symmetry in @xcite which is precisely the symmetry group @xmath44 in @xcite . the latter approach is in many ways more powerful however since , once one realizes that the law of corresponding states is a group , one can bring the full machinery of group theory to bear on the problem indeed there is a very highly developed and beautiful mathematical literature on the modular group and its sub - groups and the transformation properties of functions of @xmath15 under this group action , @xcite @xcite . thus , for example , one can deduce the selection rule @xmath54 for transitions between two states with filling factors @xmath55 and @xmath56 , with @xmath57 and @xmath58 odd , @xcite . however the law of corresponding states does serve to highlight the physics of the problem in particular the landau level addition symmetry assumes that the physics at any partially filled landau level is independent of how many lower landau levels are completely full . while this seems a reasonable assumption when the landau levels are well spaced it is not so reasonable if any two landau levels get close to one another . spin splitting is important here : if one ignores electron spin the landau levels of the unperturbed hamiltonian are all evenly spaced , but including spin and a land @xmath59-factor could result either in well spaced landau levels ( if the landau splitting is comparable to the cyclotron energy ) or in landau levels occurring in pairs which are close to one another ( if the spin splitting is very small ) . some possibilities are shown in figure 1 . the important parameter here is the ratio of the spin splitting energy to the cyclotron energy . if this ratio is precisely @xmath60 then all the split landau levels are evenly spaced at exactly half the cyclotron energy of the spinless case . experimentally the value is often near @xmath60 , @xcite . however if the ratio of the spin splitting energy to the cyclotron energy is either very small or near a half ( more generally near an integer or a half integer ) , then the landau levels lie very close to each other in pairs . in this latter case the physics in the upper member of a pair could well be influenced by the presence of electrons in its lower member . however , when the pairs are well separated it would seem reasonable to assume that the physics of any _ pair _ is independent of how many pairs below are filled . this is tantamount to replacing the landau level addition transformation with @xmath61 : @xmath62 there does not seem to be any reason to expect flux attachment to be modified when the spin splitting is small and we shall therefore concentrate on the group generated by @xmath48 and @xmath61 . this sub - group of @xmath3 is usually denoted by @xmath2 in the mathematical literature and it can can also be described by demanding that both @xmath63 and @xmath49 in ( [ gammadef ] ) be even integers . this group has already received some attention in relation to the quantum hall effect , @xcite,@xcite . it is our intention to explore the differences between @xmath3 and @xmath2 , as applied to the quantum hall effect and to the scaling flow of the conductivities in particular . to this end the next section deals with the relation between the modular group and scaling . very powerful predictions can be made by combining the two parameter scaling hypotheses of khmelnitskii @xcite , as implemented in @xcite , with modular symmetry of the quantum hall effect . one assumes that there is a physical length scale , @xmath9 , upon which the conductivities depend @xmath9 could be the electron scattering length , for example , which diverges as @xmath64 in an infinite system . it is argued in @xcite that @xmath9 should depend on the external magnetic field @xmath65 and the temperature @xmath22 through a single scaling variable @xmath66 , where @xmath67 is the deviation of the external magnetic field from the critical value @xmath68 and @xmath69 is an anomalous dimension . the derivatives @xmath70 give flow equations for the conductivities , whose integral curves , @xmath71 and @xmath72 , are the scaling functions of @xcite . more generally one would expect the conductivities to depend on more parameters , such as the density of impurities ( integer effect ) or the charge carrier density ( fractional effect ) . scaling arguments would then lead to the form @xmath73 and @xmath73 for some exponent @xmath74 , where @xmath75 is the charge carrier density , for example . thus varying @xmath65 and @xmath22 independently , while keeping @xmath75 fixed , gives two free parameters with which to explore the whole upper - half @xmath15-plane . alternatively one could keep @xmath65 fixed and vary @xmath22 and @xmath75 . pruisken s scaling form is recovered at low temperatures , provided the conductivities are well behaved functions of the second argument , which tends to zero as @xmath64 if the exponent @xmath74 is negative and diverges as @xmath64 if @xmath74 is positive . in the following it shall be assumed that @xmath22 is low enough that the second argument here can be ignored and the conductivities are functions of a single scaling variable , @xmath76 . more generally one could define @xmath1-functions to be derivatives with respect to a function of @xmath63 , or alternatively one could use filling factors , @xmath77 , and consider derivatives with respect to a function @xmath78 of @xmath79 , with @xmath80 , @xmath81 these latter are clearly not the same as the @xmath1-functions in ( [ betal ] ) . indeed it may be an abuse of language to call them @xmath1-functions the integrated form @xmath82 and @xmath83 might be more correctly called scaling functions but we shall continue to use the notation of equation ( [ betas ] ) since these functions are somewhat analogous to callan - symanzik functions of relativistic quantum field theory . one of the principle assumptions used in the following is that the action of @xmath3 ( or @xmath2 ) should commute with the flow equations ( [ betas ] ) , @xcite . when spin splitting leads to well separated landau levels ( @xmath3 case ) one can make predictions about the global phase diagram in the complex @xmath15-plane and the positions of the critical points in the crossover between two hall plateaux and between a laughlin state and the insulator , @xcite , @xcite , @xcite . moreover one can deduce the semi - circle law , @xcite , without making any further assumptions indeed without recourse to any specific microscopic model . this law was proposed on the basis of an analysis of a specific microscopic model in @xcite and has strong experimental support @xcite . there is also a duality for any transition between two hall plateaux at @xmath55 and @xmath56 , with @xmath57 and @xmath58 odd . and @xmath84 ( @xmath85 , @xmath86 and @xmath87 ) . in the sequel any mention of a hall - hall transition is implied to include the laughlin state - insulator transition , unless otherwise stated explicitly . ] this duality is given by parameterizing the semi - circle by @xmath88 in @xmath89 ( where @xmath90 ) and inverting @xmath91 . for the laughlin state insulator transition , this @xmath91 inversion leads to the observed experimental duality @xmath92 under @xmath93 @xcite . in particular a general formula for the values of the conductivities at the critical points is given by setting @xmath94 , @xmath95 , @xcite these are fixed points of @xmath3 in the sense that there exists an element of @xmath3 that leaves these points invariant . they are the only such points above the real axis in the complex @xmath15-plane . by making further assumptions about the form of the scaling functions one can make much more specific predictions about the form of the crossover between two hall states for the @xmath3 case , @xcite,@xcite,@xcite . for the @xmath2 case , when spin splitting does not separate the landau levels completely , there has been an analysis of the possible scaling flow too , @xcite,@xcite . though in this case there is no prediction of the critical conductivities for any transitions because the group @xmath2 has no fixed points above the real axis . however one can still derive the semi - circle law using exactly the same techniques as in @xcite , except applied to @xmath2 rather than @xmath3 , provided one assumes a generalisation for particle - hole symmetry for @xmath2 , @xmath96 to go further in the analysis of a flow compatible with @xmath2 we need to make further assumptions about the form of the @xmath1-functions in ( [ betas ] ) and in the next section the assumptions and techniques adopted in @xcite for the @xmath3 case will be adapted and modified to further the analysis of @xcite . in terms of the complex conductivity , equation ( [ betas ] ) can be written as @xmath97 demanding that the flow described in ( [ generalbeta ] ) commute with the action of @xmath3 or @xmath2 requires that @xcite @xmath98 this equation follows easily from ( [ gammadef ] ) and the fact that @xmath31 . to go any further requires making some further assumptions . various possibilities have been suggested in the literature , @xcite @xcite , and the line we shall follow is to investigate further the consequences of the assumption that the @xmath1-functions are _ meromorphic _ functions of @xmath15 . unfortunately this assumption has no microscopic justification at the moment , rather it is made on the basis of experience with super - symmetric qcd where the groups @xmath3 and @xmath2 also appear and the @xmath1-function for @xmath99 super - yang - mills theory are meromorphic functions of a complex coupling with positive imaginary part , @xcite @xcite . there is certainly no evidence for any real connection between the quantum hall effect and super - symmetric qcd , beyond the observation that the same discrete groups crop up as symmetries of the physics in both cases , but it is interesting to apply expertise gained from the occurrence of these groups in the latter framework to the former , just to see what happens . the occurrence of meromorphic @xmath1-functions in super - symmetric qcd is intimately related to super - symmetry and it is possible that super - symmetry may have a rle to play in the quantum hall effect , @xcite , but this is as yet unclear . however it may be worth noting in passing that the modular group was discovered in a statistical mechanical model prior to its use in the quantum hall effect @xcite indeed this work analysed a model which was motivated by ordinary qcd ( i.e. not super - symmetric ) . a second motivation for the meromorphic assumption , and perhaps more important from a practical point of view , is that there is a very highly developed mathematical literature on the theory of meromorphic functions satisfying equation ( [ modform ] ) such functions are called modular forms of weight @xmath100 in the mathematical literature and this puts very strong restrictions on the form of the @xmath1-functions . thus even though the meromorphic assumption lacks any physical foundation the predictions are extremely strong and thus the assumption is easily falsifiable by experiment an important aspect of any theoretical analysis . we shall see , in fact , that experiment agrees surprisingly well with the consequences of this assumption . for the @xmath3 case it is shown in @xcite that @xmath1-functions satisfying @xmath101 ( with @xmath49 even ) and subject to some further reasonable physical restrictions , must have a precise analytic form in terms of well known classical functions called jacobi @xmath102-functions , @xcite . the further restrictions are : that the @xmath1-functions should vanish at @xmath103 and @xmath104 as fast as possible in order to explain the fantastic stability of the hall plateaux , which must be attractive fixed points of the flow ; that the @xmath1-functions should be finite as @xmath105 the weak coupling regime of field theory models ; and finally that there are no fixed points other than those mentioned above and those representing the critical point in the transition between any two plateau . a further assumption used in @xcite , that @xmath17 should be constant as @xmath105 , is in fact not necessary since this is actually a consequence of the other assumptions , @xcite . the jacobi @xmath102-functions are defined either as infinite sums or as infinite products in the variable @xmath106 , which converge for @xmath107 : @xmath108 and satisfy the identity @xmath109 there is a very special meromorphic function that is _ invariant _ under @xmath3 and is the simplest such function , in the sense that it has the fewest possible number of zeros and poles : @xmath110 satisfies @xmath111 for any @xmath112 . the unique form of the @xmath1-function ( up to a positive constant ) that is compatible with the above assumptions is obtained in @xcite and is @xmath113 where @xmath114 . of course to make any contact with physics we still have to specify the unknown function @xmath78 that is used to define the @xmath1-functions in ( [ betas ] ) we shall say more about this later . these ideas were applied to @xmath1-functions compatible with @xmath2 in @xcite but the resulting flow had no evidence of any critical point point in the crossover between two plateau . we shall therefore re - examine this flow , with a view to introducing such a critical point . to do this recall the physical picture that @xmath3 is the relevant symmetry group when the landau levels are all well separated . now imagine varying the relative magnitudes of the spin splitting and the cyclotron energy , for example by varying the land @xmath59-factor or by varying the effective mass of the charge carriers , until the landau levels are paired off into close pairs which are well separated from other pairs , as in figure 1 . in this latter case the relevant group is @xmath2 . we thus have a continuous deformation from @xmath3 symmetry to @xmath2 symmetry , and it should be possible to deform the @xmath1-functions for @xmath3 continuously to those for @xmath2 . the analysis in @xcite requires using the function @xmath115 rather than @xmath116 , since @xmath117 is the appropriate invariant function for @xmath2 @xmath116 is certainly invariant under @xmath2 , since @xmath118 and @xmath116 is invariant under @xmath3 , but it is not the simplest such function . of course since @xmath116 is invariant under @xmath2 it can be expressed in terms of @xmath117 , explicitly @xmath119 where ( [ thetaidentity ] ) has been used but one can not express @xmath117 in terms of @xmath116 without introducing branch cuts . one can thus write the @xmath3 symmetric @xmath1-functions ( [ analyticbeta ] ) in terms of @xmath117 , @xmath120 this has a pole in @xmath117 at @xmath121 , which corresponds to @xmath122 for the transition between the integer quantum hall state at @xmath103 and the insulator at @xmath123 . of course this pole is reflected at every image of @xmath124 under the action of @xmath3 and thus all the critical points for every quantum hall transition are generated from this one in the transition @xmath125 by the symmetry group . as discussed at length in @xcite this pole is not a disaster , but is a perfectly regular manifestation of the critical point in the crossover , at @xmath126 this will be discussed in more detail a little later . the pole at @xmath122 is a feature specific to @xmath3 , it must be there because @xmath124 is a fixed point of @xmath3 , but there is nothing about @xmath2 that dictates the position of the critical point . it would therefore seem natural to postulate the @xmath2 @xmath1-functions to be the following deformation of the @xmath3 @xmath1-functions : @xmath127 where @xmath128 is a sample specific constant . for well separated , spin split landau levels @xmath129 giving the @xmath3 @xmath1-functions ( [ analyticbeta ] ) , but when pairs of landau levels lies close to each other @xmath128 can deviate from the more symmetric value of @xmath130 . if one assumes particle - hole symmetry @xmath131 of equation ( [ ph2 ] ) the semi - circle law follows and the critical point must lie on the semi - circle spanning @xmath132 and @xmath133 . now the function @xmath117 is real on this curve and therefore @xmath128 must be real . in fact @xmath134 on this semi - circle and so @xmath135 for the critical point to lie between @xmath123 and @xmath103 . equation ( [ betag2 ] ) is easily integrated with respect to @xmath136 to give @xmath137 where @xmath138 is an arbitrary integration constant ( complex in general ) . since the function @xmath78 is assumed to be real along any flow line the complex phase of the right hand side of the above equation must be constant and equal to the phase of @xmath138 . thus the trajectory is determined by the phase of @xmath138 and we can easily plot the integral curves of ( [ betag2 ] ) by doing a contour plot of the complex phase of the right hand side of ( [ exps ] ) . two such plots are shown in figure 2 , the first is for the @xmath3 case , @xmath129 , and has already been presented in @xcite , while the second corresponds to the choice @xmath139 . the distortion of the flow away from the ideal @xmath3 flow is manifest in the lower figure ( khmelnitskii s original analysis , @xcite , is only compatible with @xmath3 ) . one of the main points of this paper is to suggest that such a distortion is an inevitable consequence of having landau levels that are not completely spin split . some such distortion is a likely consequence of any flow , including those that are not meromorphic . although the quantitative shape of the flow shown in figure 2 does not depend on the unknown function @xmath78 , provided only that it is monotonic , this function must still be specified in order to make further progress . in order to pin this function down , at least for hall - hall crossovers , consider the semi - circular arches on which @xmath117 is real and greater than unity . we can always choose @xmath140 at the critical point where @xmath141 , so @xmath142 in ( [ exps ] ) . now note that the flow diagrams in figure 2 are completely unaffected by a rescaling of @xmath78 . it is @xmath143 that is the physical parameter , so consider @xmath144 . a rescaling of @xmath136 should not affect the flow diagram therefore @xmath144 must be a scale invariant function of @xmath136 and the only such function is a power law @xmath145 , and so one concludes that @xmath146 , for some real constants @xmath147 and @xmath148 . this relates to the earlier observation that the @xmath1-functions ( [ analyticbeta ] ) have a pole at the critical values of @xmath149 where @xmath150 . navely this looks disastrous a critical point should be associated with a zero of the @xmath1-function not a singularity ! indeed experimentally @xmath17 and @xmath16 are perfectly finite and smooth as the magnetic field passes through @xmath68 ( for finite @xmath22 ) , and @xmath151 is finite at @xmath152 . however , as emphasised previously , these are not wilsonian @xmath1-functions and the resolution of this problem lies in the exponent @xmath147 . with @xmath153 and @xmath154 we have @xmath155 so , with @xmath151 finite at @xmath152 , @xmath1 will diverge if @xmath156 . indeed a careful analysis of the analytic form of @xmath1 in equation ( [ analyticbeta ] ) shows that @xmath157 near the critical point @xmath122 and @xmath158 . this is also true for the spin degenerate form ( [ betag2 ] ) . the explicit analytic form of @xmath159 , within the framework of the meromorphic ansatz ( [ betag2 ] ) can now be obtained using the same methods as in @xcite . first express the invariant function @xmath115 in terms of elliptic integrals of the second kind @xmath160 using the classical formulae @xcite @xmath161 ( where the modulus @xmath162 is related to @xmath15 by @xmath163 and @xmath164 , with @xmath165 the complementary modulus ) . this gives @xmath166 and equation ( [ exps ] ) reads , with @xmath142 , @xmath167 which , for a fixed @xmath128 , defines @xmath168 in terms of @xmath136 . for @xmath3 , with @xmath129 , this is easily solved to give @xmath169 explicitly , but for a more general value of @xmath128 the relation must remain implicit . since @xmath170 on the semi - circle spanning @xmath123 and @xmath103 the elliptic integrals for the crossover are complex when written in terms of @xmath162 , which makes it difficult to pick out the real and imaginary parts of @xmath15 from the relation @xmath163 . so , as in @xcite , this relation is re - expressed in terms of @xmath171 , so that @xmath172 and @xmath173 and @xmath174 are then real , @xmath175}.\ ] ] this equation , together with the implicit definition of @xmath176 @xmath177 and the relation derived earlier for @xmath136 on the semi - circle @xmath178 completely determine the crossover conductivities as functions of the scaling variable @xmath179 in terms of two unknown , sample specific , constants , @xmath148 and @xmath128 . the exponent @xmath69 is believed to be universal and has been measured to be @xmath180 , @xcite . the conductivities , and corresponding resistivities , for the integer transition @xmath181 in the @xmath3 case ( @xmath129)were first derived in @xcite and are reproduced here in figure 3 , with the choice @xmath182 and @xmath183 , as functions of @xmath184 at the four different temperatures used in @xcite for this transition . before comparing these analytic expressions with the experimental data we must address the issue of how the experimental longitudinal conductivity is related to its theoretical cousin experimentally the longitudinal conductivity can not be measured directly . what is actually measured is the longitudinal and transverse resistances from which are inferred the longitudinal and transverse resistivities . in two dimensions , of course , resistivities and resistances have the same dimensions but they can still differ by dimensionless geometric factors . the transverse resistivity and the transverse resistance are identical , independent of any such geometrical factors a fact which is essential for the accuracy of the hall quantization but this is not the case for the longitudinal resistivity , @xmath6 and resistance , @xmath7 . when experimental resistivities are quoted it is actually @xmath7 that is measured , together with the length @xmath9 and width @xmath10 of a rectangular sample , and the formula @xmath185 is used to determine the experimental longitudinal resistivity , @xmath186 . as fabrication techniques improve @xmath10 and @xmath9 become more accurately determined and so @xmath186 becomes more accurately measured . however let us examine equation ( [ rhoexp ] ) carefully and ask what assumptions go into the derivation and how @xmath187 might be related to the @xmath6 that a theoretician might quote . the derivation of ( [ rhoexp ] ) is extremely simple because the assumptions are grossly over simplified , @xcite . in the absence of any magnetic field consider a rectangular slab carrying current density @xmath188 running parallel to its long axis , where @xmath189 is the longitudinal voltage driving the current . using @xmath190 as a cartesian co - ordinate along the width the total current is @xmath191 so @xmath192 , which is equation ( [ rhoexp ] ) . some important assumptions that go into this derivation are : * a perfectly rectangular sample . * electron trajectories which are classical , linear and parallel . * no perturbations due to source and drain contacts or hall voltage probes . * neglect of any magnetic fields generated by the currents themselves . * electrons scatter before reaching the drain . as fabrication techniques improve the first assumption may be getting better and better but the others are somewhat questionable . impurities or electron - electron interactions are liable to make electron trajectories deviate from straight lines and the last assumption may break down for very pure samples at low temperatures , for example . a more correct relation between longitudinal resistance and resistivity would read : @xmath193 where @xmath12 is some function of the geometry that tends to unity as @xmath194 , and this is all one can really say . a relation similar to ( [ rhoth ] ) has been studied when an external magnetic field is present in @xcite , however this analysis suffers from the same difficulties as listed above , and indeed some of these are discussed in that reference . if @xmath195 is not small we should expect differences between the experimentalist s @xmath186 in equation ( [ rhoexp ] ) and the theoretician s @xmath6 in equation ( [ rhoth ] ) . if there is such a renormalisation of @xmath6 how can this be reconciled with the experimentally observed semi - circle law , @xcite ? would this not be modified if the experiment had used the wrong normalisation for @xmath6 ? the answer is yes , in general , but no for laughlin - insulator transitions only and in fact the best experimental data for the semi - circle law come from the laughlin - insulator transitions , @xcite . rescaling @xmath6 for the laughlin - insulator transitions ( @xmath196 , with integral @xmath197 ) does not affect the semi - circle law because these transitions are described by the semi - circle @xmath198 , @xmath199 , which is mapped to a vertical line @xmath200 in the @xmath14-plane by @xmath52 . rescaling @xmath201 does not change the nature of these vertical lines and so merely serves to re - parameterise the semi - circle in the @xmath15-plane , but it is still a semi - circle . this is not true for hall - hall transitions , such as @xmath202 for example . the semi - circle @xmath203 is mapped to @xmath204 which is also a semi - circle . rescaling @xmath6 would distort this semi - circle into an ellipse in the @xmath14-plane which would then correspond to an ellipse in the @xmath15-plane . indeed this suggests an alternative technique for determining @xmath6 from the experimental data , which does not rely on ( [ rhoexp ] ) . if one assumes the law of corresponding states , including the particle - hole transformation , to be a symmetry at low temperatures ( possibly a more plausible assumption than classical , linear trajectories for electrons ) then the semi - circle law follows @xcite . for a hall - hall transition it has just been shown that there is at most one normalisation of @xmath6 that leads to a semi - circular transition so if we _ assume _ a semi - circular transition we can use this to determine the normalisation of @xmath6 for a hall - hall transition . this provides a technique for determining @xmath6 from the resistance @xmath7 which is completely independent of any geometrical factors or assumptions the only assumption is that the law of corresponding states is a symmetry at low enough temperatures , which is well motivated @xcite . this will not work for laughlin state - insulator transitions since these correspond to straight vertical lines in the @xmath14-plane , which are always semi - circles in the @xmath15-plane regardless of the normalisation of @xmath6 , but this very fact can be turned to our advantage . any deviation from a semi - circular arch in the @xmath15-plane for a laughlin state - insulator transition can be interpreted as a sign that the temperature is not yet low enough for the law of corresponding states to be a valid symmetry and this gives a quantitative measure of when the temperature is `` low enough '' for the normalisation of @xmath6 to be fixed for a hall - hall transition . in the following section we shall compare the analytic predictions of the previous section with the available experimental data , allowing for the possible ambiguity in the normalisation of the experimentally determined value of @xmath186 in laughlin - insulator transitions discussed above . the analytic prediction of section 4 for @xmath3 , based on the assumption of meromorphicity and derived in @xcite , is reproduced for the transition @xmath205 in figure 3 . the values of @xmath148 and @xmath69 , as well as the values of the temperature , used here were specially chosen to give a good fit to the data of @xcite and the analytic curves are displayed in conjunction with the experimental data in figure 4 . the agreement appears to be quite good without any rescaling of @xmath6 , despite the lack of any physical justification for the meromorphic ansatz . it is to be stressed that there is really only _ one _ adjustable parameter in this fit , the value of the constant @xmath148 which is just a choice of scale for the horizontal axis . the exponent @xmath69 is not independent , but is measured in this experiment to be @xmath206 so there is only a small leeway available to vary it to fit the curves . the fit is not perfect but it seems remarkable that it is so good with only one parameter at our disposal ! for the laughlin - insulator transition @xmath207 the theoretical prediction is that the critical resistivity is @xmath208 while the experimental data indicate a value some 30% larger than this . so we use the freedom to re - scale @xmath186 discussed in the last section to argue that the true normalisation for @xmath6 is such that the critical point _ should _ be at unity and re - scale the theoretical @xmath6 by @xmath209 to obtain figure 5 , which is super - imposed on the experimental data in figure 6 . again the temperatures are chosen to agree with those used in the actual experiment , but this time the value @xmath210 is found to give the best fit ( there is no reason to expect @xmath148 to be universal ) . the same sample was used in these two experiments , so the interpretation suggested here requires a different normalisation for @xmath6 in the two transitions . this could be accommodated , for example , by postulating a @xmath65 dependence in the geometrical function @xmath116 of equation ( [ rhoth ] ) , causing @xmath116 to vary by 30% as @xmath65 is increased by a factor of three , which is the ratio of the critical fields in the two transitions , but with @xmath116 being fairly constant under variations of @xmath65 of order 10% , which is the range depicted in the experimental transitions . a second experiment , with excellent support for a semi - circular transition , is that of hilke et al . , @xcite . in this experiment the critical point in the @xmath211 transition is most certainly not at @xmath212 . two possible explanations of this are are : i ) the landau levels come in closely related pairs , so that the symmetry group is @xmath2 with @xmath213 and not @xmath3 ; ii ) the group is @xmath3 but the experimental @xmath186 differs from the theoretical one by a factor of @xmath214 . using the predictions of the meromorphic ansatz we can test these two hypothesis and by far the most successful is the latter it is simply impossible to get a good fit with the former assumption while the theoretical curves for the latter , after increasing @xmath6 by 1.7 , are shown in figure 7 and compared to the experimental data in figure 8 . the data here are plotted against @xmath65 rather than @xmath18 and converting @xmath18 to @xmath65 requires knowing the charge carrier density . in terms of the magnetic field , @xmath215 , equation ( [ sdef ] ) gives the relation between @xmath65 and the scaling variable @xmath136 @xmath216 where @xmath217 is related to the constant @xmath148 in equation ( [ sdef ] ) by @xmath218 , with @xmath219 the critical filling factor . the best value of @xmath217 to fit the data of @xcite is @xmath220 while the optimal value of @xmath69 is @xmath221 which is certainly a little low for comfort , but is just compatible with three standard deviations of the value in @xcite ( no experimental value for @xmath69 is quoted in the paper @xcite , so we work with the value given in @xcite , because this is believed to be universal ) . the highest temperature at @xmath222k is a very poor fit , but all the experimental evidence is that this temperature is too high for @xmath3 symmetry to be valid there is strong deviation in the experimental data away from the semi - circle law at this temperature too . so again the agreement between the experimental data and the theoretical predictions of the meromorphic ansatz is very good , provided one re - scales @xmath186 and assumes @xmath3 symmetry . the actual values of @xmath9 and @xmath10 in this sample are @xmath223 m and @xmath224 m so @xmath225 is small but not really close to zero . of course just because one analytic suggestion appears to give a good fit to the experimental data does not mean that it has to be the correct one . the experimental data for the transition @xmath226 can also be fitted by an exponential , @xcite , but when this is done it appears to indicate that scaling , and therefore super - universality , may be in trouble . the authors of @xcite fit the form @xmath227 to their experimental data , where @xmath228 is a function of @xmath22 to be determined . they find that @xmath229 is incompatible with the data for any @xmath69 and suggest instead a linear form , @xmath230 with @xmath231 and @xmath232 non - zero constants . since @xmath233 gives the best fit this is interpreted as a violation of the scaling hypothesis . it was suggested in @xcite that it may be possible to recover scaling with a different functional fit in particular the form suggested here in terms of elliptic integrals seems promising but a definitive answer would require a quantitative analysis , such as a @xmath234-fit , which would require access to the actual experimental numbers rather than just the figures in @xcite . an important difference between the exponential and the elliptic integral forms is that , near the hall plateau at which @xmath235 , @xmath236 for the former while @xmath237 in the latter . it is difficult to distinguish visually between these two possibilities with the current data . returning to the difference between @xmath3 and @xmath2 all the experimental data considered here are well described with @xmath3 . it would be very interesting to see good experimental crossover curves at different temperatures for samples with very small spin splitting , at temperatures low enough for the semi - circle law to be valid over a range of temperatures . the arguments given in section two would imply that it may be @xmath2 that is relevant then rather than @xmath3 and the sample dependent parameter @xmath128 , which parameterizes the position of the critical point , would have to be taken from experiment rather than coming as a prediction from the symmetry . it has been argued that the rich duality structure of the quantum hall effect , as embodied in the infinite order discrete non - abelian group @xmath3 , which is the mathematical description of the law of corresponding states , should be modified when the landau levels are organized into close pairs , for example when the spin splitting is small relative to the cyclotron energy . this modification requires replacing @xmath3 with the smaller , but nevertheless still infinite order , group @xmath2 . in terms of the law of corresponding states this replacement is equivalent to replacing the landau level addition transformation , @xmath238 , with @xmath239 . the derivation of the semi - circle law from duality symmetry is still valid if one assumes that the particle - hole transformation of the spin split case , @xmath240 , is replaced with @xmath241 . the main difference between the groups @xmath3 and @xmath2 is that for the later the critical point in the crossover between two hall plateaux ( or between a laughlin state and the insulating phase ) is no longer fixed by duality , as it is in @xmath3 case , but can be anywhere on the semi - circle spanning the two states in the complex @xmath15-plane . this results in a distortion of the flow diagram shown in the lower diagram of figure 2 . the experimental data often show a critical point in the crossover which is not at the point predicted by @xmath3 symmetry , i.e. at @xmath242 for the hall - hall transition @xmath202 . two possible causes of this are investigated here : i ) it is due to the breaking of @xmath3 to @xmath2 or ii ) it is due to a renormalisation of the experimentally determined longitudinal resistivities so that they do not coincide with their theoretical values . without any further assumptions one can not distinguish between these two possibilities . with the added assumption that the @xmath1-function described in section three are meromorphic functions of the complex conductivity an analytic form is obtained for @xmath243 and @xmath244 . for the @xmath2 case there are two free parameters in the resulting expression , equations ( [ analyticba])-([sdef ] ) , apart from the exponent @xmath69 these correspond to the normalisation of the horizontal axis in figure 3 and the angular position of the critical point in the crossover . for integer transitions the critical point lies at the top of the semi - circle when the larger symmetry @xmath3 is relevant but can be anywhere on the arch for the smaller group @xmath2 . however a simplified analysis of the relation @xmath245 , used to determine the longitudinal resistivity from the experimentally measured longitudinal resistance , shows that it should only be considered to yield the correct normalisation for @xmath6 in the limit @xmath246 . it is suggested that a more reliable technique for determining the correct normalisation is to ascertain the temperature range for which the laughlin - insulator transition is semi - circular and to assume that the semi - circle law is also valid for hall - hall transitions in this regime . this would then fix a unique value for the ratio @xmath247 for the hall - hall transitions . a comparison with some of the available experimental data , gives very good fits ( figures 4 , 6 , and 8) for @xmath3 duality provided one renormalises the experimental value of @xmath6 , at least for the laughlin - insulator transitions . this author is not aware of any experimental data in support of the breaking of @xmath3 to @xmath2 but , if the ideas presented here are correct , it ought to be possible to see this breaking in samples where the spin splitting is small and the temperature is low enough for the semi - circle law to be valid . the main difficulty with the present analysis is of course the ad hoc assumption of meromorphic @xmath1-functions . there is no theoretical justification for this assumption , either microscopic or mesoscopic , but in view of the remarkable agreement with experiment that it produces ( figures 4 , 6 , and 8) it would seem to merit some consideration . the assumption of complex analyticity is a very natural one in the context of super - symmetric gauge theories , where the groups @xmath3 and @xmath2 also manifest themselves as duality symmetries , but this is a consequence of super - symmetry in these models . while it is not impossible that super - symmetry may be relevant to the quantum hall effect it is certainly not established and probably not accepted by most workers in the field . the original motivation for testing an analytic ansatz was simply mathematical elegance and it is certainly a surprise that it seems to give answers that agree so well with experiment . nevertheless it seems worthwhile pursuing this hypothesis in the light of said agreement and a possible physical justification for complex analyticity of the @xmath1-functions is currently under investigation . it is a pleasure to thank the theoretical physics group , institute for nuclear physics , orsay , where most of this work was carried out , for their hospitality and in particular yvon georgelin for useful discussions . i would also like to thank cliff burgess for helpful discussions on crossover in the quantum hall effect . jain , phys . lett . * 63 * , 199 ( 1989)j.k . jain , adv . phys . * 41 * , 105 ( 1992 ) s .- c . zhang , t.h . hansson and s. kivelson , phys . lett . * 62 * , 82 ( 1989 ) s. kivelson , d - h . lee and s - c . zhang , phys . * b46 * , 2223 ( 1992 ) c.a . ltken and g.g . ross , phys . rev . * b45 * , 11837 ( 1992 ) c.a . ltken and g.g . ross , phys . * b48 * , 2500 ( 1993 ) s .- j . rey and a. zee , b352 * , 897 ( 1991 ) j.l . cardy and e. rabinovici , nuc . * b205 * , 1 ( 1982 ) ; j. l. cardy , nuc . b205 * , 17 ( 1982 ) g. bertoldi , _ potts model : duality , uniformization and the seiberg - witten modulus _ , cond - mat/9911383 c.a . ltken , j. phys . * 26 * , l811-l817 ( 1993 ) e. kiritsis , _ supersymmetry and duality in field theory and string theory _ , hep - ph/9911525 c.p . burgess and c.a . ltken , nuc . phys . * b500 * , 367 ( 1997 ) b.p . dolan , j. phys . * a32 * , ( 1999 ) l243 ( cond - mat/9805171 ) b.p . dolan , nuc . phys . * 460b [ fs ] * , ( 1999 ) 297 ( cond - mat/9809294 ) n. taniguchi , _ nonperturbative renormalisation group function for quantum hall plateau transitions imposed by global symmetries _ , cond - mat/9810334 c.p . burgess and c.a . ltken , phys . * b451 * , 365 ( 1999 ) ( hep - th/9812396 ) y. georgelin , t. masson and j - c . wallet , _ @xmath248 modular symmetry , renormalization group flow and the quantum hall effect _ , cond - mat/9906193 c.p . burgess , r. dib and b.p . dolan , _ derivation of the semi - circle law from the law of corresponding states _ , cond - mat/9911476 n. seiberg and e. witten , nuc * b426 * , 19 ( 1994 ) ( hep - th/9407087 ) n. koblitz , _ introduction to elliptic curves and modular forms _ , 2nd ed . , ( springer , 1993 ) . rankin , _ modular forms and functions _ , ( cambridge university press , 1977 ) . ltken , nuc . phys . * b396 * , 670 ( 1993 ) y. georgelin and j - c . wallet , phys . lett . * a224 * , 303 ( 1997 ) ; y. georgelin , t. masson and j - c . wallet , j. phys . * a30 * , 5065 ( 1997 ) a. shapere and f. wilczek , nuc . phys * b320 * , 669 ( 1989 ) m.m . fogler and b.i . shklovskii , phys . rev . * b52 * , 17366 ( 1995 ) d.e . khmelnitskii , pisma zh . . fiz . * 38 * , 454 ( 1983 ) ( jetp lett . * 38 * , 552 ( 1983 ) ) a.m.m . pruisken , phys . lett . * 61 * , 1297 ( 1988 ) a.m. dykhne and i.m . ruzin , phys . rev . * b50 * , 2369 ( 1994 ) ; i. ruzin and s. feng , phys . lett . * 74 * , 154 ( 1995 ) m. hilke et al . * 46 * , 775 ( 1999 ) ( cond - mat/9810217 ) d. shahar , d.c . tsui , m. shayegan , j.e . cunningham , e. shimshoni and s.l . sondhi , solid state comm . * 102 * ( 1997 ) 817 ( cond - mat/9607127 ) ; d. shahar , d.c . tsui , m. shayegan , e. shimshoni and s.l . sondhi , science * 274 * , ( 1996 ) 589 ( cond - mat/9510113 ) a. ritz , phys . lett . * b434 * 54 , ( 1998 ) ( hep - th/9710112 ) m.r . zirnbauer _ conformal filed theory of the integer quantum hall plateau transition _ , hep - th/9905054 e.t . whittaker and g.n . watson , _ a course of modern analysis _ , ( 1940 ) d. shahar , d. c. tsui , m. shayegan , e. shimshoni and s. l. sondhi , phys . 79 * , 479 ( 1997 ) ( cond - mat/9611011 ) m.m . cage in _ the quantum hall effect _ , eds . prange and s.m . girvin , springer , ( 1987 ) r.w . rendell and s.m . girvin , phys . rev . * b23 * , 6610 ( 1981 ) d. shahar , m. hilke , c.c . li , d.c . tsui , s.l . sondhi and m. razeghi , solid sate comm . * 107 * , 19 ( 1998 ) ( cond - mat/9706045 )

generalized parton distributions ( gpds ) @xcite parametrize the non - perturbative hadron structure in hard exclusive processes ( for a recent review , see , e.g. , @xcite ) . their measurement would represent a unique way to access several crucial features of the structure of the nucleon , such as the angular momentum sum rule of the proton @xcite , as well as information on the structure of the proton in position space @xcite . therefore , relevant experimental efforts to measure gpds , by means of exclusive electron deep inelastic scattering ( dis ) off the proton , are likely to take place in the next few years @xcite . one of the most promising processes is represented by deeply virtual compton scattering ( dvcs ) @xcite . at twist - two , four helicity - even gpds occur . two of them are usually called unpolarized gpds and labeled @xmath1 and @xmath2 . the first of them gives , in dis kinematics , the usual quark density . the latter turns out to be a spin - flip distribution , in the sense that it implies a change of the spin of the target . the other two are usually called polarized or helicity - dependent gpds . they are labeled @xmath0 and @xmath3 . the first of them yields in dis kinematics the polarized quark density , the second is again a spin - flip distribution . the four of them enter the cross sections for the dvcs process , although the two non - spin - flip ones , @xmath1 and @xmath0 , give the dominant contribution . reasonable estimates are necessary for the planning of experiments . for these quantities , several theoretical predictions have been already produced by using different descriptions of hadron structure : bag models @xcite , soliton models @xcite , light - front @xcite and bethe salpeter approaches @xcite , and phenomenological estimates based on parametrizations of parton distribution functions ( pdfs ) @xcite . besides , an impressive effort has been devoted to study the perturbative qcd evolution @xcite of gpds , and the gpds at twist three accuracy @xcite . chiral - odd gpds deserve a special mention : while they have been shown to be in principle experimentally accessible @xcite , no theoretical estimates are available for them . we are interested in constituent quark model ( cqm ) calculations . in this framework , in order to compare model predictions with data taken in dis experiments , one has to evolve , according to perturbative qcd , the leading twist component of the physical structure functions obtained at the low momentum scale associated with the model , from this so called `` hadronic scale '' , @xmath4 @xcite , to the momentum scale of the data . such a procedure has proven successful in describing the gross features of the standard pdfs by using different cqms ( see , e.g. , @xcite ) . similar expectations motivated the study of gpds in ref . @xcite , where a simple formalism has been proposed to calculate the quark contribution to the gpds from any non relativistic model . as an illustration , results for the gpd @xmath5 , obtained in the non relativistic ( nr ) model of isgur and karl ( ik ) @xcite have been evolved from @xmath6 up to dis scales , to next to leading order ( nlo ) accuracy . in ref . @xcite the same quark contribution has been evaluated , at @xmath6 , using the overlap representation of the gpds @xcite in light - front dynamics , along the lines developed in @xcite . the same approach has been also applied to calculate the valence quark contribution to the helicity - dependent gpds @xcite . in all the papers dealing with cqm listed so far @xcite , only the dokshitzer - gribov - lipatov - altarelli - parisi ( dglap ) region of gpds can be evaluated . to our knowledge , no cqm estimate of the helicity - dependent gpds has been performed in the efremov - radyushkin - brodsky - lepage ( erbl ) region . this is precisely the argument of the present paper , where an approach recently proposed for the unpolarized gpds in a cqm scenario @xcite , providing us with predictions also in the erbl region , will be extended to the helicity - dependent gpds . in ref . @xcite , the procedure of ref . @xcite has been extended and generalized since , as a matter of fact , the latter , when applied to the standard forward case , has been proven to reproduce only the gross features of pdfs @xcite . in order to achieve a better agreement with data , the approach has to be improved . in a series of papers , it has been shown that unpolarized @xcite and polarized @xcite dis data are consistent with a low energy scenario , dominated by complex constituent quarks inside the nucleon , defined through a scheme suggested by altarelli , cabibbo , maiani and petronzio ( acmp ) @xcite , updated with modern phenomenological information . the same idea has been applied to demonstrate the evidence of complex objects inside the nucleon @xcite , analyzing intermediate energy data of electron scattering off the proton . a similar scenario , as old as the cqm itself @xcite , has been extensively used by other groups , starting form the concept of `` valon '' @xcite . in ref . @xcite , the description of the unpolarized forward case of ref . @xcite has been generalized and applied to the calculation scheme of ref . @xcite . as a result , more realistic predictions for the gpds have been obtained and , at the same time , the erbl region , not accessible before , has been explored . in particular , the evaluation of the sea quark contribution has become possible , so that gpds can be calculated , in principle , in their full range of definition . such an achievement will permit to estimate cross - sections that are relevant for actual gpds measurements , providing us with an important tool for planning future experiments . in here , the discussion is extended to helicity - dependent gpds . the description of the polarized forward case of ref . @xcite is generalized to calculate helicity - dependent gpds , following the path that lead to the results of ref . @xcite starting from the calculation of the unpolarized forward structure function in ref . @xcite . as in ref . @xcite , the proposed approach will be applied here in a nr framework , which allows one to evaluate the gpds only for small values of the 4-momentum transfer , @xmath7 ( corresponding to @xmath8 , where @xmath9 is the constituent quark mass ) and small values also for the skewedness parameter , @xmath10 . the full kinematical range of definition of gpds will be studied in a followup , introducing relativity in the scheme . the paper is structured as follows . in the second section , after the definition of the main quantities of interest , a formula for the helicity - dependent current quark gpd @xmath11 , in a cqm with point - like constituents , will be derived . for the same quantity , in the third section , in a scenario dominated by composite constituent quarks , a _ formula in terms of the corresponding quantity of the constituent quark and of a polarized off - diagonal momentum distribution , will be derived in impulse approximation ( ia ) . then , the helicity - dependent constituent quark gpds are built in the fourth section , according to acmp philosophy and using the double distribution ( dd s ) representation @xcite of the gpds . in the fifth section , results obtained by using cqm wave functions of the ik model and our helicity - dependent constituent quark gpds will be shown . conclusions will be drawn in the last section . let us consider a hard exclusive process , where the absorption of a high - energy virtual photon by a quark in a hadron target is followed by the emission of a particle to be later detected ; finally , the interacting quark is reabsorbed back into the recoiling hadron . if the emitted and detected particle is , for example , a real photon , the so called deeply virtual compton scattering process takes place @xcite . we adopt here the formalism used in ref . let us think of a nucleon target , with initial ( final ) momentum and helicity @xmath12 and @xmath13 , respectively . the gpds @xmath14 and @xmath15 are defined through the expression @xmath16 + where @xmath17 is the 4-momentum transfer to the nucleon , @xmath18 is the quark field and m is the nucleon mass . it is convenient to work in a system of coordinates where the photon 4-momentum , @xmath19 , and @xmath20 are collinear along @xmath21 . the @xmath10 variable in the arguments of the gpds is the so called `` skewedness '' , parametrizing the asymmetry of the process . it is defined by the relation @xmath22 , where @xmath23 is a light - like 4-vector satisfying the condition @xmath24 . as explained in @xcite , gpds describe the amplitude for finding a quark with momentum fraction @xmath25 ( in the infinite momentum frame ) in a nucleon with momentum @xmath26 and replacing it back into the nucleon with a momentum transfer @xmath27 . besides , when the quark longitudinal momentum fraction @xmath28 of the average nucleon momentum @xmath29 is less than @xmath30 , gpds describe antiquarks ; when it is larger than @xmath10 , they describe quarks ; when it is between @xmath30 and @xmath10 , they describe @xmath31 pairs . the first and second @xmath28 intervals are commonly referred to as the dglap region and the third as the erbl region @xcite , following the pattern of evolution in the factorization scale . one should keep in mind that , besides the variables @xmath32 and @xmath7 explicitly shown , gpds depend , as the standard pdfs , on the momentum scale @xmath33 at which they are measured or calculated . to make the presentation easier looking , we omit the latter dependence unless when specifically needed . the values of @xmath10 which are possible for a given value of @xmath7 are : 0 / . [ xim ] the known constraints of @xmath34 are : \i ) the so called `` forward '' limit , @xmath35 , i.e. , @xmath36 , yields the usual polarized pdfs h_q(x,0,0)=q(x ) ; [ i ) ] \ii ) the integration over @xmath28 , yields the contribution of the quark of flavor @xmath37 to the axial form factor ( ff ) of the target @xmath38 , called hereafter @xmath39 : dx h_q(x,,^2 ) = g_a^q(^2 ) ; [ ii ) ] \iii ) the polynomiality property @xcite , according to which the @xmath28-integrals of @xmath40 and of @xmath41 are polynomials in @xmath10 of order @xmath42 . in @xcite , the ia expression for the unpolarized @xmath43 , suitable to perform cqm calculations , was obtained . in a similar fashion , we derive next a formula for the helicity - dependent gpd @xmath14 from the definition ( [ eq1 ] ) . using the light - cone spinor definitions as given in the appendix b of @xcite , and defining : @xmath44 , for the light - cone helicity combination @xmath45 one obtains @xmath46 so that , for @xmath47 : @xmath48 i.e. @xmath49 according to this equation , in order to obtain the gpd @xmath34 for @xmath47 one has to evaluate @xmath50 , starting from its definition , eq ( [ eq1 ] ) . using light - cone quantized quark fields in the l.h.s . , whose creation and annihilation operators , @xmath51 and @xmath52 , obey the commutation relation \ { b(k),b^(k ) } = ( 2 ) ^3 2 k^+ ( k^+ - k^+ ) ^2(k_- k_ ) , and using light - cone states normalized as p | p = ( 2 ) ^3 2 p^+ ( p^+ - p^+ ) ^2(p_- p_ ) , one obtains , for @xmath53 @xcite f^q_++ ( x,,^2 ) = 1 2 |p^+ v b^(k + ) b(k ) , [ fqpp ] where b^(k ) b(k ) = _ sign ( ) p+ | b_^((x-)|p^+ , k_ ) b_((x+)|p^+ , k_)| p+ , and @xmath54 is a volume factor . since we want to obtain only the quark contribution to @xmath11 , the only one which can be evaluated in a cqm with three valence , point - like quarks , we are interested here in the @xmath53 region . ( [ fqpp ] ) can be written ( see @xcite where all the steps of a similar derivation are explicitely given ) : f^q_++ ( x,,^2 ) = 1 2 |p^+ v ( k^+ |p^+ - ( x+ ) ) b^(k + ) b(k ) . [ fqpp1 ] in a nr framework , states and creation and annihilation operators have to be normalized according to p | p = ( 2 ) ^3 ( p^+ - p^+ ) ( p_- p _ ) and \ { b(k^+,k_),b^(k^+ , k _ ) } = ( 2 ) ^3 ( k^+ - k^+ ) ( k_- k _ ) respectively . as a consequence , one has to consider that @xcite : | p |p , b(k ) b(k^+,k _ ) , so that in eq . ( [ fqpp1 ] ) , in terms of the new states and fields , one has to perform the substitution b^(k + ) b(k ) & = & _ sign ( ) p+ | b_^((x-)|p^+ , k_+ _ ) b_((x+)|p^+ , k_)| p + + & & 2 |p^+ + & & _ sign ( ) p | b_^(k^+ + ^+ , k_+ _ ) b_(k^+ , k_)| p , [ newnorm ] where the relation @xmath55 has been used . now , inserting eq . ( [ newnorm ] ) in eq . ( [ fqpp1 ] ) , one gets f_++^q(x,,^2 ) & = & 1 v ( k^+ |p^+ - ( x+ ) ) 2 + & & _ sign ( ) p | b_^(k^+ + ^+ , k_+ _ ) b_(k^+ , k_)| p = + & = & 1 v ( k^+ |p^+ - ( x+ ) ) + & & _ sign ( ) p | b_^(k^+ + ^+ , k_+ _ ) b_(k^+ , k_)| p + o(^2 ) . [ intm ] the constituent quarks with mass @xmath9 are taken to be on shell , @xmath56 , thus d^2 k_d k^+ ( k^+ k_0 ) d k , and from eq . ( [ intm ] ) and eq . ( [ hf++ ] ) , one gets h_q(x,,^2 ) & = & ( k^+ |p^+ - ( x+ ) ) + & & + o(^2 ) = [ int_1 ] + & = & d k ( k^+ |p^+ - ( x+ ) ) + o(^2 ) . [ inter ] in the last step , the definition of a ( non - diagonal ) polarized nr momentum distribution , @xmath57 , has been used , together with the fact that a nr momentum distribution describes the probability of finding a constituent of momentum @xmath58 in a given system up to terms of order @xmath59 @xcite . so far , we have shown that , in a nr cqm , the gpd @xmath14 can be calculated , for @xmath47 , @xmath60 and @xmath61 ( which means , in turn , @xmath62 ) , through the following expression : @xmath63 the above equation , corresponding to eq . ( 8) in @xcite if polarization is taken into account , permits the calculation of @xmath14 in any cqm , and it naturally verifies some of the properties of the gpds . in fact , the polarized quark density , @xmath64 , as obtained by analyzing , in ia , the dis off the nucleon ( see , e.g. , @xcite ) , assuming that the interacting quark is on - shell , is recovered in the forward limit , where @xmath65 : @xmath66 so that the constraint eq . ( [ i ) ] ) is fulfilled . in the above equation , @xmath67 is the nr polarized momentum distribution of the quarks in the nucleon and , integrating eq . ( [ usu ] ) over @xmath28 , one obtains the contribution of the quark of flavor @xmath37 to the ia definition of the axial ff , so that eq . ( [ ii ) ] ) is verified . the polynomiality condition is formally fulfilled by the gpd defined in eq . ( [ usu ] ) , although the present accuracy of the model , explicitly written in the latter equation , does not allow to really check polynomiality , due to the effects of the munich symmetry @xcite , which excludes @xmath68 contributions . with respect to eq . ( [ usu ] ) , a few caveats , already listed in ref . @xcite for the unpolarized case , are necessary . \i ) one should keep in mind that eq . ( [ usu ] ) is a nr result , holding for @xmath47 , under the conditions @xmath60 , @xmath62 . if one wants to treat more general processes , the nr limit should be relaxed by taking into account relativistic corrections . in this way , at the same time , an expression to evaluate @xmath15 could be obtained . \ii ) the constituent quarks are point - like objects . \iii ) if use is made of a cqm , containing only constituent quarks ( and also antiquarks in the case of mesons ) , only the quark ( and antiquark ) contribution to the gpds can be evaluated , i.e. , only the region @xmath69 ( and also @xmath70 for mesons ) can be explored . in order to introduce the study of the erbl region ( @xmath71 ) , so that observables like cross - sections , spin asymmetries and so on can be calculated , the model has to be improved . \iv ) in actual calculations , the evaluation of eq . ( [ usu ] ) requires the choice of a reference frame . in the following , the breit frame will be chosen , where one has @xmath72 and , in the nr limit we are studying , one finds @xmath73 . it happens therefore that , in the argument of the @xmath74 function in eq . ( [ usu ] ) , the @xmath28 variable for the valence quarks is not defined in its natural support , i.e. it can be larger than 1 and smaller than @xmath10 . although the support violation is small for the calculations that will be shown here , it has to be reported as a drawback of the approach . the issue iii ) will be discussed in the next sections , by relaxing the condition ii ) and allowing for a finite size and composite structure of the constituent quark . in the standard forward case , the procedure described in the previous section , has been proven to be able to reproduce the gross features of pdfs @xcite . in order to achieve a better agreement with data , the approach has to be enriched . in previous papers , it has been shown that unpolarized @xcite and polarized @xcite dis data are consistent with a low energy scenario dominated by complex systems of point - like partons . the forward parton distributions are therefore given by a convolution between constituent quark momentum distributions and constituent quark structure functions . the latter quantities were obtained by using updated phenomenological information in a scenario firstly suggested by altarelli , cabibbo , maiani and petronzio ( acmp ) already in the seventies @xcite . following the same idea , in ref . @xcite a model for the reaction mechanism of an off - forward process , such as dvcs , where gpds could be measured , has been proposed . as a result , a convolution formula giving the proton @xmath5 gpd in terms of a constituent quark off - forward momentum distribution , @xmath75 , and of a gpd of the constituent quark @xmath76 itself , @xmath77 , has been derived . in here , a similar program will be developed for the helicity - dependent gpd @xmath11 . it is assumed that the hard scattering with the virtual photon takes place on a parton of a spin @xmath78 target , such as the proton , made of complex constituents . the scenario is depicted in fig . 1 for the special case of dvcs , in the handbag approximation . one parton ( _ current _ quark ) with momentum @xmath79 , belonging to a given constituent of momentum p , interacts with the probe and it is afterwards reabsorbed , with momentum @xmath80 , by the same constituent , without further re - scattering with the recoiling system of momentum @xmath81 . we suggest here an analysis of the process which is quite similar to the usual ia approach to dis off nuclei @xcite , recently applied also to nuclear gpds @xcite . in the class of frames chosen in section 2 , and in addition to the kinematical variables , @xmath28 and @xmath10 one needs a few more to describe the process . in particular , @xmath82 and @xmath83 , for the `` internal '' target , i.e. , the constituent quark , have to be introduced . these quantities can be obtained defining the `` + '' components of the momentum @xmath79 and @xmath84 of the struck parton before and after the interaction , with respect to @xmath85 and @xmath86 : @xmath87 from the above expressions , @xmath83 and @xmath82 are immediately obtained as @xmath88 and , since @xmath89 , if @xmath90 , one also has @xmath91 in order to derive a convolution formula in ia for @xmath11 , the procedure used for standard dis will be adopted @xcite . first , in eq . ( [ int_1 ] ) , two complete sets of states , corresponding to the interacting particle and to the recoiling system , are properly inserted to the left and right - hand sides of the quark operator : @xmath92 second , since eventually use has to be made of nr wave functions , a state @xmath93 has to be normalized in a nr manner : p | p = ( 2 ) ^3 ( p - p ) , [ norr ] so that in eq . ( [ iaint ] ) one has to perform the substitution : | p > | p > . since , using ia in the intrinsic frame of the target one has , for the nr states : @xmath94 a convolution formula is readily obtained from eq . ( [ iaint ] ) : @xmath95 \nonumber \\ & \times & { \xi ' \over \xi } \tilde h_{q_0 q}(x',\xi',\delta^2 ) + { \cal{o } } \left ( \xi^2 \right ) ~. \label{spec}\end{aligned}\ ] ] in the above equations , the kinetic energies of the residual system and of the recoiling target have been neglected , and @xmath96 is a spin dependent , off - diagonal spectral function for the constituent quark in the target : @xmath97 besides , the quantity h_q_0q(x,,^2 ) & = & ( k^+ |p^+ - ( x+ ) ) + & & 1 ( 2 ) ^3 v k^+ k_0 _ sign ( ) p | b_q,^(k^+ + ^+ , k_+ _ ) b_q,(k^+ , k _ ) | p = + & = & ( k^+ |p^+ - ( x + ) ) + & & 1 ( 2 ) ^3 v k^+ k_0 _ sign ( ) p | b_q,^(k^+ + ^+ , k_+ _ ) b_q,(k^+ , k _ ) | p [ gpdb ] is , according to eq . ( [ int_1 ] ) , the helicity - dependent gpd of the constituent quark @xmath76 up to terms of order @xmath98 , and in the above equation use has been made of eqs . ( [ x1 ] ) and ( [ xi1 ] ) . the delta function in eq ( [ spectral ] ) defines @xmath2 , the removal energy , in terms of @xmath99 , the excitation energy of the recoiling system . concerning eqs ( [ spec][gpdb ] ) , two comments are in order . the first concerns the accuracy of the actual calculations which will be presented . in the following , nr wave functions will be used to evaluate eq . ( [ spec ] ) , so that the accuracy of the calculation is of order @xmath100 . in fact , if use is made of nr wave functions to calculate the non - diagonal spectral function , the result holds for @xmath101 . the same constraint can be written @xmath102 . second , from the study of forward dis off nuclei , it is known that , in going from a covariant formalism to one where use can be made of the usual nr wave functions , one has to keep the correct normalization @xcite . this procedure leads to the appearance of the `` flux factor '' , represented in eq . ( [ flux ] ) by the expression @xmath103 ( which gives , in forward dis , the usual @xmath104 term ) . this factor gives one at order @xmath100 , which is the accuracy of the present analysis , and it will not be included in the calculation described below . if one wants to be predictive in more general processes at higher momentum transfer , the accuracy of eq . ( [ spec ] ) is not good enough any more and the calculation , in order to be consistent , has to be performed taking into account relativistic corrections . ( [ spec ] ) can be written in the form @xmath105 taking into account that @xmath106 = z + \xi - { p^+ \over p^+ } ( 1 + \xi ) = z + \xi - { p^+ \over \bar p^+}~,\ ] ] eq . ( [ pnk ] ) can also be written in the form : @xmath107 where @xmath108 one should notice that eqs . ( [ main ] ) and ( [ hq0 ] ) or , which is the same , eq . ( [ spec ] ) , fulfill the constraint @xmath109 previously listed . the constraint @xmath110 , i.e. the forward limit of gpds , is verified . in fact , by taking the forward limit ( @xmath111 ) of eq . ( [ main ] ) , one gets the expression which is usually found , for the polarized parton distribution @xmath64 , in the ia analysis of polarized dis off polarized three - body systems made of composite constituents ( see , i.e. , @xcite for the @xmath112he target ) @xmath113 in the latter equation , @xmath114 is the light - cone momentum distribution of the constituent @xmath76 in the nucleon , @xmath115 is the distribution of the quark of flavour @xmath37 in the constituent @xmath76 and @xmath116 , the @xmath117 limit of eq . ( [ pnk ] ) , is the one body spin dependent spectral function , labeled @xmath118 in ref . if the system were a three - body nucleus , the constituents would be nucleons and eqs . ( [ main])-([hq0f ] ) would coincide with the corresponding ones obtained in ref . @xcite for the polarized @xmath112he target . the constraint @xmath119 , i.e. the @xmath120integral of the gpd @xmath11 , is also naturally fulfilled . in fact , by @xmath120integrating eq . ( [ main ] ) , one easily obtains : @xmath121 which can also be cast in the form of eq . ( [ main ] ) , with the light - cone off - diagonal momentum distribution given by @xmath122 which coincides with eq . ( [ usu ] ) and it is to be evaluated in a given cqm , for @xmath123 or @xmath124 , while @xmath125 is the constituent quark helicity - dependent gpd , which is still to be discussed and will be modelled in the next section . the main issue remaining is the definition of @xmath126 , the helicity - dependent constituent quark gpd , appearing in eq . ( [ main ] ) . as it has been done in ref . @xcite for the unpolarized case , this quantity will be modelled to respect that its forward limit , recovers the constituent quark parton distributions . we adopt the picture proposed in @xcite that the constituent quark is a complex system of point - like partons described by the scheme suggested by altarelli , cabibbo , maiani and petronzio ( acmp ) @xcite . in this scheme the constituent quarks are composite objects whose structure functions are described by a set of functions @xmath127 that specify the number of point - like partons of type @xmath37 which are present in the constituent of type @xmath76 , with fraction @xmath28 of the total momentum . we call these functions , generically , the structure functions of the constituent quark . the functions describing the nucleon parton distributions are expressed in terms of the independent @xmath127 and of the constituent density distributions ( @xmath128 ) as , q(x , q^2)=_q_0_x^1 dzz q_0(z , q^2 ) _ q_0q ( x z , q^2 ) , [ fx ] where @xmath37 labels the various partons , i.e. , valence quarks ( @xmath129 ) , sea quarks ( @xmath130 ) , sea antiquarks ( @xmath131 ) and gluons @xmath132 . the different structure functions of the constituent quarks are derived from three very natural assumptions @xcite : i ) the point - like partons are @xmath133 degrees of freedom , i.e. quarks , antiquarks and gluons ; ii ) regge behavior for @xmath134 and duality ideas ; iii ) invariance under charge conjugation and isospin . these considerations define in the case of the valence quarks the following structure function _ qq_v(x , q^2 ) = ( a + 1 2 ) ( 1 2 ) ( a ) ( 1-x)^a-1 . [ csf1]for the sea quarks the corresponding structure function becomes , _ qq_s(x , q^2 ) = c x ( 1-x)^d-1,[csf2 ] and , in the case of the gluons , it is taken _ qg(x , q^2 ) = g x ( 1-x)^b-1 .[csf3 ] the last input of the approach consists of the definition of the scale at which the constituent quark structure is defined . we choose the so called hadronic scale @xmath6 @xcite . this last hypothesis fixes @xmath135 the parameters of the approach ( eqs . ( [ csf1 ] ) through ( [ csf3 ] ) ) . the constants @xmath38 , @xmath136 , @xmath137 and the ratio @xmath138 are determined by the amount of momentum carried by the different partons , corresponding to a hadronic scale of @xmath139 gev@xmath140 , according to the parameterization of @xcite . @xmath141 ( or @xmath142 ) is fixed according to the value of @xmath143 at @xmath144 @xcite , and its value is chosen again according to @xcite . we stress that all these inputs are forced only by the updated phenomenology , through the 2@xmath145 moments of pdfs . the values of the parameters obtained are listed in @xcite . the other ingredients appearing in eq . ( [ fx ] ) , i.e. , the density distributions for each constituent quark , represent the unpolarized equivalent of eq . ( [ usu ] ) . in ref . @xcite , the spin dependent structure function of the constituent quark have been built generalizing the analysis of ref . @xcite , leading to eqs . ( [ csf1 ] ) ( [ csf3 ] ) . in here , we review the procedure addressed in ref . let q ( x,_0 ^ 2 ) = q_+ ( x,_0 ^ 2 ) - q_- ( x,_0 ^ 2 ) where @xmath146 label the quark spin projections and @xmath37 represents any flavor . the @xmath147 approach , generalized to the polarized case , implies q_i(x ) = _ q_0 _ x^1 _ j q_0j(z ) _ q_0jq_i ( ) where @xmath148 labels the partonic spin projections , @xmath149 the constituent quark spins and @xmath128 . using spin symmetry we arrive at q ( x ) = _ x^1 where @xmath150 , and _ q_0q = _ q_0+q+ - _ q_0+q- note at this point that the unpolarized case can be described in this generalized formalism as q ( x ) = q_+(x ) + q_-(x ) = _ q_0 _ x^1 q_0 ( z ) _ q_0q ( ) , where _ q_0q = _ q_0+q+ + _ q_0+q- , the description , reformulated in terms of the conventional valence and sea quark separation , yields @xmath151 _ q_0 _ x^1 q_0 ( z ) [ dfx ] since @xmath152 , i.e. any constituent has a leading parton with the same quantum numbers , one has q_v ( x ) = _ x^1 q_0 ( z ) _ q_0q_v ( ) , q_s ( x ) = _ q_0 _ x^1 q_0 ( z ) _ q_0q_s ( ) [ sea ] introducing @xmath153 ( spin - isospin ) symmetry as a simplifying assumption , gives _ u_0u = _ d_0d and _ u_0d = _ d_0u . furthermore they imply _ u_0 u_v + _ u_0 u_s = _ d_0 d_v + _ d_0 d_s and _ u_0 d_s = _ d_0 u_s . by adding to these the following relations _ u_0 u_s = _ d_0 u_s , [ simp1 ] _ d_0 d_s = _ u_0 d_s , [ simp2 ] which are beyond @xmath153 symmetry , but quite reasonable , one obtains , _ u_0 u_s = _ d_0 u_s = _ u_0 d_s = _ d_0 d_s = _ q_0 q_s [ phisea ] and _ u_0 u_v = _ d_0 d_v = _ q_0 q_v . in this way the structure functions for the valence and for the sea are reduced to just two independent constituent structure functions and eq . ( [ sea ] ) simplifies to q_s ( x ) = _ x^1 ( u_0 ( z ) + d_0 ( z ) ) _ q_0q_s ( ) . [ deltasea ] the same argumentation applied to gluons implies g ( x ) = _ x^1 ( u_0 ( z ) + d_0 ( z ) ) _ q_0 g ( ) thus the @xmath147 procedure can be extended to the polarized case by introducing three additional structure functions for the constituent quarks : @xmath154 , @xmath155 and @xmath156 . in order to determine the polarized constituent structure functions we add some assumptions which will determine them and those of the unpolarized case using a small number of parameters . they are : iv ) factorization assumption : @xmath157 can not depend upon the quark model used , i.e , can not depend upon the particular @xmath158 ; v ) positivity assumption : the positivity constraint @xmath159 is saturated for @xmath160 . these additional assumptions determine completely the parameters of the polarized constituent structure functions . in fact , the @xmath133 partonic picture , regge behavior and duality imply that _ q_0 f = c_f x^a_f ( 1-x)^a_f - 1 and @xmath161 , for all @xmath162 , as defined by dominant exchange of the @xmath163 meson trajectory @xcite . the positivity restriction , @xmath164 , is a natural consequence of the probability interpretation of the parton distributions . the assumption that the inequality is saturated for @xmath165 , in the spirit of ref . @xcite , implies that @xmath166 , the latter being the parameter fixed in the analysis of the unpolarized case , and therefore when @xmath167 the partons which carry all of the momentum also carry all of the polarization , i.e. , @xmath168 . as is discussed in ref . @xcite , this scheme has been changed only for the sea structure function , in order to obtain a reasonable description of the polarized sea . this led to the definition of a parameter , @xmath169 , different from the one , @xmath170 defined in the unpolarized case . from the point of view of the number of parameters , this is a minimal assumption , since it reduces the parameters of the polarized case to those of the unpolarized one , with the only exception of @xmath170 . lastly the exponents @xmath171 are also taken from the unpolarized case . to summarize the parameterization , let us stress that the main change between the polarized functions and the unpolarized ones comes only from regge behavior and the only parameter which has been changed is the one related to the polarized sea distribution . with all the above hypothesis the constituent quark functions become _ qq_v(x , _ 0 ^ 2 ) = ( a + 1 2 ) ( 1 2 ) ( a ) x^a_f ( 1-x)^a-1 [ pi ] _ 0 ^ 2 ) = c_s x^a_f ( 1-x)^d-1 _ qg(x , _ 0 ^ 2 ) = g x^a_f ( 1-x)^b-1 [ pf ] where @xmath172 are the parameters of the unpolarized case , while @xmath173 is taken to reproduce the polarized data . in what follows , we shall use @xmath174 , the range proposed in ref . @xcite in agreement with ref . @xcite . we have to generalize this scenario to describe off - forward phenomena . of course , the forward limit of our gpds formula , eq . ( [ main ] ) , has to be given by eq . ( [ dfx ] ) . by taking the forward limit of eq . ( [ main ] ) , one obtains : @xmath175 so that , in order for the latter to coincide with eq . ( [ dfx ] ) , one must have @xmath176 . in such a way , through the acmp prescription , the forward limit of the unknown constituent quark gpd @xmath125 can be fixed . the off - forward behavior of the constituent quark gpds has to be modelled and it will be done , as in ref . @xcite for the unpolarized case , using the `` @xmath177-double distributions '' ( dd s ) language proposed by radyushkin @xcite . the dd s , @xmath178 , are a representation of gpds which automatically guarantees the polynomiality property . gpds can be obtained from dd s after a proper integration , and , as is explained in @xcite , parton densities are recovered in the forward limit , while meson distribution amplitudes are obtained in the @xmath179 limit of the dd s . in some cases , such a transparent physical interpretation , together with the symmetry properties which are typical of distribution amplitudes ( @xmath180 symmetry ) , allows a direct modelling , already developed in @xcite . the relation between any gpd @xmath0 , defined _ la ji _ , for example the one we need , i.e. @xmath181 for the constituent quark target , is related to the @xmath177-dd s , which we call @xmath182 for the constituent quark , in the following way @xcite : @xmath183 with some care , the expression above can be integrated over @xmath184 and the result is explicitly given in @xcite , where a factorized ansatz is suggested for the dd s which , for the helicity - dependent one , could be written : @xmath185 with the @xmath177 dependent term , @xmath186 , having the character of a mesonic amplitude . besides , in eq . ( [ ans ] ) @xmath187 represents the forward density and , eventually , @xmath188 the constituent quark axial ff . one immediately realizes that the helicity - dependent gpd of the constituent quark , eq . ( [ hdd ] ) , with the factorized form eq . ( [ ans ] ) , fulfills the crucial constraints of gpds , i.e. , the forward limit , the first - moment and the polynomiality condition , the latter being automatically verified in the dd s description . in the following , it will be assumed for the constituent quark gpd the above factorized form , so that one needs to model the three functions appearing in eq . ( [ ans ] ) , according to the present description . for the amplitude @xmath189 , use will be made of one of the simple normalized forms suggested in @xcite , on the bases of the symmetry properties of dd s : @xmath190 in addition , since we will identify quarks for @xmath191 , pairs for @xmath192 , antiquarks for @xmath193 , and , since in our approach the forward densities @xmath194 have to be given by the standard @xmath157 functions of the @xmath147 approach , eqs . ( [ pi])([pf ] ) , one has , for the dd of flavor @xmath37 of the constituent quark : @xmath195 the above definition , when integrated over @xmath177 gives the correct limits @xcite : @xmath196 and @xmath197 eventually , as axial ff we will take a simple monopole form corresponding to a constituent quark size @xmath198 : @xmath199 showing i.e. the same @xmath33 dependence of the electromagnetic ff used in @xcite , the latter scenario being strongly supported by the analysis of @xcite . by using such a ff and eq . ( [ h ] ) , together with the standard acmp @xmath157 s , eqs . ( [ pi ] ) ( [ pf ] ) , in eq . ( [ dd ] ) , and inserting the obtained @xmath200 into eq . ( [ hdd ] ) , the constituent quark gpd in the acmp scenario can be eventually calculated . let us discuss our factorized choice for the dd of the constituent quark , eq . ( [ ans ] ) . such a factorization does not hold for the gpd of the proton @xcite . no microscopic calculation supports it @xcite . the present approach is consistent with the absence of factorization of the _ proton _ gpd in @xmath201dependent and @xmath201independent terms since an additional , crucial , @xmath7 dependence comes from the quantity @xmath75 in eq . ( [ main ] ) , i.e. , from the proton wave function of the quark model used . the factorized form is assumed for the gpd of the constituent quark for simplicity . we do not know what the helicity - dependent gpd of the constituent quark is and we start modelling it by using the simplest possible structure which fulfills the required constraints . clearly this choice deserves a deeper explanation , which we do not have at the moment . however , one expects that the structure of the constituent quark gpd is closer to the asymptotic behavior , than that of the proton , because it probes higher energies . the non perturbative contributions arise in the precise shape of the three functions building the gpd , which we take from general qcd based arguments and from data . it is true that even the forward structure of the constituent quark is not known , but we use a parameterization which performs well in describing the data @xcite . in this sense , the forward limit is well reproduced , as shown in the next section . on the contrary , the profile eq . ( [ h ] ) is _ assumed _ to be valid also for the constituent quark , even in the helicity dependent case , simply because it satisfies some of the required properties . these arguments and a small set of parameters , fixed once forever , make our approach able , if a proper cqm is used , to describe dis data for parton distributions , elastic form factors , and to predict unpolarized and polarized gpds . it allows for an effective description of the known structure of the nucleon and its predictive power needs to be tested with new observables . in this section we present the results obtained for the helicity - dependent gpd @xmath11 of the proton , for @xmath47 and @xmath62 , according to the approach described so far . the main equation to be evaluated is eq . ( [ main ] ) , written again here below for the sake of clarity : @xmath202 in the above equation , the quantity @xmath181 , the helicity - dependent constituent quark gpd , is modelled according to the arguments described in the previous section . this means that it is obtained evaluating eq . ( [ hdd ] ) , where the dd of the constituent quark , @xmath182 , is given by eq . ( [ dd ] ) , calculated in turn through the ff , eq . ( [ ffacmp ] ) , the function @xmath189 , eq . ( [ h ] ) , together with the forward acmp @xmath157 s , eqs . ( [ pi ] ) ( [ pf ] ) . the other ingredient in eq . ( [ main ] ) , @xmath203 , has been evaluated according to eq . ( [ usu ] ) : @xmath204 the calculation has been performed in the breit frame , where one has , in the nr limit studied , @xmath205 . the off - diagonal momentum distribution appearing in the formula above , @xmath206 , defined in eq . ( [ clo ] ) , has been evaluated within the isgur and karl ( ik ) model @xcite , according to the definition n_q_0(p , p + ) = n_q_0+(p , p + ) - n_q_0-(p , p + ) , where @xmath207 is the off - diagonal momentum distribution for the valence quark momentum @xmath208 and spin projection parallel or antiparallel to that of the nucleon . @xmath209 can be evaluated projecting out the appropriate spin and flavor component of the constituent quark and in the corresponding baryonic state is given by n_u_0/d_0 , ( p , p + ) & = & 3 ( _ sf < n , j_z = + | |n , j_z = + > _ sf ) + & & d p_1 d p_2 ^*(p_1,p_2,p + ) ( p_1,p_2,p ) , where @xmath210 is the spin - flavor state and @xmath211 is the proton wave function in momentum space ( cf . @xcite for the definition in the unpolarized case ) . in the ik cqm @xcite , including contributions up to the @xmath212 shell , the proton state is given by the following admixture of states @xmath213 where the spectroscopic notation @xmath214 , with @xmath215 being the symmetry type , has been used . the coefficients were determined by spectroscopic properties to be @xcite : @xmath216 , @xmath217 , @xmath218 , @xmath219 . + the results for the gpd @xmath220 , neglecting in ( [ ikwf ] ) the small @xmath142-wave contribution , are found to be : @xmath221~ , \label{iku}\end{aligned}\ ] ] @xmath222~ , \label{ikd}\end{aligned}\ ] ] for the flavors @xmath223 and @xmath224 , respectively , with @xmath225 @xmath226 @xmath227 @xmath228 \nonumber \\ & + & { 4 \over 9 } a_m^2 \left [ { 5 \over 8 } - { 3 \over 4 } { k^2 \over \alpha^2 } + { 3 \over 8 } { k^4 \over \alpha^4 } { 1 \over { 2 \alpha^2 } } \left ( -{\delta^2 \over 3 } + \vec \delta \cdot \vec k \right ) \left ( { k^2 \over \alpha^2 } -1 \right ) \right ] \\ & - & { 4 \over 9 } \sqrt{3 } a_s a_{s ' } \left [ \left ( - 1 + { k^2 \over \alpha^2 } \right ) - { 2 \over { 3 \alpha^2 } } \left ( { \delta^2 \over 3 } - \vec \delta \cdot \vec k \right ) \right ] ~ , \nonumber\end{aligned}\ ] ] @xmath229~,\end{aligned}\ ] ] @xmath230 and @xmath231 , being @xmath232 the constituent quark mass . here we have used the notation @xmath233 . evaluating the forward limit , @xmath234 , of eqs . ( [ iku ] ) and ( [ ikd ] ) , the expressions of the polarized quark densities in the ik model , obtained in @xcite , are consistently recovered . the harmonic oscillator parameter , @xmath177 , of the ik model , has ben chosen so that the experimental r.m.s . of the proton is reproduced by the slope , at @xmath235 , of the charge ff , so that a value of @xmath236 is obtained @xcite . such a ff reproduces well the data at the low values of @xmath7 which are accessible in the present approach . for higher values of @xmath7 , it would not be realistic @xcite . the same quality of results is expected for the proton axial form factor , which is obtained by integrating eq . ( [ main ] ) over @xmath28 , according to eq . ( [ ffc ] ) . the experimental dipole trend @xcite is qualitatively recovered at the low values of @xmath7 under scrutiny here . results for the @xmath223-quark @xmath237 distribution , at the scale of the model @xmath238 , are shown in figs 2 to 4 . in fig . 2 , it is shown for @xmath239 gev@xmath140 and @xmath240 . one should remember that the present approach does not allow to estimate realistically the region @xmath241 gev@xmath140 , so that we are showing here the result corresponding to the highest possible @xmath7 value . accordingly , the maximum value of the skewedness is therefore @xmath242 ( cf . ( [ xim ] ) ) , fulfilling the requirement @xmath47 . the dashed curve represents what is obtained in the pure ik model , i.e. , by evaluating eq . ( [ iku ] ) . one should notice that such a result , obtained in a pure valence cqm , should vanish for @xmath243 and for @xmath244 . the small tails which are found in these forbidden regions represent the amount of support violation of the approach . in particular , for the shown values of @xmath7 and @xmath10 , a violation of 2 @xmath245 is found . in general , in the accessible region the violation is never larger than few percents . the full curves in fig . 3 represents the complete result of the present approach , i.e. , the evaluation of eq . ( [ main ] ) following the steps and using the ingredients described in this section and in the previous one . the two curves are obtained in correspondence of the two extreme values of the regge intercept @xcite , @xmath246 ( upper curve ) and @xmath247 ( lower curve ) , used to define the helicity dependent constituent quark structure functions @xcite . a relevant contribution is found to lie in the erbl region , in agreement with other estimates @xcite . as already explained , the knowledge of gpds in the erbl region is a crucial prerequisite for the calculation of all the cross - sections and the observables measured in the processes where gpds contribute . we notice that the erbl region is accessed here , with respect to the approach of ref . @xcite generalized to the helicity - dependent case , which gives the dashed curve , thanks to the constituent structure which has been introduced implementing the acmp procedure . a similar result has been found in @xcite in the unpolarized case . in fig . 3 , special emphasis is devoted to show the @xmath10-dependence of the results . for the allowed @xmath10 values , @xmath248 , evaluated using our main formula , eq . ( [ main ] ) , is shown for four different values of @xmath28 . it is clearly seen that such a dependence is stronger in the erbl region , than in the dglap region , in good agreement with other estimates @xcite and with our previous analysis , performed in the unpolarized case @xcite . to allow for a complete view of the outcome of our approach , in fig . 4 the @xmath28 and @xmath10 dependences are shown together in a 3-dimensional plot . the results shown so far are associated with the low scale of the model , the hadronic scale @xmath6 , fixed to the value 0.34 gev@xmath140 , as discussed in section 4 . in order to have a feeling of the quality of the present predictions , since no data are available at the moment for @xmath249 , we evaluate in the following the forward limit of the obtained quantities . in figs . 5 and 6 , the proton and neutron spin dependent structure function @xmath250 , built from the forward limit of eq . ( [ main ] ) evaluated for @xmath251 , are compared with experimental data . in ref . @xcite , the succesful analysis had been performed using the model of ref . @xcite , so that the encouraging results shown there have to be confirmed when the ik model is considered , as it is being done here . to evaluate @xmath250 at the experimental scale , a nlo evolution of the model parton distributions has been performed . it is known that perturbative @xmath133 to this order allows the proposal of varied factorization schemes @xcite , whereby one is able to define the partonic distributions in different ways without altering the physical observables . in our analysis the parton distributions are determined by the quark model through the acmp procedure . it is evident that different factorization schemes lead to different results . following ref . @xcite , we have adopted on physical grounds the @xmath252 scheme as defined in ref.@xcite . scheme , in order to have the axial anomaly governing the first moment of @xmath253 , as proposed in ref @xcite . ] in it some relevant physical observables , such as @xmath254 , the total spin carried by the quarks , are scale independent , i.e. , they behave as conserved quantities , and therefore the partons have a well defined interpretation in terms of constituents . fig . 5 and 6 show that the present approach describes well the data @xcite for the proton structure function @xmath253 , and the trend of the neutron data @xcite is also reasonably reproduced . the quality of the agreement is comparable with that obtained in ref . @xcite using the model of @xcite . the present approach in the forward limit , i.e. the description given in ref . @xcite , also allows for the predictions of axial charges of the proton . by using the ik model , the following values are obtained ( the first listed value corresponds to the regge slope @xmath255 , the second to @xmath256 ) : for the octet charge , @xmath257 ( experimental value : @xmath258 @xcite ) , for the isospin charge , @xmath259 ( experimental value : @xmath260 @xcite ) , while the predictions for the singlet charge and for the spin carried by the gluons lead to @xmath261 ( from the analysis of ref . @xcite : @xmath262 ) . again , the agreement with data of these results , obtained by using the ik wave functions , is comparable to the one obtained in ref . @xcite by using the wave functions of ref . @xcite . this finding deserves a further comment . the lack of high momentum components in the ik wave function with respect to that of the model of ref . @xcite would indicate the latter as a better tool to describe the data . nevertheless , the low-@xmath28 behavior of the model results turns out to be mainly governed by the constituent quark structure rather than by the constituent quark wave function . the correct regge - behavior given by our parametrization of the constituent quark structure functions provides us with a model independent prediction . such a feature has not to do with the use of a relativistic model , which would improve only the high @xmath28 behavior . as discussed in ref @xcite , relativity and constituent quark structure are independent features , both of them necessary if model calculations are used for a proper description of data in the full @xmath28 range . since no data are presently available for helicity - dependent gpds , it is certainly useful to compare the results of model calculations with model independent constraints , if available . it has been proposed that they could be obtained from analysis of data in particular kinematical conditions . very recently , this issue has been addressed in two papers , where gpds have been obtained , at @xmath263 , by fitting the first moment of a reasonable ansatz of gpds to ff data @xcite . anyway , such a procedure is difficult to be realized for the quantity under scrutiny here , due to the lack of a satisfactory experimental knowledge of the axial ff , which can not provide us with a clear constraint on @xmath264 . as a matter of fact , in ref . @xcite no prediction has been produced for it , so that no comparison is possible between the present results and those of @xcite . on the contrary , some curves reproducing @xmath264 have been shown in @xcite . nevertheless , as the authors of that paper clearly state , the lack of precise data for the axial ff does not allow a realistic parameterization of @xmath264 , but rather a simple ansatz for it . this ansatz is given by the product of the polarized parton distributions , taken from the `` scenario 1 '' of ref . @xcite at @xmath265 gev@xmath140 , and an exponential term , giving the @xmath7 dependence and additional @xmath28 dependence , taking into account the different regge behavior at different values of the momentum transfer . the latter dependence is borrowed from the analysis performed for the helicity independent gpds , giving the so called default fit of @xcite , and no argument is given to support this procedure in the helicity dependent case . anyway , we show in fig . 7 the comparison of our results for the valence distribution @xmath266 , evaluated at @xmath267 gev@xmath140 , evolved to nlo from the hadronic scale up to @xmath268 gev@xmath140 , with the ansatz proposed in @xcite , evaluated in the same kinematical conditions . it is seen that the latter and the present model predict a similar scenario . the very low @xmath28 behavior is governed in both cases by the regge phenomenology , which is left here free to vary between the two full lines of the figure , while it is fixed in ref . @xcite in the way described above , producing a big enhancement at low values of @xmath269 . summarizing this section , we have developed here a scheme which provides us with the helicity - dependent gpd @xmath0 in the full @xmath28 range . this is obtained thanks to the constituent quark structure , implemented dressing the three quarks of a cqm , where initially only the dglap region of gpds was accessible . this is an important development , a prerequisite for any attempt to calculate cross sections and asymmetries of related processes . the next step of our studies will be indeed to use the obtained gpds for the evaluation of cross sections . a thorough analysis of both polarized and unpolarized data have shown , in previous work , that constituent quarks can not be considered elementary when studied with high - energy probes . certainly evolution is needed , i.e. the constituent quarks at the hadronic scale have to be undressed by incorporating bremsstrahlung in order to reach the bjorken regime . besides , the constituent quarks should be endowed with soft structure in order to approach the data . thus , the constituent quarks , when under scrutiny by high - energy probes , seem to behave as complex systems , with a very different behavior from the current quarks of the basic theory . these features , which we found in structure functions , have also been recently discussed in form factors @xcite . the formalism of composite constituent quarks has been recently applied to the study of generalized parton distributions , developing a formalism which expresses the hadronic gpds in terms of constituent quarks gpds by means of appropriate convolutions @xcite . the approach has been extended here to treat helicity - dependent gpds . to our knowledge , this represents the first calculation which gives , in a constituent quark framework , estimates of helicity - dependent gpds in the erbl region . following the spirit of ref . @xcite , this is done incorporating phenomenological features of various kinematical regimes . in particular , use has been made of the radyushkin s factorization ansatz , thus our constituent quark gpds are defined in terms of the product of three functions : i ) the constituent quark polarized structure function , where we use the parameterization of @xcite following the acmp proposal @xcite ; ii ) radyushkin s double distributions @xcite ; iii ) constituent quark form factor as suggested in ref . once the helicity - dependent gpds are defined in this way , we have developed the scheme to incorporate them into any nucleon model by appropriate convolution . in order to show the type of predictions to which our proposal leads , we have used here , as an illustration , the ik @xcite model . however , in this latest step of our scheme , any non - relativistic ( or relativized ) model can be used to define the hadronic gpds . as in the unpolarized case @xcite , the present scheme transforms a hadronic model , in whose original description only valence quarks appear , into one containing all kinds of partons ( i.e. , quarks , antiquarks and gluons ) . moreover , the starting model produces no structure in the erbl region , while after the structure of the constituent quark has been incorporated , it does . the completeness of the @xmath28-range , for the allowed @xmath7 and @xmath10 ( @xmath270 and @xmath47 ) , of the present description , is a prerequisite for the calculation of cross - sections and other observables in a wide kinematical range . relativistic corrections , which permit to access a wider kinematical region , could be included in the approach . we have reminded the reader of the calculation for the forward polarized structure functions to see how in this case , where experimental data are available , our scheme leads , even with a naive quark model , to a reasonable description of the data . thereafter , we have proceeded to calculate the gpds of physical interest to guide the preparation and analysis of future experiments . our calculation seems to be consistent with an ansatz recently suggested in ref . @xcite . this work is the continuation of an attempt to describe the properties of hadrons in different kinematical and dynamical scenarios . although our description can never be a substitute of quantum chromodynamics , it may serve to guide experimenters to physical processes where the theory might show interesting features , worthy of a more fundamental effort . useful comments by b. pire are gratefully acknowledged . thanks the department of theoretical physics of the valencia university , where part of this work has been done , for a warm hospitality and financial support . this work is supported in part by gv - grupos03/094 , mcyt - fis2004 - 05616-c02 - 01 and by the italian miur through the prin theoretical studies of the nucleus and the many body systems . 99 d. mller , d. robaschik , b. geyer , f.m . dittes , and j. hoeji , fortsch . 42 , 101 ( 1994 ) ; hep - ph/9812448 . a. radyushkin , phys . b 385 , 333 ( 1996 ) ; phys . rev . d 56 , 5524 ( 1997 ) . x. ji , phys . 78 , 610 ( 1997 ) ; phys . d 55 , 7114 ( 1997 ) . m. diehl , phys . 388 , 41 ( 2003 ) . jaffe and a.v . manohar , nucl . b 337 , 509 ( 1990 ) . m. burkardt , phys . d 62 , 071503 ( 2000 ) ; int a 18 , 173 ( 2003 ) ; m. diehl , eur . j. c 25 , 223 ( 2002 ) ; j.p . ralston and b. pire , phys . rev . d 66 , 111501 ( 2002 ) ; a.v . belitsky and d. mller , nucl . a711 , 118 ( 2002 ) ; x. ji , phys . 91 , 062001 ( 2003 ) . b.a . mecking , nucl . a711 , 330c ( 2002 ) ; k. rith , nucl . a711 , 336c ( 2002 ) ; d. von harrach , nucl . phys . a711 , 342c ( 2002 ) . guichon and m. vanderhaeghen , prog . 41 , 125 ( 1998 ) . x. ji , w. melnitchouk , and x. song , phys . d 56 , 5511 ( 1997 ) ; i.v . anikin , d. binosi , r. medrano , s. noguera , and v. vento , eur . j. a 14 , 95 ( 2002 ) . k. goeke , m.v . polyakov , and m. vanderhaeghen , prog . 47 , 401 ( 2001 ) . petrov , p.v . pobylitsa , m.v . polyakov , i. bornig , k. goeke , and c. weiss , phys . rev . d 57 , 4325 ( 1998 ) ; m. penttinen , m.v . polyakov , and k. goeke , phys . d 62 , 014024 ( 2000 ) . b.c . tiburzi and g.a . miller , phys . c 64 , 065204 ( 2001 ) ; phys . d 65 , 074009 ( 2002 ) . a. mukherjee and m. vanderhaeghen , phys d 67 , 085020 ( 2003 ) . l. theussl , s. noguera , and v. vento , eur.phys.j.a20 , 483 ( 2004 ) . a. freund and v. guzey , phys . b 462 , 178 ( 1999 ) ; l. frankfurt , a. freund , v. guzey , and m. strikman , phys . lett . b 418 , 345c ( 1998 ) . radyushkin , phys . b 449 , 81 ( 1999 ) ; i.v . musatov and a.v . radyushkin , phys . rev . d 61 , 074027 ( 2000 ) . belitsky , d. mller , l. niedermeier , and a. schfer phys . b 474 , 163 ( 2000 ) , and references therein ; a. freund and m.f . mcdermott , eur . j. c 23 , 651 ( 2002 ) , and references therein . a.v . belitsky and d. mller , nucl . b589 , 611 ( 2000 ) ; n. kivel , maxim v. polyakov , a. schfer , and o.v . teryaev , phys . b 497 , 73 ( 2001 ) ; a.v . belitsky , a. kirchner , d. mller , and a. schfer , phys . b 510 , 117 ( 2001 ) . d.yu . ivanov , b. pire , l. szymanowski , and o. teryaev , phys . b 550 , 65 ( 2002 ) . g. parisi and r. petronzio , phys . b 62 , 331 ( 1976 ) ; r.l . jaffe and g.g . ross , phys . b 93 , 313 ( 1980 ) . m. glck , e. reya , and a. vogt , eur . j. c 5 , 461 ( 1998 ) and references therein . m. traini , v. vento , a. mair , and a. zambarda , nucl . a614 , 472 ( 1997 ) . s. scopetta and v. vento , eur . j. a 16 , 527 ( 2003 ) . n. isgur and g. karl , phys . d 18 , 4187 ( 1978 ) ; phys . d 19 , 2653 ( 1979 ) . s. boffi , b. pasquini , and m. traini , nucl . b649 , 243 ( 2003 ) . m. diehl , t. feldmann , r. jakob , and p. kroll , nucl . b596 , 33 ( 2001 ) ; eur . j. c 8 , 409 ( 1999 ) . p. faccioli , m. traini , and v. vento , nucl . a656 , 400 ( 1999 ) . s. boffi , b. pasquini , and m. traini , nucl . b680 , 147 ( 2004 ) . s. scopetta and v. vento , phys . d 69 , 094004 ( 2004 ) . s. scopetta , v. vento , and m. traini , phys . b 421 , 64 ( 1998 ) . s. scopetta , v. vento , and m. traini , phys . b 442 , 28 ( 1998 ) . g. altarelli , n. cabibbo , l. maiani , and r. petronzio , nucl . b69 , 531 ( 1974 ) ; g. altarelli , s. petrarca , and f. rapuano , phys . b 373 , 200 ( 1996 ) . r. petronzio , s. simula , and g. ricco , phys . rev . d 67 , 094004 ( 2003 ) . g. morpurgo , physics 2 , 95 ( 1965 ) . hwa , phys . d 22 , 759 ( 1980 ) ; r.c . hwa and c.b . yang , phys . c 66 , 025204 ( 2002 ) and references therein . radyushkin , phys . rev . d 59 , 014030 ( 1999 ) . radyushkin , in m. shifman , ( editor ) : at the frontier of particle physics , vol . 2 , ( world scientific , singapore , 2001 ) pp . 1037 - 1099 , hep - ph/0101225 ; x. ji , j. phys . g 24 , 1181 ( 1998 ) . p. mulders , phys . rept . 185 , 83 ( 1990 ) . frankfurt and m.i . strikman , phys . rept . 160 , 235 ( 1988 ) . m. ropele , m. traini and v. vento , nucl . phys . a 584 , 634 ( 1995 ) . l. mankiewicz , g. piller , and t. weigl , eur . j. c 5 , 119 ( 1988 ) . s. scopetta , phys . c 70 , 015205 ( 2004 ) . c. ciofi degli atti , s. scopetta , e. pace , and g. salm , phys . c48 , r968 ( 1993 ) . c. ciofi degli atti , e. pace , and g. salm , phys . c46 , r1591 ( 1992 ) . f. cano and b. pire , nucl . phys . a 711 , 133c ( 2002 ) ; eur . phys . j. a 19 , 423 ( 2004 ) . v. guzey and m.i . strikman , phys . c 68 , 015204 ( 2003 ) . r. l. heimann , nucl . b64 ( 1973 ) 429 ; j. ellis and m. karliner , phys . b213 ( 1988 ) 73 . r. carlitz and j. kaur , phys . 38 , 673 ( 1977 ) ; j. kaur , nucl . phys . b 128 , 219 ( 1977 ) ; a. schaefer , phys . b 208 , 175 ( 1988 ) . g. altarelli , r.d . ball , s. forte and g. ridolfi , b 496 , 337 ( 1997 ) . giannini , rep . 54 , 453 ( 1991 ) . ahrens et al . , phys . d 35 , 785 ( 1987 ) . r. bijker , f. iachello and a. leviatan , ann . 236 , 69 ( 1994 ) ; phys rev . c 54 , 1935 ( 1996 ) ; phys . rev d 55 , 2862 ( 1997 ) . w. furmansky and r. petronzio , z. phys . c 11 , 293 ( 1982 ) . ball , s. forte and g. ridolfi , phys . b 378 , 255 ( 1996 ) . r. mertig , w. l. van neerven , z.phys . c70 , 637 ( 1996 ) ; w. vogelsang , phys . rev d54 , 2023 ( 1996 ) . g. altarelli and g. g. ross , phys . b 212 , 391 ( 1988 ) . f.e . close and r.g . roberts , phys . b 316 , 165 ( 1993 ) . particle data group , j. ashman _ et al . , _ phys d 54 , 1 ( 1996 ) . emc collaboration , j. ashman _ et al . b 328 , 1 ( 1989 ) . smc collaboration , d. adams _ et al . d 56 , 5330 , ( 1997 ) . e154 collaboration , k. abe _ et al . , _ phys . 79 , 26 ( 1997 ) . s. noguera , s. scopetta and v. vento , phys . d70 , 094018 ( 2004 ) . m. diehl , th . feldmann , r. jakob , and p. kroll , hep - ph/0408173 . m. guidal , m.v . polyakov , a.v . radyushkin , and m. vanderhaeghen , hep - ph/0410251 . j. blmlein and m. bttcher , nucl . b 636 , 225 ( 2002 ) . * fig . 1 * : the handbag contribution to the dvcs process in the present approach . * fig . 2 * : the gpd @xmath0 for the flavor @xmath223 , for @xmath239 gev@xmath140 and @xmath240 , at the momentum scale of the model . dashed curve : result in the pure isgur and karl model , eq . ( [ iku ] ) . the small tails which are found in the forbidden regions , @xmath243 and @xmath244 , represent the amount of support violation of the approach . full curve : the complete result of the present approach , eq . ( [ main ] ) . the two curves are obtained in correspondence of the two extreme values of the regge intercept @xcite , @xmath246 ( upper curve at low @xmath28 ) and @xmath247 . * fig . 3 * : for the @xmath10 values which are allowed at @xmath267 gev@xmath140 , @xmath248 , evaluated using our main equation , eq . ( [ main ] ) , is shown for four different values of @xmath28 , at the momentum scale of the model . from top to bottom , the dash - dotted line represents the gpd at @xmath271 , the full line at @xmath272 , the dashed line at @xmath273 , and the long - dashed line at @xmath274 . the curves are obtained in correspondence of the value of the regge intercept @xmath247 ( see text ) . * fig . 4 * : the @xmath28 and @xmath10 dependences of @xmath248 , for @xmath267 gev@xmath140 , at the momentum scale of the model . the value of the regge intercept has been taken to be @xmath247 ( see text ) . * fig . 5 * : we show the structure function @xmath275 obtained at @xmath276 gev@xmath140 by evolving at nlo the model calculation @xcite considering the structure of the constituents , for the two extreme regge behaviors mentioned in the text ( @xmath277 is the upper curve , here and also in the following figure ) . the data from refs . @xcite ( full dots ) and @xcite at @xmath278 gev@xmath140 are also shown . * fig . 6 * : the structure function @xmath279 for the neutron evolved at nlo to @xmath280 gev@xmath140 , for the two extreme regge behaviors mentioned in the text , is shown by the two full curves . the data from refs . @xcite at @xmath280 gev@xmath140 are also shown . * * : comparison of the results of the present approach for the valence distribution @xmath266 ( full curves ) , evaluated at @xmath267 gev@xmath140 , evolved to nlo from the hadronic scale up to @xmath268 gev@xmath140 , with the ansatz proposed in @xcite ( dotted curve ) , evaluated in the same kinematical conditions . the two full curves are obtained in correspondence of the two extreme values of the regge intercept @xcite , @xmath246 ( upper curve at low @xmath28 ) and @xmath247 .

blue straggler stars ( bsss ) were first identified in the galactic globular cluster ( ggc ) m3 by @xcite , as a group of stars bluer and brighter than the cluster turn - off stars , thus being more massive than the stars currently evolving at the turn - off ( to ) . since then , bsss have been identified in a variety of stellar systems in very different environmental conditions including : the field @xcite , open clusters @xcite , globular clusters ( e.g. , * ? ? ? * ) , and dwarf galaxies @xcite . their puzzling position in the color - magnitude diagram ( cmd ) suggested that they do not fit in the traditional scheme of stellar evolution for single stars . in fact , in order to populate that region of the cmd , these exotic objects must have experienced a physical process able to allow them to still stay in the core h - burning stage despite of their mass and the cluster age ( for a detailed historical review , see * ? ? ? * ) . presently , two different physical mechanisms are favored for explaining the formation of bsss : _ i ) _ coalescence through direct stellar collisions ( col - bs , * ? ? ? * ) , and _ ii ) _ evolution of primordial binary systems , in which mass transfer between the two components allows the rejuvenation of the secondary component ( mt - bss , * ? ? ? it is also widely accepted that these two scenarios are not in competition , but they could both occur simultaneously in a stellar system . in fact , the coexistence of bsss formed through both the collisional and mass transfer channels is commonly invoked to interpret the radial distribution of bsss in many ggcs . the higher central concentration is explained by an efficient rate of stellar collisions in the densest regions , while the rise at large distance would be due to mt - bsss @xcite surviving in the outskirts . while the scenario for bsss in galactic clusters is well settled , at least from the observational point of view , in contrast , this is not true for dwarf galaxies . in fact , even though the first indication for the presence of bss candidates in the sextans dwarf dates back to @xcite , and , since then , several observational findings suggest the presence of bss candidates in several dsphs ( see , e.g. , * ? ? ? * and references therein ) , we are still faced with the problem of understanding if these candidates are _ genuine _ bsss or _ normal _ core h - burning stars belonging to an , albeit sparse , intermediate - age stellar population . recently , a few studies have focused on the properties of bsss in nearby dsph galaxies in the local group ( lg ) . @xcite presented wide field data for the galaxies draco and ursa minor , concluding that the population of bsss is , in both systems , compatible with the mt - bss scenario . @xcite discussed the properties of bsss in sculptor and fornax . sculptor presents similarities with both draco and ursa minor , as its bss population is also compatible with a mt - bss population . on the contrary , the relatively strong central radial concentration of the bss candidate stars in fornax favors the presence of an intermediate - age stellar population ( in agreement with previous studies of the star formation history ( sfh ) of fornax ) . unraveling the nature of the bss candidates has important implications for our understanding of the sfhs of these galaxies . for example , whenever it is possible to discriminate between a sequence of bsss and a young ms population , it is possible to set a constraint on the age of the last star formation event in a galaxy . therefore , the bss populations in draco , ursa minor , and sculptor suggest that no star formation occurred in the last few gyr in these galaxies . intriguingly , @xcite found a statistically significant anti - correlation between the frequency of bsss and the absolute visual magnitude ( m@xmath0 ) of 8 nearby dsphs . the anti - correlation is valid over a large range of luminosities , and presents a different slope than that of ggcs . in addition , @xcite ruled out the possibility that collisional binaries could have contributed to the observed bss populations and concluded that the bsss in dwarf spheroidal galaxies are mainly formed via mass exchange in primordial binaries . in view of the important implications of the presence of a sizable sample of bsss in a dwarf galaxy , it is worthwhile to apply the approach outlined by @xcite to other dwarfs . the hope is to eventually produce independent constraints useful for addressing the true nature of the bss candidates . l|cc @xmath1 & @xmath2 & @xmath3 + @xmath4 & @xmath5 & @xmath6 + n@xmath7 & @xmath8 & @xmath9 + n@xmath10 & @xmath11 & @xmath12 + f@xmath13 & @xmath14 & @xmath15 + n@xmath16 & @xmath17 & @xmath18 + in this paper , we investigate these issues further by analyzing the properties of candidate bsss in the isolated dsphs cetus and tucana . the data sets are part of the lcid project ( _ local cosmology from isolated dwarfs _ ) , aimed at recovering the full sfhs of six isolated lg galaxies : cetus and tucana ( dsph ) , lgs 3 and phoenix ( dirr / dsph ) , ic 1613 and leo a ( dirr ) . the plan of the paper is as follows : in the next section we briefly describe the photometric data set and outline the criteria for selecting the candidate bsss in the observed cmds . in 3 we investigate the properties of the bss populations . 4 presents the results of dynamical simulations . the discussion and final remarks close the paper . the observations used in this paper have already been presented in ( * ? ? ? * cetus ) and ( * ? ? ? * tucana ) . here we summarize the main points relevant to the following discussion . the images were collected with the acs camera aboard the hst , under the project _ the onset of star formation in the universe : constraints from nearby isolated dwarf galaxies _ ( pid 10505 , pi c. gallart ) . a total of 25 and 32 orbits have been devoted to cetus and tucana , respectively , using the @xmath19 and @xmath20 filter pass bands . the photometric reduction has been performed using the daophot / allframe package @xcite . the final catalog has been calibrated to the vegamag system using the transformations presented in @xcite . the adopted distance modulus and reddening are summarized in table [ tab : tab01 ] , together with other quantities calculated in this work . figure [ fig : samples ] shows the cmds of the two galaxies , with the regions adopted to select the various samples studied superimposed . note that we adopted the same boxes for both galaxies , applying a shift to take into account the differences in distance modulus and reddening . the cmds of both galaxies are @xmath21 mag deeper than the to , reaching @xmath22 mag . the blue plume of candidate bsss emerges clearly in the cmd , for @xmath23 mag , @xmath24 . we defined a box in this region to include the bulk of these stars . note that there are a number of objects between the bsss region and the red giant branch ( rgb ) . we decided not to include these stars for two reasons : _ i ) _ it is possible that blended objects from the most populated to region are polluting this part of the cmd ; _ ii ) _ in this color range , we expect to find stars evolved off the main sequence . also , we did not include the brightest objects ( @xmath25 ) because it is possible that they are extreme horizontal branch ( hb ) stars . also note that we estimated a negligible contamination from both foreground galactic stars ( due both to the small area covered and the high galactic latitude of the two galaxies ) and background galaxies ( thanks to the careful cleaning of the catalogs , see @xcite ) . figure [ fig : samples ] also shows the regions adopted to select stars in the hb , the rgb , and the progeny of bsss during the central helium burning phase . this last box has been placed above the red hb , paying attention not to include the agb clump , which clearly appears in the cmds of both galaxies ( @xmath26 mag ) . 0.8 cm in the box covering the blue plume , we have identified 940 and 1214 objects in cetus and tucana , respectively , which we consider as bss candidates . note that the present data cover a significant fraction of the body of tucana ( tidal radius @xmath27 ) , which was centered in the acs field ( field - of - view @xmath28 ) . however , cetus occupies a larger area in the sky ( @xmath29 , @xcite ) , and the sampled field covers a smaller fraction . moreover , the observed field was offset from the center so that the innermost regions of cetus were not observed @xcite . if we take this into account , following the reasoning of @xcite to estimate the total number of rr lyrae stars , the total number of bss candidates in cetus is estimated to be @xmath30 5,000 stars . the most important question is to assess the nature of these stars . bsss have been found in very different environments , from the field , to low - density stellar systems such as open clusters and dwarf spheroidals , and in most globular clusters . they are therefore common products , and it is reasonable to expect to find them in tucana and cetus as well . therefore , the basic question we are trying to address is : are these candidates genuine bsss , or is there also a significant component of truly young stars ? in the following sections we will try to address this using different observables . 0.4 cm , but we present a zoom of the cmd of cetus ( left panel ) and tucana ( right panel ) , corresponding to the core he - burning stage . in each panel the box represents the region used to select the candidate progeny of the blue plume stars . four tracks are overplotted , for different mass and metallicity assumptions . the filled stars mark the location of the anomalous cepheid variables detected in these galaxies @xcite . , title="fig:",width=321 ] the main objective of the lcid project was to derive accurate sfhs for a sample of six isolated galaxies in the local group , in order to understand their evolution in a cosmological context . to do this , we adopted well - established techniques based on the comparison of the observed cmd with a synthetic one . we devoted significant effort to investigating the error budget and also possible systematics affecting our solution , due to the use of different photometric packages , stellar evolutions libraries , sfh codes . the main conclusion ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) of this analysis was that we can rely , within the error bars , on both the uniqueness and stability of the derived solutions . therefore , the most important features recovered , such as the epoch and the duration of the main star formation events , including those at the oldest epoch , and the age - metallicity relations are solid results . the sfhs presented in @xcite and @xcite , for cetus and tucana respectively , show that both galaxies are made of old and metal - poor populations . in particular , both galaxies formed 90% of their stars at epochs older than 9 gyr ago , and the bulk of the stars have metallicities in the range @xmath31 . in addition , both galaxies present a well defined age - metallicity relation in the sense of increasing metallicity with decreasing age . the sfh reconstructions indicate a small population ( @xmath32 in mass ) of stars significantly younger than the majority of stars ( @xmath33 gyr ) but with metallicities consistent with the oldest stars ( @xmath34 ) . thus , these stars do not follow the general age - metallicity relations in these two galaxies . the analysis of the best - fit cmds show that they nicely populate the blue plume region of the cmd , brighter and bluer than the oldest ms turn - off . the possibility that the metallicity of these stars has been understimated by the code is excluded with high confidence level . in fact , higher metallicity and an age of @xmath35 gyr , would produce also a redder rgb sequence that is not observed . on the basis of the retrieved sfh , @xcite made the hypothesis that the sequence of blue stars , brighter than the old ms turn - off , are consistent with a population of bsss , characterized by the same metal content as the metal - poor dominant stellar component . an alternative interpretation would be that these are truly young metal - poor ms stars , formed as a consequence of gas accretion . however , it seems very unlikely that two galaxies , that spent most of their life in isolation , accreted gas and formed stars of the same metallicity and at the same epoch , in one event that created roughly the same amount of mass ( 3% of the total stellar mass ) . figure [ fig : track ] shows the comparison of the tucana cmd with theoretical tracks from the basti database of stellar evolution models @xcite . here we note that the analysis of the sfh for the lcid galaxies has been performed by using the same evolutionary framework ; so the present analysis is fully consistent with those concerning the sfhs . the selected tracks span the range of metallicities suggested by the sfhs , z = 0.0001 , 0.0003 , 0.0006 , and 0.001 as an upper limit . the figure discloses some interesting clues . first , the black lines in the left panel show the zero age main sequence ( zams ) for two metallicities : z=0.0001 ( solid line ) and z = 0.001 ( dashed ) . the comparison with the observed cmds shows that both zamss border the blue envelope of the bss sequence , independent of the assumed metal content . moreover , the same panel shows that the maximum mass fitting the brightest objects is equal to about 1.5 @xmath36 ( red lines ) , again with a negligible dependence on the assumed metallicity ( z=0.0001 , 0.001 for the red solid and dashed lines , respectively ) . the same figure also shows the location of two theoretical tracks corresponding to @xmath37 stellar models ( green lines ) , taken as representative of the stellar structures currently evolving at the to and sub - giant branch of both galaxies . the filled symbols show the first point corresponding to the central helium burning phase . it is evident - as also fully supported by the sfh analysis - that for the to and the sgb , a range of metallicities in the adopted evolutionary framework is required to properly interpret the observed morphology of the cmd . this comparison reveals that the maximum mass of the stars in the blue plume is at most @xmath38 2 times larger than the mass of the stars at the to . this occurrence is in good agreement with the properties of bsss in other stellar systems such as ggcs ( see , i.e. , * ? ? ? * and references therein ) . the central panel shows the comparison with stellar tracks for different masses ( @xmath39 , green and red lines , respectively ) , and different metallicities ( @xmath40-solid , @xmath41-dashed , @xmath42-dot - dashed ) . the shaded area encloses the region corresponding to the central hydrogen burning phase , whose end is marked by the asterisks . the bulk of bss candidates clearly occupies this region of the cmd . the plot clearly shows that a spread in mass is needed to fit the observed distribution of blue plume stars . a spread in metallicity is also possible but , as shown in the right panel of the same figure , values larger than @xmath43 seem unlikely . the 1.0 @xmath44 track is too faint and too red to represent the bulk of the blue plume objects . the 1.2 @xmath44 still gives a good representation , and can be considered as an upper limit both in term in of mass and metallicity . in fact , despite the 1.5 @xmath44 still limits the brightest blue plume stars , the age of this model at the turn - off ( @xmath381.3gyr ) is in contrast with the finding of the sfh . on the other hand , higher metallicities would shift the tracks to redder color and fainter magnitude , significantly worsening the agreement in particular in the rgb phase . thus , relatively metal - poor and relatively massive ( @xmath45 ) stellar structures are preferred to better represent the peak of the distribution . therefore , the metallicity derived for the bss candidates is in good agreement with that estimated for the bulk of the older populations , as discussed in [ sec : sfh ] , and not in agreement with the higher metallicities found at the end of the initial episodes of star formation in cetus and tucana . this result can be considered a circumstantial evidence supporting the idea that - at least a large fraction of - the stars located along the blue plume could be genuine bsss ; in fact if they were true young stars formed in a late star formation burst , they would have a metallicity lower than that one could expect of the basis of a standard age - metallicity relation . additional information can be provided by the study of the bss progeny in the core he - burning stage . evolutionary models for ms stars predict that stars more massive than about @xmath46 are located in the red side of the hb at a magnitude level depending on their he core mass @xcite . in particular , the right - hand panel of figure [ fig : track ] suggests that the evolved stars in this mass range all clump in the same region of the cmd . this implies that the progeny of bsss of similar mass , independent of the formation mechanism ( see , e.g. , * ? ? ? * ) , should occupy a similar region in the cmd , redder and brighter than their low - mass counterparts . this means that , on the one hand , the distribution of bss progeny in the cmd provides very little information about the mass distribution of the bss population above @xmath46 . on the other hand , the ratio of the number of stars can be compared with theoretical expectations based on the evolutionary lifetimes in the core h- and he - burning stages . figure [ fig : trackevol ] shows an expanded view of the hb region of the cetus ( left panel ) and tucana ( right panel ) cmds . superimposed to both plots are the evolutionary tracks , starting with the onset of the central helium burning , for two masses ( 1.0 -blue line- and 1.3 -red-@xmath36 ) and two metallicities ( @xmath40 -solid- and @xmath42 -dashed- ) . the boxes corresponding to our selections are shown . both panels show the presence of a sparse sequence above the red hb . while a few of them , which are located close to theoretical prediction for the most metal - poor stars , fall outside our box , the majority of stars , inside the box , look more compatible with slightly more metal - rich tracks . note that this is in agreement with the sfh results , which gives for the two galaxies a mean metallicity of the order of @xmath47 ( cetus ) and @xmath42 ( tucana ) . to further investigate the interpretation that the stars located inside the box showed in fig . [ fig : trackevol ] can be really considered the progeny of the stars located along the ms blue plume , we compared the star counts ratio between the objects located inside the two boxes with theoretical predictions obtained taking into the account the information from the sfh . * their fig 18 ) have already shown a synthetic cmd built from the sfh solution of cetus , where stars with colors and magnitudes typical of a population of bsss clearly appear . we therefore counted the stars using the same boxes defined here , and estimated the same ratio using a best - fit cmd for both cetus and tucana . the derived ratios are 0.08@xmath480.01 and 0.08@xmath480.01 for cetus and tucana respectively . analogous estimates can be made from the observed cmd . the number of bsss in cetus and tucana are 940 and 1214 , respectively , while the number of objects in the evolved phase are 90 and 123 . this gives ratios of 0.10@xmath480.01 for cetus and 0.10@xmath480.01 for tucana . this shows that there is good agreement with the empirical values . this suggests that the selected stars are the evolved counterpart of blue plume objects . moreover , this test strongly suggests that , if we are dealing with a pure bsss population , the analysis performed using models calculated for normal ms stars is only marginally affecting the results . it is also worth noting that the progeny of bss - in particular the more massive ones - can cross the instability strip for radial pulsation during the core he - burning stage ; in this case they pulsate as anomalous cepheids @xcite . when studying the populations of variable stars in the lcid galaxies , @xcite discovered three anomalous cepheids in cetus , and six in tucana . the locations of these variable stars in the cmds of both galaxies are shown in fig . [ fig : trackevol ] . they appear bluer and brighter than the expected bulk of the bss progeny . this suggests that anomalous cepheids might have slightly larger masses than the typical bss . 0.4 cm [ fig : cumul ] the luminosity function of blue plume stars is a powerful diagnostic to both infer information on their nature and , in case of a bss population , to distinguish between the two main formation mechanisms . in fact , if these stars belong to an intermediate - age population , we expect to find them more centrally concentrated ( see next section ) and also that the brightest ( youngest ) are preferentially located to the innermost regions . on the other hand , it has been found that in ggcs the bss luminosity function changes as a function of radial distance @xcite , in the sense that the brightest bss stars are preferentially located in the central regions . since there is no doubt that in gc we only have bsss , with no contamination from intermediate - age stars , this occurrence is explained with a combination of two different formation mechanisms . collisions , which are efficient in the inner regions of ggc , tend to create bluer and brighter bsss @xcite . in the outer parts of the ggcs , the bsss are predominantly created from primordial binaries . in dsph galaxies the formation of col - bss is highly disfavored due to the very low stellar density , so that we do not expect to observe a similar behavior @xcite . therefore , in case a central concentration of the brightest blue plume stars were observed , this would point to the presence of a genuine intermediate - age population . figure [ fig : lumfun_rad ] shows the bss candidates luminosity function for cetus ( left panel ) and tucana ( right panel ) . the plot shows the luminosity function of stars inside ( dotted line ) and outside ( dashed line ) a distance of 1.5 times the core radius , @xmath49 . due to the different number of stars and area covered by the two samples , the curves were normalized to their area . the shape of the luminosity function does not change significantly as a function of radius , and similarly we do not find any evidence for an obvious shift in the magnitude of the peak . we find that the fraction of the brightest bsss ( @xmath50 , i.e. , within 0.5 mag of the brightest ) , is also constant with radius . for the inner and outer regions we found @xmath51 and @xmath52 in the case of cetus , and @xmath53 and @xmath54 for tucana . note that the errors are solely poisson statistical errors . thus , we find no evidence for segregation in either galaxy , supporting the absence of centrally concentrated bright stars , and supporting that the population of bsss in cetus and tucana is formed as a result of the evolution of binary systems , even in the innermost regions . this last result can be further verified studying the radial distribution of candidate bsss compared to that of other stellar tracers . it has been found in many ggcs that the radial profile of bsss has a central cusp ( c.f . * ) , which falls to a minimum ( the zone of avoidance * ? ? ? * ) , and then increases with radius . this phenomenon has again been interpreted with the different formation mechanisms at work : the col - bsss are more centrally concentrated than mt - bss @xcite , which instead dominate in the outskirts of clusters . if col - bsss are not present in dsphs , we expect no central concentration , as verified by @xcite and @xcite for draco , ursa minor and sculptor . figure [ fig : grad_rad ] presents the ratios @xmath55 , @xmath56 , and @xmath57 for cetus ( left ) and tucana ( right ) , as a function of the galactocentric distance expressed in units of core radius . we divided the complete catalog into five ( cetus ) and six ( tucana ) radial bins , containing the same total number of stars , according to the position of stars in elliptical annuli . the points in figure [ fig : grad_rad ] show the ratio in each elliptical radius , plotted as a function of the mean galactocentric distance of each annulus . figure [ fig : grad_rad ] clearly shows the absence of strong central concentrations of bsss in both galaxies . we tested the hypothesis that the ratios of the various samples are consistent with a flat distribution by comparing the ratios to the mean of the observed distribution . in the case of cetus , we estimated non - reduced @xmath58 of 6.6 , 7.3 , 9.3 for @xmath59 , @xmath60 , and @xmath61 , respectively . the null hypothesis therefore has corresponding probabilities of 0.15 , 0.12 , and 0.05 . this suggests that @xmath59 and @xmath60 are consistent with a flat distribution , while @xmath61 is only marginally consistent . in the case of tucana , we derive @xmath58 values of 14.8 , 10.6 , 12.8 ( @xmath59 , @xmath60 , @xmath61 ) , corresponding to probabilities of 0.01 , 0.06 , and 0.03 . this means that @xmath59 and @xmath61 are not consistent with a flat distribution , while @xmath60 is only marginally consistent . although @xmath61 has larger uncertainties than the other ratios , it is interesting to note that , qualitatively , @xmath61 shows larger fluctuations than the others . moreover , in the range @xmath62 the trend for the two galaxies is opposite . cetus presents an initial increase followed by a drop , while in tucana an initial decrease within the core radius is followed by the increase in the ratio of bsss relative to hb stars . at large distance , there is mild evidence of an increase of @xmath61 , especially for tucana where the field covers a large area in terms of @xmath49 . this is in agreement with the predictions by @xcite . another way to represent the radial distribution of different samples is to use the cumulative distributions , shown in figure [ fig : cumul ] for the bss , rgb , sgb , and hb stars . a kolmogorov - smirnov test gives probabilities of 0.861 ( 0.004 ) , 0.48 ( 0.02 ) , and 0.58 ( 0.28 ) for cetus ( tucana ) that the bsss and the rgb / hb / sgb samples are drawn from the same parent population . however , it is interesting to note that , starting from 3 @xmath49 , the bss cumulative distribution is systematically lower , in both galaxies , than the others . this indicates higher bss frequencies at larger distances . @xcite found that a linear correlation exists between the logarithmic frequency of bsss ( f@xmath63 , normalized to the number of hb stars ) and the absolute visual magnitude of nearby lg dwarfs . interestingly enough , they found that this correlation is verified over a large range of * magnitudes * , and that its slope is significantly different , i.e. , less steep , than a similar relation found by @xcite for ggcs . presently , an underlying physical mechanism for this trend has not been identified based on the selections presented in figure [ fig : samples ] , we estimated that f@xmath63 is equal to 0.05@xmath480.05 and 0.02@xmath480.04 for cetus and tucana , respectively . in figure [ fig : fm ] , these values are plotted together with data from other galaxies given in figure 2 of @xcite . we adopted the absolute @xmath64 magnitude from @xcite for cetus ( @xmath65 mag ) and from @xcite for tucana ( @xmath66 mag ) . [ fig : fm ] shows that tucana and cetus are consistent with the relationship derived from other lg dwarfs . using the analytic relation for @xmath67 from @xcite , we obtain @xmath680.08 and 0.03@xmath480.08 for cetus and tucana , respectively , again , showing very good agreement with the observed values . if we accept that the @xmath69 relation is significant , then this is an independent indication that the we are dealing with a genuine bss population . if the blue plume stars were due to fluctuations in the sfh producing younger stars , one would not expect any relationship between @xmath67 and @xmath64 . on the other hand , if blue plume stars are bsss , they essentialy belong to the same population generating the hb stars . as already noted by @xcite , the trend is similar in the case of globular clusters , but with significantly shallower slope . the existence of an anti - correlation with the luminosity ( * the higher the luminosity , the lower the bss frequency * ) , and hence with the mass of the cluster , might suggest that in more massive globular clusters the mechanisms affecting the binary systems are more efficient then in less massive ones , and this cause a deficiency of bss with respect of hb stars ( see * ? ? ? * ) . in dsph , the shallower slope might indicate that the same mechanisms are still at work , but are less efficient due to the lower density of dwarf galaxies . on the basis of present results , it appears evident that if the reliability of @xmath70 relation was confirmed by increasing the number of lg dwarfs in the sample , it could represent a valuable tool for discriminating stellar systems hosting truly young ms stars from those with a significant population of old , genuine bsss . for example , a preliminary analysis of a deep cmd of fornax dsph ( del pino et al . 2011 , in preparation , also shown in figure [ fig : fm ] as a crossed - circle symbol ) suggests that the f@xmath63 estimate for this galaxy ( f@xmath71 ) is significantly higher than expected from its absolute magnitude ( f@xmath72 ) . this is in agreement with the conclusion presented in @xcite , and with the fact that fornax hosts a population as young as 100 myr ( see also * ? ? ? further clues about the origin of bsss can be inferred from their dynamical evolution ( see , e.g. , * ? ? ? we thus perform a wide sample of dynamical simulations , following the evolution of mt - bss and col - bss populations in the potential of cetus and tucana . we use the code bev @xcite , in the upgraded version described by @xcite . the same code has been used to model bsss in draco , ursa minor @xcite , sculptor and fornax @xcite . bev integrates the dynamical evolution of bsss under the influence of the host - galaxy potential , dynamical friction , distant encounters with other stars , and three - body encounters ( which are negligible for dsphs ) . the potential of the host galaxy is assumed to be time - independent and is represented by a multi - mass king model . the classes of mass are the same as in @xcite , and the to mass is assumed to be 0.8 @xmath73 ( see figure [ fig : track ] ) for both cetus and tucana . to calculate the potential , we input the number density of stars in the core ( @xmath74 and @xmath75 stars pc@xmath76 for cetus and tucana , respectively ) the velocity dispersion in the core ( we assume @xmath77 km s@xmath78 for both cetus and tucana ) , and we modify the central adimensional potential @xmath79 ( defined in * ? ? ? * ) , equating the simulated concentration and radial density profile with the observed values . [ fig : simprofile ] shows the resulting models for cetus ( obtained for @xmath80 and @xmath81 ) and for tucana ( obtained for @xmath82 and @xmath83 ) . in the case of tucana , the simulated profile and the concentration of the model match the observations reported in @xcite . for the concentration of cetus , we adopt @xmath84 , which best matches our data . we note that the model for cetus has a relatively high value of @xmath79 , with respect to other dsphs ( generally , @xmath85 for globular clusters and @xmath86 for dsphs ) . in our model , bsss are generated with a given position , velocity , mass and lifetime . the fundamental differences between col - bsss and mt - bsss are the following : \(i ) the distribution of initial positions : the initial positions of mt - bsss are randomly chosen according to a probability distribution homogeneous in radial distance from the center . this means that mt - bsss are initially distributed according to an isothermal sphere , since mt - bsss are expected to follow the same distribution as the primordial binaries ( see * ? ? ? * ; * ? ? ? ( the spherical nature of the dsphs is not conclusively established , but is chosen as a reasonable starting point . ) the minimum and the maximum value of the distribution of initial radial distances of mt - bsss , @xmath87 and @xmath88 ( both defined as three - dimensional quantities ) , have been tuned in order to find the best - fitting simulation . on the other hand , col - bsss are generated only inside @xmath49 ( i.e. , @xmath89 and @xmath90 in all the simulations ) , the only region were collisions might occur , although collisions are highly unlikely in dsphs ( the probability of a three - body interaction in the core of cetus and tucana is @xmath91 times lower than in a moderately dense globular cluster ) . the initial distribution of col - bsss is not necessarily homogeneous in the radial distance , as they are not expected to follow the same distribution as binaries ( in our simulations , we assume a constant probability distribution between the center of the cluster and @xmath49 ) . \(ii ) the distribution of initial velocities : the initial velocities of mt - bsss are drawn from the equilibrium distributions of dsph stars with the same masses as the mt - bsss . col - bsss are assumed to be born with a natal kick ( see * ? ? ? we adopt a kick velocity , @xmath92 , @xmath93 . for kick velocities higher than @xmath94 most of col - bsss are ejected from the dsph , whereas for kick velocities lower than @xmath95 , col - bsss are almost indistinguishable from mt - bsss from a dynamical point of view . in most of the runs , the masses of the bsss are assumed to be @xmath96 = 1.3 m@xmath97 . we also calculated simulations with @xmath96 = 0.9 , 1.1 m@xmath97 , but the results are substantially unchanged . each bss is evolved for a time @xmath98 , randomly selected from a homogeneous distribution between @xmath99 and @xmath100 . the parameter @xmath101 is the lifetime of bsss ( see * ? ? ? we calculated simulations with @xmath102 2 and 4 gyr ( 4 gyr is the age inferred from the derived sfh ) . tables [ tab : tab02 ] and [ tab : tab03 ] show the parameters of the simulations and report also the fraction @xmath103 of col - bsss present in each run ( @xmath104 , where @xmath105 and @xmath106 are the number of col - bsss and of mt - bsss , respectively ) . the results of the @xmath58 analysis are also included . for each run ( we made 38 runs for cetus and 37 for tucana ) 10000 bsss have been simulated . we show three values of the non - reduced @xmath58 , referring to @xmath107 ( @xmath108 , column 8 of tables [ tab : tab02 ] and [ tab : tab03 ] ) , @xmath109 ( @xmath110 , column 9 ) and @xmath111 ( @xmath112 , column 10 ) . [ fig : dyn ] shows the radial distributions of @xmath113 , @xmath109 , and @xmath111 obtained for the best - matching simulations ( run c3 and run t7 for cetus and tucana , respectively ) , compared with the observed distributions ( the same as in fig . [ fig : grad_rad ] ) . we stress that only the radial distribution of @xmath114 comes from the simulations , and it is compared with the observed radial distribution of @xmath115 , @xmath116 and @xmath117 . in fact , the adopted simulation method does not allow to evolve these specific stellar types , but only stars in a given class of mass ( e.g. , mapelli et al . we also remind that the underlying potential of the dsph is analytic and time - independent . the best - matching simulations reproduce well the observed radial distributions of @xmath107 , @xmath118 , and @xmath111 . in fact , the values of the minimum non - reduced @xmath119 are @xmath120 , @xmath121 and @xmath122 for cetus ( run c3 ) and @xmath123 , @xmath124 and @xmath125 for tucana ( runs t5 , t7 , t14 , t15 , and t16 ) . fig [ fig : chisq ] shows the behavior of @xmath108 ( that of @xmath110 and of @xmath112 is very similar , see tables [ tab : tab02 ] and [ tab : tab03 ] ) as a function of the fraction of col - bsss ( @xmath103 ) . as it was reasonable to expect , adding a population of col - bsss significantly reduces the agreement between data and model : for both cetus and tucana , only simulations with @xmath126 give an acceptable @xmath119 . the dependence of the results on the bsss mass ( in the range allowed by the observations ) is negligible . this is in agreement with the findings of previous papers @xcite for bsss in draco , ursa minor , and sculptor . the dependence on the lifetime of bsss is negligible for cetus ( where the runs with @xmath127 gyr are almost equivalent to the runs with @xmath128 gyr ) and slightly more important for tucana ( where the runs with @xmath127 gyr have generally lower values of @xmath119 with respect to the runs with @xmath128 gyr , but the difference is not statistically significant ) . given the range of masses expected for the bss , suggested by the comparison with stellar tracks , we assume 4 gyr as fiducial value . in the case of cetus , for the best - matching runs @xmath129 and @xmath130 , indicating that primordial binaries were initially distributed ranging from the center of the dsph out to ( at least ) @xmath131 . the tidal radius of cetus ( derived from our data ) is 24@xmath132 , i.e. , @xmath133 , much larger than @xmath134 . however , this difference likely does not mean that bsss do not form beyond @xmath131 , but is a consequence of the fact that we do not have data beyond @xmath135 ( furthermore , the coverage of cetus is not complete even inside @xmath136 ) . in the case of tucana , for which data are available almost up to the tidal radius , the best - matching value of @xmath137 is equal to the tidal radius . interestingly , the best - matching values of @xmath138 for tucana are all larger than one core radius . in summary , the results of dynamical simulations show that it is possible to explain the observed distribution of bss candidates , in both cetus and tucana , with a mt - bss population . any population of col - bsss more numerous than the 10% of the total number of bsss reduces the agreement between data and simulations . this result is not surprising , as basic dynamical calculations tend to exclude the existence of col - bsss in dsphs , but it is important to point out that the simulations can almost rule out the unphysical scenario where @xmath139% of bsss are col - bsss in both cetus and tucana . cccccccccc c1 & 0.0 & 5.0 & 0.9 & 4 & & 0 & 7.7 & 6.5 & 4.7 + c2 & 0.0 & 5.0 & 1.1 & 4 & & 0 & 9.5 & 8.1 & 5.8 + * c3 * & * 0.0 * & * 5.0 * & * 1.3 * & * 4 * & * * & * 0 * & * 5.8 * & * 4.9 * & * 3.5 * + c4 & 0.1 & 5.0 & 1.3 & 4 & & 0 & 8.3 & 7.0 & 5.0 + c5 & 0.2 & 5.0 & 1.3 & 4 & & 0 & 7.3 & 6.2 & 4.5 + c6 & 0.3 & 5.0 & 1.3 & 4 & & 0 & 10 & 8.7 & 6.2 + c7 & 0.4 & 5.0 & 1.3 & 4 & & 0 & 14 & 12 & 8.5 + c8 & 0.5 & 5.0 & 1.3 & 4 & & 0 & 19 & 16 & 12 + c9 & 1.0 & 5.0 & 1.3 & 4 & & 0 & 51 & 42 & 31 + c10 & 0.0 & 5.0 & 0.9 & 2 & & 0 & 6.7 & 5.6 & 4.1 + c11 & 0.0 & 5.0 & 1.1 & 2 & & 0 & 7.2 & 6.1 & 4.4 + c12 & 0.0 & 5.0 & 1.3 & 2 & & 0 & 9.8 & 8.3 & 5.9 + c13 & 0.1 & 5.0 & 1.3 & 2 & & 0 & 9.0 & 7.6 & 5.4 + c14 & 0.2 & 5.0 & 1.3 & 2 & & 0 & 7.0 & 5.9 & 4.2 + c15 & 0.3 & 5.0 & 1.3 & 2 & & 0 & 7.5 & 6.3 & 4.6 + c16 & 0.4 & 5.0 & 1.3 & 2 & & 0 & 12 & 10 & 7.5 + c17 & 0.5 & 5.0 & 1.3 & 2 & & 0 & 20 & 17 & 12 + c18 & 1.0 & 5.0 & 1.3 & 2 & & 0 & 50 & 41 & 30 + c19 & 0.0 & 4.0 & 1.3 & 4 & & 0 & 10 & 8.7 & 6.2 + c20 & 0.2 & 4.0 & 1.3 & 4 & & 0 & 9.4 & 7.8 & 5.5 + c21 & 0.5 & 4.0 & 1.3 & 4 & & 0 & 7.9 & 6.7 & 4.8 + c22 & 1.0 & 4.0 & 1.3 & 4 & & 0 & 41 & 34 & 25 + c23 & 0.0 & 6.0 & 1.3 & 4 & & 0 & 13 & 11 & 7.7 + c24 & 0.2 & 6.0 & 1.3 & 4 & & 0 & 16 & 13 & 9.1 + c25 & 0.5 & 6.0 & 1.3 & 4 & & 0 & 26 & 22 & 16 + c26 & 1.0 & 6.0 & 1.3 & 4 & & 0 & 77 & 64 & 46 + c27 & 0.0 & 5.0 & 1.3 & 4 & 1 & 0.1 & 7.2 & 6.0 & 4.3 + c28 & 0.0 & 5.0 & 1.3 & 4 & 1 & 0.2 & 13 & 11 & 7.8 + c29 & 0.0 & 5.0 & 1.3 & 4 & 1 & 0.3 & 20 & 17 & 12 + c30 & 0.0 & 5.0 & 1.3 & 4 & 1 & 0.4 & 28 & 23 & 16 + c31 & 0.0 & 5.0 & 1.3 & 4 & 1 & 0.5 & 43 & 36 & 26 + c32 & 0.0 & 5.0 & 1.3 & 4 & 1 & 0.6 & 56 & 47 & 33 + c33 & 0.0 & 5.0 & 1.3 & 4 & 1 & 0.7 & 76 & 63 & 45 + c34 & 0.0 & 5.0 & 1.3 & 4 & 1 & 0.8 & 94 & 78 & 56 + c35 & 0.0 & 5.0 & 1.3 & 4 & 1 & 0.9 & 110 & 91 & 65 + c36 & & & 1.3 & 4 & 1 & 1.0 & 122 & 101 & 72 + c37 & 0.0 & 5.0 & 1.3 & 4 & 2 & 0.1 & 9.4 & 7.9 & 5.7 + c38 & 0.0 & 5.0 & 1.3 & 4 & 2 & 0.2 & 14 & 12 & 8.3 + cccccccccc t1 & 0.0 & 5.4 & 1.3 & 4 & 0 & & 101 & 78 & 60 + t2 & 0.5 & 5.4 & 1.3 & 4 & 0 & & 51 & 40 & 30 + t3 & 1.0 & 5.4 & 1.3 & 4 & 0 & & 21 & 17 & 12 + t4 & 1.5 & 5.4 & 1.3 & 4 & 0 & & 5.7 & 4.6 & 3.4 + t5 & 1.7 & 5.4 & 0.9 & 4 & 0 & & 2.6 & 2.1 & 1.6 + t6 & 1.7 & 5.4 & 1.1 & 4 & 0 & & 3.2 & 2.6 & 1.9 + * t7 * & * 1.7 * & * 5.4 * & * 1.3 * & * 4 * & * 0 * & * * & * 2.4 * & * 1.9 * & * 1.5 * + t8 & 2.0 & 5.4 & 1.3 & 4 & 0 & & 7.7 & 6.1 & 4.8 + t9 & 2.5 & 5.4 & 1.3 & 4 & 0 & & 18 & 14 & 11 + t10 & 0.0 & 5.4 & 1.3 & 2 & 0 & & 106 & 83 & 63 + t11 & 0.5 & 5.4 & 1.3 & 2 & 0 & & 55 & 43 & 33 + t12 & 1.0 & 5.4 & 1.3 & 2 & 0 & & 23 & 18 & 14 + t13 & 1.5 & 5.4 & 1.3 & 2 & 0 & & 2.5 & 2.1 & 1.5 + t14 & 1.7 & 5.4 & 0.9 & 2 & 0 & & 2.3 & 1.8 & 1.4 + t15 & 1.7 & 5.4 & 1.1 & 2 & 0 & & 1.9 & 1.5 & 1.1 + t16 & 1.7 & 5.4 & 1.3 & 2 & 0 & & 1.6 & 1.3 & 0.95 + t17 & 2.0 & 5.4 & 1.3 & 2 & 0 & & 4.8 & 3.7 & 2.9 + t18 & 2.5 & 5.4 & 1.3 & 2 & 0 & & 16 & 13 & 9.8 + t19 & 0.0 & 4.0 & 1.3 & 4 & 0 & & 139 & 109 & 83 + t20 & 0.5 & 4.0 & 1.3 & 4 & 0 & & 83 & 65 & 49 + t21 & 1.0 & 4.0 & 1.3 & 4 & 0 & & 41 & 32 & 24 + t22 & 1.5 & 4.0 & 1.3 & 4 & 0 & & 15 & 12 & 8.6 + t23 & 1.7 & 4.0 & 1.3 & 4 & 0 & & 5.5 & 4.5 & 3.3 + t24 & 2.0 & 4.0 & 1.3 & 4 & 0 & & 6.3 & 4.9 & 3.8 + t25 & 2.5 & 4.0 & 1.3 & 4 & 0 & & 11 & 8.7 & 6.7 + t26 & 1.7 & 5.4 & 1.3 & 4 & 1 & 0.1 & 7.8 & 6.0 & 4.5 + t27 & 1.7 & 5.4 & 1.3 & 4 & 1 & 0.2 & 24 & 18 & 14 + t28 & 1.7 & 5.4 & 1.3 & 4 & 1 & 0.3 & 45 & 34 & 26 + t29 & 1.7 & 5.4 & 1.3 & 4 & 1 & 0.4 & 98 & 76 & 58 + t30 & 1.7 & 5.4 & 1.3 & 4 & 1 & 0.5 & 140 & 109 & 84 + t31 & 1.7 & 5.4 & 1.3 & 4 & 1 & 0.6 & 199 & 156 & 120 + t32 & 1.7 & 5.4 & 1.3 & 4 & 1 & 0.7 & 247 & 195 & 149 + t33 & 1.7 & 5.4 & 1.3 & 4 & 1 & 0.8 & 297 & 236 & 180 + t34 & 1.7 & 5.4 & 1.3 & 4 & 1 & 0.9 & 330 & 262 & 200 + t35 & & & 1.3 & 4 & 1 & 1.0 & 377 & 301 & 229 + t36 & 1.7 & 5.4 & 1.3 & 4 & 2 & 0.1 & 5.8 & 4.5 & 3.3 + t37 & 1.7 & 5.4 & 1.3 & 4 & 2 & 0.2 & 24 & 18 & 15 + the results discussed in the previous sections provide evidence that the two isolated dwarf galaxies cetus and tucana host a significant population of bsss , whose properties strongly suggest that they are the product of mass exchange in binary systems . we did not find any evidence supporting the hypothesis that part of the objects populating the blue plume are intermediate - age or young ms stars . on the basis of these findings , it appears that both cetus and tucana are similar to ursa minor , draco @xcite , and sculptor @xcite . in the following , we first focus on the comparison of tucana and sculptor , two dwarfs which show close similarities in terms of their sfh and present - day stellar content . sculptor is a well established case of a dsph galaxy hosting two different old populations . @xcite first noted evidence of two rgb bumps in sculptor . @xcite presented a photometric and spectroscopic ( 235 rgb stars ) investigation of this galaxy , showing the presence of strong radial gradients and two stellar populations with different kinematic and chemical properties . moreover , the radial distributions of the red and blue hb components are correlated with the properties of the rgb stars , suggesting that red stars are more centrally concentrated , more metal - rich , and have a higher velocity dispersion , while the blue component has a broader distribution , lower metallicity , and a smaller velocity dispersion . @xcite showed the first evidence of stellar radial gradients in tucana ; after dividing both the hb and the rgb into red and blue components , the number of stars belonging to the red components decreases faster than the corresponding blue sequence with increasing galaxy radius . moreover , the analysis of the properties of rr lyrae variables stars @xcite , the sfh @xcite , and the double rgb bump @xcite strongly support the coexistence of two old ( @xmath140 10 gyr ) populations in tucana , with slightly different ages , metallicities , and spatial distributions . 0.4 cm figure [ fig : scltuc ] compares the ratio of the number of bsss versus red ( asterisks ) and blue ( squares ) hb stars as a function of galactocentric radius for the two galaxies . the values for sculptor are taken from table 2 of @xcite . interestingly , in both cases the bsss are more centrally concentrated than the blue component of the hb , and less concentrated than the red component . we note also that , for each galaxy , the two curves cross at a similar galactocentric distance , @xmath141 @xmath49 . in the case of tucana , this trend is confirmed also for the bsss and the hb stars when compared to red and blue rgb stars ( figure [ fig : grad_rad2 ] , top left and top right , respectively ) . therefore , if we compare the global n@xmath142/n@xmath143 ratio in figure [ fig : grad_rad ] with figure [ fig : scltuc ] and figure [ fig : grad_rad2 ] ( top panels ) , it can be inferred that the radial trends of the bsss are indeed driven by the fast decrease of the red hb stars in the central regions , associated with the younger , more metal - rich component , while the increase at larger radii is driven by the more homogeneously distributed bluest stars . this means that the peculiar radial trend of the candidate bsss when they are normalized to the hb stars is not due to a peculiar distribution of these objects inside the galaxy but only to the existence of a significant metallicity gradient in tucana , a gradient that is well traced by both the red and blue hb stellar populations . the similarities between these two galaxies are particularly intriguing because of the remarkably different environmental conditions : tucana spent most of its life in isolation , while sculptor is a close satellite of the milky way . if all the bsss are descendants of primordial binaries , this suggests that the different environments of the two galaxies did not alter their binary distributions . in contrast , figure [ fig : grad_rad2 ] ( bottom panels ) shows that the properties of different stellar samples in cetus appear more homogeneous . this would be in agreement with other findings , as the properties of the rr lyrae variable stars and of the rgb bump , which indicate a single dominant population characterized by a more uniform distribution of ages and metallicities . interesting features appear when comparing the distribution of the bsss with those of the red and blue stars of both the rgb and hb . we defined these samples as follows . the region selecting the rgb stars was simply split in two , similarly to what has been done for the analysis of the rgb bump in @xcite . the red and blue hb are defined considering only the stars redder than @xmath144 and bluer than @xmath145 in the hb region , therefore mostly avoiding the rr lyrae instability strip . figure [ fig : cumul2 ] compares the distributions of these four samples with the bsss for both tucana and cetus . in the case of cetus , we find that the bsss are significantly less concentrated than the red hb and more concentrated than the blue hb , while they more closely follow the distributions of both of the rgb components . in the case of tucana , the bsss closely follow the distributions of both blue components ( possibly more concentrated in the central regions ) , while they are significantly less concentrated than the red sequences . we have demonstrated that the populations of bss candidates in tucana and cetus are consistent with all predictions of creation from mt - bsss . in fact , it would be unprecedented to have a truncation in star formation without an associated population of mt - bsss . unfortunately , our limited knowledge of the evolution of binary stellar systems prevents us from making accurate predictions of the expected numbers of mt - bsss . in sum , we expect to see mt - bsss in dsph galaxies , and we detect them with consistent properties . the remaining question is whether some fraction of the bss candidates could still be due to newly formed ms stars . the inconsistency with the age - metallicity relationship could be due to infall of a gas cloud with a low metallicity coincident with that of the main population which then produces subsequent star formation . however , the truncation at luminosities corresponding to twice the to mass implies a second unlikely coincidence if some of the bss candidates were true ms stars . while very difficult to rule out such a scenario completely , it is clear that newly formed ms stars are not necessary to explain all of the observations . in this paper , we have presented the analysis of candidate bsss in two isolated dsphs in the periphery of the local group , cetus and tucana . deep hst / acs data allowed us to identify 940 and 1214 bss candidates in cetus and tucana , respectively . different indicators suggest that these are true bsss ; and , thus , there are no young main sequence stars in either galaxy . the first evidence comes from the sfh we derived for both objects @xcite . not allowing for the presence of bsss , then a small population of relatively young ( 3 to 5 gyr old ) stars is detected , with metallicities similar to the oldest most metal - poor stars , and thus , much lower than expected from the age - metallicity relation of the dominant population . the comparison with stellar tracks presented in this paper suggests that the bsss have typical masses @xmath146 , and smaller than @xmath147 , in agreement with the fact that the maximum expected mass of a bss is twice the mass presently evolving at the to ( @xmath148 ) . moreover , the limited spread in metallicity found for the bss populations , perfectly in agreement with that of turn - off and subgiant stars , suggests that the use of stellar evolution models for single stars does not introduce strong biases in the analysis . the analysis of the luminosity functions and radial distributions support the bss interpretation . on one hand , we showed that the luminosity function does not change as a function of radius , and in particular the brightest ( youngest ) stars do not appear more centrally concentrated , as one would expect from a young main sequence . on the other hand , the radial profiles do not show any central peak , as observed in globular clusters due to col - bss stars or as it would be in the case of a young component . thus , the present analysis also suggests that the bsss in cetus and tucana formed from primordial binaries and , as in the case of other dsphs , col - bsss are unlikely to form . this conclusion is strongly supported by the dynamical simulations we performed , taking into account the properties of the dwarfs under scrutiny . these simulations indicate that the observed bss distributions are well reproduced when only mt - bsss are taken into account , while the fit worsens when even a small number of col - bsss ( @xmath149% ) is included in the dynamical simulation . we can therefore safely conclude that it is highly likely that cetus and tucana host a genuine population of bss stars , and we confirm the conclusions presented in previous studies that both galaxies did not experience any star formation episodes in the last 8 gyr . it is particularly interesting that the positions of cetus and tucana in the f@xmath63 vs. @xmath64 plane fit very well in the relation presented by @xcite . this , in turn , could be a powerful instrument to identify young main sequence stars in resolved spheroidal galaxies , discriminating from systems with only old populations from those also hosting younger stars . the authors thank the anonymous referee for the useful comments , yazan momany for sharing the data used in figure [ fig : fm ] , and giuliana fiorentino for useful discussions on the acs . support for this work was provided by nasa through grant go-10515 from the space telescope science institute , which is operated by aura , inc . , under nasa contract nas5 - 26555 , the iac ( grant 310394 ) , the education and science ministry of spain ( grants aya2004 - 06343 , aya2007 - 3e3507 , and aya2010 - 16717 ) . s.c . acknowledges the financial support of inaf through the prin inaf 2009 ( p.i . : r. gratton ) . this research has made use of nasa s astrophysics data system bibliographic services and the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration .

in fluid mechanics , the non - slip boundary condition ( nsbc ) states that fluid at the interface between a fluid and a solid surface is stationary with respect to that surface . surface slip arises when there is non - zero relative motion between the interfacial fluid and the solid surface , in which case the nsbc does not apply . in the vast majority of experiments described historically , the nsbc has been an adequate boundary condition @xcite . however , finite surface slip has been the subject of much recent interest and promises to play a significant role in the current and future development of micro- and nanofluidic devices @xcite . navier @xcite first described surface slip as early as the 1820s and his mathematical definition of the slip length has been widely used , for example in recent reviews @xcite and analyses of slip - dependent flow @xcite . slip is a well - studied phenomenon in non - newtonian fluids such as polymer solutions @xcite . for slip of newtonian fluids ( newtonian slip ) , theory is well developed @xcite , but interpretation of experiments has been much more controversial . newtonian slip is important because technologies can usefully be applied to newtonian fluids , but also because understanding of newtonian slip will provide a more profound understanding of fundamental slip mechanics . the physical mechanisms governing slip are affected by variables that include surface chemistry , shear rate , surface structure , newtonian viscosity , non - newtonian behaviour at the molecular scale and combinations of each of these variables @xcite . nanoengineered superhydrophobic surfaces trap gas between the fluid and the surface , producing large slip lengths of the order @xmath410 @xmath5 m @xcite . experimentally , newtonian slip has been investigated using a variety of methods @xcite . the most widely used and precise technique involves measurement of the drainage forces between closely - spaced surfaces in an atomic force microscope ( afm ) or surface force apparatus ( sfa ) and applying an analysis based on reynolds theory of lubrication @xcite . robust , non - zero values of @xmath0 for newtonian fluids at relatively smooth surfaces are typically of the order 10 - 100 nm @xcite . experiments have not established firm quantitative correlations between slip and major experimental variables , such as contact angle @xcite . recently , it has emerged with some clarity that in afm or sfa experiments , the nsbc can be expected to hold on smooth hydrophilic surfaces @xcite , while finite slip lengths of order 10 - 20 nm have been measured on some hydrophobic surfaces @xcite . additionally , the long - held idea that surface roughness should inhibit slip @xcite is now supported by reasonable experimental evidence @xcite . shear - mode oscillations can be used to drive fluid flow , and such systems pose scientific questions that are distinct from those relating to the study of steady fluid flows . oscillating piezoelectric components are likely to prove very useful in future applications . ultrasonic oscillators , well established as important tools for investigating thin films and bulk fluids , can also be used to probe the interfacial forces that are relevant in the study of slip . kanazawa and gordon @xcite analytically coupled a half - space of newtonian fluid to a piezoelectric quartz crystal microbalance ( qcm ) , in which a solid interface oscillates in shear mode , using the nsbc . further reports have presented experiments @xcite and analysis @xcite describing slip of a newtonian fluid adjacent to an oscillating surface . a relatively large set of slip lengths measured using a qcm was obtained by ellis and hayward @xcite . torsional quartz crystal oscillators operating at khz frequencies were initially developed and used for viscoelasticity measurements @xcite and have been used to measure the complex and viscoelastic behaviour of polymer solutions . more recent work has investigated dispersions of colloidal @xcite and micellar @xcite spheres . the shear rates accessible using torsional oscillators are considered advantageously high for such rheological measurements @xcite . a particular configuration , the composite torsional ultrasonic oscillator , was developed and analysed by robinson et al . @xcite and robinson and smedley @xcite , who also derived an equivalent circuit . recently , the first experimental results investigating newtonian slip at a smooth surface using such a device have been reported @xcite , for oscillations at 40 khz . there has been some confusion relating to fundamental aspects of slip , perhaps best characterised by neto et al . @xcite , who noted that dialogue between the qcm community and other researchers interested in slip has been scarce . in order to address this issue , navier s definition of the slip length is used consistently in the current work . an intuitive , consistent and general methodology for slip measurement at an oscillating surface is presented . analytic inconsistencies in the literature are identified and clarified , especially those concerning an oscillating interface . the composite torsional oscillator is introduced along with a full analysis for measurement of newtonian slip using this device . preliminary experiments are presented to demonstrate how this device is used to probe interfacial forces in oscillatory flow . we discuss the interpretation of measurements and the analytical assumptions on which they are based . finally , methods of slip measurement are compared and the advantages of using the composite torsional oscillator are explained . by definition , viscosity @xmath6 is constant and independent of shear rate throughout a newtonian fluid . the velocity field @xmath7 of an incompressible newtonian fluid of density @xmath8 in the presence of pressure gradient @xmath9 and body force @xmath10 can be determined by solving the navier - stokes equation , @xmath11 microfluidic flows have characteristically low reynolds numbers , so the convective derivative @xmath12 is insignificant when compared with the viscous term @xmath13 . solutions to the navier - stokes equation are dependent on flow geometry and boundary conditions . the relevant solution for the current paper concerns a smooth surface oscillating in shear mode in contact with a fluid half - space . in the absence of significant body forces or pressure differentials , the equation of the shear wave induced in the fluid is @xmath14 where @xmath15 is the fluid velocity in the direction of the surface oscillations , @xmath16 is the cartesian direction perpendicular to the solid surface and @xmath17 . the value of @xmath15 at the interface is @xmath18 , which depends on the fluid - surface boundary condition . the time dependence of @xmath15 and @xmath18 is @xmath19 , where the frequency of oscillation is @xmath20 and @xmath21 is time . this solution is relevant for ultrasonic oscillators . boundary slip occurs when the nsbc is contravened , so that there is finite relative motion between a solid surface and an adjacent fluid . navier @xcite characterised the slip boundary condition by defining the slip length @xmath0 . for flow adjacent to a planar , impermeable solid surface , @xmath22 here @xmath23 is the shear velocity relative to and parallel with the surface and @xmath24 is the value of @xmath23 at @xmath25 . navier s definition of slip has been widely used over a long period and has a strong physical basis relating to viscous force at the interface . the fluid shear rate at the interface that appears in eq . [ eq : gen slip length ] also determines the viscous force , @xmath26 slip therefore determines the shear stress at the interface , which is analogous to friction in character . the coefficient @xmath27 , defined as @xmath28 is finite only when there is slip . using eqs . [ eq : gen slip length ] , [ eq : gen viscous stress ] and [ eq : friction define ] , we find @xmath29 it is noted that the nsbc and boundary slip are , respectively , analogous to the regimes of static and sliding solid - solid friction . extending the analogy , we might expect a wide variety of material and interfacial properties to determine the onset and magnitude of solid - liquid friction . if slip length depends on shear rate , eq . [ eq : gen slip length ] becomes non - linear and rate - dependent formulations of slip ( e.g. @xcite ) become relevant . currently , there is conflicting experimental evidence relating to possible shear rate dependence of @xmath0 @xcite . a different physical interpretation of slip length relates @xmath0 to the distance into the solid over which the tangent to the flow profile at the surface must be extrapolated for the velocity to reach zero . in this context , navier s slip length has correctly been described as a fictional distance @xcite and is labelled in fig . [ fig : slip definition ] . the definition of slip length used in some recent slip studies @xcite , and formalised by ellis and hayward @xcite as ( their notation ) @xmath30 , or @xmath31 is distinct from navier s definition in eq . [ eq : gen slip length ] . ellis and hayward s definition corresponds to extrapolation of the flow profile a distance @xmath32 ( fig . [ fig : slip definition ] ) relative to the surface when slip occurs . the value of @xmath32 is equal to @xmath0 only in the case of a linear velocity flow profile . an often - used schematic diagram @xcite , which uses a linear flow profile to depict slip length , appears to have been an important factor in the emergence of eq . [ eq : b_1 ] . we wish to point out that when the flow of a newtonian fluid is driven by an oscillating surface ( as introduced above ) , the shear rate introduces non - zero phase into the value of @xmath0 , which is defined using @xmath23 ( eq . [ eq : gen slip length ] ) . the existence of this phase highlights the limitations of direct quantitative comparison of measurements of @xmath0 using different experimental techniques . when measuring slip at an oscillating surface , it is both intuitive and useful to use the dimensionless complex parameter @xmath33 . this parameter , which simply quantifies the velocity discontinuity between the solid surface and the first liquid layer , simplifies the algebra describing the effect of damping on the equivalent circuit of a piezoelectric oscillator . it is defined by interfacial properties and does not depend on the flow profile , @xmath34 the velocity of the solid surface is @xmath35 and the fluid oscillation is assumed to be dominated by a sinusoidal oscillation of angular frequency equal to that of the solid surface . under this assumption , @xmath36 is equivalent to parameters introduced by ferrante et al . @xcite and used elsewhere @xcite for modelling of slip at an oscillating surface ; to our knowledge , it has not previously been applied to experimental measurements with full consideration of phase dependence . more generally , @xmath36 can be decomposed into fourier components , an approach that has been used when modelling nanobubble - induced slip @xcite . the nsbc is consistent with the values @xmath37 and @xmath38 . for an oscillating surface , flow is relative to surface motion , so @xmath39 the magnitude and phase of @xmath40 relative to @xmath35 are plotted in fig . [ fig : slip phasor ] , for both slip and non - slip cases . figure [ fig : c ] shows that the relative amplitude of oscillation is uniformly non - zero at the surface when slip is present , but varies significantly near the interface ( @xmath41 ) depending on the values of @xmath42 and @xmath43 . further from the surface ( @xmath44 ) , the relative amplitude tends towards a stable value close to 1 but dependent on @xmath42 . the relative phase follows a similar trend in fig . [ fig : d ] , showing strong dependence on the combination of @xmath42 and @xmath43 near the interface . there is a phase discontinuity at @xmath45 for the cases in which @xmath46 . the plane of discontinuous phase is consistent with a negative slip length ( fig . [ fig : alpha vs b ] ) and could , for example , be interpreted as the edge of a stagnant boundary layer . further from the surface ( @xmath44 ) , the relative phase converges towards the nsbc value , which tends towards zero a long way from the interface . using this flow profile , slip length can be directly related to @xmath36 ( eqs . [ eq : gen fluid wave ] , [ eq : gen slip length ] , [ eq : gen alpha define ] and [ eq : gen osc flow profile ] ) : @xmath47 equation [ eq : gen alpha equals ] is plotted in fig . [ fig : alpha vs b ] using the notation @xmath48 . figure [ fig : alpha_b_amp ] shows that @xmath42 decreases when @xmath49 is within @xmath50 of @xmath51 , so that the slip length @xmath0 is positive . the phase plot ( fig . [ fig : alpha_b_phase ] ) demonstrates that @xmath52 when the fluid oscillates in phase with the surface ( @xmath53 ) . for the nsbc , @xmath54 , so @xmath49 can take any value . the maximum size of the phase shift @xmath43 is limited by slip length magnitude . the complicated nature of the flow profile near the interface supports the use of the parameter @xmath36 , which does not depend on the flow other than at the interface itself , in contrast to @xmath0 . however , calculation of the shear rate using the flow profile is still necessary for derivation of @xmath36 from surface force experiments . the damping force on an oscillating surface is equal and opposite to the viscous force in the adjacent newtonian fluid . when slip occurs , the damping stress is ( eqs . [ eq : gen fluid wave ] , [ eq : gen viscous stress ] and [ eq : gen osc flow profile ] ) , @xmath55 analysis of ultrasonic oscillators is greatly simplified by using the equivalent circuit analogy , in which mechanical components are replaced by their electronic equivalents . we now incorporate the damping stress into the analysis of a mechanical circuit . the stress at the interface ( eq . [ eq : gen interface stress ] ) , which has components both in phase and out of phase with surface velocity , can be written as @xmath56 the impedance of the system is @xmath57 , as previously applied to a torsional oscillator by bergenholtz et al . @xcite , who used the nsbc . this impedance is dependent on the liquid , the surface , and the interaction between them . the real part of the impedance ( @xmath58 ) determines the stress opposing , and in phase with , the surface velocity , @xmath59 this impedance is viscous in nature and dissipative of energy . in an equivalent circuit , velocity is analogous to current and @xmath58 is like a resistance . energy dissipation is determined by the quality factor of the oscillation . the inertial term ( @xmath60 ) gives the stress that is in phase with the acceleration of the surface , @xmath61 an increase in the inertial term is equivalent to the addition of mass to the surface of the oscillator . in the equivalent circuit , the time derivative of current is analogous to acceleration ( modified by a factor of @xmath20 ) , so @xmath60 is an inductance . a change in @xmath60 is measured by considering the change in the period of oscillation . under the nsbc , @xmath58 and @xmath60 are equal for newtonian fluid damping . slip - induced change in the relative sizes of these terms indicates a finite value of @xmath43 . the theory that has been presented implicitly addresses two conceptual pitfalls that have caused confusion and misinterpretation in a range of previous work . the first pitfall concerns the variant slip length @xmath32 , defined in eq . [ eq : b_1 ] , which is only equal to @xmath0 in the limit of a linear flow profile . this limit is generally not applicable for slip measurements . the difference between @xmath0 and @xmath32 is greater than 10% for typical slip measurements using afm , sfa or ultrasonic oscillation methods . we further note that ellis and hayward s definition has been used inconsistently ; both with and without considering a taylor expansion @xcite . navier s slip length @xmath0 should be used at the present time due to its strong physical basis and to promote consistency across experimental studies . the second conceptual pitfall relates to those experimental studies of slip at oscillating surfaces which have imposed a restriction on the phase of slip @xcite , especially the assumption that the slip length @xmath0 is real - valued . it is equivalent to assume that @xmath27 or its reciprocal @xmath62 is real - valued , or more generally , that the damping force is in phase with the surface velocity ( eq . [ eq : b friction ] ) . several lines of reasoning lead us to emphasize that @xmath0 must be allowed to have non - zero phase . we have shown that , because the slip length is defined using the shear rate ( eq . [ eq : gen slip length ] ) , the @xmath16-dependence of the fluid velocity ( a complex exponential , eq . [ eq : gen fluid wave ] ) introduces phase to the value of @xmath0 . the usual ( nsbc ) form of solid - fluid friction force is viscous ( eq . [ eq : gen interface stress ] ) and therefore @xmath63 out of phase with the velocity above an oscillating surface . figure [ fig : alpha vs b ] shows that the phase of @xmath0 in the limit approaching the nsbc is non - zero and that both magnitude and phase must be determined for either @xmath0 or @xmath36 to be fully described . restricting @xmath0 to real values is equivalent to restricting @xmath36 to one degree of freedom in eq . [ eq : gen alpha equals ] . furthermore , experiments @xcite have indicated that a fluid can be expected to oscillate in phase with , or slightly lag , the surface oscillation . neither interpretation is consistent with a real value of @xmath0 . having established that @xmath0 must be considered complex , we address specific arguments made in previous studies in support of a real - valued @xmath0 . firstly , there has been a direct assumption @xcite that the value of @xmath27 ( or @xmath62 ) should not be phase dependent . this assumption has been based on rodahl and kasemo s argument @xcite , which links @xmath27 to the ratio of force to velocity for low reynolds number drag on small bodies ( e.g. stokes drag ) , in which case @xmath27 is a real constant . this argument ignores the fact that important factors in the determination of stokes drag include the contribution of pressure surrounding the particle , the requirement for steady non - oscillatory flow and use of the nsbc . we note that friction and drag parameters are usually real because they are not used to describe oscillatory flow . in conjunction with this point , we note that the term coefficient of friction , which has frequently been used to describe @xmath27 in slip studies , is distinct from the usual parameter of that name , which relates the shear force to the normal force in the case of solid - solid sliding friction . secondly , an argument for phase restriction @xcite has drawn on the correlation between bond strength and bond length , which should determine magnitude and phase of slip respectively . such an argument does not explain why the phase should take one particular value , or why such a correlation between magnitude and phase should not be experimentally investigated . non - zero phase is not significant in previous work relating to lower - frequency oscillations in afm or sfa devices @xcite , in which @xmath64100 @xmath5 m for 39 hz oscillations in water , as opposed to @xmath4500 nm for a 1 mhz oscillator . similarly , periodically patterned microchannels @xcite are adequately described by a real - valued slip length when the ratio of the flow velocity to the pattern wavelength does not exceed @xmath41 khz , corresponding to patterns of extremely short spatial wavelength . the previous discussion of slip is now applied to the composite torsional ultrasonic oscillator ( fig . [ fig : to_schematic ] ) . the damping torque when the specimen rod is submerged in a newtonian fluid was derived by robinson et al . @xcite under the nsbc , using an equivalent circuit analysis . in that derivation , the shear damping stress on a flat surface is integrated over the surface of the specimen rod . appendix a , in this paper , gives the same derivation with slip included and appendix b discusses an assumption that is used . similar derivations have considered damping of other torsional oscillators by viscoelastic fluids under the nsbc , yielding similar functional forms @xcite . for the oscillator of specimen rod radius @xmath65 described by fig . [ fig : to_schematic ] , the total torque for @xmath66 or @xmath67 immersed flat surfaces is @xmath68 where the subscript max refers to the value at the antinode of the standing wave , which has wavelength @xmath69 . when a specimen rod is immersed , the resistance and inductance added to the equivalent circuit are @xmath70 and @xmath71 respectively , so @xmath72 equation [ eq : to t equivalent ] is directly analogous to eq . [ eq : components of interface stress ] for this specific configuration if we substitute notations @xmath73 and @xmath74 , consistent with the resistive and inductive descriptions of these parameters . comparing eqs . [ eq : to nsbc t ] and [ eq : to t equivalent ] and introducing a constant @xmath75 , we find @xmath76 the equivalent circuit reduces to a series lcr circuit @xcite . @xmath71 is therefore related to the change in period upon immersion @xmath77 and the resonant period @xmath78 , using @xmath79 the inverse quality factor of the oscillation @xmath80 is given by the total resistance of the composite oscillator @xmath81 and the total inductance @xmath82 : @xmath83 @xmath70 is calculated by assuming that @xmath82 is dominated by the mechanical inductance ( moment of inertia ) of the rods @xcite , so that ( @xmath84 ) @xmath85 @xmath86 is the change in @xmath80 upon immersion . when piezoelectric elements are incorporated , the equivalent circuit for the torsional oscillator gives the inverse quality factor , @xmath87 where @xmath88 and @xmath89 are voltages applied to the drive and gauge crystals respectively , and @xmath90 is a calibrated constant for a particular oscillator arrangement @xcite . the quality factor is also given by @xmath91 where @xmath78 is the resonant period and @xmath92 and @xmath93 are measurements of the period when @xmath88 is double the resonance value with @xmath89 held constant . by plotting the right hand side of eq . [ eq : to q^-1c ] against the ratio @xmath88/@xmath89 , eq . [ eq : to q^-1 ] can be checked and the value of @xmath90 determined . @xmath90 is independent of specimen rod damping , but it is dependent on the rigid moment of inertia of the specimen rod , so changes if mass is added to the specimen rod . we now consider comparison of a base case ( subscript 0 ) in which the nsbc holds ( @xmath94 , @xmath38 ) , with a case in which slip is present ( subscript 1 ) . in practice , it is difficult to determine whether there is slip in the base case , so @xmath36 can be used to measure the relative amplitude and phase of slip between two cases investigated . @xmath75 remains constant between measurements as long as @xmath95 , @xmath69 , @xmath6 , @xmath8 and @xmath20 are constant , so using eqs . [ eq : r_eta ] , [ eq : to nsbc tor ] and [ eq : to r_eta ] , we define @xmath96 and @xmath97 as @xmath98 and @xmath99 rearranging eqs . [ eq : r_foralpha ] and [ eq : l_eta ] , @xmath36 is calculated directly from measurement : @xmath100 and @xmath101 equations [ eq : alpha measure mag ] and [ eq : alpha measure phase ] are plotted in fig . [ fig : alpha from measurement ] . figure [ fig : alpha amp ] shows that @xmath42 increases with both @xmath96 and @xmath97 , indicating that if mass is decoupled from the surface , the amplitude of fluid oscillation will decrease unless more power is applied . figure [ fig : alpha phase ] shows that to retain the nsbc value of phase ( @xmath38 ) , @xmath96 must decrease if @xmath97 decreases . therefore , an oscillation generating relatively low power will tend to produce a value of @xmath102 ( fluid oscillation leading ) unless there is some decoupling of mass from the surface , and therefore reduced inertia . similarly , if more power is generated , the fluid will lag unless balanced by greater inertia . considered together , fig . [ fig : alpha amp ] and fig . [ fig : alpha phase ] show that a non - zero measurement of either @xmath96 or @xmath97 necessarily indicates a departure from @xmath37 , or @xmath38 or both . the introduction of a slip length is mathematically equivalent to considering a viscoelastic fluid . by inspection of eq . [ eq : gen interface stress ] , and remembering that @xmath103 , we find that instead of using @xmath36 , we can introduce a complex viscosity @xmath104 to replace the newtonian viscosity @xmath6 such that @xmath105 . due to coupling to the decaying wave in the fluid , we have @xmath106 , where @xmath107 is the complex shear modulus . we find that the storage modulus @xmath108 and the loss modulus @xmath109 are given by combinations of the inertial and dissipative terms in the oscillation @xcite , @xmath110 and @xmath111 it is therefore misleading to use the viscoelastic description of storage and loss moduli for the current situation . it is also misleading to describe a newtonian fluid as viscoelastic until there is experimental evidence for viscoelastic behaviour rather than some other cause of slip . some previous qcm work based on fully phase - dependent equivalent circuit analyses @xcite has considered slip in conjunction with a viscoelastic fluid and layering , thereby confusing possible newtonian slip with other effects ; for further effects of layering see the discussion ( below ) . such an approach may be motivated by layering of a qcm electrode substrate , that is not present when using a torsional oscillator . one previous investigation @xcite has noted the equivalence of viscoelasticity and slip in interpretation of slip measurements . to demonstrate investigation of slip using a torsional oscillator , preliminary experiments have been carried out using a 40 khz oscillator , as described by fig . [ fig : to_schematic ] . the fused quartz specimen rod was uniformly ground using al@xmath112o@xmath113 grits to remove surface features of magnitude @xmath1145 @xmath5 m , and polished using cerium oxide powder . polishing debris was removed by sonication in acetone , isopropanol , 10% nitric acid and ethanol . any remaining organic material was removed using piranha solution ( 70 vol% sulphuric acid , 30 vol% hydrogen peroxide ) . the rod was then thoroughly washed in deionised water ( 0.2 @xmath5 m filtered , 18.2 m@xmath115 cm ) and lightly flame - polished to minimize surface roughness . the slip case was configured by coating the specimen rod with a hydrophobic fluoropolymer surfactant ( rs components ) , evenly dispensed from an aerosol can . the base case was retrieved by dissolving the surfactant in isopropanol prior to re - cleaning the rod using piranha solution and washing with deionised water and ethanol . a pre - optically polished , flame - polished piece of fused quartz has rms roughness @xmath116 nm and peak to trough roughness @xmath117 nm , as measured over several widely - distributed areas of size 5 @xmath5m@xmath118 using afm . previously @xcite , scanning electron microscopy ( sem ) revealed peak to trough roughness of 24 @xmath5 m on rods left unpolished following grinding using al@xmath112o@xmath113 , with no significant alteration of the texture following the addition of the surfactant . however , a flame - polished surface coated with the surfactant has additional texture characterised by features of spacing @xmath119 @xmath5 m with rms roughness 24 nm and peak to trough roughness @xmath120 nm , observed using optical microscopy and afm . the advanced static contact angle of deionised water on fused quartz is @xmath121 . the advanced contact angle on the same surface coated with the surfactant is @xmath122 . the texture of the coating causes some hysteresis , and the receded contact angle is @xmath123 . contact angles were measured using the sessile drop method ( cam 200 , ksv instruments ) . prior to each immersion run , the specimen rod was washed in deionised water and dried in air . for the base case , the uncoated rod was also rinsed with ethanol prior to drying . deionised water was used as the test fluid in all experiments . the oscillator was contained in an insulated bell jar standing on a plate maintained at 303 @xmath124 1 k and situated on a large mechanically damped pad usually used for a transmission electron microscope . measurements from recent work are presented alongside previously reported experiments @xcite in table [ tab : results ] and explained in this section . the data points for @xmath96 and @xmath97 are each the result of several immersion events in each of the base and slip cases . the specimen rod was immersed by slowly raising a filled vessel to the rod . a value of @xmath86 or @xmath77 for any particular immersion run is the difference between the dry value and the value when the water was in contact with the node of the standing wave , adjusted to account for any background drift . the fluid meniscus was aligned with the vertical mid - point of a rubber o - ring that was fitted half way along the specimen rod . the o - ring , which fixes the position of the node , was located at a point where surface motion is zero or minimised , and is considered rigid during calibration . each immersion was carried out after the oscillator had been left for at least 90 minutes , so that instrumental drift per second was @xmath125 for @xmath88/@xmath89 and @xmath126 s for @xmath78 , as measured over 10 minute intervals . the standard error of the mean measurement was combined with error ascribed to drift to give the total experimental error in @xmath86 and @xmath77 in each case . uncertainty is dominated by variation between different immersion events , probably due to the extreme thermal and mechanical sensitivity of the instrument . the lesser contribution due to instrumental drift is accounted for by assuming that the mean drift correction is a random error . earlier experiments @xcite have established that uncertainty due to surface tension effects and location of the node is small relative to the standard error in the mean value , because very little damping occurs at or near the node . in order to calculate @xmath86 from measured voltages , the half width of the oscillation was measured directly following each immersion . equations [ eq : to q^-1 ] and [ eq : to q^-1c ] were then used to calculate a precise value of @xmath90 . it has been shown @xcite that a plot having gradient @xmath90 is typically linear ( @xmath127 for all fits ) and independent of temperature , the immersion level and the presence of surfactant . the two former variables are held constant during immersions , but the slip case experiments were calibrated separately from the base case in the most recent experiments . the value of @xmath90 is slightly lower for the slip case , as expected from the first principles analysis @xcite in which @xmath90 is inversely proportional to the oscillator s moment of inertia . the maximum error in @xmath90 , conservatively estimated as the difference between the slip case and the base case , is not significant in comparison with the experimental error of @xmath36 . @xmath90 changes when the torsional oscillator is mechanically altered , but was constant for each individual shear rate presented in fig [ fig : alpha_expt ] . for the nsbc , it is expected that @xmath128/@xmath78 @xcite . this relation was followed to within 5% for the majority of immersions . even accounting for the possibility of slip , immersion events with @xmath11410% deviation clearly indicate a significant instability in the oscillation and such data were discarded . experiments have been performed at shear rate amplitudes @xmath129 between 230 and 6840 s@xmath3 , corresponding to linear displacement amplitudes of 3.2 and 96 nm at the antinode . the linear amplitude of the oscillation is equal to @xmath130/@xmath20 . the shear rate at the surface of the rod is calculated as described by robinson et al . @xcite . for this calculation , rod diameters and masses were measured using a micrometer screw gauge and a mettler ae 200 mass balance , respectively . for each strain rate , intrinsic gauge capacitance @xmath131 was measured to better than @xmath132 by finding the gradient of straight line plots ( @xmath133 ) of the fractional change in resonant period as a function of total gauge capacitance @xcite . the transformer ratio linking the electronic and mechanical parts of the equivalent circuit was calculated using @xmath90 . the alternative calculation ( from first principles ) incorporates large error relating to the electrode separation angle 2@xmath134 ( fig . [ fig : to_schematic ] , @xcite ) . the final uncertainty in shear rate amplitude is @xmath135 . however , the uncertainty in each value relative to the others depends only on @xmath89 , which is precise to better than @xmath136 . the results for the five experiments in table [ tab : results ] are plotted as slip parameters in fig . [ fig : alpha_expt ] . figures [ fig : ex alpha amp ] and [ fig : ex alpha phase ] appear similar because , for all five data points , mean values of @xmath137 and @xmath138 have the same sign . all data lie within experimental uncertainty of the nsbc . for the three data points from ref . @xcite , the consistency of mean values ( @xmath139 , @xmath140 ) previously led to the conclusion that finite slip had likely been observed . although the newer results are not directly comparable due to differences in surface preparation , the mean values are now scattered around @xmath37 , @xmath38 , suggesting that the nsbc is observed with a high degree of accuracy . the most recent experiments have covered a greater range of strain rates and afforded greater precision due to improved stability of the oscillation and better thermal and mechanical isolation . the corresponding mean values of @xmath141 are now less than 30 nm , and the upper limits bounded by experimental uncertainty are less than 60 nm . for each data point , the lower error bar of @xmath142 is at zero because the nsbc falls within experimental uncertainty . because of the phase discontinuity at the nsbc ( fig . [ fig : alpha vs b ] ) , any value of @xmath49 is possible for near - nsbc combinations of the measured variables . previously , qcm measurements @xcite have obtained values of @xmath0 of similar precision , although in that case finite slip lengths of magnitudes between 10 and 80 nm were observed on bare as well as a range of hydrophobic surfaces . as discussed previously , the analysis used to derive these values contains inconsistencies . when considering other results in the literature , we emphasize that comparisons are drawn from data sets taken from measurements of very different flows ( see discussion ) and that the current data are observed over a well - defined , wide range of shear rates . it is therefore remarkable that the data is consistent with emerging trends from high - precision afm studies . namely , slip at a hydrophobic smooth surface occurs on scales @xmath143 nm , if at all ; and surface roughness such as that observed for the slip case in the current work may play a part in suppressing any slip at a hydrophobic surface . in order to discuss the interpretation of experimental slip measurements , we first clarify the terminology used in the discussion . lauga et al.s definitions @xcite of indirect measurement , apparent slip and effective slip are used , although we extend the definition of effective slip to include temporal as well as spatial averaging . true slip is an oft - used term not defined by lauga et al . we define it here as the case when the mean tangential velocity component of fluid is different from that of a solid surface immediately in contact with it . experimental methods are predominantly indirect at present . even the most direct optical methods of slip measurement have been labelled indirect @xcite , in the sense that the velocity of tracers rather than the fluid itself is measured . in the case of the oscillator , the measured damping force is not necessarily consistent with the idealised analysis presented in the theory section . in any surface force measurement , the flow profile is inferred rather than explicitly determined . under these conditions , slip parameters @xmath0 and @xmath36 only describe the slip dynamics in full when rate - independent , true slip occurs and the fluid is entirely newtonian . caution is required when analytically linking postulated mechanisms to apparent or effective slip measurements . for example , it often appears convenient to assume the presence of layering at or near the slip interface @xcite . typically , layers of unknown thickness and interfacial properties only introduce more modelling parameters . as discussed previously , introduced viscoelasticity is indistinguishable from slip . in two studies employing @xmath36 @xcite , multiple viscoelastic layers with multiple boundary conditions were modelled and several parameters were fitted . the resulting magnitudes of @xmath36 ( ranging from @xmath144 to @xmath145 ) were larger than reasonable for a newtonian fluid . at present , mechanisms of apparent slip are poorly characterized , so empirical measurement of slip parameters remains the most important step . newtonian fluids should be considered newtonian right up to the interface to avoid introducing error . sfa and afm techniques are currently the most well developed and widely - employed methods of measuring surface slip , and can reach resolution of better than @xmath146 nm for @xmath0 @xcite . optical methods such as the fluorescent recovery after photobleaching ( frap ) @xcite and particle image velocimetry ( piv ) @xcite techniques allow resolution of @xmath124100 nm , limited by diffraction and tracer particle size . recently , a nano - piv system has reportedly resolved true slip lengths of less than 100 nm with uncertainty less than @xmath12420 nm @xcite . several groups have measured slip using capillary techniques @xcite , but accuracy has been limited by capillary fabrication issues , the difficulty of controlling surface chemistry within a capillary and ( in some cases ) the effects of electro - osmotic flow . a recently - developed technique derives slip from measurement of thermal motion of confined colloidal particles @xcite . a slip length of 18 @xmath124 5 nm has been reported at near - zero shear rate . torsional oscillator experiments presented in the current work confirmed the nsbc to within @xmath147 for @xmath42 , @xmath148 for @xmath43 and to within 60 nm of @xmath2 . we stress that there are no significant limitations on the precision of the oscillator other than control of atmospheric conditions , vibration and electrical noise . further , the technique offers advantages and points of difference when compared with other techniques . a particular advantage of the torsional oscillator is characterization of experimental shear rate . the shear rate amplitude is easily varied by adjusting the applied gauge voltage and can at least cover the range @xmath149 to @xmath150 s@xmath3 . a major reason for use of torsional oscillators in rheological experiments is the large and difficult - to - reproduce range of accessible shear rates @xcite . in contrast , the variable velocity distribution across a qcm causes analytic complications @xcite . in the afm and sfa techniques @xcite , the shear rate varies with respect to surface separation and distance from an axis passing through the centres of the approaching curved surfaces . the shear rate is maximized away from this axis and also depends on the approach or withdrawal rate of the two spheres @xcite . the second major advantage of the torsional oscillator is versatility . surface treatments can easily be applied to the specimen rod , either while remote from the oscillator or attached to it . if the specimen rod is detached , the response of the oscillator is easily recalibrated . electronic voltage and period measurements are captured near - instantaneously , so time - dependent trends can be measured . the oscillator is perfectly suited to measurement of in - situ changes of surface chemistry or other parameters , and experimental precision would be significantly enhanced for such experiments . the oscillator can readily be used at variable temperatures and even variable pressures . although it is best suited to differential measurements ( similarly to optical techniques ) , the zero - slip response can be determined by careful calibration of all the parameters described in the original equivalent circuit analysis @xcite . in contrast , use of qcms and other torsional oscillator designs @xcite for slip measurement presents difficulties with regard to surface modification and incorporating piezoelectric equations into an equivalent circuit analysis . an important distinction between damping - oscillation methods and other techniques is that the shear damping force is measured . sfa and afm methods measure drainage forces caused by pressure in the flow around curved surfaces ; viscous forces are not significant . another difference is that , when using the torsional oscillator , slip occurs over a relatively large fluid - solid interface , so measurement of molecular - scale effects is limited by surface homogeneity . however , measurement of effective slip will give more accuracy and precision than a value derived from a small system . it is not true that measurements of effective slip must take place on a scale @xmath151 in order to have a measurable impact @xcite , because any fundamental slip mechanism should be consistent over the whole interface . regardless , we note that the linear displacement amplitude for the torsional oscillator is of comparable magnitude to the slip length , so molecular - scale interactions are certainly probed using this device . in this paper , we have drawn particular attention to the occurrence of slip at an oscillating interface . the ultrasonic composite torsional oscillator is a versatile instrument for measurement of viscous damping forces , and therefore slip . in order to fully characterize this device for slip measurements , we have revisited fundamental aspects of the slip analysis . the analysis has then been applied at a generalised oscillating surface , followed by the surface of the oscillator s specimen rod in particular . preliminary experiments at shear rates between 230 and 6800 s@xmath3 have probed slip at a hydrophobic surface of rms roughness 24 nm , obtaining resolution better than @xmath147 for @xmath42 , @xmath148 for @xmath43 and within 60 nm of @xmath2 . navier s slip parameter , while retaining the units of length , is best suited to interpretation as a surface force effect . measurement of surface forces has played an important role in developing the understanding of slip to date , while applications are likely to utilise the lubricating effect at a slipping surface . the actual velocity discontinuity at an oscillating interface , as encapsulated by the parameter @xmath36 for oscillating surfaces , is being directly observed with increasing precision using optical methods . the link between a velocity discontinuity and a surface force necessarily utilizes the spatial gradient of an analytical flow profile , unless both types of experiment can be performed on the same system . ongoing experiments will aim to significantly develop the stock of published slip measurements at oscillating surfaces . particular experimental topics will be the influence of shear rate , hydrophobicity and surface texture , while the development and in - situ characterisation of surfaces or fluids with switchable properties will also be a priority . thanks are due to dr . shen chong for assistance with setting up the apparatus described here and used previously @xcite . in this section , the damping torque on a composite torsional ultrasonic oscillator rod , as illustrated in fig . [ fig : to_schematic ] , is calculated . we follow robinson and smedley @xcite and incorporate the case in which there is finite slip , as introduced in the theory section . using the cylindrical polar co - ordinate system defined in fig . [ fig : to_schematic ] , the velocity in the azimuthal ( @xmath152 ) direction is given by @xmath153 where the subscript max refers to the value at @xmath154 and at the antinode of the standing wave , which has wavelength @xmath69 . there is a velocity gradient in the fluid in both the @xmath155- and @xmath16-directions . viscous forces in the fluid exert azimuthal shear stresses on the rod , which dampen the torsional oscillation . forces due to the @xmath16-gradient are neglected ( appendix b ) . the total damping torque is derived by integrating the tangential stress ( eq . [ eq : gen interface stress ] ) along the immersed , curved length of the rod @xmath156 . it is assumed that the fluid surface coincides with a node and the free end is an antinode , so that @xmath157 . the torque is @xmath158 the torque on the oscillating flat end of a rod is derived similarly . the surface velocity at radius @xmath155 from the cylindrical axis is given by @xmath159 and the torque is then integrated over the surface of the free end , so that @xmath160 neglecting edge effects between the flat end and the curved surfaces , the total torque for @xmath161 immersed flat surfaces is @xmath162 the velocity flow field around the torsional rod has gradients in both the @xmath155- ( radial ) and @xmath16- ( axial ) directions . the azimuthal stress due to these gradients are given by and @xmath32 defined in the text , with the shaded area representing solid material . navier s slip length @xmath0 ( eq . [ eq : gen slip length ] ) is only equal to the variable @xmath32 ( eq . [ eq : b_1 ] ) when the velocity profile is linear . the labels @xmath166 and @xmath16 denote cartesian coordinate axes.,width=321 ]

variational methods ( vm ) have long been known as powerful tools in theoretical physics , especially in the context of reduced modeling . as opposed to approximating differential equations , approximating the action functional guarantees that the corresponding model too has a variational structure by definition . this implies that conservation laws are also generally preserved , so a model inherits key features of the original system notwithstanding the reduction @xcite . one of the areas that particularly benefit from this fact is wave theory @xcite , where even crude approximations to an action functional often yield insightful reduced models @xcite . the advantages of vm become especially obvious when one deals with media such as plasmas , which can be nonstationary , inhomogeneous , anisotropic , and exhibit both temporal and spatial dispersion simultaneously @xcite . first - principle differential equations are often unrealistic to handle in this case , whereas vm allow for simple and intuitive modeling . specifically , a wave theory can be constructed as an axiomatic field theory without even specifying the waves nature ( electromagnetic , acoustic , etc . ) . defining the action functional explicitly is needed only to connect the resulting general equations with specific quantities of interest @xcite . for example , this provides a convenient alternative to using maxwell s equations for electromagnetic waves @xcite , which are generally complicated @xcite ( as opposed to the corresponding vm ) even in the geometrical - optics ( go ) limit @xcite . but which exact variational principle does one begin with ? the microscopic least action principle ( lap ) is typically available in the form @xmath0 = 0 $ ] , where @xmath1 describes the wave field , @xmath2 describes the medium , and @xmath3 is a known functional . however , practical applications in the context described above require a different variational principle , where @xmath1 is the _ only _ independent variable , while the medium s degrees of freedom are `` integrated out '' , i.e.,somehow expressed through @xmath1 . if the medium response is adiabatic ( e.g.,if @xmath4 $ ] can be approximated with a local function ) , an effective lap for @xmath1 can be formulated simply as @xmath5 = 0 $ ] , as will also be discussed below . this principle has enjoyed insightful applications , e.g.,in the theory of generalized ponderomotive forces @xcite and nonlinear plasma waves @xcite . however , when the medium response is not adiabatic , the dynamics of @xmath1 becomes dissipative . then , the mentioned effective lap becomes inapplicable , and the problem requires a more subtle approach . this issue has been recently receiving attention in plasma physics @xcite and is also important for advancing variational formulations of plasma and wave kinetics such as in ref . the purpose of the present paper is to find the appropriate modification of the effective lap that would incorporate dissipation under the above assumptions . accommodating dissipative effects within the lap has a long history ( e.g.,cf . @xcite ) both as a formal problem of finding a variational principle for a given equation @xcite and also in modeling physical systems , for example , quantizing radiation in dissipative media @xcite . it is hardly possible to overview the topic here , so the given references are intended as representative but not exhaustive . in any case , the subject of our paper is different from those discussed in literature , particularly , in the following aspects : * we do not address the problem of finding a lap for a system with general dissipation . instead , we consider only emergent dissipation in nonconservative subsystems of systems that are overall conservative . * we avoid the standard approach ( known under many different names @xcite ) where additional independent functions are introduced that serve as lagrange multipliers @xcite . in our formulation , only time - independent lagrange multipliers are used . * we avoid variable transformations that are used ( as in , e.g.,refs . @xcite ) to compensate the loss of the phase space volume caused by dissipation . * in contrast to theories discussed in quantum contexts , our theory is not statistical . instead , we search for a lap that describes deterministic dynamics on a finite time interval for a given initial state of the medium . thus , we can not describe wave fields using the standard fourier representation , and we also have to specify boundary conditions ( bc ) in time , which are often ignored . in fact , it is the attention to these bc that helps accommodate dissipation within a lap . * instead of modeling media as collections of stationary oscillators ( as in , e.g.,refs . @xcite ) , we allow for media that are nonstationary , inhomogeneous , anisotropic , and have both temporal and spatial dispersion simultaneously . moreover , our approach allows treating classical and quantum degrees of freedom on the same footing . * our theory is formulated within the standard calculus of variations . we avoid exotic constructs such as complex actions or fractional derivatives @xcite . feynman integrals , which are used in related quantum theories ( e.g.,see ref . @xcite ) , are also avoided . our main results are as follows . we formulate a variational principle for a dissipative subsystem of an overall - conservative system by introducing constant lagrange multipliers and lagrangians nonlocal in time . we call it the variational principle for projected dynamics ( vppd ) . although the work is originally motivated by needs of the plasma wave theory , our results are applicable to general dynamical systems just as well . in the first part of the paper ( secs . [ sec : notation]-[sec : routh ] ) , we introduce the general approach where @xmath4 $ ] is an arbitrary nonlinear functional and derive an exact effective lap for @xmath1 . in the second part ( sec . [ sec : losc ] ) , we elaborate on applications of this approach to the important special case of linear @xmath4 $ ] , which corresponds to linear media . our formulation is equally applicable to both classical and quantum media . in the third part ( secs . [ sec : asym ] and [ sec : cont ] ) , we discuss various asymptotic approximations of the effective lap for oscillations and waves in linear media , including the quasistatic and go approximation . to our knowledge , this is the _ first variational formulation of dissipative go _ in a general linear medium , which is allowed to be nonstationary , inhomogeneous , nonisotropic , and exhibits both temporal and spatial dispersion simultaneously . also notably , the `` spectral representation '' of the dielectric tensor @xmath6 that enters the go equations is shown to be , strictly speaking , the weyl symbol of the medium dielectric permittivity . ( previously , it has been known that weyl symbols emerge naturally in wave kinetics @xcite . ) this can be considered as an invariant definition of @xmath6 , as opposed to _ definitions used in literature that have been a source of a continuing debate @xcite . our work is also intended as a stepping stone for extending the variational theory of modulational stability in general wave ensembles @xcite to dissipative systems . details will be described in a separate publication . the following notation is used throughout the paper . the symbol @xmath7 denotes definitions , and @xmath8 denotes transposition . for example , for a column vector @xmath1 [ eq . ( [ eq : q ] ) ] , one has @xmath9 , which is a row vector ; i.e.,in this case , @xmath8 merely lowers the index . we assume @xmath10 for any @xmath11 , where @xmath12 is the kronecker delta . ( summation over repeated indices is assumed unless specified otherwise . ) hence , in principle , @xmath13 and @xmath14 could be considered equivalent , but we prefer to retain the distinction because some quantities ( such as coordinates ) are naturally defined with upper indexes , while others ( such as momenta ) are naturally defined with lower indexes . introducing a more fundamental geometrical interpretation will not be needed for our purposes . note also that some indexes will be used _ just _ to introduce new symbols . this applies to all scalars ( e.g.,@xmath15 and @xmath16 ) and to some vectors ( e.g.,@xmath17 , @xmath18 , @xmath19 , @xmath20 ) . we also define @xmath21 . the symbol @xmath22 will denote @xmath23 . the notation `` @xmath24 '' will denote that the equation is obtained by extremizing a corresponding action functional with respect to @xmath11 . the remaining notation is introduced within the main text . finally , the abbreviations used in the text are summarized as follows : + [ cols= " > , < " , ] + we start by introducing the general formalism of lagrangian mechanics . although this formalism is commonly found in textbooks ( e.g.,see ref . @xcite ) , it is essential to restate it here , so we can explicitly reference specific definitions and equations later in the text . consider a lagrangian system @xmath25 characterized by some real coordinate vector @xmath26 the dynamics of @xmath25 on a time interval @xmath27 is governed by the lap . we call this interval @xmath28 , and we also use the symbol @xmath28 to denote its duration , @xmath29 . we assume the action functional @xmath30 to have the form @xmath31 = \int^{t_2}_{t_1 } l^{(q)}(t , q , \dot{q})\,dt,\end{gathered}\ ] ] where @xmath32 is called a lagrangian . then , the standard formulation of the lap is as follows : @xmath33 = 0 , \\ \delta q(t_{1,2 } ) = 0.\label{eq : dqzero2}\end{gathered}\ ] ] [ here and further , the notation @xmath34 means `` both @xmath35 and @xmath36 '' ] . note that eq . ( [ eq : dqzero2 ] ) implies bc @xmath37 where @xmath38 and @xmath39 are some given constants . from eq . ( [ eq : aq ] ) , one obtains @xmath40\,\delta q\,dt,\end{gathered}\ ] ] where we introduced the `` canonical momentum '' @xmath41 with @xmath42 , or , more compactly , @xmath43 . the first term on the right in eq . ( [ eq : aux1 ] ) vanishes because @xmath34 are fixed . hence , the lap leads to the following euler - lagrange equations ( ele ) : @xmath44 this is a second - order equation for the @xmath45-dimensional coordinate @xmath1 , so the @xmath46 bc ( [ eq : bcq ] ) ( two per each component of @xmath1 ) provide just the right number of parameters to define a solution of eq . ( [ eq : ele ] ) . alternatively , one can also approach eq . ( [ eq : ele ] ) as an initial - value problem ; then , the condition on on @xmath47 would need to be replaced with a condition on @xmath48 . it is straightforward to extend this formulation to lagrangians that depend also on higher - order derivatives , but we will not consider such extensions for the sake of brevity . now suppose that @xmath25 is coupled to another lagrangian system @xmath49 , which we call the `` medium '' . suppose also that @xmath49 is characterized by some @xmath50-dimensional coordinate @xmath2 . the dynamics of the resulting system @xmath51 is governed by the lap @xmath0 = 0 $ ] , where @xmath3 is the total action . for clarity , we adopt it in the form @xmath52 , where @xmath53 is the lagrangian given by @xmath54 the assumed bc are @xmath55 which introduce @xmath56 bc similar to eq . ( [ eq : bcq ] ) , namely , @xmath57 like in sec . [ sec : single ] , one then obtains @xmath58 = ( p \cdot \delta q)\,\big|_{t_1}^{t_2 } + ( \eta \cdot \delta \xi ) \,\big|_{t_1}^{t_2}\\ + \int^{t_2}_{t_1 } \big [ ( { \partial}_q l^{(q ) } - \dot{p})\,\delta q + ( { \partial}_\xi l^{(\xi ) } - \dot{\eta})\,\delta \xi\ , \big]\,dt,\end{gathered}\ ] ] where we introduced @xmath59 the first two terms on the right side of eq . ( [ eq : auxa2 ] ) vanish due to eqs . ( [ eq : dqzero ] ) , so the lap yields the following ele : @xmath60 equations ( [ eq : ele21 ] ) and ( [ eq : ele22 ] ) are second - order equations for @xmath61 and @xmath62 , so the mentioned bc provide just enough parameters to define a solution . although eqs . ( [ eq : ele21 ] ) and ( [ eq : ele22 ] ) form a self - consistent system , it is also convenient to approach eq . ( [ eq : ele22 ] ) formally as if @xmath1 were as a prescribed function . in that case , one can , in principle , solve for @xmath2 and express it as some functional @xmath63 $ ] , provided that one is given a @xmath64-dimensional integration constant @xmath65 to determine a solution unambiguously . the resulting system is characterized by @xmath1 alone and is not conservative . our goal is to derive an effective variational principle for this subsystem , or , in other words , for the dynamics of @xmath51 `` in projection '' on @xmath25 . first , consider the variation of @xmath3 evaluated on @xmath66 $ ] : @xmath67 = ( p \cdot \delta q)\ , \big|_{t_1}^{t_2 } + ( \hat{\eta } \cdot \delta \hat{\xi } ) \,\big|^{t_2}_{t_1 } + \delta { \mathscr{f}}^{(q ) } + \delta { \mathscr{f}}^{(\xi)},\notag\\ \delta { \mathscr{f}}^{(q ) } \doteq \int^{t_2}_{t_1 } f^{(q)}(t , q , \dot{q } , \hat{\xi } , \dot{\hat{\xi}})\,\delta q\,dt,\notag\\ \delta { \mathscr{f}}^{(\xi ) } \doteq \int^{t_2}_{t_1 } f^{(\xi)}(t , q , \dot{q } , \hat{\xi } , \dot{\hat{\xi}})\,\delta \xi\,dt.\notag\end{gathered}\ ] ] by definition of @xmath68 , one has @xmath69 . we will also adopt @xmath70 , as usual . then , @xmath67 = ( \hat{\eta } \cdot \delta \hat{\xi})\,\big|^{t_2}_{t_1 } + \delta { \mathscr{f}}^{(q)}.\label{eq : a10}\end{gathered}\ ] ] if @xmath65 is a local function of @xmath71 , then the first term on the right of eq . ( [ eq : a10 ] ) vanishes . in this case , by requiring that @xmath3 is extremized , one arrives at the correct equation ( [ eq : ele21 ] ) . however , such choice of @xmath65 is inconvenient . it is more practical to choose @xmath65 such that it characterizes the _ initial _ state of the medium , namely , @xmath72 [ for example , if @xmath1 is electric field , and @xmath2 is linear polarization , then @xmath73 , where @xmath74 describes free oscillations independent of @xmath1 , and @xmath75 is some kernel ( secs . [ sec : losc]-[sec : cont ] ) . ] in this case , @xmath76 is zero , but @xmath77 is not even though @xmath34 are fixed . hence , @xmath67 = ( \hat{\eta } \cdot \delta \hat{\xi})\big|^{t_2 } + \delta { \mathscr{f}}^{(q)},\label{eq : a1}\end{gathered}\ ] ] where the first term on the right generally does not vanish . [ an exception is the special case when @xmath78 is a local function of @xmath1 ; then , fixing @xmath36 implies fixing @xmath79 too . ] this means that satisfying eq . ( [ eq : ele21 ] ) at fixed @xmath34 is _ not _ sufficient to extremize @xmath80}$ ] . in order to ensure that the action is extremized at fixed @xmath34 , we must constrain the value of @xmath81 separately . the problem of extremizing @xmath3 under an additional constraint @xmath82 is a standard isoperimetric problem that is solved by introducing a lagrange multiplier ( * ? ? ? ii.14 ) . in our case , it is an @xmath50-dimensional multiplier , which we will call @xmath83 . [ the fact that @xmath83 itself needs to be found increases the number of independent variables in the system , which is why @xmath84 can be constrained independently from @xmath34 . ] specifically , the dynamics `` projected '' on @xmath25 is governed by the `` effective '' action @xmath85 \doteq a[q , \hat{\xi } ] + \lambda \cdot \hat{\xi}(t_2),\end{gathered}\ ] ] and the corresponding variational principle , vppd , is @xmath86 = 0 , \quad \delta q(t_{1,2 } ) = 0.\end{gathered}\ ] ] the value of @xmath83 can be readily found in the general case as follows . under the bc ( [ eq : bceff ] ) , one has @xmath87 = \delta { \mathscr{f}}^{(q ) } + ( \eta_t + \lambda ) \cdot \delta \hat{\xi}(t_2).\end{gathered}\ ] ] here , @xmath88 is the value of the functional @xmath89 corresponding to the physical trajectory for given @xmath17 , @xmath18 , @xmath19 , and @xmath90 . the fact that the assumed fixed value of @xmath84 must be consistent also with the assumed @xmath36 implies @xmath91 . hence , to satisfy eq . ( [ eq : bceff ] ) , we adopt @xmath92 this ensures that @xmath93 = 0 $ ] , so the variational principle ( [ eq : bceff ] ) indeed leads to the correct equation ( [ eq : ele21 ] ) . the vppd ( [ eq : bceff ] ) , combined with the solution for @xmath83 given by eq . ( [ eq : aux71 ] ) , is the main conceptual result of this paper . in what follows , we discuss various modifications of this principle and its applications to specific types of problems of practical interest . first , consider the following alternative formulation . we start by introducing the hamiltonian of @xmath94 , @xmath95 as a function of @xmath96 . at fixed @xmath97 , one has @xmath98 where we substituted eqs . ( [ eq : petadef ] ) and ( [ eq : ele22 ] ) . this leads to @xmath99 which are known as hamilton s equations . they are equivalent to the combination of eqs . ( [ eq : petadef ] ) and ( [ eq : ele22 ] ) , as one can also recheck by direct calculation using eq . ( [ eq : hxi ] ) . consider rewriting @xmath3 in terms of these new variables : @xmath100\,dt,\end{gathered}\ ] ] where @xmath101 can be recognized as a routhian ( * ? ? ? 41 ) . then , @xmath102\,dt.\notag\end{gathered}\ ] ] let us assume that @xmath34 are kept fixed , as in sec . [ sec : coupled ] . then , the requirement @xmath103 leads to correct equations [ eqs . ( [ eq : ele21 ] ) and ( [ eq : he ] ) ] if @xmath104 are treated as @xmath105 independent variables . for this reason , the lap for our coupled systems can be reformulated as @xmath106 = 0\end{gathered}\ ] ] with bc still given by eqs . ( [ eq : dqzero ] ) . ( note that , although @xmath107 is now an independent variable too , it yet does _ not _ need to be fixed at the end points . otherwise , the problem would be overdefined . ) accordingly , @xmath108 is defined as @xmath109 \doteq a[q , \hat{\xi } , \hat{\eta } ] + \lambda \cdot \hat{\xi}(t_2),\end{gathered}\ ] ] where @xmath83 is a constant lagrange multiplier . like earlier , one can show that the solution for @xmath83 is given by eq . ( [ eq : aux71 ] ) . let us also introduce a `` symmetrized '' representation @xcite of the above equations that we will need in sec . [ sec : losc ] . notice that the action ( [ eq : apq ] ) can be rewritten as @xmath110 = \frac{1}{2}\,\eta \cdot \xi\,\big|_{t_1}^{t_2 } + { \mathscr{a}}[q , \xi , \eta],\\ { \mathscr{a}}[q , \xi , \eta ] \doteq \int^{t_2}_{t_1 } \big[\lambda^{(q ) } + \frac{1}{2}\,(\eta \cdot \dot{\xi } - \dot{\eta } \cdot \xi ) - h^{(\xi)}\big]\,dt.\label{eq : aux33}\end{gathered}\ ] ] using @xmath111 and eq . ( [ eq : mla ] ) , one gets @xmath112\big|_{t_1}^{t_2 } + \delta { \mathscr{f}}^{(q ) } + \delta { \mathscr{f}}^{(\xi)}.\label{eq : aux34}\end{gathered}\ ] ] as usual , we require @xmath70 . then , we have @xmath50 bc per end point , and the first term in the square brackets vanishes . the remaining bc can be defined , e.g. , as @xmath113 ( summation over indexes is _ not _ assumed here ; also , the symbol @xmath114 denotes that the equality must be satisfied independently at @xmath115 and @xmath116 . ) . with obvious reservations , eq . ( [ eq : aux102 ] ) is also equivalent to @xmath117 this provides precisely @xmath50 bc per each end point , as needed , and eliminates the second term in the square brackets in eq . ( [ eq : aux34 ] ) . hence , the requirement of extremal @xmath118 leads to correct equations [ eqs . ( [ eq : ele21 ] ) and ( [ eq : he ] ) ] . therefore , the lap can be reformulated as @xmath119 = 0\end{gathered}\ ] ] with bc given by eqs . ( [ eq : dqzero2 ] ) and ( [ eq : aux102 ] ) . since the phase space variables @xmath2 and @xmath107 enter @xmath118 on the same footing , we call @xmath118 a phase - space action . for @xmath118 , the analog of the constrained variational principle that we introduced in sec . [ sec : constrained ] is @xmath120 = 0 , \quad \delta q(t_{1,2 } ) = 0,\end{gathered}\ ] ] where the action can be taken in the form @xmath121 \doteq { \mathscr{a}}[q , \hat{\xi } , \hat{\eta } ] + \sum_m \lambda_m ( \hat{\xi}^m/\hat{\eta}_m)\big|^{t_2}.\end{gathered}\ ] ] it is easy to see that this vppd leads to correct equations for @xmath1 provided that @xmath122 satisfy @xmath123 where @xmath124 is the value of @xmath125 corresponding to the physical trajectory for given @xmath17 , @xmath18 , @xmath19 , and @xmath90 . let us also introduce an alternative form of @xmath126 $ ] that we will need further on . instead of fixing @xmath127 , one can fix @xmath128 and @xmath79 independently at the expense of doubling the number of unknown lagrange multipliers . specifically , one can introduce two @xmath50-dimensional lagrange multipliers , @xmath129 and @xmath130 , and define the effective action as follows : @xmath131 \doteq { \mathscr{a}}[q , \hat{\xi } , \hat{\eta } ] + \hat{\eta}(t_2 ) \cdot \lambda^{(\eta ) } + \lambda^{(\xi ) } \cdot \hat{\xi}(t_2).\end{gathered}\ ] ] then , one can show that eq . ( [ eq : aux101 ] ) yields correct equations for @xmath1 provided that @xmath132 where @xmath133 and @xmath134 are the values of @xmath125 and @xmath135 corresponding to the physical trajectory for given @xmath17 , @xmath18 , @xmath19 , and @xmath90 . these equations can also be presented in a more compact form , as discussed in appendix [ app : psp ] . let us now consider , as an example of practical interest , the case when @xmath49 is an ensemble of linear oscillators . in this case , the above phase - space representation facilitates the formulation of a general theory of linear dispersion in a convenient `` quantumlike '' form that is introduced below . note that truly quantum equations are also subsumed under this formulation . as a side note , the special case of _ adiabatic _ oscillators can be described by a simpler machinery discussed in appendix [ app : osc ] . when @xmath49 is an ensemble of linear oscillators , there exists @xcite a linear transformation of the variables @xmath136 to some variables @xmath137 ( roman indexes are used to distinguish them from standard italic indexes that denote vector components ) such that @xmath138 = \int_{t_1}^{t_2 } \lambda^{(\xi)}\,dt}$ ] , @xmath139 where @xmath140 and @xmath141 . ( here , we assume that the underlying oscillators have positive energies for simplicity . negative - energy oscillations @xcite , which are somewhat exotic and typically unstable yet not impossible , could be included by generalizing the notation as explained in ref . the matrix @xmath142 is hermitian and serves as a hamiltonian . also , @xmath143 is a bilinear form of @xmath144 , and @xmath145 is the complex - conjugate bilinear form of @xmath146 . the terms @xmath143 and @xmath145 are determined by the rate at which parameters of the medium evolve in time and vanish when those parameters are time - independent . also @xcite , @xmath147 so the bc ( [ eq : aux102 ] ) can be expressed as @xmath148\big|^{t_{1,2}},\end{gathered}\ ] ] or , equivalently , as @xmath149_{t_{1,2 } } = 0 $ ] @xcite . the corresponding ele are as follows : @xmath150 yet @xmath151 and @xmath152 can be expressed as linear combinations of @xmath144 and @xmath146 . hence , one can also rewrite eqs . ( [ eq : aux64 ] ) in an equivalent complex form @xmath153 where @xmath154 is considered a functional of @xmath144 and @xmath146 . in this sense , @xmath144 and @xmath146 can be formally treated as independent variables . then , it is easy to see that , under the bc ( [ eq : bcpsi ] ) , one has @xmath155\,dt.\end{gathered}\ ] ] accordingly , the ele can be cast as follows : @xmath156 an additional equation is obtained for @xmath146 , but it is simply the adjoint of eq . ( [ eq : aux65 ] ) . ( this is understood because we have only two independent real functions , @xmath151 and @xmath152 , so there can be only one independent complex equation . ) thus , the equation for @xmath146 will be omitted for brevity . the terms @xmath143 and @xmath145 describe effects caused by parametric resonances . ( for example , if @xmath144 were the complex probability amplitude of a free quantum particle , they would describe particle annihilation and production . ) we ignore such effects for simplicity , so @xmath143 and @xmath145 are henceforth omitted . then , the lagrangian of @xmath49 can be represented simply as follows : @xmath157 this leads to a quantumlike equation for @xmath144 , @xmath158 which conserves @xmath159 even though @xmath142 may depend on time . the quantity @xmath159 is understood as the total number of `` quanta '' , or `` particles '' , in @xmath160 @xcite . [ another name of this quantity is the total action of oscillations in @xmath160 . we prefer not to use this term here to avoid confusion , since @xmath161 is also called the action of @xmath160 . ] now , let us introduce the interaction between @xmath49 and @xmath25 . we will describe it by an additional hamiltonian @xmath162 . assuming that the coupling is sufficiently weak , one can model @xmath162 with a function linear in @xmath2 and @xmath107 . since @xmath162 is also real , its representation through @xmath144 and @xmath146 must then have the form @xmath163 where @xmath164 is some complex vector and the minus sign is added for convenience . we will assume @xmath164 to depend on @xmath165 for clarity , but the dependence on derivatives of @xmath1 could also be included . as a side note , this particular model does not conserve @xmath159 . a conservative model , which describes parametric interactions and quantum interactions in particular , is discussed in appendix [ app : q ] . since @xmath142 in this model can depend only on time , it is convenient to eliminate it using a variable transformation . specifically , consider a new variable @xmath166 defined via @xmath167 . let us choose the operator @xmath168 ( a `` propagator '' ) such that @xmath169 and require that @xmath168 be a unit matrix at @xmath115 . this ensures that @xmath168 remains unitary at all times , so it can be cast as @xmath170 , where @xmath171 is some hermitian operator . more specifically , the propagator can be expressed in terms of the following time - ordered exponential @xmath172 : @xmath173.\end{gathered}\ ] ] also importantly , for any @xmath174 , @xmath175 , and @xmath176 , one has @xmath177 in the new variables , the symmetrized action has the form @xmath178 = \int_{t_1}^{t_2 } l\,dt$ ] , where @xmath179 and @xmath180 . the corresponding lap is formulated as @xmath181 = 0\end{gathered}\ ] ] with bc given by eq . ( [ eq : dqzero2 ] ) and @xmath182\big|^{t_{1,2 } } = 0.\end{gathered}\ ] ] it is easy to see that the corresponding ele are @xmath183 now suppose that one has been able to express @xmath166 through @xmath1 . we denote the resulting functional @xmath184 and assume that it is parameterized by some @xmath50 complex constants ( i.e.,@xmath64 real constants ) , say , @xmath185 . hence , we can introduce a vppd for @xmath25 as in sec . [ sec : sym ] with @xmath186 = \int^{t_2}_{t_1}\big\ { \lambda^{(q ) } ( t , q , \dot{q } ) \\ + \frac{1}{2}\,[\gamma^\dag(t , q ) \cdot \hat{\psi } + \hat{\psi}^\dag \cdot \gamma(t , q ) ] \big\}\,dt + { \mathcal{b}},\end{gathered}\ ] ] @xmath187.\label{eq : aux334}\end{gathered}\ ] ] here , @xmath188 is a complex lagrange multiplier introduced such that the action remains real . this representation implies that we seek to fix the values of @xmath189 and @xmath190 independently from @xmath36 by introducing just the right number of lagrange multipliers ( namely , the @xmath64 components of @xmath188 , where the real and imaginary parts are counted separately ) . the values of these multipliers are to be found from the condition that the values of @xmath189 and @xmath190 must be consistent with the value of @xmath36 . it is easily seen that choosing @xmath191 where @xmath192 is the value of @xmath193 corresponding to the physical trajectory for given @xmath194 , leads to a correct equation for @xmath1 [ eq . ( [ eq : qfq ] ) ] . hence , @xmath195 is indeed the sought effective action for the projected dynamics of @xmath25 . by integrating eq . ( [ eq : psiq ] ) , one can express @xmath184 as follows : @xmath196 here , @xmath197 is a linear integral operator , @xmath198\,dt',\end{gathered}\ ] ] which can be understood as the polarizability of @xmath49 with respect to @xmath199 . ( the symbol @xmath200 means `` applied to the expression on the right '' . ) hence , the effective lagrangian can be cast as follows : @xmath201 = \int^{t_2}_{t_1 } \lambda^{(q)}_{0 } \,dt + \xi + { \mathcal{b}},\\ \xi \doteq \text{re } \int^{t_2}_{t_1 } \gamma^\dag \cdot ( \alpha_\gamma \circ \gamma)\,dt.\end{gathered}\ ] ] the difference between @xmath202 and @xmath203 produces an effect similar to that of an external force , while @xmath204 can be described as follows . suppose a `` small - amplitude approximation '' , namely , that the coordinate @xmath1 is chosen such that @xmath205 , where @xmath11 is a small parameter . assuming that the interaction is weak , a typical interaction hamiltonian will then be linear in @xmath1 , so we assume @xmath206 , where @xmath207 is an @xmath208 complex matrix that may depend only on @xmath176 , if at all . ( the coefficient @xmath209 is introduced to simplify the interpretation of the final result . ) this gives @xmath210 where @xmath211 is another @xmath208 complex matrix , namely , @xmath212 ( so that , while @xmath213 is typically slow , @xmath211 is oscillatory ) . then , @xmath214 and @xmath204 can be written as follows : @xmath215 where @xmath216 is an integral operator defined as @xmath217 $ ] . it is convenient to express this operator in terms of a response ( or green s ) function @xmath216 as @xmath218 here , @xmath219 is the heaviside step function , and @xmath216 is a real matrix function given by @xmath220 \notag \\ & = - \text{im } \left[v^\dag(t ) \cdot v(t')\right ] \notag \\ & = ( i/2)\,\big[v^\dag(t ) \cdot v(t ' ) - v^{\rm t}(t ) \cdot v^*(t')\big ] , \label{eq : akern}\end{aligned}\ ] ] which satisfies @xmath221 note that `` re '' and `` i m '' above actually mean real and imaginary parts , _ not _ hermitian and anti - hermitian parts , and , in general , @xmath216 is _ not _ hermitian . as a side remark , the polarizability can be understood as a pseudodifferential operator expressible through @xmath176 and @xmath222 . the fact that the polarizability operator is not necessarily hermitian can be related to the fact that @xmath222 is not hermitian on a finite time interval , as opposed to the whole time axis . a related discussion for spatial ( rather than temporal ) dispersion can be found in refs . @xcite . also note that , using eq . ( [ eq : alphaq ] ) , one can cast @xmath204 as @xmath223 or , equivalently , as @xmath224 where we used eq . ( [ eq : attpsym ] ) and @xmath225 . these formulas will be used below . if @xmath1 is understood as a vector , then @xmath216 serves as a tensor of rank @xmath226 . hence , it is convenient to introduce also the related covariant tensor of rank @xmath227 , i.e.,``lower the index '' of the matrix @xmath228 . we denote such index lowering by underlining . using the corresponding matrix @xmath229 , one can write @xmath230 for any @xmath11 and @xmath231 . accordingly , we can lower the index also in @xmath216 , i.e.,define @xmath232 this allows one to write @xmath233 ( because transposing a vector is the same as lowering its index ) , and also @xmath204 can be expressed without involving @xmath234 : @xmath235 if one does not need to distinguish upper and lower indexes , the distinction between @xmath216 and @xmath236 can be ignored . it is instructive to explain at this point how the ele ( [ eq : qfq ] ) that we anticipate from the general theory can be derived also by a brute - force calculation using @xmath238 = \int_{t_1}^{t_2 } \left[{\partial}_q \lambda^{(q ) } - \dot{p } + \frac{\text{re}\,(v^\dag \cdot \psi_0)}{\sqrt{2}}\right ] \cdot \delta q\,dt \\ + \delta \xi + \delta{\mathcal{b}}.\end{gathered}\ ] ] first , using eq . ( [ eq : xittp ] ) and the fact that @xmath239 for any scalar @xmath11 , one gets @xmath240\,dt\,dt ' \notag\\ & = \frac{1}{2 } \int_{t_1}^{t_2 } \int^{t_2}_{t_1 } \left [ { q^{\rm t}}(t ' ) \cdot { \alpha^{\rm t}}(t , t')\,\theta(t - t ' ) \cdot \delta q(t ) + { q^{\rm t}}(t ) \cdot \alpha(t , t')\,\theta(t - t ' ) \cdot \delta q(t ' ) \right]\,dt\,dt ' \notag\\ & = \frac{1}{2 } \int_{t_1}^{t_2 } \int^{t_2}_{t_1 } \left [ { q^{\rm t}}(t ' ) \cdot { \alpha^{\rm t}}(t , t')\,\theta(t - t ' ) \cdot \delta q(t ) + { q^{\rm t}}(t ' ) \cdot \alpha(t ' , t)\,\theta(t ' - t ) \cdot \delta q(t ) \right]\,dt\,dt ' \notag\\ & = \frac{1}{2 } \int_{t_1}^{t_2 } \left [ \int_{t_1}^{t } { q^{\rm t}}(t ' ) \cdot { \alpha^{\rm t}}(t , t')\,dt ' + \int_{t}^{t_2 } { q^{\rm t}}(t ' ) \cdot \alpha(t ' , t)\,dt ' \right ] \cdot \delta q(t ) \ , dt \notag\\ & \equiv \int_{t_1}^{t_2 } \xi^{(q){\rm t}}(t ) \cdot \delta q(t)\,dt.\label{eq : xiq}\end{aligned}\ ] ] using eq . ( [ eq : attpsym ] ) , we can also rewrite @xmath241 as follows : @xmath242 next , one gets from eq . ( [ eq : aux334 ] ) that @xmath243,\end{gathered}\ ] ] where @xmath244 can be obtained from eq . ( [ eq : fnpsi ] ) ; namely , @xmath245 let us substitute eq . ( [ eq : mupsi ] ) and again use eq . ( [ eq : fnpsi ] ) , now to express @xmath192 . this gives @xmath246 \\ \cdot \frac{1}{\sqrt{2 } } \int^{t_2}_{t_1 } v(t)\cdot \delta q(t)\,dt\bigg\},\end{gathered}\ ] ] or , equivalently , @xmath247}{\sqrt{2 } } + { \mathcal{b}}^{(q){\rm t}}\bigg\}\cdot \delta q(t)\,dt,\label{eq : dbq}\\ { \mathcal{b}}^{(q ) } \doteq \frac{1}{2}\int^{t_2}_{t_1 } \alpha(t , t ' ) \cdot q(t')\,dt',\end{gathered}\ ] ] and we also notice that @xmath248 by combining eqs . ( [ eq : xiq ] ) and ( [ eq : dbq ] ) , we obtain @xmath249}{\sqrt{2}}\\ + { [ \xi^{(q)}(t ) + { \mathcal{b}}^{(q)}(t)]^{\rm t}}\bigg\ } \cdot \delta q(t)\,dt,\end{gathered}\ ] ] and eqs . ( [ eq : xiq ] ) and ( [ eq : biq ] ) give @xmath250 . thus , @xmath251}{\sqrt{2 } } + ( \underline{\alpha } \circ q)(t ) \bigg\}\cdot \delta q(t)\,dt,\end{gathered}\ ] ] so eq . ( [ eq : aeffxib ] ) becomes @xmath252 = \int_{t_1}^{t_2 } \big[{\partial}_q \lambda^{(q ) } - \dot{p } + \underline{\alpha } \circ q \\ + \frac{v^\dag \cdot \psi_0}{\sqrt{2 } } + \frac{\psi_0^\dag \cdot v}{\sqrt{2 } } \big ] \cdot \delta q\,dt.\end{gathered}\ ] ] then , the requirement @xmath253 = 0 $ ] leads to @xmath254 considering the assumed expression for @xmath199 [ eq . ( [ eq : gamma ] ) ] , this agrees with the general ele [ eq . ( [ eq : qfq ] ) ] , as expected . in this section , we make approximations to the vppd and derive reduced theories for certain limiting cases . although the calculations below may seem cumbersome , producing equivalent results by a brute - force reduction of eq . ( [ eq : pg ] ) ( as opposed to the underlying functional ) would still have been harder to do in the general case . consider a representation of the response function in terms of the `` symmetrized '' time coordinates @xmath255 specifically , we define @xmath256 which has the following important property : @xmath257 using eqs . ( [ eq : akern])-([eq : gdeftau ] ) along with eq . ( [ eq : vdef ] ) , one can write @xmath258,\label{eq : avu}\\ { \mathcal{u}}(\bar{t } , \tau ) \doteq u(\bar{t } + \tau/2 , t_1 ) \cdot u^\dag(\bar{t } - \tau/2 , t_1).\end{gathered}\ ] ] since @xmath168 is unitary , one has @xmath259 . using this along with eq . ( [ eq : mcu2 ] ) , we can rewrite @xmath260 as follows : @xmath261 or , in other words , @xmath262,\end{gathered}\ ] ] where @xmath263 is a time - ordered exponential . first , consider a stationary medium , i.e.,a medium where @xmath213 and @xmath142 are constant . then , eq . ( [ eq : mcu3 ] ) gives @xmath264 , and eq . ( [ eq : avu ] ) gives @xmath265 for simplicity , assume the basis in which @xmath142 is diagonal ( `` energy basis '' ) ; i.e.,@xmath266 . then , matrix elements of @xmath267 can be written as @xmath268 provided that the set of @xmath269 is dense enough , the sum over @xmath270 can be replaced with an integral [ @xmath271 : @xmath272 where @xmath273 $ ] . since @xmath274 is thereby a fourier image of @xmath275 ( or vice versa ) , it approaches zero at @xmath276 , at least if @xmath275 is continuous . if the characteristic gap @xmath277 between neighboring eigenfrequencies is nonzero ( and @xmath278 ) , the same conclusion applies at @xmath279 , when the spectrum discreteness is inessential . thus , as long as larger @xmath280 are not of interest ( as will be assumed below ) , there is no need to distinguish continuous and discrete spectra . the effect @xmath281 is known as phase mixing . although introduced here for a stationary medium , it extends also to media evolving at large enough time scales @xmath282 , namely , @xmath283 where @xmath284 is the characteristic eigenvalue of @xmath142 . the former requirement ensures that @xmath213 and @xmath142 be approximately constant on the time scale @xmath285 . the latter requirement allows approximating eq . ( [ eq : mcu3 ] ) much like in the case of a stationary medium : @xmath286.\end{gathered}\ ] ] our intention is to discuss approximations to the vppd that become possible in this limit . in order to do that , let us introduce some definitions first . let us introduce an auxiliary matrix function @xmath287 where @xmath288 is assumed real , @xmath267 is given by eq . ( [ eq : gdeftau ] ) , and @xmath289 is some constant . ( in later sections , we will treat @xmath289 as a function of @xmath290 , but this distinction is not important here . ) let us also introduce the hermitian and anti - hermitian parts of @xmath291 , namely , @xmath292 such that @xmath293 . ( in this notation , both @xmath294 and @xmath295 are hermitian , so @xmath296 and @xmath297 are anti - hermitian . ) as shown in appendix [ app : alphas ] , @xmath298 provided efficient phase mixing , the function @xmath291 has a well - defined limit at large @xmath289 . we call this limit @xmath299 ; i.e.,@xmath300 the function @xmath299 is recognized as the laplace image of @xmath267 evaluated at real @xmath288 and also , more generally , as the _ weyl image of the polarizability operator _ @xcite . like for @xmath291 , the hermitian and anti - hermitian parts of @xmath301 satisfy @xmath302 approximate formulas for @xmath291 and @xmath301 in the limit ( [ eq : vro ] ) can be found in appendix [ app : alphas ] . it is also shown there that @xmath303 and @xmath304 satisfy the kramers - kronig relations . first , let us consider the case when @xmath1 and , possibly , the medium evolve on time scales @xmath282 that satisfy @xmath305 ( to simplify the notation , we will assume all the three small parameters to be comparable to each other , but this is not essential . ) below , we derive an asymptotic expression for @xmath195 using @xmath306 as a small parameter . to approximate the term @xmath204 as given by eq . ( [ eq : xisgn ] ) , let us first map the integration variables @xmath307 in the double integral to @xmath308 using eqs . ( [ eq : ttau ] ) . the integration domain is then mapped as shown in figs . [ fig : domain](a ) and ( b ) , namely , to @xmath309 , where @xmath310 is a piecewise - linear function . this gives @xmath311 where tr stands for `` trace '' , and @xmath312 is a matrix given by @xmath313 to the lowest ( zeroth ) order in @xmath306 , one has @xmath314 the function @xmath315 [ given by eq . ( [ eq : aproph ] ) at @xmath316 can be approximated with @xmath317 everywhere except for an @xmath318 part of the integration domain . neglecting this @xmath318 correction , we get @xmath319 \cdot q(\bar{t})\,d\bar{t},\end{gathered}\ ] ] which corresponds to the approximation illustrated in fig . [ fig : domain](c ) . we have omitted the index @xmath320 in @xmath317 because , in the zero - frequency limit , the polarizability s weyl symbol ( [ eq : aw ] ) becomes hermitian ( appendix [ app : alphas ] ) ; i.e.,@xmath321 . specifically , @xmath322 as seen from eq . ( [ eq : zerow ] ) in the limit when the oscillating term is eliminated by phase mixing . using eq . ( [ eq : fnpsi ] ) , let us express @xmath184 as @xmath324 the function @xmath171 is such that , at zero @xmath306 , it has a constant derivative equal to @xmath142 . hence , at small nonvanishing @xmath306 , one can expect @xmath325 to be slow and close to @xmath142 . then , the integrand in eq . ( [ eq : intpsi0 ] ) is a rapidly oscillating exponent times a slow function @xmath326 . such integral is entirely determined by values of the integrand at the ends of the integration domain and is of order @xmath327 @xcite . this gives the following estimate for the term @xmath323 in eq . ( [ eq : aux334 ] ) : @xmath328 assuming the effect of @xmath194 vanishes due to phase mixing . hence , @xmath329 . by combining the above results and neglecting corrections @xmath318 and smaller , one obtains @xmath121 = \int^{t_2}_{t_1 } [ \lambda^{(q)}_{0 } ( t , q , \dot{q } ) - \phi_0(t , q)]\ , dt,\\ \phi_0(t , q ) \doteq - \frac{1}{2}\,[\underline{\alpha}_0(t ) \cdot q ] \cdot q.\label{eq : udef}\end{gathered}\ ] ] clearly , the corresponding ele is as follows : @xmath330 these results indicate that the interaction of a quasistatic system with a linear medium @xmath49 is equivalent to the interaction with an effective potential @xmath331 given by eq . ( [ eq : udef ] ) . the symmetric matrix @xmath332 represents the covariant form of @xmath333 given by eq . ( [ eq : alpha0est ] ) and serves as the polarizability of @xmath49 with respect to @xmath1 . also note that , for positive - definite hamiltonian @xmath142 , the polarizability @xmath333 is positive - definite too , and thus @xmath334 . here , we derive an asymptotic variational principle for the case when @xmath1 oscillates rapidly compared to the evolution of the medium . the method is akin to whitham s averaging @xcite , but the averaging procedure is justified differently . specifically , this is done as follows . suppose @xmath203 is bilinear in @xmath335 . then , eq . ( [ eq : pg ] ) is linear and allows complex solutions . let us consider a complex solution @xmath336 of the form @xmath337 , where @xmath338 is a complex vector and @xmath339 is a real phase . provided that the time scale @xmath282 at which the medium evolves is large enough , in principle it is possible to construct asymptotic @xmath338 and @xmath339 such that @xmath340 is slow compared to @xmath339 at all @xmath176 ; i.e.,@xmath1 is quasimonochromatic . ( the effect of @xmath194 will be assumed negligible , like in sec . [ sec : qs ] . ) more specifically , harmonics of @xmath288 remain suppressed with exponential accuracy in the small parameter @xmath341 . ( if the system supports more than one mode , we assume that only one mode is excited , and the spectral distance to other modes is @xmath342 . ) here , we assume that @xmath282 is also the time scale of the evolution of @xmath338 and @xmath288 , and @xmath343 . clearly , if @xmath337 is a solution , then @xmath344 a solution too , and so is @xmath345 . since the latter is also real , it can be adopted as an _ exponentially accurate asymptotic model _ of a realizable physical process , which we can also express as follows : @xmath346.\end{gathered}\ ] ] let us substitute this model into eq . ( [ eq : lgamma ] ) . as usual @xcite , the rapidly oscillating terms that couple @xmath336 and @xmath347 can be eliminated , which is also expected from the definition of @xmath336 and @xmath347 . ( such `` @xmath339-averaging '' is denoted below with angular brackets , @xmath348 . ) this implies that the medium responds to @xmath336 and @xmath347 independently , as if there were two independent subsystems @xmath49 that interact with @xmath336 and @xmath347 , respectively . the corresponding responses will be denoted @xmath349 . hence , instead of the lagrange multipliers @xmath188 and @xmath350 that we introduced before , now we must introduce twice as much lagrange multipliers . we denote them @xmath351 and @xmath352 . hence , like in sec . [ sec : pdaeff ] , the effective action ( [ eq : lgq ] ) can be represented as follows : @xmath353 @xmath354 [ where we used eq . ( [ eq : xisgn ] ) ] , and @xmath355 , where @xmath356,\end{gathered}\ ] ] assuming the notation @xmath357 and @xmath358 . also like in sec . [ sec : pdaeff ] , one finds @xmath359 [ the quantity @xmath351 must not be confused with @xmath188 given by eq . ( [ eq : mupsi ] ) . in particular , note the extra factor @xmath360 . ] as usual , the functions @xmath339 , @xmath338 , and @xmath361 are considered independent . then , @xmath362\,dt.\end{gathered}\ ] ] we seek to approximate @xmath363 to the zeroth order in the small parameter @xmath364 where @xmath285 is the phase - mixing time ( sec . [ sec : pm ] ) . ( as in sec . [ sec : qs ] , we adopt for clarity that all the three parameters are comparable to each other . ) we assume @xmath365 and also the standard ordering for @xmath299 ( sec . [ sec : weyl ] ) , which will be be verified _ a posteriori _ : @xmath366 then , the individual terms on the right side of eq . ( [ eq : aeffgo1 ] ) can be calculated as follows . since @xmath203 is a local function of @xmath335 , the function @xmath368 can not depend on @xmath339 other than through @xmath288 . hence , @xmath369\ , dt,\end{gathered}\ ] ] where we used @xmath370 and integrated by parts with @xmath371 . since @xmath372 while @xmath373 , @xmath374 , and @xmath375 , all the three terms in the square brackets are @xmath376 . hence , within the accuracy of our model , it is enough to approximate @xmath367 to the zeroth order in @xmath377 . since @xmath203 was assumed to be a bilinear function of @xmath335 , the functional @xmath367 then has the following form : @xmath378\,dt,\end{gathered}\ ] ] where the factor @xmath360 is introduced for convenience , and @xmath379 is some matrix that is determined by the local frequency @xmath380 and , in nonstationary systems , also @xmath176 . [ below , we also refer to this matrix using a shortened notation @xmath381 . ] the anti - hermitian part of this matrix provides zero contribution to @xmath382}$ ] , so we can assume that @xmath381 is hermitian for all real @xmath288 without loss of generality . then , `` re '' in eq . ( [ eq : ad ] ) can be omitted . let us now approximate @xmath383 given by eq . ( [ eq : xiqsgn ] ) . first , note that , due to eq . ( [ eq : ttau ] ) , we have @xmath384 hence , @xmath385 where the second term on the right is estimated as @xmath386 = o(\epsilon_\omega^2 ) \ll 1.\end{gathered}\ ] ] this gives @xmath387 omitting @xmath388 , one can rewrite eq . ( [ eq : xiqsgn ] ) as follows : @xmath389 here , the matrix @xmath143 is given by @xmath390 since the integrand in eq . ( [ eq : xiaux ] ) does not depend on @xmath339 explicitly , the correction @xmath391 can be omitted for the same reason as in sec . [ sec : aq ] . hence , we switch from integration in @xmath307 to the integration in @xmath308 and get @xmath392 note that @xmath393 ( sec . [ sec : gobasic ] ) . this function can be approximated with @xmath303 everywhere except for an @xmath391 part of the integration domain . since corrections @xmath391 in @xmath383 are deemed negligible , we obtain @xmath394 as a side note , one can also rewrite eq . ( [ eq : xixi ] ) as @xmath395 , so @xmath383 is understood as a contribution of an effective potential @xmath396 that the subsystem @xmath25 experiences due to the interaction with the subsystem @xmath49 . the potential @xmath397 represents the ponderomotive energy ( or , more precisely , the ponderomotive term in the lagrangian ) of @xmath49 . in the context of particle dynamics with hermitian @xmath301 , the obtained linear relation between @xmath397 and @xmath301 is also known as the @xmath398-@xmath399 theorem @xcite . since @xmath383 has the same form as @xmath367 given by eq . ( [ eq : ad ] ) , let us combine these two terms into a `` dressed action '' @xmath400 . for that , let us introduce @xmath401 , whose hermitian and anti - hermitian parts are @xmath402 then , @xmath403 this can also be written as @xmath404 , where @xmath405 is understood as the dressed lagrangian of @xmath25 . accordingly , one gets @xmath406\ , dt,\end{gathered}\ ] ] where @xmath407 is understood as the number of quanta of the oscillation mode , or the action of this mode @xcite [ again , not to be confused with the action such as @xmath195 ] . a straightforward calculation ( appendix [ app : db ] ) gives @xmath410.\label{eq : dy}\end{gathered}\ ] ] using @xmath411 , one further gets @xmath412 { \partial}_\omega \big\ } \cdot \widetilde{\alpha}_{\omega , a}.\end{gathered}\ ] ] note that @xmath413 ( sec . [ sec : gobasic ] ) and @xmath414 [ eq . ( [ eq : dgosc ] ) ] . since we need to calculate @xmath415 to the zeroth order in @xmath377 , the contribution of @xmath416 can then be neglected . hence , @xmath417\,\delta\theta(\bar{t})\,d\bar{t}.\end{gathered}\ ] ] similarly , one obtains that @xmath418 , so @xmath419 is entirely negligible . in summary , substituting this result into eq . ( [ eq : aux3001 ] ) gives @xmath420 where we omitted corrections @xmath391 and , for the same reason , replaced @xmath295 with @xmath304 . using @xmath421 along with eqs . ( [ eq : dadr ] ) and ( [ eq : dbgo ] ) and also requiring @xmath422 , we get the following set of ele : @xmath423 plus the conjugate equation for @xmath361 . note that eq . ( [ eq : it ] ) implies @xmath424 , which is in agreement with the original assumption ( [ eq : alphaord ] ) . it is also instructive to rewrite these in terms of the unit polarization vector @xmath425 and the scalar amplitude @xmath11 , assuming @xmath426 : @xmath427 = -2({\mathsf{e}}_\omega^\dag\cdot { \mathsf{d}}_{\omega , a}\cdot { \mathsf{e}}_\omega)\,a^2.\label{eq : ampleq}\end{gathered}\ ] ] equation ( [ eq : dhe ] ) can be understood as a local dispersion relation , which determines @xmath428 and the local frequency @xmath429 . equation ( [ eq : ampleq ] ) can be understood as an amplitude equation . in a dissipationless medium , which we _ define _ as a medium with @xmath430 , this equation becomes @xmath431 , which is known as the action conservation theorem @xcite . for example , consider a one - dimensional stationary system . in this case , eq . ( [ eq : dhe ] ) becomes @xmath432 and the amplitude equation gives @xmath433 , where @xmath434 ( this quantity is not to be confused with @xmath164 that was introduced in sec . [ sec : co ] . ) one hence obtains that @xmath1 has a well - defined complex frequency ; i.e.,@xmath435 , where @xmath436 . within the accuracy of our theory , this can be understood as a solution of the complex dispersion relation @xmath437 as seen from the fact that @xmath438 combined with eqs . ( [ eq : dh ] ) and ( [ eq : gammainst ] ) . hence , @xmath439 can be understood as the complex dispersion function . although the above results were derived for discrete oscillators , they can be readily extended to waves in continuous media . a continuous medium is understood as a system where the scalar product can be expressed as an integral over some continuous space @xmath440 ( we assume that @xmath440 is euclidean for clarity , but see refs . @xcite for a more general case ) , specifically as @xmath441 here , @xmath442 , and @xmath443 and @xmath444 are fields of some finite dimension @xmath445 . for example , they can be electromagnetic fields in physical space ; then @xmath446 . ( we use bold symbols to distinguish @xmath447-dimensional coordinates and @xmath445-dimensional vectors from @xmath45- and @xmath50-dimensional coordinates and vectors that we introduced earlier , since @xmath45 and @xmath50 are infinite in the continuous limit . ) accordingly , the polarizability operator can be rewritten as @xmath448 the symmetrized kernel is introduced as @xmath449 , where @xmath290 and @xmath280 are defined as usual [ eq . ( [ eq : ttau ] ) ] , and @xmath450 we also introduce the corresponding weyl symbol ( the term is used with the same reservations as earlier @xcite ) , @xmath451 the modification of the remaining notation is obvious , so it will not be presented in detail . let us consider a special case of practical interest where the medium is weakly inhomogeneous in both time and space . ( anisotropy and general dispersion , including both temporal and spatial dispersion , are implied . ) unlike in sec . [ sec : fastglobal ] , where all oscillators were assumed to have the same @xmath288 at given @xmath176 , we now allow the frequency to be spatially inhomogeneous . specifically , we adopt @xmath452\end{gathered}\ ] ] and assume that the frequency and the wave vector , @xmath453 are slow functions of @xmath454 . explicitly , we require @xmath455 where @xmath377 is defined in eq . ( [ eq : epsilonom ] ) , and @xmath456 is the analogous parameter that involves spatial scales instead of temporal scales . then , one obtains @xmath457 , where @xmath458 , and @xmath459 is the lagrangian density given by @xmath460 this leads to @xcite @xmath461 + \frac{1}{4}\,[{{\boldsymbol{\rm q}}}^\dag \cdot { { \boldsymbol{\rm { \mathsf{d}}}}}_{(\omega , { { \boldsymbol{\rm k } } } ) , h } ] \cdot \delta { { \boldsymbol{\rm q}}}\bigg\},\end{gathered}\ ] ] where @xmath462 and @xmath463 . the expression for @xmath464 is obtained like in sec . [ sec : fastglobal ] and can be written as @xmath465\,\delta \theta.\end{gathered}\ ] ] hence , one obtains the following ele : @xmath466,\label{ref : act}\\ \delta { { \boldsymbol{\rm q}}}^\dag & : \quad { { \boldsymbol{\rm { \mathsf{d}}}}}_{(\omega,{{\boldsymbol{\rm k } } } ) , h}\cdot { { \boldsymbol{\rm q } } } = 0,\label{eq : qdisp}\end{aligned}\ ] ] plus the conjugate equation for @xmath467 . it is also instructive to rewrite these equations as @xmath468 here , @xmath469 is the unit polarization vector . we also added an equation for @xmath470 that flows from its definition [ eq . ( [ eq : defwk ] ) ] , introduced @xmath471 , which represents the group velocity , and also introduced @xmath472 which represents the local dissipation rate . equations ( [ eq : dhe2])-([eq : gammago ] ) form a complete set of equations that describe dissipative go waves in inhomogeneous nonstationary media . they are in agreement with results of ref . @xcite , where dissipative effects were described as an addition to the variational formulation rather than as a part of it . ( the variational formulation of _ dissipationless _ wave dynamics is also discussed in refs . @xcite . ) applications of these general equations to electromagnetic waves are discussed in detail in ref . @xcite . here , we only point out that one can choose @xmath473 to be the complex amplitude of the electric field , @xmath474 ; then , @xmath475 the second term in eq . ( [ eq : lem ] ) is understood as @xmath476 , where we used faraday s law @xmath477 to express the complex amplitude of the magnetic field @xmath478 , and @xmath479 is the speed of light . also , @xmath6 is the dielectric tensor in the spectral representation ( or , more specifically , the weyl symbol of the dielectric permittivity operator ) , and the indexes @xmath320 and @xmath3 denote its hermitian and anti - hermitian parts , as usual . accordingly , eq . ( [ ref : act ] ) can be understood as a restatement of poynting s theorem @xcite . as a side note , an even simpler variational derivation of the results reported in this section is possible by means of the weyl calculus combined with a technique _ la _ ref . we leave details to future publications . in summary , we have formulated a variational principle for a dissipative subsystem of an overall - conservative system by introducing constant lagrange multipliers and lagrangians nonlocal in time . we call it the variational principle for projected dynamics , or vppd . we have also elaborated on applications of the vppd to the special case of linear systems , particularly in the context of wave propagation in general linear media . the focus was on how the variational formulation helps in deriving reduced models , such as the quasistatic and go models . in particular , we have proposed a variational formulation of dissipative go in a linear medium that is inhomogeneous , nonstationary , nonisotropic , and exhibits both temporal and spatial dispersion simultaneously . the `` spectral representation '' of the dielectric tensor @xmath6 that enters go equations is shown to be , strictly speaking , the weyl symbol of the medium dielectric permittivity . this can be considered as an invariant definition of @xmath6 , as opposed to _ ad hoc _ definitions that are commonly adopted in literature and have been a source of a continuing debate . our work is also intended as a stepping stone for extending the variational theory of modulational stability in general wave ensembles @xcite to dissipative systems . details will be described in a separate publication . the authors thank d. bnisti for stimulating discussions . the work was supported by the nnsa ssaa program through doe research grant no . de - na0002948 , by the u.s . doe through contract no . de - ac02 - 09ch11466 , and by the u.s . dod ndseg fellowship through contract no . as a side note , equations of sec . [ sec : sym ] can be expressed in a more compact form in terms of the phase space coordinate @xmath480 and the poincar two - form @xmath481 , @xmath482 \eta \end{array } \right ) , \quad \omega_{\alpha\beta } \doteq \left ( \begin{array}{c @{\quad } c } 0 & - \mathbb{i}\\[3pt ] \mathbb{i } & 0 \end{array } \right),\end{gathered}\ ] ] where @xmath483 is a @xmath484 unit matrix . in particular , one can cast eq . ( [ eq : aux33 ] ) as @xmath485 = \int^{t_2}_{t_1 } \big[\lambda^{(q ) } + \frac{1}{2}\,\omega_{\alpha\beta } z^\alpha \dot{z}^\beta - h^{(z)}\big]\,dt\end{gathered}\ ] ] ( greek indexes span from 1 to @xmath64 ) and eq . ( [ eq : aux34 ] ) as @xmath486 = \big[p \cdot \delta q + \frac{1}{2}\,\omega_{\alpha\beta}z^\alpha\,\delta z^\beta\big]\,\big|_{t_1}^{t_2 } + \delta { \mathscr{f}}^{(q ) } + \delta { \mathscr{f}}^{(z ) } , \notag\\ \delta { \mathscr{f}}^{(z ) } \doteq \int^{t_2}_{t_1}(\omega_{\alpha\beta}\dot{z}^\beta - { \partial}_\alpha h^{(z)})\,\delta z^\alpha\,dt,\end{gathered}\ ] ] where @xmath487 and @xmath488 . accordingly , hamilton s equations ( [ eq : he ] ) are cast as follows : @xmath489 also , eq . ( [ eq : aeffsr ] ) can be represented as follows : @xmath490 = { \mathscr{a}}[q , \hat{z } ] + \frac{1}{2}\,\omega_{\alpha\beta } \hat{z}^\alpha(t_2 ) z^\beta_t,\end{gathered}\ ] ] where @xmath491 \eta_t \end{array } \right).\end{gathered}\ ] ] on physical trajectories one has @xmath492 . since @xmath481 is antisymmetric , this implies that the second term on the right side of eq . ( [ eq : auxx ] ) vanishes on such trajectories . here , we discuss an instructive special case , namely , the situation when the hamiltonian @xmath493 in eq . ( [ eq : apq ] ) does not depend on @xmath2 explicitly . in this case , @xmath107 is conserved and can be treated as a constant parameter , which we denote @xmath494 . hence , one can write @xmath495 = \int_{t_1}^{t_2 } [ \lambda^{(q)}(t , q , \dot{q } ) - h^{(\xi)}(t , q , \dot{q } ; j)]\,dt.\end{gathered}\ ] ] an alternative derivation of the same result , also known as the routh reduction can be found , e.g.,in ref . @xcite . in particular , eq . ( [ eq : aeffl ] ) can be applied @xcite when @xmath49 is an _ adiabatic oscillator _ in which @xmath136 are the angle - action variables . ( then , @xmath494 is called an adiabatic invariant @xcite . ) in the special case when @xmath49 is a _ linear _ oscillator , one can further write @xmath496 , where @xmath497 is the corresponding canonical frequency ( i.e.,such that @xmath498 ) , which is independent of @xmath494 but may depend on @xmath97 . then , @xmath499 = \int_{t_1}^{t_2 } [ \lambda^{(q)}(t , q , \dot{q } ) - j \cdot \omega(t , q , \dot{q})]\,dt.\end{gathered}\ ] ] for example , one can use this to describe a charged particle in a weakly inhomogeneous field @xmath500 . in this case , @xmath497 is the gyrofrequency , and @xmath494 is proportional to the particle magnetic moment @xmath501 , such that @xmath502 @xcite . then , the last term in eq . ( [ eq : leff ] ) is nothing but the effective potential responsible for the diamagnetic force . in addition to the linear coupling discussed in sec . [ sec : co ] , it is instructive to consider parametric interactions , particularly because they subsume quantum interactions and thus can be understood as most general . specifically , let us consider @xmath503 where @xmath504 is some hermitian operator . ( we assumed that @xmath162 is independent of derivatives of @xmath1 for brevity , but the corresponding generalization is straightforward . ) the total lagrangian in this case can be written as @xmath505 and the corresponding ele are @xmath506 notably , eq . ( [ eq : psiq2 ] ) conserves @xmath507 even in the presence of coupling , in contrast with eq . ( [ eq : psiq ] ) . for projected dynamics , the effective action @xmath195 is introduced just like in sec . [ sec : pdaeff ] . when the solution for @xmath166 and @xmath508 are substituted into eq . ( [ eq : aux103 ] ) , the second term and the third term cancel each other due to eq . ( [ eq : psiq2 ] ) , so one ends up with @xmath121 = \int^{t_2}_{t_1 } \lambda^{(q ) } ( t , q , \dot{q})\,dt + { \mathcal{b } } , \\ { \mathcal{b } } \doteq \frac{i}{2}\,\big[\nu^\dag \cdot \hat{\psi}(t_2 ) - \hat{\psi}^\dag(t_2 ) \cdot \nu\big].\end{gathered}\ ] ] let us show how a correct ele for @xmath1 is obtained from here by a straightforward calculation . first , @xmath252 = \int_{t_1}^{t_2 } [ { \partial}_q \lambda^{(q ) } - \dot{p}]\,\delta q\,dt + \delta{\mathcal{b}},\label{eq : daeffq}\\ \delta { \mathcal{b } } = \frac{i}{2}\,\big[\nu^\dag \cdot \delta \hat{\psi}(t_2 ) - \delta \hat{\psi}^\dag(t_2 ) \cdot \nu\big].\end{gathered}\ ] ] in order to calculate @xmath244 , let us use the fact that @xmath509 , where @xmath510 is a unitary propagator given by the following time - ordered exponential : @xmath511.\end{gathered}\ ] ] if the variation @xmath512 is localized to an infinitesimal time interval @xmath513 following some time @xmath280 , one can readily write @xmath514 where one can further substitute @xmath515 . then , for a general variation @xmath512 , one gets @xmath516\,\delta q(\tau)\,d\tau.\notag\end{gathered}\ ] ] for @xmath517 at any given time @xmath176 , this gives @xmath518.\end{gathered}\ ] ] then , after substituting eq . ( [ eq : mupsi ] ) , one gets @xmath519 \\ = - \text{re}\,\big[\hat{\psi}^\dag(t ) \cdot { \partial}_q h(t , q(t ) ) \cdot \hat{\psi}(t)\big].\label{eq : aux112}\end{gathered}\ ] ] here , we also used @xmath520 and the fact that @xmath521^\dag \cdot u_h(t_2 , t)$ ] is a unit operator , because @xmath510 is unitary . since @xmath522 is hermitian , one can omit `` re '' , so eq . ( [ eq : aux112 ] ) becomes @xmath523 after substituting this into eq . ( [ eq : daeffq ] ) and requiring @xmath524 = 0 $ ] , one obtains @xmath525 this is equivalent to eq . ( [ eq : qleffq ] ) , as expected . suppose that the coordinate @xmath1 is chosen such that @xmath205 , where @xmath11 is a small parameter . assuming that the interaction is weak , a typical interaction hamiltonian is then linear in @xmath1 . like @xmath142 , any zeroth - order term can be removed by a variable transformation , so we assume @xmath526 , where @xmath527 may depend only on @xmath176 , if at all . we now seek to derive an approximation of @xmath53 accurate up to @xmath528 ; i.e.,terms @xmath529 will be neglected . to construct such an approximation , we adopt an asymptotic representation of @xmath166 as a power series in @xmath11 ; i.e.,@xmath530 , where @xmath531 . according to eq . ( [ eq : psiq2 ] ) , @xmath532 , so truncating this series as @xmath533 is sufficient . since @xmath194 is constant , @xmath49 is then fully described by @xmath534 , which serves as the new independent variable instead of the original @xmath166 . hence , after omitting complete time derivatives ( which do not affect ele ) , eq . ( [ eq : aux103 ] ) becomes similar to eq . ( [ eq : lgamma ] ) , namely , @xmath535 where the new @xmath536 and @xmath199 are given by @xmath537 ( one may also recognize this as the leading - order born approximation of the original problem . ) this result shows that theory of linear dispersion for interaction hamiltonians of the form ( [ eq : hint2 ] ) can be constructed identically to that for interaction hamiltonians of the form ( [ eq : hint ] ) . in particular , this means that general predictions of linear dispersion theory are independent of whether oscillators comprising a medium are classical or quantum . here , we discuss some properties of the functions @xmath291 and @xmath301 that are introduced in sec . [ sec : weyl ] . first , eq . ( [ eq : aproph ] ) is proved as follows : @xmath541 where we used that @xmath267 is real by definition . equation ( [ eq : apropa ] ) is proved similarly . second , let us explicitly derive an approximation for @xmath291 in the limit ( [ eq : vro ] ) . using eq . ( [ eq : avu ] ) for @xmath267 together with eq . ( [ eq : mcuasym ] ) for @xmath260 , we obtain @xmath542.\notag\end{gathered}\ ] ] then , using @xmath543 , one gets @xmath544}{h_0 - \omega } \\ + \frac{1- \cos[(h_0 + \omega)\tau_m]}{h_0 + \omega } \bigg\ } \cdot v,\notag\end{gathered}\ ] ] @xmath545}{h_0 - \omega } - \frac{\sin[(h_0 + \omega)\tau_m]}{h_0 + \omega } \right\ } \cdot v,\notag\end{gathered}\ ] ] or , in the continuous - spectrum limit ( sec . [ sec : pm ] ) , @xmath546}{2(\omega - \omega ) } \,[f(\bar{t } , \omega ) - f(\bar{t } , -\omega)]\,d\omega,\label{eq : ath}\\ \widetilde{\alpha}_{\omega , a } = \int \frac{\sin[(\omega - \omega)\tau_m]}{2(\omega - \omega ) } \,[f(\bar{t } , \omega ) - f(\bar{t } , -\omega)]\,d\omega.\label{eq : ata}\end{gathered}\ ] ] notably , the anti - hermitian part of @xmath291 vanishes in the zero-@xmath288 limit ; i.e.,@xmath547 , so @xmath548 , and @xmath549 \cdot v \\ \to \int \omega^{-1}[1 - \cos(\omega \tau_m)]\,f(\bar{t } , \omega)\,d\omega . \end{gathered}\ ] ] asymptotic expressions for @xmath301 under the condition ( [ eq : vro ] ) can be obtained by taking the large-@xmath289 limit of eqs . ( [ eq : ath ] ) and ( [ eq : ata ] ) . the oscillating terms vanish for @xmath550 . for @xmath551 , one has @xmath552 \equiv 0 $ ] , so , instead of averaging the corresponding cosine to zero , we entirely exclude the point @xmath551 from the integration domain . also , @xmath553/(\omega - \omega ) \to \pi \delta(\omega - \omega)$ ] . hence , one gets @xmath554,\end{gathered}\ ] ] where @xmath555 denotes the cauchy principal value . equation ( [ eq : ahaa ] ) can be understood also as @xmath556 , where @xmath557 is the hilbert transform . using that @xmath558 , we then obtain @xmath559 . thus , in summary , @xmath560 which can be recognized as the kramers - kronig relations . to calculate @xmath562 used in sec . [ sec : fastglobal ] , we proceed as follows . first , using eqs . ( [ eq : auxb ] ) and ( [ eq : nupm ] ) , we get @xmath563 \right\}.\end{gathered}\ ] ] using the notation @xmath564 , we further rewrite this as follows : @xmath565 \right\ } \notag\\ & = \frac{1}{16 } \bigg\{\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,q^{\dag}(t ) \cdot v^{\dag}(t ) \cdot v(t ' ) \cdot q(t')\,\delta \theta ( t ' ) + \int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , [ e(\bar{t } , \tau)\,q^{\dag}(t)\cdot v^{\dag}(t)\cdot v(t')\cdot q(t')]^{\dag}\,\delta \theta ( t ' ) \notag \\ & \mbox { } \quad - i\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,q^{\dag}(t)\cdot v^{\dag}(t)\cdot v(t')\cdot \delta q(t ' ) + i\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , [ e(\bar{t } , \tau)\,q^{\dag}(t)\cdot v^{\dag}(t)\cdot v(t')\cdot \delta q(t')]^{\dag } \bigg\ } \notag \\ & = \frac{1}{16}\bigg\ { \int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,q^{\dag}(t)\cdot v^{\dag}(t)\cdot v(t')\cdot q(t')\,\delta \theta ( t ' ) + \int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , -\tau)\,q^{\dag}(t ' ) \cdot v^{\dag}(t ' ) \cdot v(t)\cdot q(t)\,\delta \theta ( t ' ) \notag \\ & \mbox { } \quad - i\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,q^{\dag}(t ) \cdot v^{\dag}(t)\cdot v(t')\cdot \delta q(t ' ) + i\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , -\tau)\,\delta q^{\dag}(t')\cdot v^{\dag}(t')\cdot v(t)\cdot q(t ) \bigg\ } \notag \\ & = \frac{1}{16}\bigg\ { \int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,q^{\dag}(t)\cdot v^{\dag}(t)\cdot v(t ' ) \cdot q(t')\,\delta \theta ( t ' ) + \int_{t_1}^{t_2 } dt ' \int_{t_1}^{t_2 } dt\ , e(\bar{t } , \tau)\,q^{\dag}(t)\cdot v^{\dag}(t)\cdot v(t')\cdot q(t')\,\delta \theta ( t ) \notag \\ & \mbox { } \quad - i\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,q^{\dag}(t)\cdot v^{\dag}(t)\cdot v(t')\cdot \delta q(t ' ) + i\int_{t_1}^{t_2}dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,\delta q^{\dag}(t)\cdot v^{\dag}(t)\cdot v(t')\cdot q(t ' ) \bigg\ } \notag \\ & = \frac{1}{8}\bigg\ { \int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,q^{\dag}(t ) \cdot v^{\dag}(t)\cdot v(t')\cdot q(t')\,\frac{\delta \theta ( t')+\delta \theta(t)}{2 } \notag \\ & \mbox { } \quad -\frac{i}{2}\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,q^{\dag}(t ) \cdot v^{\dag}(t ) \cdot v(t')\cdot \delta q(t ' ) + \frac{i}{2}\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,\delta q^{\dag}(t)\cdot v^{\dag}(t)\cdot v(t')\cdot q(t ' ) \bigg\ } \notag \\ & = \frac{1}{8}\,\text{tr } \int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau ) \left\{w(\bar{t } , \tau)\ , \frac{\delta \theta ( t')+\delta \theta ( t)}{2 } + \frac{i}{2 } \left[q(t')\cdot \delta q^\dag ( t)-\delta q(t')\cdot q^\dag ( t)\right ] \right\}\cdot v^{\dag}(t)\cdot v(t').\notag\end{aligned}\ ] ] next , we notice that @xmath566 = \delta { \mathcal{b}}_+[q^ * , -\theta ] = ( \delta { \mathcal{b}}_+[q^ * , -\theta])^*$ ] . hence , @xmath567\right\ } \cdot v^{\dag}(t)\cdot v(t')\right]^*\\ = -\frac{1}{8}\,\text{tr}\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau)\,\left[w(\bar{t } , \tau)\,\frac{\delta \theta ( t')+\delta \theta ( t)}{2 } + \delta y(\bar{t } , \tau)\right ] \cdot v^{\rm t}(t)\cdot v^*(t'),\notag\end{gathered}\ ] ] where @xmath143 is given by eq . ( [ eq : ww ] ) and @xmath568 is given by eq . ( [ eq : dy ] ) . for @xmath569 , this leads to @xmath570 \cdot \left[v^{\dag}(t)\cdot v(t')-v^{\rm t}(t)\cdot v^*(t')\right]\notag\\ & = \frac{1}{8}\,\text{tr}\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau ) \left[w(\bar{t } , \tau)\,\frac{\delta \theta ( t')+\delta \theta(t)}{2 } + \delta y(\bar{t } , \tau)\right ] \cdot \left[-2i \alpha(t , t')\right]\notag\\ & = - \frac{i}{4}\,\text{tr } \int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e(\bar{t } , \tau ) \left[w(\bar{t } , \tau)\,\frac{\delta \theta ( t')+\delta \theta(t)}{2 } + \delta y(\bar{t } , \tau)\right ] \cdot \alpha(t , t')\notag\\ & = - \frac{i}{4}\,\text{tr}\int_{t_1}^{t_2 } dt \int_{t_1}^{t_2 } dt'\ , e^{i \omega \left(\bar{t}\right ) \tau } \left[w(\bar{t } , \tau)\,\delta \theta \left(\bar{t}\right)+\delta y(\bar{t } , \tau)\right ] \cdot \bar{\alpha } ( \bar{t } , \tau ) + o\left(\epsilon_\omega^2\right)\notag\\ & \approx \delta { \mathcal{b}}_{\theta}+\delta { \mathcal{b}}_q,\end{aligned}\ ] ] where @xmath571 and @xmath572 are given by eqs . ( [ eq : dbtheta ] ) and ( [ eq : dbq ] ) , respectively . the only exception are media with extremely simple dispersion , such as those described by a local refraction index . although often presented as characteristic models of dielectric media , they , in fact , do not provide a truly representative picture of general wave dynamics . d. bnisti , _ envelope equation for the linear and nonlinear propagation of an electron plasma wave , including the effects of landau damping , trapping , plasma inhomogeneity , and the change in the state of wave _ , phys . plasmas * 23 * , 102105 ( 2016 ) . c. m. bender , m. gianfreda , n. hassanpour , and h. f. jones , _ comment on `` on the lagrangian and hamiltonian description of the damped linear harmonic oscillator '' [ j. math . 48 , 032701 ( 2007 ) ] _ , j. math . phys . * 57 * , 084101 ( 2016 ) . r. l. dewar , r. f. abdullatif , and g. g. sangeetha , _ complex nonlinear lagrangian for the hasegawa - mima equation _ , in proceedings of the 20th iaea fusion energy conference ( 2004 ) , th / p6 - 1 , http://www-naweb.iaea.org/napc/physics/fec/fec2004/datasets/th_p6-1.html . this implies @xmath50 _ real _ bc per each end point , as in the original problem ( sec . [ sec : coupled ] ) . this is in contrast with ref . @xcite , where @xmath50 _ complex _ bc were assumed per end point in the quantumlike representation [ namely , @xmath578 , so the problem was overdefined . a brief overview of the weyl calculus can be found , for instance , in refs . @xcite . here , we only notice that , for an operator @xmath3 given by @xmath579 , its weyl image is @xmath580 . the polarizability operator can be brought to the form assumed for @xmath581 by making two replacements in eq . ( [ eq : alphaq ] ) : ( i ) @xmath582 , which is justified provided we consider the operator on functions that are identically zero at @xmath583 ; and ( ii ) @xmath584 , which is justified by the presence of the heaviside function in the integrand . as asymptotic approximation of @xmath585 \doteq \int_a^b g(t)\,e^{i\vartheta(t ) } \,dt}$ ] , where @xmath586 evolves at some rate @xmath587 , can be found via integration by parts . this gives @xmath588 = \smash{\int_a^b } [ ( d / dt)(ig e^{i\vartheta}/\omega ) - i g'(t)/\omega]\,dt$ ] or , equivalently , @xmath588 = \smash{(ig e^{i\vartheta}/\omega)|_a^b } - \smash{{\mathscr{j}}[i g'(t)/\omega]}$ ] , where the functional on the right is smaller than the functional on the left by a factor @xmath589 . the calculation can also be iterated . then , @xmath588 $ ] can be represented as an asymptotic series of terms that are determined by @xmath590 , @xmath339 and their derivatives evaluated locally at the ends of the integration domain . in linear systems , @xmath591 , where @xmath592 is the energy . thus , the action conservation law that we presented is also known as the conservation of an adiabatic invariant @xmath593 . for details , see , e.g. , ref . @xcite . h. qin , j.liu , j. xiao , r. zhang , y. he , y. wang , y. sun , j. w. burby , l ellison , and y zhou , _ canonical symplectic particle - in - cell method for long - term large - scale simulations of the vlasov - maxwell equations _ , nucl . fusion * 56 * , 014001 ( 2015 ) . a. stern , y. tong , m. desbrun , and j. e. marsden , _ geometric computational electrodynamics with variational integrators and discrete differential forms _ , in _ geometry , mechanics , and dynamics : the legacy of jerry marsden _ , edited by d. e. chang , d. d. holm , g. patrick , and t. ratiu ( springer , new york , 2015 ) , p. 437 .

empirical evidence is mounting that living systems and communities of them might operate at the vicinity of a critical point @xcite . criticality , with its concomitant scale invariance , power laws distributions , and extremely large correlations and response @xcite could be a possible source of functional advantages for biological systems@xcite . in particular , in the brain one of the flagships of the criticality hypothesis it could lead to maximal dynamic ranges , high sensitivity to stimuli , optimal transmission and storage of information , and very diverse dynamical repertoires @xcite . different mechanisms and scenarios have been described in the recent literature to explain how such a critical behavior comes about @xcite . on the other hand , some authors have argued that apparent criticality could be just an artifact of the attempt to fit overly simplified models to complex and highly heterogeneous systems @xcite . in a recent work with a fresh perspective @xcite , criticality has been shown to emerge in communities of individuals / agents trying to communicate with each other and creating a collective entity . this approach considers a community of individuals equipped with a ( genetic , neural , regulation ... ) network representing the internal configuration of each individual agent . the state of each of such networks is controlled by some parameters that completely determine the steady state probability distribution function of internal configurations . it is assumed that the state of each single agent is defined by such a probability distribution . each individual of the community tries to mimic , i.e. to infer or `` understand '' , the state of others within a community . with unexpected generality , under this dynamics , the community experiences a drift toward the critical point of the network dynamics , i.e. at the edge between ordered and disordered states . remarkably , this emerging criticality entails a large variability among individuals ; indeed , in the critical regime , small variations in parameter values are reflected in large state changes . this leads to the somehow surprising conclusion that individuals aiming at understanding each other in the best possible way thus having the possibility to coordinate their behavior and reactions end up exhibiting a large variability . naively , one could have anticipated that such individuals could have converged to an `` ordered phase '' in which a deterministic and fixed output would be easily predictable or , alternatively , to a `` disordered phase '' in which they all would be essentially random and alike . however , as shown in @xcite , such hypothetical solutions turn out to be unstable against fluctuations or noise , and the criticality is the only feasible outcome . in this paper , we present a variant of the model proposed in @xcite , which maintains the same phenomenology , but is formulated in a simpler setting . this allows us to characterize quantitatively the dynamics , being able to describe mathematically both the steady state and the transient under different conditions . using this simpler model , we elucidate how and why criticality comes about and investigate what are the roles of cooperation and competition among agents . the manuscript is structured as follows : in section [ sec : framework ] we outline the framework introduced in @xcite in a simple and concise way . in section [ sec : model ] , we describe a variant of the original model susceptible to mathematical analysis , while in section [ sec : results ] we study the attractors of such dynamics by employing a mathematical formalism which allows to map a community of interacting agents into an effective fokker - planck equation @xcite . finally , in section [ sec : nu ] we study a generalization of the model in which individuals can also cooperate or compete . consider a community of @xmath0 individuals / agents trying to imitate as much as possible the state of each other . each of these individuals is characterized by a set of @xmath1 internal computing units ( i.e. nodes of a network ) that , for simplicity , we assume to take binary values . therefore , the internal state of each single agent is fully characterized by the string @xmath2 . this state changes in time , @xmath3 , obeying some dynamical rules leading to a stationary probability distribution @xmath4 , where the @xmath5-dimensional variable @xmath6 accounts for all the parameters modulating the dynamics ( where @xmath5 is the dimension of the parameter space ) , as for instance the strength of couplings between network nodes . each individual agent ( i.e. its internal state ) is completely determined by the value of @xmath7 and , for simplicity , all individuals are assumed to be identical and to obey the same internal dynamical rules except , possibly , for the specific values of their internal parameters . in other words , each agent is characterized by its coordinates in a common parameter space . let us emphasize that the framework introduced here consists in a rather sketched idealization of specific biological systems , which aims at generality rather than specificity . however , to provide the reader with some intuition , one could think of a community of bacteria in which individuals are sensitive and responsive to others , and they reconfigure their internal state ( as described for instance by their gene regulatory networks , with intricate feedback mechanism between signals and gene state switches ) in order to behave similarly to others . this may be relevant in a variety of processes that require of ( or benefit from ) a collective response at the community level . the main goal of this work is to underline the non - trivial attractors that could emerge out of this type of `` imitation '' game at its relationship with criticality . certainly , our framework should be carefully adapted to describe specific biological problems . before proceeding and for the sake of self - consistency we briefly recall two basic concepts of information theory : the kullback - leibler divergence and the fisher information . readers familiar with these concepts can safely skip the following two paragraphs . * the kullback - leibler ( kl ) divergence * @xcite allows for quantifying the difference between two probability distributions . for instance , given the parameter sets @xmath7 and @xmath8 characterizing two probability distribution functions , the kl divergence from the second distribution to the first is defined as @xcite : @xmath9 where the average @xmath10 is taken over @xmath11 . in short , ( [ eq : kl ] ) measures the deficit of information when @xmath12 is used to approximate @xmath4 @xcite . importantly , the kl divergence constitutes a pseudo - distance , as it does not obey the triangle inequality , and is not symmetric as in general @xmath13 , except in the case in which both distribution are identical ( in which case the divergence vanishes ) . * the fisher information ( fi ) * is a measure of how _ distinguishable _ is a ( finite ) dataset extracted from a probability distribution from another one obtained with slightly different parameter values . for example , there could be a region in @xmath7 space in which @xmath4 are mostly invariant as we change @xmath7 , while in another regions the distribution could be highly sensitive to parameter changes . the fi is defined as @xcite : @xmath14 for @xmath15 . the determinant @xmath16 is a measure of the density of distinguishable distributions in parametric space ( as a function of @xmath7 ) @xcite . in particular , if @xmath16 peaks at @xmath17 , this corresponds to a region in the space of parameters @xmath7 in which distributions are highly sensitive to changes of the parameters . thus , it is not surprising that the fi exhibits a peak at critical points @xcite . to illustrate these concepts , without loss of generality , we can parametrize the internal probability distribution of agents as : @xmath18 where @xmath19 are the so - called `` observable '' functions ( i.e. functions of the internal configuration ) and @xmath20 . if , in particular , @xmath21 , the internal state of each agent corresponds to a mean - field ising model at some temperature @xcite . for such a parametrization , eq . ( [ eq : parametrization ] ) , the fi is @xmath22 , which corresponds to the generalized susceptibility , that is well - known to diverge at critical points . consequently , in terms of information theory , most distinguishable patterns are concentrated around the critical point @xcite . in what follows , we mainly work with two different parametrizations . the simplest one corresponds to the ( zero - field ) ising mean - field model @xcite : @xmath23 which has the advantage of having just only one free parameter @xmath24 ( usually interpreted as the inverse temperature ) and has a fi peaked at @xmath25 , which diverges at @xmath26 in the ( thermodynamic ) limit , @xmath27 . similarly , we also consider the mean - field ising model with an external field ( which has two parameters ) : @xmath28 which has a fi that diverges as @xmath27 at @xmath29 . this corresponds to i ) a line of first order phase transitions for @xmath30 and @xmath31 and ii ) a second order phase transition at @xmath30 and @xmath32 . as it will become clear in what follows , we are more interested in the second case as , for any finite system size @xmath1 , the maximum of the determinant of the fi is closer to @xmath33 ( i.e. the critical point of the second - order phase transition ) than to any other point along the discontinuous phase transition line . we use the notation @xmath34 for the critical point defined in the thermodynamic limit @xmath27 while @xmath35 is used for the maximum of @xmath16 in parameter space for any given size @xmath1 . we introduce a model , which is identical in spirit to the co - evolutionary model in @xcite but which , in contrast , enables for analytical understanding . here we discuss both the original and the modified models , stressing their similitudes and differences . both models consist of a community of @xmath0 individuals aiming at having an internal state / distribution as similar as possible to the others as quantified by their kl divergences to other individuals i.e. aiming at optimizing the information they have from the rest of the community , i.e. at minimizing its information deficit ( see fig . [ fig : model ] ) . in both cases , individual agents are characterized by an internal state ( probability distribution function ) parametrized as described in the previous section . all individuals are identical in principle , but they may differ in their parameter values . in the original model ( see `` coevolutionary model '' in @xcite ) the dynamics proceeded by randomly selecting at each time step a pair of agents ; given that the kl divergence is not symmetric , one of the two agents has a larger `` fitness '' ( i.e. it infers better the state of the other than the other way around ) and thus , as a consequence of this informational advantage , it has a larger probability of generating progeny . reversely , the less fit agent is more likely to die and be removed from the community , while the fittest one generates an offspring that inherits its parameter values with some small variability from it . this dynamical process is iterated in time , defining in this way a genetic algorithm . after sufficiently long time the stable output of this dynamics turned out to be that the community of agents evolved to have intrinsic parameters located around the corresponding critical point . results in @xcite were mostly computational , even if some heuristic understanding was also provided . analytical progress was hindered by the intrinsic difficulty of dealing mathematically with the genetic algorithm as defined above . for this reason , our first goal in this paper is to construct a version of the original model that , while leaving intact the main phenomenology allows for analytical treatment . with this motivation , we propose to analyze a slightly different model . this is not an evolutionary one , in the sense that agents do not die nor reproduce . instead it is an adaptive one in which agents slightly change their parameters trying to enhance their fitness . thus , we model each agent @xmath36 by its position in parameter space @xmath37 ; it experiences an _ adaptive force _ which is a function of its information - deficit respect to the other agents , plus some stochastic noise and that can be written as the derivative of some ( pseudo)potential , @xmath38 . in particular , the adaptive model is defined by means of the set of langevin equations @xmath39 for @xmath40 , where @xmath41 is the force that an individual agent @xmath42 produces on an agent @xmath36 , which depends solely on the kl divergence from the second to the first . the first term in the r.h.s . ( [ eq : langevin ] ) is the averaged force acting upon agent @xmath36 , which points in the direction of the potential gradient . observe that , in the most general case , @xmath38 is a pseudo - potential and not a true potential , because , owing to the asymmetry of the kl divergence , @xmath43 , the forces in eq . ( [ eq : langevin ] ) do not obey the potentiality ( schwarz ) condition . still , even in such cases , the force can be defined as the derivative of such a pseudo - potential . the second term in eq . ( [ eq : langevin ] ) represents a thermal noise which is an ineluctably present ingredient of any biological process and that as we will see plays an important role in the dynamical process ( getting rid of unstable solutions ) . in particular , @xmath44 modulates the noise amplitude , and @xmath45 is a white noise with @xmath46 and @xmath47 ( @xmath48 , and @xmath15 ) . to complete the model definition we still need to specify the functional form of the pseudo - potential @xmath49 . the main constraint we need to impose is that @xmath38 has to increase monotonically with @xmath43 ( i.e. @xmath50 ) in order to guarantee that the dynamical process , which converges to minima of @xmath38 , leads also to minimal values of the kl divergence . in this way , at each time step , each agent @xmath36 changes its parameter , @xmath51 , in the direction of the gradient of the averaged @xmath38 , with some added stochasticity . here for reasons explained in detail in [ sec : concavity] we restrict our analysis to the case of a convex potential @xmath52 . having specified the shape of @xmath38 and thus the full model , the question is : where does the community evolve to as a consequence of this information - based dynamics ? in the next section we derive a mathematical approach to compute the attractors of the dynamics in the space of parameters @xmath7 for different choices of the ( pseudo)potential @xmath53 . is composed of @xmath1 internal binary units ; its internal state is represented by the probability distribution @xmath54 for each configuration @xmath55 . the free parameter @xmath51 controls the internal dynamics and all individuals obey the same dynamical rules ( some type of probability distribution function , but possibly with different parameter values ) . at each time , individual @xmath36 changes its parameter @xmath51 in the direction that minimizes its average information - deficit respect to the community , @xmath56 , where @xmath57 is the kullback - leibler divergence ( eq . ( [ eq : kl ] ) ) from @xmath54 to @xmath58 , and @xmath53 is some increasing ( pseudo - potential ) function of @xmath43 . additionally , there is some randomness in the dynamical process . the focus is in the adaptation of the whole community in the space of parameters @xmath7 ; does it converge , and if so , where ? ] we are interested in the evolution of the distribution of agents in parameter space , defined as : @xmath59 luckily enough , there exists an analytical method to map a set of particle - like agents ( interacting among themselves by means of langevin dynamics ) into an equation for the probability distribution of their positions in the space in which they move . such a method was proposed by dean in @xcite ( an earlier reference to this equation can be found in the work of kawasaki @xcite ) , and it allows us to derive an equation for @xmath60 from eq . ( [ eq : langevin ] ) . this calculation can be simply understood as the change of variables @xmath61 using the ito s calculus ( see [ sec : dean ] or ref . @xcite for the explicit derivation ) , leading to the following equation for the probability distribution : @xmath62 where @xmath63 is a new gaussian noise with zero mean and correlation @xmath64 ( @xmath15 ) , interpreted in the ito s sense @xcite . observe that , in the limit @xmath65 , eq . ( [ eq : dean ] ) becomes deterministic , and we find an equation for @xmath66 , which , roughly speaking , is a sort of non - linear fokker - planck equation @xcite in which the drift term depends on the distribution @xmath67 itself . observe that eq . ( [ eq : dean ] ) synthesizes the evolution of the community in a single dynamical equation . before proceeding , some remarks are in order : ( i ) the global gradient ensures the conservation of the total probability ( no individuals are created or annihilated ) ; ( ii ) the diffusive term comes from fluctuations in the dynamics ( controlled by @xmath68 ) ; ( iii ) at any region where there are no individuals ( i.e. , where @xmath69 ) , the distribution does not change ( both the deterministic and the stochastic terms vanish ; i.e. it is an absorbing state / region @xcite ) . we can infer the shape of @xmath67 from eq . ( [ eq : dean ] ) ; in the case in which the drift force is small ( which occurs , for instance , when individuals are very close to each other , as expected if the dynamics is cohesive ) , the r.h.s . of eq . ( [ eq : dean ] ) becomes negligible , and @xmath67 is expected to look like a gaussian distribution . also , for a narrow distribution , we can anticipate that , if we expand the drift term ( eq . ( [ eq : drift ] ) ) in taylor series around the centroid of the community , first order contributions will cancel after integrating in the community ( as individuals approach each other , the overall force almost vanishes ) , and second order terms controlling concavity / convexity of @xmath38 may play an important role of the dynamics ( see [ sec : concavity ] for further details ) . anyway , a detailed calculation needs to be performed in order to clarify these questions for each specific case . employing eq . ( [ eq : dean ] ) it is possible to derive a set of deterministic equations for the evolution of its moments for @xmath65 . the mean @xmath70 can be obtained multiplying eq . ( [ eq : dean ] ) by @xmath7 and integrating by parts : @xmath71 where @xmath72 . similarly for the covariance matrix , defined as @xmath73 ( @xmath15 ) , we find : @xmath74 these equations can not be integrated as the evolution of the mean and covariance matrix still depends on the whole distribution @xmath67 , i.e. they are not a closed set of equations . however , in the next section we derive an approximation allowing us to circumvent this difficulty and characterize the evolution of the community by the mean and covariance matrix of the distribution in parameter space . here we develop an approximate scheme giving rise to a closed set of explicit equations for the evolution of the mean and covariance matrix of parameters in the community . to this end , we first expand @xmath75 around @xmath76 in eqs . ( [ eq : mu1 ] ) and ( [ eq : k1 ] ) , and integrate over @xmath7 and @xmath8 ( see supplementary material for a detailed derivation ) ; the result is then implicitly given as a function of moments of @xmath67 . this expansion can be truncated at some level of approximation ; the first contributing terms are ( see supplementary material ) : @xmath77 where @xmath78 is the susceptibility as defined in eq . ( [ eq : fi ] ) , @xmath79 , and @xmath80 is the @xmath81 moment of the distribution of parameters , @xmath66 . for the case in which @xmath38 has a local minimum at @xmath82 ( i.e. , @xmath83 , @xmath84 ) , the second term in eq . ( [ eq : secondmom1 ] ) vanishes , and one needs to keep further terms in the expansion , obtaining a slightly more complex expression ( see [ sec : v0 ] ) . equations ( [ eq : firstmom1 ] ) and ( [ eq : secondmom1 ] ) can not be integrated , as we still need additional equations for the 4-th moment @xmath80 . we circumvent this problem by approximating the 4-th moment in terms of the second ones ( moment closure ) . thus we assume that @xmath67 is approximately gaussian . to justify this ansatz we can argue that , for small noise variance @xmath68 , individuals stay close to each other . as revealed by eq . [ eq : firstmom1 ] , the effect of the drift is weaker for narrower distributions , and the dynamics is essentially diffusive ( at least locally ) , which leads to a gaussian distribution . furthermore , we have corroborated that the results obtained from this approach match well numerical simulations for small values of @xmath68 , while the approximation breaks down for larger values . therefore , we simply take @xmath85 , obtaining closed equations for the first and second moments . in particular , for the case of just one parameter ( @xmath86 ) we find : @xmath87 or a slightly different second equation if @xmath83 ( [ sec : v0 ] ) . from this set of equations , we can see that : * the centroid of the community , @xmath88 , follows the direction of the gradient of the fi , and the sign of @xmath89 controls the direction in which individuals move : while the community follows the direction which maximizes the fi for @xmath84 , the opposite occurs for @xmath90 , i.e. , individuals follow the direction which minimizes it ( see [ sec : concavity ] ) . if the fi experiences an abrupt change of its shape , this should be translated into an abrupt change of the dynamics ( see , for instance , how the community rapidly goes from the transient to the stationary regime in figs . [ fig:1dim ] and [ fig:2dim ] ) . * the variance of the distribution , @xmath91 , changes owing to two opposite mechanisms : it increases owing to fluctuations ( controlled by @xmath68 ) , while it shrinks as the result of interactions ( faster in regions with higher fi ) . therefore , if the fi diverges in the thermodynamic limit , all individuals should be concentrated _ at _ the critical point . * in a noiseless scenario ( @xmath92 ) , the variance of the distribution vanishes in the stationary limit , @xmath93 ( i.e. all individuals have the same parameter values ) , and the dynamics can stop before the community has reached the global maximum of the fi . noise destabilizes these kind of attractors , and the only stable solution is the critical point . imposing stationarity in the previous equations , it is straightforward to derive the steady - state distribution , which is centered at the maximum of the fi , @xmath94 , and has variance @xmath95 ( see [ sec : v0 ] for the case @xmath83 ) . thus , as @xmath96 diverges for @xmath97 , all individuals will locate exactly _ at _ the critical point , @xmath98 , in the thermodynamic limit , and very close to it for large but finite internal system sizes . [ fig:1dim ] represents the mean and variance of parameters of 10 independent computer simulations of the dynamics ( eq . ( [ eq : langevin ] ) ) , in a community of @xmath99 individuals with internal parametrization of eq . ( [ eq : ising_1d ] ) , for @xmath100 ( the potential has a minimum at @xmath82 ) . solid dark lines represent the numerical integration of our theoretical prediction for @xmath83 ( see eqs . ( [ eq : firstmom1 - 1dim ] ) and ( [ eq : secondmom2 - 1dim ] ) in the [ sec : v0 ] ) . as expected , individuals are attracted towards the maximum of the fi , represented by the solid line at @xmath101 . [ fig:1dim ] illustrates the perfect agreement of the theoretical approach with the numerical integration of the dynamics . let us remark that deviations stem from the finite number of individuals @xmath0 in computer simulations . analogous simulations with @xmath102 are shown in [ sec : concavity ] , revealing again an excellent agreement between simulation an theory . ) , with the one - dimensional ising - mean internal parametrization eq . ( [ eq : ising_1d ] ) . in this example , the local potential is @xmath100 . colored fluctuating lines represent @xmath103 independent computer simulations of the dynamics , the solid black curves correspond to the numerical integration of the theoretical prediction for @xmath65 given by eqs . [ eq : firstmom1 - 1dim ] and [ eq : secondmom2 - 1dim ] . parameters : number of individuals @xmath99 , number of internal nodes @xmath104 , noise amplitude @xmath105 . the maximum of the fisher information is located at @xmath106 for @xmath104 , ( approaching to @xmath107 in the thermodynamic limit ) . ] let us remark that the analytical calculation becomes much more involved for the case of more parameters , @xmath108 , but our simulations ( see below ) confirm that the same phenomenology i.e . convergence towards criticality emerges . [ fig:2dim ] represents the mean and covariance matrix in time ( obtained via numerical integration of eq . ( [ eq : langevin ] ) ) for the two - parameter ising mean - field - like parametrization , eq . ( [ eq : ising_2d ] ) , in a community of @xmath99 individuals and @xmath100 . solid dark lines represent the numerical integration of eqs . [ eq : firstmom1 ] and [ eq : secondmom2 ] ( using the gaussian approximation ) . an interesting observation is that , as in the one - dimensional case , the community evolves near the critical point of the second - order phase transition ( i.e. to the global maximum of the determinant of the fi ) rather than along the first - order phase transition line . in this case , the exact attractor ( for finite internal system sizes , @xmath1 ) can not be computed , as the stationary solution of eq . ( [ eq : firstmom1 ] ) becomes harder to solve for @xmath109 . certainly , for a finite internal system size , @xmath104 , the attractor @xmath110 slightly differs from @xmath111 for @xmath104 , but both approach to the critical point , @xmath33 , in the thermodynamic limit . ) , with the two - dimensional ising mean - field internal parametrization , eq . ( [ eq : ising_2d ] ) . as in fig . [ fig:1dim ] , the local potential is @xmath100 . colored fluctuating lines correspond to 10 independent realizations of the dynamics , and the solid black ones to the numerical integration of the theoretical prediction for @xmath65 , given by eqs . [ eq : firstmom1 ] and [ eq : secondmom2 ] ( using the gaussian approximation @xmath112 ) . parameters : number of individuals @xmath99 , number of internal nodes @xmath104 , noise amplitude @xmath113 . the maximum of determinant of the fisher information is located at @xmath114 for @xmath104 , ( approaching to @xmath115 in the thermodynamic limit ) . ] let us now generalize the dynamics to embrace the following mechanisms , commonly studied in evolutionary game theory @xcite : * * neutralism * : each individual tends to minimize its information - deficit regardless of others ( already analyzed case ) . * * cooperation * : each individual tends to minimize its information - deficit but also , simultaneously , that of all the other individuals in the community . * * competition * : each individual tries to minimize its information - deficit , and , at the same time , tries to maximize the information - deficit of the others . to account for all these possibilities , we redefine the potential between individuals @xmath36 and @xmath42 as a linear combination of their respective information - deficits : @xmath116 where the parameter @xmath117 $ ] tunes the interaction from competition ( @xmath118 ) to cooperation ( @xmath119 ) , including the intermediate neutral case ( @xmath120 ) . the generalized dynamics is simply obtained by substituting @xmath49 with @xmath121 in eq . ( [ eq : langevin ] ) and eq . ( [ eq : drift ] ) . we call @xmath122 the `` symmetry coefficient '' , as @xmath123 and @xmath124 are anti - symmetric and symmetric functions , respectively , under the exchange @xmath125 . we can perform a similar expansion to that carried in section [ sec : expansion ] for the mean and covariance matrix of the distribution of parameters ( see supporting material ) . the first non - vanishing terms are : @xmath126 notice that eq . ( [ eq : firstmom1-nu ] ) becomes much simpler than its counterpart in the neutral case , @xmath120 , eq . ( [ eq : firstmom1 ] ) . further terms in the expansion have to be computed in eq . ( [ eq : firstmom1-nu ] ) for the case @xmath120 or @xmath83 , and similarly for eq . ( [ eq : secondmom1-nu ] ) for @xmath118 ( see supporting material ) . some remarks are in order : ( i ) eq . ( [ eq : firstmom1-nu ] ) tells us that the community moves in the direction of @xmath127 , where the sign is given by @xmath128 . therefore , agents move toward the critical point for any @xmath129 , while they move in the opposite direction for any @xmath130 . ( ii ) eq . ( [ eq : firstmom1-nu ] ) does not depend on the sign of @xmath89 , but on the sign of @xmath131 , which is always positive as @xmath38 is an increasing function of @xmath43 ( observe also that , the previously discussed issue about the concavity / convexity has no relevance for @xmath132 ) . summarizing , individuals move to the critical point for any `` mostly '' antisymmetric potential ( @xmath129 ) , while they escape from it for any `` mostly symmetric '' potential ( @xmath130 ) . the case @xmath120 represents a borderline case , which has to be analyzed as shown before keeping higher - order terms in the expansion . [ fig : potential ] ( top - left ) illustrates the generalized potential for the case of just two individuals , in the symmetrical case , @xmath133 , with the parametrization of eq . ( [ eq : ising_1d ] ) in the linear case @xmath134 . observe that in this case the local potential @xmath135 is identical to the jensen - shannon entropy @xcite . as the figure shows , there is a flat region in the space of parameters along the principal diagonal . as near the center , larger potential values ( lighter colors ) are observed , one can expect that the potential is minimal at @xmath136 , suggesting that agents should move away from criticality toward highly ordered or highly disordered states . indeed , this is confirmed by fig . [ fig : potential ] ( bottom ) which shows simulation results for this case and reveals that agents tend to scape toward regions of low fi , i.e , away from the critical point ( similar results are obtained for larger values of @xmath0 ; not shown ) . the top - right panel of fig . [ fig : potential ] illustrates the opposite scenario , i.e. the anti - symmetrical case , @xmath118 , for the same parametrization ( eq . ( [ eq : ising_1d ] ) ) and @xmath134 . as before , we can see a flat region in the space of parameters for @xmath137 ( i.e. , away from the critical point ) . however , there exists a region of lower information - deficit for one individual ( blue region ) , which is simultaneously maximal for the other one ( red region ) . therefore , both individuals want to be close enough to their respective blue regions , but can not reach it , and the stable strategy is found at the maximum of fi , i.e , at the critical point . this is confirmed by our simulations , as illustrated by the top panel of fig . [ fig : potential ] . as a consequence , competition pushes agents to the critical region , with highest variability of internal patterns . some questions arise from a game - theory perspective : what is the best strategy for individuals : to cooperate , to compete , or just to be neutral ? what occurs if we take the optimal strategy for any specific case an introduce a small amount of individuals with different strategies ? is the previous optimal strategy still stable ? these questions require of further investigation , that we leave for a future work . in the case of cooperation ( @xmath119 , left ) and of competition ( @xmath118 , right ) . the first is symmetric under the exchange of @xmath138 , while the second is antisymmetric . once individuals approach each other , arrows indicate the region in which the centroid is expected to move . _ bottom : _ two individuals evolve their parameters @xmath139 and @xmath140 with the dynamics given by of eq . ( [ eq : langevin ] ) ( substituting @xmath49 with @xmath141 ) . for both @xmath119 and @xmath118 , we have plotted the trajectories of each parameter ( using the same color - code ) starting from two different initial conditions ( @xmath142 and @xmath143 , respectively ) . simulations illustrate that individuals scape towards regions of lower fi ( away from the critical point ) in the cooperative case , whereas the opposite occurs in the case of competition , in which they end at the region of highest fi , i.e. , the critical point . in all cases , we have implemented the dynamics using the ising mean - field like parametrization of eq . ( [ eq : ising_1d ] ) and @xmath134 , @xmath144 and @xmath145 . ] in this paper we have studied the dynamics of a community of individuals evolving to mimic other fellow agents in its community , i.e. to enhance the information each one captures from the others . the framework employed here is similar to that in our precedent work @xcite : each individual is constituted by an internal boolean network obeying an internal dynamics , which can be modulated by changing its internal parameters in the direction which maximizes its similitude to others . the model introduced here is adaptive as opposed to the one in @xcite which was evolutionary ; its main advantage of the model presented here is that allows for a mathematical treatment , which was mostly missing in @xcite . as a first step , using a well established method for the study of interacting stochastic processes@xcite and through several approximations we are able to compute analytically the attractors of the dynamics , confirming our previous findings : the community self - tunes to the neighborhood of the critical point , separating ordered from disordered internal states . it is important to stress this result : the critical point can be identified with the region of parameter space with maximum variability in terms of information theory ; the `` imitation game '' studied here thus leads to , surprisingly , maximum complexity . this is similar to previous results by kaneko and suzuki @xcite , who studied two abstract `` birds '' that play to imitate each other s `` songs '' . the attractor of such dynamics is located at the `` edge of chaos '' , exhibiting maximum variability @xcite . in the simplest version of the model introduced here , individuals adapt with the only aim of improving their individual knowledge . as a second step , we have studied a more general scenario introducing a few basic mechanisms usually studied in game theory : cooperation ( adaptation to improve the amount of information collected by each individual agent as well as by the others in the community ) , competition ( adaptation to improve each agent s information and reduce that of the others ) . in these cases , we observe that , while cooperation leads the community to regions with scarce variability ( i.e. to the ordered or disordered regions ) , criticality emerges for dominantly - competing communities . we hypothesize that the mechanisms reported here and their relationship with social interactions may be related with the emergence of criticality at a community level , as for instance with flocks of birds @xcite . more ambitiously , the interplay between cooperation and competition and the emergence of complex patterns in communities might be related with the large phenotypic variability observed in some bacterial communities @xcite , a survival strategy by which individuals diversify their behavior to minimize long - term extinction risks , usually referred to as `` bet - hedging '' @xcite . however , clarifying this point would require further investigation . summing up , the question about whether biological organisms are poised to criticality is being harshly debated , but setting the problem from an information - theory point of view opens new and stimulating questions that seem worth exploring . we are grateful to j. r. banavar for helpful discussions and valuable suggestions . j.h . thanks r. cerezo for constant support . m.a.m . and j.h . acknowledge support from the spanish mineco excellence project fis2013 - 43201-p . thanks the support of the university of padova , physics and astronomy department ( senior grant 129/2013 prot . in this appendix we illustrate the role of the concavity / convexity of the information - deficit ( pseudo)potential , @xmath146 . for the sake of simplicity let us consider the case of a community consisting of just two individuals with parameters @xmath7 and @xmath8 , respectively . introducing the parametrization of eq . ( [ eq : parametrization ] ) in the definition of the kl divergence ( eq . ( [ eq : kl ] ) ) , it can be easily checked that : @xmath147 with this , we can rewrite the force @xmath148 ( eq . ( [ eq : drift ] ) ) as : @xmath149 and consequently , the drifts exerted on the two agents have opposite directions i.e . individuals tend to approach each other to exhibit similar patterns and their relative strength , @xmath150 , is given by @xmath151 . this means that the individual with large value of @xmath152 experiences a larger drift . the forthcoming argument is illustrated in fig . [ fig : concavity ] . without loss of generality , suppose that @xmath153 . then , in our framework , the individual with parameter @xmath7 has more information than the individual with @xmath8 , and it is more fit . if @xmath53 is a convex function ( @xmath154 ) , as @xmath53 is monotonically increasing , it follows that @xmath155 ( see left top panel in fig . [ fig : concavity ] ) . consequently , by eq . ( [ eq : drift2 ] ) , individual @xmath8 experiences a higher drift than @xmath7 , and their centroid moves towards the individual with lower kl divergence , in this case @xmath7 . the opposite occurs for a concave @xmath53 ( corresponding to right top panel in fig . [ fig : concavity ] ) . if @xmath156 , we can see that @xmath157 , and the centroid approaches to the individual with higher kl divergence , here @xmath8 . this is confirmed by simulations of the dynamics ( i.e. , by numerical integration of eq . ( [ eq : langevin ] ) ) : in bottom panel of fig . [ fig : concavity ] we represent 10 independent realizations for two choices of @xmath53 ( eq . ( [ eq : langevin ] ) ) with opposite concavity , in a community of @xmath99 individuals ( blue curves correspond to @xmath158 , and red curves to @xmath159 ) . in the case of @xmath154 , the community moves toward the critical point , while it escapes from it for @xmath156 . this result is understood analytically by means of eq . ( [ eq : firstmom1 - 1dim ] ) . and @xmath8 , respectively , exert onto each other is given by @xmath160 , which means that the one with higher @xmath152 experiences a higher drift ( derivatives have been represented with tangent lines ) . consequently , if @xmath154 ( left top panel ) , individual with less @xmath43 also moves less , while the opposite occurs for @xmath156 ( right top panel ) , where the individual with less @xmath43 experiences a higher drift and is more likely to scape . ( bottom panel ) the concavity / convexity of @xmath38 switches the dynamics . we represent the mean parameter of the community in time for choices of @xmath53 with opposite concavity : we plot 10 independent realizations of the dynamics , eq . ( [ eq : langevin ] ) , for the case ( blue curves ) @xmath161 ( @xmath162 ) and ( red curves ) @xmath163 ( @xmath164 ) , and the ising mean - field - like parametrization , eq . ( [ eq : ising_1d ] ) . while individuals approach to the critical point , @xmath94 , for @xmath154 , they scape from it for @xmath156 , confirming that the sign of @xmath165 inverts the dynamics . solid black lines represent the theoretical prediction , given by the numerical integration of eqs . [ eq : firstmom1 - 1dim ] and [ eq : secondmom1 - 1dim ] . the initial condition is @xmath166 , and parameter values are @xmath167 and @xmath168 . ] the linear case , @xmath134 , can be solved exactly : @xmath169 and , as both individuals perform equal jumps in opposite directions , the centroid does not change at all . summarizing , the dynamics is not completely determined with the statement that `` individuals change their parameters in the direction which maximizes their information '' ; _ one needs also to specify how much individuals move_. under an adaptive / evolutionary framework , the more plausible scenario corresponds to the case in which fitter individuals ( i.e. with more information ) experience lower drifts than poorly fitted individuals ( i.e. with low information ) , which corresponds to the choice of @xmath154 . the opposite choice , @xmath156 , becomes senseless in our model , as it represents the case in which fittest individuals are more likely to scape to other regions in the space of parameters . therefore , we restrict our analysis to the choice with @xmath154 . in this appendix we briefly review the calculation done by dean to obtain a dynamical equation for the density of a system of interacting `` particles '' each one obeying a langevin dynamics @xcite . for this , we define the contribution of each individual @xmath36 to the global distribution of parameters as @xmath170 , so that @xmath171 . then , any function of the form @xmath172 can be written as : @xmath173 and its time derivative as : @xmath174 alternatively , the integral can be also computed expanding the argument using the ito s calculus ; by doing so one finds : @xmath175 where we have integrated by parts at the second step . identifying the arguments of equations [ eq : dean2 ] and [ eq : dean3 ] , one obtains : @xmath176 that , averaging over individuals , transforms to : @xmath177 however , this is not a closed equation for @xmath66 , as it still depends on @xmath178 . as dean shows , one can simply substitute it with @xmath179 , in which @xmath63 is a new delta - correlated gaussian noise , as it has exactly the same correlation function @xcite . after doing that , we finally recover eq . ( [ eq : dean ] ) . note that , compared to dean s result , we have got an extra @xmath180 factor because we have normalized the global density function to @xmath181 rather than @xmath0 . for the case in which @xmath53 has a local minimum at @xmath82 ( i.e. , @xmath83 , @xmath84 ) , eq . ( [ eq : secondmom1 ] ) vanishes , and we have to keep further terms in the expansion , obtaining a slightly more complex expression : @xmath182 in particular , for the case of just one - parameter , @xmath86 , we obtain : @xmath183 the stationary variance in this case is @xmath184 . 10 url # 1#1urlprefix[2][]#2 beggs j and plenz d 2003 _ the journal of neuroscience _ * 23 * 1116711177 nykter m , price n , aldana m , ramsey s , kauffman s a , hood l e , yli - harja o and shmulevich i 2008 _ proceedings of the national academy of sciences _ * 105 * 18971900 petermann t , thiagarajan t a , lebedev m , nicolelis m , chialvo d r and plenz d 2009 _ proc natl acad sci usa _ * 106 * 1592115926 chen x , dong x , beer a , swinney h and zhang h 2012 _ physical review letters _ * 108 * 148101 bialek w , cavagna a , giardina i , mora t , silvestri e , viale m and walczak a 2012 _ proceedings of the national academy of sciences _ * 109 * 47864791 tagliazucchi e , balenzuela p , fraiman d and chialvo d r 2012 _ frontiers in physiology _ * 3 * issn 1664 - 042x furusawa c and kaneko k 2012 _ phys rev lett _ * 108 * 208103 issn 1079 - 7114 haimovici a , tagliazucchi e , balenzuela p and chialvo d r 2013 _ physical review letters _ * 110 * 178101 bialek w , cavagna a , giardina i , mora t , pohl o , silvestri e , viale m and walczak a m 2014 _ proceedings of the national academy of sciences _ * 111 * 72127217 de arcangelis l , lombardi f and herrmann h 2014 _ journal of statistical mechanics : theory and experiment _ * 3 * p03026 schuster h g , plenz d and niebur e 2014 _ criticality in neural systems _ ( john wiley & sons ) shew w l , clawson w p , pobst j , karimipanah y , wright n c and wessel r 2015 _ nature physics _ binney j , dowrick n , fisher a and newman m 1993 _ the theory of critical phenomena _ ( oxford : oxford university press ) lesne a 1998 _ renormalization methods _ ( wiley , new york ) mora t and bialek w 2011 _ journal of statistical physics _ * 144 * 268302 beggs j m 2008 _ phil . r. soc . a _ * 366 * 329343 chialvo d r 2010 _ nature physics _ * 6 * 744750 kinouchi o and copelli m 2006 _ nature physics _ * 2 * 348351 mora t , walczak a , bialek w and callan jr c 2010 _ proceedings of the national academy of sciences _ * 107 * 54055410 shew w l and plenz d 2013 _ the neuroscientist _ * 19 * 88100 chat h and muoz m 2014 _ physics _ * 7 * 120 levina a , herrmann j m and geisel t 2007 _ nature physics _ * 3 * 857860 bonachela j a and muoz m a 2009 _ journal of statistical mechanics : theory and experiment _ * 9 * p09009 bonachela j , de franciscis s , torres j j and muoz m 2010 _ journal of statistical mechanics : theory and experiment _ * 2 * p02015 millman d , mihalas s , kirkwood a and niebur e 2010 _ nature physics _ * 6 * 801805 stepp n , plenz d and srinivasa n 2015 _ plos computational biology _ * 11 * mastromatteo i and marsili m 2011 _ journal of statistical mechanics : theory and experiment _ * 2011 * p10012 marsili m , mastromatteo i and roudi y 2013 _ journal of statistical mechanics : theory and experiment _ * 2013 * p09003 hidalgo j , grilli j , suweis s , muoz m a , banavar j r and maritan a 2014 _ proceedings of the national academy of sciences _ * 111 * 1009510100 dean d s 1996 _ journal of physics a : mathematical and general _ * 29 * l613 cover t m and thomas j 1991 _ elements of information theory _ ( wiley ) balasubramanian v 1997 _ neural computation _ * 9 * 349368 kawasaki k 1994 _ physica a : statistical mechanics and its applications _ * 208 * 3564 gardiner c w _ et al . _ 1985 _ handbook of stochastic methods _ vol 4 ( springer berlin ) marro j and dickman r 2005 _ nonequilibrium phase transitions in lattice models _ ( cambridge university press ) nowak m a 2006 _ evolutionary dynamics _ ( harvard university press ) kaneko k and suzuki j 1994 evolution to the edge of chaos in an imitation game _ artificial life iii _ ( citeseer ) kussell e and leibler s 2005 _ science _ * 309 * 20752078 ripa j , olofsson h and jonzn n 2009 _ proceedings of the royal society of london b : biological sciences _ rspb20092023 * supplementary material : + cooperation , competition and the emergence of criticality in communities of adaptive systems * j. hidalgo , j. grilli , s. suweis , a. maritan and m. we present the explicit calculations developed in section iv ( simple model ) and section v ( cooperation / competition model ) in the main text . as the former can be understood as a particular case of the latter ( taking the symmetry coefficient @xmath120 ) , here we only refer to the more the general case . we start from the dynamics , given by the set of equations : @xmath185 where @xmath186 , @xmath44 modulates the noise amplitude , and @xmath45 is a white noise with @xmath46 and @xmath47 , with @xmath48 , and @xmath15 . following dean s method as explained in section iva and in appendix b , for an infinite community , @xmath65 , the evolution of the probability distribution of parameters is given by : @xmath187 where @xmath188 is the generalized attraction force that an individual with parameter @xmath8 produces on an individual with parameter @xmath7 , defined as : @xmath189 from eq . ( [ eq : dean - nu ] ) , it is easy to obtain the evolution of the mean and covariance matrix in the community multiplying by @xmath190 and @xmath191 , respectively , and integrating over @xmath7 . after integrating by parts , one gets : @xmath192 let start by expanding @xmath193 around @xmath194 : @xmath195 where we have defined the product @xmath196 and the coefficients @xmath197 as : @xmath198 introducing this expression in eq . ( [ eq : mu1-nu ] ) , we obtain @xmath199 let us note that we do not include the elements of the form @xmath200 and @xmath201 as they are multiplied by @xmath202 and @xmath203 , respectively , which are identically 0 by definition . the first contributing terms are ( see section [ sec : appendix2 ] ) : @xmath204 and , introducing in eq . ( [ eq : taylor ] ) , we finally obtain eq . ( 17 ) in the main text . in the case in which @xmath120 or @xmath83 , the first non - vanishing terms are ( see section [ sec : appendix2 ] ) : @xmath205}\\ \fl f_\epsilon^{\nu\{\alpha\beta\gamma\delta\}_4\{\}_0 } + f_\epsilon^{\nu\{\}_0\{\alpha\beta\gamma\delta\}_4 } & = & 4(1 - 2\nu ) v'(0)\chi_{\alpha\beta\gamma\delta\epsilon}({\boldsymbol{\mu } } ) + 2(1-\nu ) v''(0 ) \chi_{\alpha\epsilon}({\boldsymbol{\mu}})\chi_{\beta\gamma\delta}({\boldsymbol{\mu } } ) + \nonumber\\ & & 2(1 - 2\nu ) v''(0)\chi_{\alpha\beta}({\boldsymbol{\mu}})\chi_{\gamma\delta\epsilon}({\boldsymbol{\mu}})+{\mathrm{sym}[\alpha,\beta,\gamma,\delta ] } \label{eq : f40_f04}\end{aligned}\ ] ] where @xmath206}$ ] represents the minimal set of terms , obtained via index permutations of its precedent elements , that we have to include to symmetrize the expression under the exchange of indexes @xmath207 , and where @xmath208 ( see appendix [ sec : appendix2 ] ) . note that , as the `` @xmath209 '' term in eq . ( [ eq : taylor ] ) is multiplied by @xmath210 and summed over @xmath211 , we have explicitly symmetrized it . for the cases of interest , i.e. @xmath120 and/or @xmath83 , the term proportional to @xmath212 in eq . ( [ eq : f40_f04 ] ) vanishes , and we obtain the following relation : @xmath213 introducing this expression in eq . ( [ eq : taylor ] ) , we finally obtain @xmath214 } \right ) \left(k_{\alpha\beta}k_{\gamma\delta } + \frac{k_{\alpha\beta\gamma\delta}}{3 } \right ) + \nonumber \\ \fl & & ( 1 - 2\nu)\frac{v''(0)}{4}\sum_{\alpha,\beta,\gamma,\delta=1}^d \left ( \chi_{\alpha\beta}({\boldsymbol{\mu } } ) \chi_{\gamma\delta\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\delta ] } \right ) \left(k_{\alpha\beta}k_{\gamma\delta } + \frac{k_{\alpha\beta\gamma\delta}}{3 } \right)\nonumber\\ \fl&&\label{eq : firstmom2-nu}\end{aligned}\ ] ] that , for @xmath120 , it reduces to eq . ( 12 ) in the main text . now we proceed analogously for the expansion of eq . ( [ eq : k1-nu ] ) : @xmath215 where we have omitted the terms multiplied by @xmath216 , which is identically zero by definition . if @xmath217 , we can just take the first term in the expansion of eq . ( [ eq : taylor2 ] ) , and then we arrive to eq . ( 18 ) in the main text . in the case @xmath83 or @xmath118 , the whole first row of eq . ( [ eq : taylor2 ] ) vanishes and we have next - order contributions . for @xmath83 and @xmath218 we obtain : @xmath219}\right ) \left(k_{\alpha\beta}k_{\gamma\tau } + \frac{k_{\alpha\beta\gamma\tau}}{3}\right)+\nonumber\\ \fl & & -\nu{v''(0)}\sum_{\alpha,\beta,\gamma=1}^d\left(\chi_{\alpha\beta}({\boldsymbol{\mu}})\chi_{\gamma\tau}({\boldsymbol{\mu}})+{\mathrm{sym}[\alpha,\beta,\gamma,\tau]}\right ) \left(k_{\alpha\beta}k_{\gamma\epsilon } + \frac{k_{\alpha\beta\gamma\epsilon}}{3}\right ) \label{eq : secondmom2-nu}\end{aligned}\ ] ] while , for the case @xmath220 and @xmath118 , we have : @xmath221 where @xmath222 ( see section [ sec : appendix2 ] ) . here we give the expressions for the derivatives of the generalized potential used in the previous section . to simplify the nomenclature , we introduce , for any function @xmath224 ( as for instance @xmath225 ) , the notation : @xmath226 from this definition , it follows that @xmath227 let start by computing the first derivatives of the fi in terms of the observables functions , @xmath228 . using the parametrization of eq . ( 3 ) in the main text , we have : @xmath229\left[\langle \phi_\beta \rangle_{\mathbf{x}}- \phi_\beta\right ] \rangle_{\mathbf{x}}\\ \fl \chi_{\alpha\beta\gamma}({\mathbf{x } } ) & = \partial_{x_\alpha } \chi_{\beta\gamma}({\mathbf{x } } ) & = \langle \left[\langle \phi_\alpha \rangle_{\mathbf{x}}- \phi_\alpha\right]\left[\langle \phi_\beta \rangle_{\mathbf{x}}- \phi_\beta\right]\left[\langle \phi_\gamma \rangle_{\mathbf{x}}- \phi_\gamma\right ] \rangle_{\mathbf{x}}\\ \fl \chi_{\alpha\beta\gamma\delta}({\mathbf{x } } ) & = \partial_{x_\alpha } \chi_{\beta\gamma\delta}({\mathbf{x } } ) & = \langle \left[\langle \phi_\alpha \rangle_{\mathbf{x}}- \phi_\alpha\right]\left[\langle \phi_\beta \rangle_{\mathbf{x}}- \phi_\beta\right ] \left[\langle \phi_\gamma \rangle_{\mathbf{x}}- \phi_\gamma\right ] \left[\langle \phi_\delta \rangle_{\mathbf{x}}- \phi_\delta\right ] \rangle_{\mathbf{x}}\nonumber\\ \fl & & \qquad - \chi_{\alpha\beta}({\mathbf{x}})\chi_{\gamma\delta}({\mathbf{x } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\delta]}\\ \fl \chi_{\alpha\beta\gamma\delta\epsilon}({\mathbf{x } } ) & = \partial_{x_\alpha } \chi_{\beta\gamma\delta\epsilon}({\mathbf{x } } ) & = \langle \left[\langle \phi_\alpha \rangle_{\mathbf{x}}- \phi_\alpha\right]\left[\langle \phi_\beta \rangle_{\mathbf{x}}- \phi_\beta\right ] \left[\langle \phi_\gamma \rangle_{\mathbf{x}}- \phi_\gamma\right ] \left[\langle \phi_\delta \rangle_{\mathbf{x}}- \phi_\delta\right ] \left[\langle \phi_\epsilon \rangle_{\mathbf{x}}- \phi_\epsilon\right ] \rangle_{\mathbf{x}}\nonumber\\ \fl & & \qquad - \chi_{\alpha\beta}({\mathbf{x}})\chi_{\gamma\delta\epsilon}({\mathbf{x } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\delta,\epsilon]}\end{aligned}\ ] ] where @xmath206}$ ] represents the minimal set of terms , obtained via index permutations of its precedent elements , that we have to include to symmetrize the expression under the exchange of indexes @xmath207 . let us note that each of the definitions above is symmetric under the exchange of any two indexes . now , we compute the cross partial derivatives of the kl divergence , @xmath230 , that are used in the next step : * @xmath235 : @xmath236 * @xmath237 : @xmath238 * @xmath239 : @xmath240 } = v'(0)\chi_{\alpha\beta\epsilon}({\boldsymbol{\mu}})\\ \fl { { { v}^{{\{\}_{0}\{\alpha\beta\epsilon\}_{3 } } } } } & = & v^{(3)}(0 ) { { { d}^{{\{\}_{0}\{\alpha\}_{1 } } } } } { { { d}^{{\{\}_{0}\{\beta\}_{1 } } } } } { { { d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + v''(0 ) { { { d}^{{\{\}_{0}\{\alpha\beta\}_{2 } } } } } { { { d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + \nonumber\\ \fl & & v'(0 ) { \textcolor{blue } { { { { d}^{{\{\}_{0}\{\alpha\beta\epsilon\}_{3 } } } } } } } + { \mathrm{sym}[\alpha,\beta,\epsilon ] } = 2v'(0)\chi_{\alpha\beta\epsilon}({\boldsymbol{\mu}})\\ \fl \nonumber\\ \fl { { { f_\epsilon^\nu}^{{\{\alpha\beta\}_{2}\{\}_{0 } } } } } & = & -\left ( { { { v}^{{\{\alpha\beta\epsilon\}_{3}\{\}_{0 } } } } } + ( 2\nu-1){{{v}^{{\{\}_{0}\{\alpha\beta\epsilon\}_{3 } } } } } \right ) = ( 1 - 4\nu)v'(0 ) \chi_{\alpha\beta\epsilon}({\boldsymbol{\mu}})\end{aligned}\ ] ] * @xmath241 : @xmath242 } = -v'(0)\chi_{\alpha\beta\epsilon}({\boldsymbol{\mu}})\\ \fl { { { v}^{{\{\alpha\beta\}_{2}\{\epsilon\}_{1 } } } } } & = & v^{(3)}(0 ) { { { d}^{{\{\alpha\}_{1}\{\}_{0 } } } } } { { { d}^{{\{\beta\}_{1}\{\}_{0 } } } } } { { { d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + v''(0 ) \left ( { { { d}^{{\{\alpha\}_{1}\{\epsilon\}_{1 } } } } } { { { d}^{{\{\beta\}_{1}\{\}_{0 } } } } } + { { { d}^{{\{\alpha\beta\}_{2}\{\}_{0 } } } } } { { { d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } \right ) + \nonumber\\ \fl & & v'(0 ) { { { d}^{{\{\alpha\beta\}_{2}\{\epsilon\}_{1 } } } } } + { \mathrm{sym}[\alpha,\beta ] } = 0\\ \fl \nonumber\\ \fl { { { f_\epsilon^\nu}^{{\{\}_{0}\{\alpha\beta\}_{2 } } } } } & = & -\left ( { { { v}^{{\{\epsilon\}_{1}\{\alpha\beta\}_{2 } } } } } + ( 2\nu-1){{{v}^{{\{\alpha\beta\}_{2}\{\epsilon\}_{1 } } } } } \right ) = v'(0 ) \chi_{\alpha\beta\epsilon}({\boldsymbol{\mu}})\end{aligned}\ ] ] * @xmath243 : @xmath244 } + \nonumber\\ \fl & = & v'(0)\chi_{\alpha\beta\gamma\epsilon}({\boldsymbol{\mu } } ) + v''(0)\chi_{\alpha\beta}({\boldsymbol{\mu}})\chi_{\gamma\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\epsilon]}\\ \fl { { { v}^{{\{\}_{0}\{\alpha\beta\gamma\epsilon\}_{4 } } } } } & = & v^{(4)}(0 ) { { { d}^{{\{\}_{0}\{\alpha\}_{1 } } } } } { { { d}^{{\{\}_{0}\{\beta\}_{1 } } } } } { { { d}^{{\{\}_{0}\{\gamma\}_{1 } } } } } { { { d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + v^{(3)}(0 ) { { { d}^{{\{\}_{0}\{\alpha\beta\}_{2 } } } } } { { { d}^{{\{\}_{0}\{\gamma\}_{1 } } } } } { { { d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + \nonumber\\ \fl & & v''(0 ) \left ( { { { d}^{{\{\}_{0}\{\alpha\beta\gamma\}_{3 } } } } } { { { d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + { \textcolor{blue } { { { { d}^{{\{\}_{0}\{\alpha\beta\}_{2 } } } } } { { { d}^{{\{\}_{0}\{\gamma\epsilon\}_{2 } } } } } } } \right ) + v'(0 ) { \textcolor{blue } { { { { d}^{{\{\}_{0}\{\alpha\beta\gamma\epsilon\}_{4 } } } } } } } + { \mathrm{sym}[\alpha,\beta,\gamma,\epsilon ] } + \nonumber\\ \fl & = & 3v'(0)\chi_{\alpha\beta\gamma\epsilon}({\boldsymbol{\mu } } ) + v''(0)\chi_{\alpha\beta}({\boldsymbol{\mu}})\chi_{\gamma\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\epsilon]}\\ \fl \nonumber\\ \fl { { { f_\epsilon^\nu}^{{\{\alpha\beta\gamma\}_{3}\{\}_{0 } } } } } & = & -\left ( { { { v}^{{\{\alpha\beta\gamma\epsilon\}_{4}\{\}_{0 } } } } } + ( 2\nu-1){{{v}^{{\{\}_{0}\{\alpha\beta\gamma\epsilon\}_{4 } } } } } \right ) = 2(1 - 3\nu)v'(0)\chi_{\alpha\beta\gamma\epsilon}({\boldsymbol{\mu } } ) \nonumber\\ \fl & & -2\nu v''(0)\chi_{\alpha\beta}({\boldsymbol{\mu}})\chi_{\gamma\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\epsilon]}\end{aligned}\ ] ] * @xmath245 : @xmath246 } \nonumber\\ \fl & = & -v'(0)\chi_{\alpha\beta\gamma\epsilon}({\boldsymbol{\mu } } ) - v''(0)\chi_{\alpha\beta}({\boldsymbol{\mu}})\chi_{\gamma\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\epsilon]}\\ \fl { { { v}^{{\{\alpha\beta\gamma\}_{3}\{\epsilon\}_{1 } } } } } & = & v^{(4)}(0 ) { { { d}^{{\{\alpha\}_{1}\{\}_{0 } } } } } { { { d}^{{\{\beta\}_{1}\{\}_{0 } } } } } { { { d}^{{\{\gamma\}_{1}\{\}_{0 } } } } } { { { d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + \nonumber\\ \fl & & v^{(3)}(0 ) \left({{{d}^{{\{\alpha\beta\}_{2}\{\}_{0 } } } } } { { { d}^{{\{\gamma\}_{1}\{\}_{0 } } } } } { { { d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + { { { d}^{{\{\epsilon\}_{1}\{\alpha\}_{1 } } } } } { { { d}^{{\{\beta\}_{1}\{\}_{0 } } } } } { { { d}^{{\{\gamma\}_{1}\{\}_{0 } } } } } \right ) + \nonumber\\ \fl & & v''(0 ) \left({{{d}^{{\{\alpha\beta\gamma\}_{3}\{\}_{0 } } } } } { { { d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + { { { d}^{{\{\alpha\beta\}_{2}\{\epsilon\}_{1 } } } } } { { { d}^{{\{\gamma\}_{1}\{\}_{0 } } } } } + { \textcolor{blue } { { { { d}^{{\{\alpha\beta\}_{2}\{\}_{0 } } } } } { { { d}^{{\{\gamma\}_{1}\{\epsilon\}_{1 } } } } } } } \right ) + \nonumber\\ \fl & & v'(0 ) { { { d}^{{\{\alpha\beta\gamma\}_{3}\{\epsilon\}_{1 } } } } } + { \mathrm{sym}[\alpha,\beta,\gamma ] } = - v''(0)\chi_{\alpha\beta}({\boldsymbol{\mu}})\chi_{\gamma\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\epsilon]}\nonumber\\ \\ \fl { { { f_\epsilon^\nu}^{{\{\}_{0}\{\alpha\beta\gamma\}_{3 } } } } } & = & -\left({{{v}^{{\{\epsilon\}_{1}\{\alpha\beta\gamma\}_{3}}}}}+(2\nu-1){{{v}^{{\{\alpha\beta\gamma\}_{3}\{\epsilon\}_{1}}}}}\right)= v'(0)\chi_{\alpha\beta\gamma\epsilon}({\boldsymbol{\mu}})\nonumber\\ \fl & & + 2\nu v''(0)\chi_{\alpha\beta}({\boldsymbol{\mu}})\chi_{\gamma\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\epsilon]}\end{aligned}\ ] ] * @xmath247 : + @xmath248[\gamma,\delta ] } \nonumber \\ \fl & = & v''(0)\left ( \chi_{\alpha\beta\epsilon}({\boldsymbol{\mu}})\chi_{\gamma\delta}({\boldsymbol{\mu } } ) - \chi_{\alpha\gamma\delta}({\boldsymbol{\mu}})\chi_{\beta\epsilon}({\boldsymbol{\mu } } ) \right ) + { \mathrm{sym}[\alpha,\beta,\epsilon][\gamma\delta]}\end{aligned}\ ] ] + @xmath249[\gamma,\delta ] } \nonumber \\ \fl & = & v''(0)\left(2\chi_{\alpha\beta\epsilon}({\boldsymbol{\mu}})\chi_{\gamma\delta}({\boldsymbol{\mu}})+\chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu}})\right)+{\mathrm{sym}[\alpha,\beta,\epsilon][\gamma,\delta]}\\ \fl \nonumber\\ \fl { { { f_\epsilon^\nu}^{{\{\alpha\beta\}_{2}\{\gamma\delta\}_{2 } } } } } & = & -\left({{{v}^{{\{\alpha\beta\epsilon\}_{3}\{\gamma\delta\}_{2}}}}}+(2\nu-1){{{v}^{{\{\gamma\delta\}_{2}\{\alpha\beta\epsilon\}_{3}}}}}\right)=v''(0 ) \chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu}})+{\mathrm{sym}[\alpha,\beta,\gamma,\delta,\epsilon]}\nonumber\\ \fl & & -2\nu v''(0)\left(2\chi_{\alpha\beta\epsilon}({\boldsymbol{\mu}})\chi_{\gamma\delta}({\boldsymbol{\mu}})+\chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu}})\right)+{\mathrm{sym}[\alpha,\beta,\epsilon][\gamma,\delta]}\end{aligned}\ ] ] * @xmath250 : + @xmath251 } \nonumber \\ \fl & = & v'(0)\chi_{\alpha\beta\gamma\delta\epsilon}({\boldsymbol{\mu } } ) + v''(0 ) \chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\delta,\epsilon]}\\ \fl { { { v}^{{\{\}_{0}\{\alpha\beta\gamma\delta\epsilon\}_{5 } } } } } & = & v^{(5)}(0 ) { { { d}^{{\{\}_{0}\{\alpha\}_{1}}}}}{{{d}^{{\{\}_{0}\{\beta\}_{1}}}}}{{{d}^{{\{\}_{0}\{\gamma\}_{1}}}}}{{{d}^{{\{\}_{0}\{\delta\}_{1}}}}}{{{d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + \nonumber\\ \fl & & v^{(4)}(0 ) { { { d}^{{\{\}_{0}\{\alpha\beta\}_{2}}}}}{{{d}^{{\{\}_{0}\{\gamma\}_{1}}}}}{{{d}^{{\{\}_{0}\{\delta\}_{1}}}}}{{{d}^{{\{\}_{0}\{\epsilon\}_{1}}}}}+\nonumber\\ \fl & & v^{(3)}(0 ) \left ( { { { d}^{{\{\}_{0}\{\alpha\beta\gamma\}_{3}}}}}{{{d}^{{\{\}_{0}\{\delta\}_{1}}}}}{{{d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + { { { d}^{{\{\}_{0}\{\alpha\beta\}_{2}}}}}{{{d}^{{\{\}_{0}\{\gamma\delta\}_{2}}}}}{{{d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } \right ) + \nonumber\\ \fl & & v''(0 ) \left ( { { { d}^{{\{\}_{0}\{\alpha\beta\gamma\delta\}_{4}}}}}{{{d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + { \textcolor{blue } { { { { d}^{{\{\}_{0}\{\alpha\beta\gamma\}_{3}}}}}{{{d}^{{\{\}_{0}\{\delta\epsilon\}_{2 } } } } } } } \right ) + \nonumber\\ \fl & & v'(0 ) { \textcolor{blue } { { { { d}^{{\{\}_{0}\{\alpha\beta\gamma\delta\epsilon\}_{5 } } } } } } } + { \mathrm{sym}[\alpha,\beta,\gamma,\delta,\epsilon ] } \nonumber \\ \fl & = & 4v'(0)\chi_{\alpha\beta\gamma\delta\epsilon}({\boldsymbol{\mu}})+2v''(0)\chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\delta,\epsilon]}\\ \fl \nonumber\\ \fl { { { f_\epsilon^\nu}^{{\{\alpha\beta\gamma\delta\}_{4}\{\}_{0 } } } } } & = & -\left({{{v}^{{\{\alpha\beta\gamma\epsilon\}_{5}\{\}_{0}}}}}+(2\nu-1){{{v}^{{\{\alpha\beta\gamma\delta\epsilon\}_{5}\{\}_{0}}}}}\right)=\nonumber\\ \fl & = & ( 3 - 8\nu)v'(0)\chi_{\alpha\beta\gamma\delta\epsilon}({\boldsymbol{\mu}})+(1 - 4\nu)v''(0)\chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\delta,\epsilon]}\nonumber\\\end{aligned}\ ] ] * @xmath252 : @xmath253 } \nonumber \\ \fl & = & -v'(0)\chi_{\alpha\beta\gamma\delta\epsilon}({\boldsymbol{\mu } } ) -v''(0)\left ( \chi_{\alpha\beta\epsilon}({\boldsymbol{\mu}})\chi_{\gamma\delta}({\boldsymbol{\mu } } ) + 2\chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu } } ) \right ) + { \mathrm{sym}[\alpha,\beta,\gamma,\delta ] } \nonumber\\ \fl & = & -v'(0)\chi_{\alpha\beta\gamma\delta\epsilon}({\boldsymbol{\mu } } ) -v''(0 ) \chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\delta,\epsilon ] } \nonumber\\ \fl & & - v''(0 ) \chi_{\alpha\beta\gamma}({\boldsymbol{\mu } } ) \chi_{\delta\epsilon } + { \mathrm{sym}[\alpha,\beta,\gamma,\delta]}\\ \fl { { { v}^{{\{\alpha\beta\gamma\delta\}_{4}\{\epsilon\}_{1 } } } } } & = & v^{(5)}(0 ) { { { d}^{{\{\alpha\}_{1}\{\}_{0}}}}}{{{d}^{{\{\beta\}_{1}\{\}_{0}}}}}{{{d}^{{\{\gamma\}_{1}\{\}_{0}}}}}{{{d}^{{\{\delta\}_{1}\{\}_{0}}}}}{{{d}^{{\{\}_{0}\{\epsilon\}_{0 } } } } } + \nonumber \\ \fl & & v^{(4)}(0 ) \left ( { { { d}^{{\{\alpha\beta\}_{2}\{\}_{0}}}}}{{{d}^{{\{\gamma\}_{1}\{\}_{0}}}}}{{{d}^{{\{\delta\}_{1}\{\}_{0}}}}}{{{d}^{{\{\}_{0}\{\epsilon\}_{0 } } } } } + { { { d}^{{\{\alpha\}_{1}\{\epsilon\}_{1}}}}}{{{d}^{{\{\beta\}_{1}\{\}_{0}}}}}{{{d}^{{\{\gamma\}_{1}\{\}_{0}}}}}{{{d}^{{\{\delta\}_{1}\{\}_{0 } } } } } \right ) + \nonumber\\ \fl & & v^{(3)}(0 ) \left ( { { { d}^{{\{\alpha\beta\gamma\}_{3}\{\}_{0}}}}}{{{d}^{{\{\delta\}_{1}\{\}_{0}}}}}{{{d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + { { { d}^{{\{\alpha\beta\}_{2}\{\epsilon\}_{1}}}}}{{{d}^{{\{\delta\}_{1}\{\gamma\}_{0}}}}}{{{d}^{{\{\delta\}_{1}\{\}_{0 } } } } } \right . + \nonumber\\ \fl & & \qquad\qquad \left . { { { d}^{{\{\alpha\beta\}_{2}\{\}_{0}}}}}{{{d}^{{\{\gamma\delta\}_{2}\{\}_{0}}}}}{{{d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + { { { d}^{{\{\alpha\}_{1}\{\epsilon\}_{1}}}}}{{{d}^{{\{\beta\gamma\}_{2}\{\}_{0}}}}}{{{d}^{{\{\delta\}_{1}\{\}_{0 } } } } } \right ) + \nonumber\\ \fl & & v''(0 ) \left ( { { { d}^{{\{\alpha\beta\gamma\delta\}_{4}\{\}_{0}}}}}{{{d}^{{\{\}_{0}\{\epsilon\}_{1 } } } } } + { { { d}^{{\{\alpha\beta\gamma\}_{3}\{\epsilon\}_{1}}}}}{{{d}^{{\{\delta\}_{1}\{\}_{0}}}}}\right . + \nonumber \\ \fl & & \qquad\qquad\left . { { { d}^{{\{\alpha\beta\}_{2}\{\epsilon\}_{1}}}}}{{{d}^{{\{\gamma\delta\}_{2}\{\}_{0 } } } } } + { \textcolor{blue } { { { { d}^{{\{\alpha\beta\gamma\}_{3}\{\}_{0}}}}}{{{d}^{{\{\delta\}_{1}\{\epsilon\}_{1 } } } } } } } \right ) + \nonumber\\ \fl & & v'(0 ) { { { d}^{{\{\alpha\beta\gamma\delta\}_{4}\{\epsilon\}_{1 } } } } } + { \mathrm{sym}[\alpha,\beta,\gamma,\delta ] } \nonumber \\ \fl & = & -v''(0 ) \chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu}})+{\mathrm{sym}[\alpha,\beta,\gamma,\delta ] } \\ \fl \nonumber\\ \fl { { { f_\epsilon^\nu}^{{\{\}_{0}\{\alpha\beta\gamma\delta\}_{4 } } } } } & = & -\left({{{v}^{{\{\epsilon\}_{1}\{\alpha\beta\gamma\delta\}_{4}}}}}+(2\nu-1){{{v}^{{\{\alpha\beta\gamma\delta\}_{4}\{\epsilon\}_{1}}}}}\right)=v'(0 ) \chi_{\alpha\beta\gamma\delta\epsilon}({\boldsymbol{\mu } } ) + \nonumber\\ \fl & & v''(0 ) \chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\delta,\epsilon ] } + 2\nu v''(0 ) \chi_{\alpha\beta\gamma}({\boldsymbol{\mu}})\chi_{\delta\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\delta]}\nonumber\\\end{aligned}\ ] ] * @xmath254 : + @xmath255[\beta,\gamma ] } \nonumber \\ \fl & = & v''(0 ) \chi_{\alpha\beta}({\boldsymbol{\mu}})\chi_{\gamma\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\epsilon]}\\ \fl \nonumber\\ \fl { { { f_\epsilon^\nu}^{{\{\alpha\}_{1}\{\beta\gamma\}_{2}}}}}&=&-\left({{{v}^{{\{\alpha\epsilon\}_{2}\{\beta\gamma\}_{2}}}}}+(2\nu-1){{{v}^{{\{\beta\gamma\}_{2}\{\alpha\epsilon\}_{2}}}}}\right ) = -2\nu v''(0 ) \chi_{\alpha\beta}({\boldsymbol{\mu}})\chi_{\gamma\epsilon}({\boldsymbol{\mu } } ) + { \mathrm{sym}[\alpha,\beta,\gamma,\epsilon]}\nonumber\\\end{aligned}\ ] ]

finite - dimensional manifolds , whether smooth , real- or complex - analytic , are commonly defined via charts and atlases . the other standard way of defining them relies on a dual point of view , focussing on the functions rather than on the points themselves , and this is achieved via a sheaf - theoretical approach . more precisely , a smooth @xmath0-dimensional manifold @xmath1 is then defined as a locally ringed space @xmath2 that is locally isomorphic to the locally ringed space @xmath3 . the sheaf - theoretical approach is hardly avoidable when one wants to deal with singular generalizations of manifolds ( varieties or schemes for instance ) , or with non - reduced situations " , such as supermanifolds , where the rings of functions " have nilpotents . in this last example , a section of the structural sheaf is not determined by its values on the points of the underlying topological space , which makes the sheaf - theoretical approach particularly relevant in defining supermanifolds . + in finite dimensions , the two definitions of manifolds ( via atlases and as certain locally ringed spaces ) are well - known to be equivalent . in infinite dimensions , the situation is quite different . infinite - dimensional manifolds , whether locally modeled on banach spaces , frchet spaces or general locally convex spaces , have almost always been defined in terms of charts and atlases . one reason for that is the belief , following the thesis of douady @xcite , that the sheaf of scalar - valued functions does not give sufficient information to define the morphisms ( contrary to the finite - dimensional case , where defining the smooth functions valued in @xmath4 suffices to determine the morphisms valued in @xmath5 for every natural number @xmath6 ) . in @xcite , mazet defines a category of infinite - dimensional analytic spaces , precisely in terms of ringed spaces . however , his category leads to pathologies ( such as the sum of two analytic maps not necessarily being analytic ) . douady avoids these pathologies by introducing a third approach for capturing the notion of space , which he uses to define his category of banach analytic spaces . namely , given a category @xmath7 , douady defines a _ @xmath7-functored space _ @xmath8 to be a pair @xmath9 where @xmath10 is a topological space , and @xmath11 is a covariant functor from @xmath7 to the category of sheaves of sets on @xmath10 . in this way , for every object @xmath12 in @xmath7 ( thought of as a possible target ) , one associates a sheaf of sets @xmath13 ( thought of as the sheaf of @xmath12-valued morphisms on @xmath10 ) . compared to a ringed space , a functored space encodes already in its structural functor " the definition of the morphisms valued in any target space ( from a certain category ) . + the functored space approach obviously adds a supplementary technical layer " , which can be felt already when defining the local models for banach analytic spaces ( as functored spaces ) . thus , unless the recourse to functored spaces is absolutely necessary , it is preferable to deal with the more traditional setting of ringed spaces , i.e. to associate only a single structure sheaf instead of a sheaf - valued functor to each space . we are thus lead to the question whether the insufficiency of the sheaf of scalar - valued functions pointed out by douady ( and the related pathologies ) appears also in the non - singular setting of infinite - dimensional smooth manifolds . + in this paper , we address this question by observing that the obstruction to define infinite - dimensional manifolds as ringed spaces boils down to the failure of a _ scalarwise smooth _ map between open sets of locally convex spaces to be smooth . more precisely , given two locally convex spaces @xmath14 and @xmath12 , and given an open subset @xmath15 of @xmath14 , we adopt a standard notion of smoothness for maps @xmath16 ( _ cf . _ definition [ bhmn ] ) , following bastiani , hamilton , milnor , neeb , ... then , a map @xmath17 is said to be _ scalarwise smooth _ if the function @xmath18 is smooth for every smooth function @xmath19 . the chain rule implies clearly that smooth maps are scalarwise smooth . the converse is easily seen to be true in finite dimensions ( just take the linear forms @xmath20 dual to a basis @xmath21 ) . in infinite dimensions , the converse is non - trivial . in the first main result of this paper ( theorem [ scalarwise smoothness implies smoothness ] ) , we prove that it is true whenever @xmath12 is mackey - complete and the mackey - closure topology on @xmath14 coincides with the given topology ( the necessary definitions will be recalled throughout the paper , but let us already note that the assumption on @xmath14 corresponds to a large class of locally convex spaces which contains frchet spaces and duals of frchet - schwartz spaces , and that every complete locally convex space is mackey - complete ) . in the second main result of the paper ( theorem [ smooth equals ringed morphism ] and corollary [ manifolds = structure sheaf - smooth manifolds ] ) , we show that for infinite - dimensional manifolds modeled on locally convex spaces @xmath14 satisfying the assumptions of theorem [ scalarwise smoothness implies smoothness ] , the definition via charts and atlases is equivalent to the definition via ringed spaces . finally , we complete our comparison of the various definitions of infinite - dimensional manifolds by showing that the definition via charts and atlases is equivalent to the one based on functored spaces . + our paper is organized as follows . in section 2 , we present a proof of the special case of theorem [ scalarwise smoothness implies smoothness ] , when the domain is @xmath4 . more precisely , we show that for a mackey - complete locally convex space @xmath14 , a curve @xmath22 is smooth if and only if it is scalarwise smooth . while this result and its idea of proof are not new ( compare @xcite ) , they are crucial to our proof of theorem [ scalarwise smoothness implies smoothness ] : we recall them in a concise but self - contained way for the convenience of the reader , which gives us also the opportunity to introduce our notations for the rest of the paper . in section 3 , we recall the definition of smooth maps that we will be using , and prepare and prove theorem [ scalarwise smoothness implies smoothness ] , using as an intermediate step the calculus of _ convenient smoothness _ studied by kriegl and michor . in section 4 , we present in detail the three definitions of infinite - dimensional manifolds under investigation here , and prove the comparison results ( theorem [ smooth equals ringed morphism ] and corollary [ manifolds = structure sheaf - smooth manifolds ] ) mentioned above . + _ acknowledgements . _ we would like to thank alexander alldridge for discussions on douady s espaces foncts " . the goal of this section is to recall the proof of the following result : if @xmath23 is a map from @xmath4 to a complete real ( hausdorff ) locally convex space @xmath14 , and @xmath24 is the continuous dual of @xmath14 , then smoothness of @xmath25 for every @xmath26 implies the smoothness of @xmath23 ( theorem [ weak smoothness of curves ] ) . while this is obvious if @xmath14 is finite - dimensional ( it is enough to take the projections @xmath27 where @xmath28 is an arbitrary basis of @xmath14 ) , proving that it remains true in the general case requires more work . the strategy ( essentially taken from @xcite ) , is the following . + in any locally convex space , there is a natural notion of bounded set . the collection of these bounded sets ( the von neumann bornology " ) is not very sensitive to the locally convex topology : a classical theorem of mackey in functional analysis shows that if one varies the topology while keeping the same dual space , the bounded sets remain the same . as a consequence , one can view the bounded sets from the perspective of the weak topology instead of the given topology . in the weak topology , it is natural and immediate that a subset of @xmath14 whose image by every linear functional is bounded must be itself bounded . + on the other hand , for curves , being @xmath29 is ultimately a bornological concept : the @xmath29 curves remain the same if one changes the locally convex topology , while keeping the same underlying bornology . this follows from the fact that a @xmath30 curve is locally lipschitz ( by the mean value theorem ) , and the lipschitz condition ( which is essentially bornological ) , implies continuity . translating smoothness in terms of lipschitz conditions ( involving bounded sets ) , it becomes possible to use the dual characterization of boundedness given by mackey s theorem , to obtain a dual characterization of smoothness . + in what follows , we recall , for the convenience of the reader and for later reference , the details of the above arguments , starting with mackey s theorem , the cornerstone in proving theorem [ weak smoothness of curves ] as well as other results in this paper . for a proof of the former theorem , see , e.g. , theorem 36.2 in @xcite or theorem 8.3.4 in @xcite . + ( mackey s theorem ) let @xmath14 be a locally convex space , and @xmath31 a subset of @xmath14 . if @xmath32 is bounded for every @xmath26 , then @xmath31 is bounded . + let @xmath14 be a locally convex space , and @xmath22 a curve . 1 . if @xmath33 is an open subset of @xmath4 , we say that @xmath23 is * lipschitz * on @xmath33 if the @xmath34 is bounded in @xmath14 . we say that @xmath23 is * locally lipschitz * if every point in @xmath4 has a neighborhood on which @xmath23 is lipschitz . + let @xmath14 be a locally convex space , and @xmath22 a curve . for @xmath35 , we say that @xmath23 is * of class @xmath36 * if all the derivatives of @xmath23 up to order @xmath6 exist , and @xmath37 is locally lipschitz . + if @xmath38 is any subset of @xmath14 , we will denote by @xmath39 the absolute convex hull of the closure of @xmath38 . we will need the following version of the mean value theorem , for curves in a locally convex space . + ( mean value theorem ) let @xmath14 be a locally convex space , and @xmath40\longrightarrow e$ ] a curve which is continuous on @xmath41 $ ] , and differentiable on @xmath42a , b[$ ] . then @xmath43 * proof :* _ cf . _ @xcite , i.1.4 . @xmath44 + let @xmath14 be a locally convex space , and @xmath22 a curve which is differentiable on an open interval @xmath45 . if @xmath46 is bounded on @xmath33 , then @xmath23 is lipschitz on @xmath33 . * proof :* let @xmath47 with @xmath48 . by the mean value theorem , @xmath49 so @xmath50 , and this last set is bounded since the absolute convex hull of any bounded set is bounded . @xmath44 + [ c^k and lip^k ] let @xmath14 be a locally convex space , and @xmath22 a curve . 1 . if @xmath23 is of class @xmath51 , then @xmath23 is of class @xmath52 . if @xmath23 is of class @xmath52 , then @xmath23 is of class @xmath53 . * proof :* 1 . if @xmath23 is @xmath51 , then @xmath54 is @xmath30 . this implies ( via the preceding corollary ) that @xmath54 is locally lipschitz . thus , @xmath23 is @xmath52 . if @xmath23 is @xmath52 , then @xmath54 is locally lipschitz . this implies that @xmath54 is continuous . thus , @xmath23 is @xmath53 . @xmath44 + let @xmath14 be a locally convex space , and @xmath22 a curve . then @xmath23 is of class @xmath29 if and only if @xmath23 is of class @xmath55 for all @xmath35 . * proof :* if @xmath23 is smooth , then @xmath23 is @xmath51 for every @xmath6 . the first part of proposition [ c^k and lip^k ] implies then that @xmath23 is @xmath52 for every @xmath6 . conversely , if @xmath23 is @xmath55 for every @xmath6 , then the second part of proposition [ c^k and lip^k ] implies that @xmath23 is @xmath56 for every @xmath6 , and so @xmath23 is smooth . @xmath44 + let @xmath14 be a locally convex space . a curve @xmath22 is said to be * weakly smooth * if @xmath25 is smooth for every @xmath26 . + let @xmath14 be a locally convex space . 1 . a sequence @xmath57 in @xmath14 is said to be * mackey - convergent * to a point @xmath58 if there exists an absolutely convex bounded set @xmath59 , and a sequence @xmath60 of real numbers converging to 0 , such that @xmath61 for all @xmath0 . 2 . a sequence @xmath57 in @xmath14 is said to be * mackey - cauchy * if there exists an absolutely convex bounded set @xmath59 , and a double sequence @xmath62 of real numbers converging to 0 , such that @xmath63 for all @xmath64 . + we also have the same notions for nets ( just replace sequence " by net " everywhere in the preceding definition ) . @xmath14 is called * mackey - complete * if every mackey - cauchy net in @xmath14 converges . + every complete locally convex space is sequentially complete , and in turn sequential completeness implies mackey - completeness . for metrizable locally convex spaces the three notions of completeness coincide . given a locally convex space @xmath14 , the completion @xmath65 yields a complete ( and thus mackey - complete ) locally convex space together with a continuous linear embedding @xmath66 having dense image . ( compare , e.g. theorem 5.2 in @xcite . ) a continuous curve @xmath67 yields then a continuous curve @xmath68 , and @xmath23 is weakly smooth if and only if @xmath69 is weakly smooth . + [ weak smoothness of curves ] let @xmath14 be a mackey - complete locally convex space and @xmath22 a curve . then @xmath23 is smooth if and only if it is weakly smooth . * proof :* if @xmath23 is smooth , then it is clear by chain rule that @xmath25 is smooth for every @xmath26 . to prove the converse , suppose that for every @xmath26 , the function @xmath25 is smooth . we first prove that @xmath23 is differentiable . let @xmath70 , and @xmath33 a compact interval about @xmath71 . set @xmath72 for all @xmath73 . we need to show that @xmath74 has a limit as @xmath75 . let @xmath76 . then @xmath77 . now for each @xmath73 , we have + @xmath78 . since @xmath25 is smooth , the same must be true for @xmath79 . in particular , the mean value theorem implies the existence of @xmath800,1[$ ] such that @xmath81 . now @xmath82 for every @xmath830,1[$ ] ( by convexity of @xmath33 ) . since @xmath84 is continuous and @xmath33 is compact , we deduce that there is @xmath85 such that @xmath86 for all @xmath870,1[$ ] . but then , @xmath88 . thus , @xmath32 is bounded . since @xmath89 was arbitrary , we deduce by mackey s theorem that @xmath31 is bounded as well , i.e. @xmath74 is lipschitz on @xmath90 . otherwise stated , the net @xmath91 with @xmath92 is mackey - cauchy . mackey - completeness of @xmath14 implies that @xmath74 has a continuous extension to @xmath33 , which allows us to define @xmath93 as @xmath94 . thus , @xmath23 is differentiable , and for every @xmath26 , we have @xmath95 ( by chain rule and linearity of @xmath89 ) . + second , we prove that @xmath46 is locally lipschitz . let @xmath70 , and @xmath33 a compact interval about @xmath71 . then for every @xmath26 , the function @xmath96 is lipschitz on @xmath33 ( since @xmath97 , being continuous , is bounded on @xmath33 ) . let @xmath98 . then , since @xmath99 , we have @xmath100 . since @xmath96 is lipschitz on @xmath33 , we have that @xmath101 is bounded . but @xmath89 is arbitrary . using again mackey s theorem , we deduce that @xmath102 is bounded , hence @xmath46 is lipschitz on @xmath33 , and so @xmath46 is locally lipschitz , that is , @xmath23 is @xmath103 . + replacing @xmath23 by @xmath46 in the above chain of arguments , we arrive at the conclusion that @xmath46 is @xmath103 as well , i.e. @xmath23 is @xmath104 . by induction , we conclude that @xmath23 is @xmath36 for all @xmath6 , i.e. @xmath23 is smooth . in this section , we start by recalling the notion of a smooth map from an open subset @xmath15 of a locally convex space @xmath14 into a locally convex space @xmath12 . among the various existing notions of ( differentiable and ) smooth maps , we choose what is called @xmath105 in the book of keller ( @xcite ) , going back at least to bastiani ( compare definitions ii.3.1 and ii.2.2 in @xcite ) . this definition of smoothness turns out to be appropriate for the construction of an applicable theory of infinite - dimensional manifolds and lie groups , compare notably @xcite , @xcite , @xcite , @xcite . the goal of this section is to generalize theorem [ weak smoothness of curves ] to maps having infinite - dimensional source . namely , we want to show that if @xmath14 is for example a frchet space or the continuous dual of a frchet - schwartz space ( _ cf . _ below for details ) , a map @xmath106 is smooth if and only if it is scalarwise smooth . by _ scalarwise smooth _ , we mean that @xmath107 for every @xmath108 . + as a matter of fact , this result is very easy to prove if we replace the notion of smoothness of bastiani et al . by another one : the convenient smoothness " considered by frlicher , kriegl and michor ( compare notably @xcite ) . a map @xmath16 is said to be _ conveniently smooth _ if it sends smooth curves to smooth curves . since the notion of convenient smoothness relies on curves , theorem [ weak smoothness of curves ] immediately yields the equivalence between convenient smoothness and scalarwise convenient smoothness of maps . + for applications , it is thus highly desirable to establish that convenient smoothness coincides with smoothness in the sense of bastiani - hamilton - milnor - neeb , for relevant classes of locally convex spaces ( generalized boman theorem " ) . that this is indeed the case was already observed in @xcite in case where the source @xmath14 is a frchet space . below we will first prove a more general result of this type , before giving our crucial characterization of smoothness in terms of weak resp . scalarwise smoothness . + let @xmath14 and @xmath12 be locally convex spaces , and @xmath15 an open subset of @xmath14 . for a continuous map @xmath16 , the _ gteaux derivative _ of @xmath17 at a point @xmath109 in the direction of a vector @xmath110 is defined by @xmath111 provided the limit exists . + [ bhmn ] the map @xmath17 is said to be * of class @xmath30 * if @xmath112 , @xmath113 exists and is continuous . we define inductively a map @xmath17 to be * of class @xmath114 * if it is of class of class @xmath30 and @xmath115 is of class @xmath56 . furthermore , a map is called being * of class @xmath29 * or * smooth * if it is @xmath56 for all @xmath116 . + if @xmath14 and @xmath12 are banach spaces , being @xmath30 in the above sense is weaker than the usual notion of @xmath30 in the sense of frchet differentiability , which requires the map @xmath117 to be continuous as map from @xmath15 to @xmath118 , equipped with the operator norm topology . however , @xmath119 in the above sense implies @xmath30 in the usual frchet differentiability sense ( _ cf . _ proposition 2.7.1 in @xcite , or theorem i.7 in @xcite ) , so that in banach spaces , the two definitions lead to the same smooth maps . + next we turn to the notion of convenient smoothness . before we recall its definition and main properties , let us already note that in general , it is possible to find conveniently smooth maps which are not even continuous ! this hints to the fact that there should be a different topology which is more adapted to convenient smoothness , and for which conveniently smooth maps are automatically continuous . since the definition of convenient smoothness relies on smooth curves , it is natural to use smooth curves to define this topology . + let @xmath14 be a locally convex space . the * mackey - closure topology * ( also called * @xmath120-topology * ) is the finest topology on @xmath14 making all the smooth curves @xmath22 continuous . + the @xmath120-topology on @xmath14 is clearly finer than the given locally convex topology . note that if @xmath14 is a frchet space or the dual of a frchet - schwartz space , then the two topologies coincide ( @xcite , theorem i.4.11 ) . for frchet spaces , we give a short and elementary derivation of this fact in this appendix . note also that there are locally convex spaces such that the mackey - closure topology is not even a vector space topology and thus _ a fortiori _ does not equal the initially given topology . examples of this phenomenon are strict inductive limits of strictly increasing sequences of infinite - dimensional frchet spaces , as e.g. @xmath121 , the space of compactly supported smooth functions on a noncompact finite - dimensional smooth manifold . ( compare @xcite , proposition i.4.26 . ) + in section 2 , we have seen that smoothness of curves is ultimately a bornological concept , in the sense that the requirement of continuity of the derivatives can be replaced by that of being locally lipschitz , a notion that relies on the bounded sets . consequently , it should not be very surprising that the @xmath120-topology can be characterized , as the next two propositions will show , in terms of mackey - convergent sequences ( which , by the following lemma , can be thought of as the sequential counterpart of locally lipschitz maps ) . + [ lipschitz preserves mackey - convergence ] let @xmath14 be a locally convex space , and @xmath22 a locally lipschitz curve . for any convergent sequence of real numbers @xmath122 , the sequence @xmath123 in @xmath14 is mackey - convergent . * proof :* let @xmath124 . since @xmath23 is locally lipschitz , there exists a neighborhood @xmath33 of @xmath125 such that the set @xmath126 is bounded . let @xmath127 be such that @xmath128 implies @xmath129 . then @xmath130 is bounded , which proves that @xmath123 mackey - converges to @xmath131 . @xmath44 + recall that a sequence @xmath132 of points of a locally convex space @xmath14 _ converges fast _ to a point @xmath58 if for every @xmath133 , the set @xmath134 is bounded in @xmath14 . therefore , a sequence @xmath132 converges fast to @xmath135 if and only if , for each @xmath133 , there exists a bounded subset @xmath136 of @xmath14 such that @xmath137 for every @xmath138 . in particular , @xmath132 mackey - converges to @xmath135 . + [ special curve lemma ] ( special curve lemma ) let @xmath14 be a locally convex space , @xmath132 a sequence in @xmath14 , and @xmath135 a point of @xmath14 . if @xmath132 converges fast to @xmath135 , then there exists a smooth curve @xmath22 such that @xmath139 for every @xmath138 and @xmath140 . * proof :* _ cf . _ @xcite , i.2.8 . @xmath44 + let @xmath14 be a locally convex space . every sequence in @xmath14 that mackey - converges to some point @xmath58 has a subsequence that converges fast to @xmath135 . * proof :* let @xmath57 be a sequence in @xmath14 that mackey - converges to some point @xmath58 . by definition , there exists an absolutely convex bounded set @xmath59 , and a sequence @xmath60 of real numbers converging to 0 , such that @xmath141 for all @xmath0 . the @xmath142 can be chosen to be positive . since @xmath60 converges to 0 , there exists a positive integer @xmath143 such that @xmath144 . then , there exists a positive integer @xmath145 such that @xmath146 . by induction , one obtains in this way a strictly increasing sequence of positive integers @xmath147 such that @xmath148 for every @xmath149 . since @xmath31 is balanced , this implies that @xmath150 for every @xmath149 . for each @xmath133 , choose @xmath151 such that @xmath152 for every @xmath149 , and set @xmath153 . using again the fact that @xmath31 is balanced , we deduce that @xmath154 . since @xmath136 is bounded , this shows that the subsequence @xmath155 converges fast to @xmath135 . @xmath44 + let @xmath14 be a locally convex space . a subset @xmath15 of @xmath14 is @xmath120-open if and only if for every point @xmath109 and every sequence @xmath57 in @xmath14 that mackey - converges to @xmath135 , there exists @xmath127 such that @xmath128 implies @xmath156 . * proof :* suppose @xmath15 is @xmath157-open , let @xmath109 and @xmath57 a sequence in @xmath14 that mackey - converges to @xmath135 . assume that the set @xmath158 is infinite . then there exists @xmath159 , a subsequence of @xmath57 , such that none of its points @xmath160 is in @xmath15 . a fortiori , @xmath155 mackey - converges to @xmath135 as well . it follows from the above lemma that @xmath155 has a subsequence @xmath161 that converges fast to @xmath135 . by lemma [ special curve lemma ] , there exists a smooth curve @xmath22 such that @xmath162 for every @xmath163 and @xmath140 . since @xmath15 is @xmath157-open , @xmath164 is open in @xmath4 . moreover , @xmath165 . therefore , there exists @xmath127 such that @xmath166 implies @xmath167 . consequently , @xmath166 implies @xmath168 . this contradicts the fact that none of the @xmath160 belongs to @xmath15 . conversely , suppose that for every point @xmath109 and every sequence @xmath57 in @xmath14 that mackey - converges to @xmath135 , there exists @xmath127 such that @xmath128 implies @xmath156 . let @xmath23 be a smooth curve , @xmath169 , and @xmath122 a sequence of real numbers converging to @xmath170 . being smooth , @xmath23 is locally lipschitz , and so by lemma [ lipschitz preserves mackey - convergence ] , the sequence @xmath171 mackey - converges to @xmath172 . by our assumption , there exists @xmath127 such that @xmath128 implies @xmath173 . thus , @xmath128 implies @xmath174 , which shows that @xmath164 is open in @xmath4 . @xmath44 + let @xmath14 be a locally convex space . a subset @xmath38 of @xmath14 is @xmath120-closed if and only if for every mackey - convergent sequence @xmath57 of points of @xmath38 , the limit of @xmath57 belongs to @xmath38 . * proof :* suppose @xmath38 is @xmath120-closed , and let @xmath135 be the limit of a mackey - convergent sequence @xmath57 of points of @xmath38 . let @xmath175 , so that @xmath15 is @xmath120-open . if @xmath176 , then @xmath109 , and so there exists @xmath127 such that @xmath128 implies @xmath156 . this clearly contradicts the fact that @xmath177 for all @xmath0 . thus , @xmath178 . conversely , suppose @xmath38 has the stated property . if @xmath38 is not @xmath120-closed , then @xmath175 is not @xmath120-open . this implies the existence of a point @xmath109 and a sequence @xmath57 in @xmath14 that mackey - converges to @xmath135 , such that the set @xmath179 is infinite . this is tantamount to the existence of a subsequence @xmath155 of @xmath57 whose terms belong to @xmath38 . since @xmath57 mackey - converges to @xmath135 the same holds true for @xmath155 . since @xmath180 for each @xmath6 , the property assumed on @xmath38 implies that @xmath178 , contradiction . @xmath44 + we are ready to recall the precise definition and some of the properties of conveniently smooth maps . _ notably @xcite ) . + let @xmath14 and @xmath12 be locally convex spaces , @xmath15 a @xmath120-open subset of @xmath14 , and @xmath16 a map . we say that @xmath17 is * conveniently smooth * if for every smooth curve @xmath181 , the curve @xmath182 is smooth . + it is easy to see that the composition of two conveniently smooth maps is conveniently smooth . + [ convenient smoothness implies c - infinity continuity ] let @xmath14 and @xmath12 be locally convex spaces , and @xmath15 an open subset of @xmath14 . any conveniently smooth map @xmath16 is continuous when @xmath14 is equipped with the @xmath120-topology . * proof :* let @xmath183 be an open subset of @xmath12 . to show that @xmath184 is @xmath120-open in @xmath15 , we need to show that @xmath185 is open in @xmath4 for every @xmath186 . but @xmath187 , which is clearly open in @xmath4 since the curve @xmath188 is smooth , and therefore continuous . @xmath44 + we denote by @xmath189 the set of conveniently smooth maps from @xmath15 to @xmath12 . + the equivalence between smoothness and convenient smoothness , when it holds , is not trivial to establish , even for functions from @xmath190 to @xmath4 . in this case , it was first proved by boman in 1967 @xcite . + ( boman s theorem ) @xmath191 * proof :* _ cf . _ @xcite or @xcite , i.3.4 . @xmath44 + now we concentrate on the structure of @xmath189 . + let @xmath14 and @xmath12 be locally convex spaces , @xmath15 a @xmath120-open subset of @xmath14 . then @xmath189 is a locally convex space for the coarsest topology making the maps @xmath192 , @xmath193 for all @xmath194 continuous . * proof :* it is not difficult to check that this topology on @xmath189 is defined by the family of seminorms @xmath195 , where we set @xmath196 for @xmath197 . moreover , if @xmath198 , let @xmath109 be such that @xmath199 , and @xmath200 the constant curve at @xmath135 . there exists a continuous seminorm @xmath74 on @xmath12 such that @xmath201 . take @xmath202 and @xmath203 . then @xmath204 thus , @xmath205 is separating . @xmath44 + one of the major benefits of working with conveniently smooth maps is the fact that they give rise to a cartesian closed category , as the next theorem will show . + [ cartesian closedness ] let @xmath206 , @xmath207 and @xmath12 be locally convex spaces , @xmath208 and @xmath209 @xmath120-open subsets of @xmath206 and @xmath207 respectively . then , as sets , @xmath210 * proof :* _ cf . _ theorem i.3.12 in @xcite . @xmath44 + a first consequence of cartesian closedness is a the following generalization of boman s theorem . + [ conveniently smooth on r^n = smooth on r^n ] let @xmath14 be a locally convex space . then , @xmath211 * proof :* let @xmath212 . by theorem [ cartesian closedness ] , we have @xmath213 @xmath214 . in particular , for every @xmath215 , the partial map @xmath216 , @xmath217 lies in @xmath218 . then @xmath219 + @xmath220 + @xmath221 . thus , all first - order partial derivatives of @xmath17 exist . inductively , one obtains the existence of all higher - order partial derivatives of @xmath17 . this proves that @xmath222 . conversely , if @xmath223 , then @xmath212 by the chain rule . @xmath44 + another consequence of cartesian closedness is the convenient smoothness of the differential . + let @xmath14 and @xmath12 be locally convex spaces , @xmath15 a @xmath120-open subset of @xmath14 , and @xmath224 . for every @xmath109 and @xmath110 , set @xmath225 . then @xmath226 . * proof :* we claim that the map @xmath227 defined by @xmath228 is conveniently smooth . indeed , let @xmath229 be a smooth curve , and set @xmath230 . then @xmath231 + @xmath232 . since @xmath233 is clearly conveniently smooth ( and smooth by corollary [ conveniently smooth on r^n = smooth on r^n ] ) , we deduce that the curve @xmath234 is smooth . thus , @xmath235 , and therefore @xmath236 . in particular , @xmath226 . @xmath44 + an immediate consequence of the above proposition , already observed in @xcite for frchet spaces , is a further generalization of boman s theorem , stating that for a reasonable class of locally convex spaces , convenient smoothness coincides with smoothness ( in the sense of the above definition ) . + let @xmath14 and @xmath12 be locally convex spaces , and @xmath15 an open subset of @xmath14 . assume that the given topology on @xmath14 coincides with the mackey - closure topology . then @xmath237 * proof :* by the chain rule , smoothness implies convenient smoothness . for the nontrivial direction , suppose @xmath16 is conveniently smooth . by the preceding proposition , @xmath112 is conveniently smooth as well . proposition [ convenient smoothness implies c - infinity continuity ] implies then that @xmath115 is continuous for the @xmath120-topology , therefore continuous by our assumption on @xmath14 . thus , @xmath17 is @xmath30 , and by induction , one obtains that @xmath17 is smooth . @xmath44 + we are ready to state and prove the main result of this section . + [ scalarwise smoothness implies smoothness ] let @xmath14 and @xmath12 be locally convex space spaces , and @xmath15 an open subset of @xmath14 . assume that the given topology on @xmath14 coincides with the mackey - closure topology and that f is mackey - complete . let @xmath16 be a continuous map . then the following are equivalent : 1 . @xmath17 is smooth . 2 . for every open subset @xmath183 of @xmath12 containing @xmath238 and every @xmath239 , we have @xmath240 . 3 . for every @xmath241 , we have @xmath240 ( i.e. @xmath17 is scalarwise smooth ) . 4 . for every @xmath242 , we have @xmath243 ( i.e. @xmath17 is weakly smooth ) . * proof :* @xmath244 is evident by chain rule , and @xmath245 follows upon taking @xmath246 . @xmath247 follows from the fact that every continuous linear map is smooth . it remains to show @xmath248 . suppose that for every @xmath242 , we have @xmath249 . for every smooth curve @xmath181 , the function @xmath250 is smooth since @xmath251 . by theorem [ weak smoothness of curves ] , the curve @xmath182 is smooth . this means that the map @xmath16 is conveniently smooth . by the above proposition , we conclude that @xmath17 must be smooth . a * class of l.c . model spaces or l.c . models * is a subclass @xmath252 of the class of real ( hausdorff ) locally convex spaces such that @xmath252 contains the numerical spaces @xmath253 for all @xmath254 . + typical examples are the class of numerical spaces @xmath253 for all @xmath254 , the class of finite - dimensional vector spaces , the class of banach spaces , the class of frchet spaces , and the class of all locally convex spaces . + [ atlas ] let @xmath252 be a class of l.c . model spaces , and @xmath255 a hausdorff topological space . 1 . a * smooth @xmath252-atlas * on @xmath255 is a family of pairs @xmath256 such that @xmath257 is an open cover of @xmath255 and for every @xmath258 , there exists a space @xmath259 from the class @xmath252 and a homeomorphism @xmath260 such that the following compatibility condition is satisfied : for every @xmath261 such that @xmath262 , the transition map @xmath263 is smooth ( in the sense of definition [ bhmn ] ) . 2 . given a * smooth @xmath252-atlas * @xmath264 on @xmath255 , a * ( compatible ) chart * on @xmath255 is a pair @xmath265 where @xmath15 is an open subset of @xmath255 and @xmath266 is a homeomorphism from @xmath15 onto an open subset @xmath267 of some space from the class @xmath252 such that the transition map between @xmath265 and every @xmath268 is smooth . a smooth @xmath252-atlas @xmath264 is said to be * maximal * if any chart @xmath265 compatible with @xmath264 belongs already to @xmath264 . a maximal atlas is also called a * smooth structure * , the class @xmath252 being understood . + let @xmath252 be a class of l.c . model spaces . a * smooth @xmath252-manifold * is a pair @xmath269 where @xmath255 is a hausdorff topological space and @xmath264 is a maximal smooth @xmath252-atlas on @xmath255 . + 1 . especially in finite dimensions , @xmath255 is often required to be second countable . however , as we are mainly interested in the generalization to infinite dimensions , we do not insist on this condition here . let us remark that we can not insist on first countability neither , since a non - metrizable locally convex space is not first countable ( see e.g. @xcite , proposition 25.1 ) . 2 . given a smooth @xmath252-atlas @xmath264 , there is always a unique maximal smooth @xmath252-atlas containing @xmath264 , obtained by adjoining to @xmath264 all the charts that are compatible with @xmath264 . 3 . if @xmath252 and @xmath270 are classes of l.c . models such that @xmath271 , then a smooth @xmath252-manifold is obviously a smooth @xmath270-manifold . denoting the class of all locally convex spaces by @xmath272 , every smooth @xmath252-manifold is then a smooth @xmath272-manifold . + given classes @xmath252 and @xmath273 of l.c . models , let @xmath269 , resp . @xmath274 be a smooth @xmath252- resp . @xmath273-manifold , and @xmath275 a continuous map . we say that @xmath17 is * smooth * if the following condition is fulfilled : for every @xmath276 and every @xmath277 such that @xmath278 , the following map ( between open sets in locally convex spaces ) is smooth : @xmath279 1 . if @xmath280 resp . @xmath281 is a smooth atlas ( not necessarily maximal ) of @xmath1 resp . @xmath282 , it is enough to check the above condition for charts in @xmath283 and @xmath284 . we notably obtain the notion of a smooth function @xmath285 on @xmath1 , namely a function @xmath286 such that for every chart @xmath287 ( atlas contained in @xmath264 ) , @xmath288 is a smooth function on the open set @xmath267 ( contained in some space from the class @xmath252 ) . since an open subset @xmath289 inherits obviously a smooth structure upon restricting the smooth structure to @xmath15 , we have a notion of smooth map / function defined on @xmath15 . + given a class of l.c . model spaces @xmath252 , let @xmath269 be a smooth @xmath252-manifold . the * sheaf @xmath290 of smooth functions * on @xmath1 is the subsheaf of the sheaf @xmath291 of continuous ( real - valued ) functions on the topological space @xmath255 , defined as the contravariant functor @xmath292 assigning to every open subset @xmath15 of @xmath255 the commutative @xmath4-algebra @xmath293 of smooth functions on @xmath15 . here , @xmath294 is the category of open subsets of @xmath255 with inclusions as morphisms , and @xmath295 is the category of unital commutative associative @xmath4-algebras with unital @xmath4-algebra homomorphisms as morphisms . + [ manifolds are ringed ] obviously , @xmath296 is a ringed space , with the stalks @xmath297 ( for @xmath298 ) being local unital @xmath4-algebras with maximal ideals @xmath299 . in short , @xmath296 is a locally ringed space . furthermore , @xmath296 is , as a locally ringed space , locally isomorphic to models @xmath300 , where @xmath14 is a space from the class @xmath252 , and @xmath301 is an open subset of @xmath14 . the atlas @xmath302 , where @xmath303 is the canonical inclusion , is obviously smooth and thus contained in a maximal atlas @xmath304 . calling @xmath305 the smooth manifold given by @xmath306 , we obviously have @xmath307 . + recall that if @xmath308 and @xmath309 are locally ringed spaces , a _ morphism of locally ringed spaces _ between them is a pair @xmath310 , where @xmath311 is a continuous map and @xmath312 is a morphism of sheaves of unital @xmath4-algebras such that for every @xmath313 , the induced unital @xmath4-algebra homomorphism @xmath314 is _ local _ , i.e. sends maximal ideal to maximal ideal . let us also recall that @xmath315 can be equivalently viewed upon as a sheaf morphism @xmath316 . in the sequel , we apply both formulations without further comment . + a locally ringed space @xmath308 is here said to be _ reduced _ if @xmath317 is a subsheaf of the sheaf of continuous functions @xmath318 . if @xmath319 is a morphism between _ reduced _ locally ringed spaces , then @xmath315 is given by @xmath320 , the pullback by @xmath321 ( and so the morphism @xmath17 is completely determined by the underlying continuous map @xmath321 ) . consequently , we often write @xmath17 instead of @xmath321 when the spaces are reduced . of course , we never do so in the non - reduced case , since @xmath315 is then part of the data . + let @xmath252 be a class of l.c a * structure sheaf - smooth @xmath252-manifold * is a reduced locally ringed space which , as a locally ringed space , is locally isomorphic to models @xmath322 , where @xmath301 is an open subset of a space from the class @xmath252 . + by remark [ manifolds are ringed ] , every smooth @xmath252-manifold is , in a natural way , a structure sheaf - smooth @xmath252-manifold . the converse is nontrivial . using the preceding section , we can nevertheless show the following results . + the class of l.c . models made of all the mackey - complete locally convex spaces will be denoted by @xmath323 . + [ smooth equals ringed morphism ] let @xmath252 be the class of locally convex spaces whose topology coincides with the mackey - closure topology . if @xmath269 is a smooth @xmath252-manifold , @xmath274 a smooth @xmath323-manifold , and @xmath275 a continuous map , then @xmath17 is smooth if and only if @xmath324 is a morphism of locally ringed spaces . * proof :* by the chain rule , smooth maps are morphisms of locally ringed spaces . assume now that @xmath17 is a morphism of locally ringed spaces . let @xmath265 and @xmath325 be charts of @xmath1 ( resp . @xmath282 ) with values in @xmath326 ( resp . @xmath327 ) such that @xmath278 . the continuous map @xmath328 is then a morphism of locally ringed spaces as well ( with respect to the natural structure sheaves ) . thus , @xmath329 satisfies condition ( ii ) of theorem [ scalarwise smoothness implies smoothness ] , which in turn implies that @xmath329 is smooth . it follows that @xmath17 itself is smooth . @xmath44 + the above theorem has the following very important consequence . + [ manifolds = structure sheaf - smooth manifolds ] let @xmath252 be again the class of locally convex spaces whose topology coincides with the mackey - closure topology . for every smooth @xmath252-manifold in the structure - sheaf sense @xmath2 , there is a canonical maximal atlas @xmath264 on @xmath255 such that @xmath269 is a smooth @xmath252-manifold , and @xmath330 . * proof :* let @xmath331 be a family of pairs such that @xmath257 is a covering of @xmath255 and for all @xmath258 , @xmath332 ( with @xmath333 and @xmath334 ) is an isomorphism of locally ringed spaces . furthermore , the continuous transition map @xmath335 , the latter map is an isomorphism of the locally ringed spaces @xmath336 and @xmath337 . by the preceding theorem , the map @xmath338 is then already smooth . it follows that @xmath283 is a smooth atlas on @xmath255 . we observe that the unique maximal atlas @xmath264 containing @xmath283 is canonically associated to the given locally ringed space . furthermore , the corresponding sheaf of smooth functions @xmath290 is equal to the structure sheaf @xmath339 . @xmath44 + the preceding theorem and its corollary show that for important classes of l.c . model spaces , as e.g. the class of frchet spaces , we can encode smoothness completely in sheaf - theoretic language . this approach simplifies the verification and application of smoothness in infinite dimensions , and allows generalizations to the non - reduced case , as e.g. for frchet supermanifolds " . in general , i.e. , for arbitrary classes of l.c . model spaces , resp . for our ( regular ) local models replaced by more general local models , as e.g. , complex - analytic sets in open subsets of complex locally convex spaces , this encoding might not be possible anymore . in order to circumvent this problem , douady introduced in @xcite the notions of the next definition . + let @xmath7 be a category , and denote , for a topological space @xmath340 , the category of sheaves ( of sets ) on @xmath340 by @xmath341 . if @xmath311 is a continuous map of topological spaces and @xmath342 is a covariant functor , then @xmath343 defined by @xmath344 is a covariant functor , called the * inverse image of @xmath345 by @xmath321*. + let @xmath7 be a category . 1 . a * @xmath7-functored space * is a pair @xmath346 where @xmath10 is a topological space , and @xmath347 is a covariant functor from @xmath7 to the category of sheaves on @xmath10 . we say that @xmath348 is the * structure functor * of @xmath346 . if @xmath9 and @xmath349 are @xmath7-functored spaces , a * morphism of @xmath7-functored spaces * between them is a pair @xmath310 where @xmath311 is a continuous map and @xmath350 is a natural transformation between the two functors @xmath351 and @xmath352 . + given a class @xmath273 of l.c . models , we continue to use the symbol @xmath273 , by a slight abuse of notation , to denote the category whose objects are open subsets of spaces from the class @xmath273 , and whose morphisms are the smooth maps between them ( in the sense of definition [ bhmn ] ) . such a category @xmath273 will be referred to as a category of l.c . models " . + given a category of l.c . models @xmath273 , an @xmath273-functored space @xmath353 is said to be * reduced * if for every @xmath354 and every open set @xmath355 , we have @xmath356 , and @xmath357 is a sheaf having local unital @xmath4-algebras as stalks . + if @xmath353 is a reduced @xmath273-functored space , then @xmath308 is a reduced locally ringed space . + given a class of l.c . models @xmath252 , let @xmath269 be a smooth @xmath252-manifold . for any category of l.c . models @xmath273 , consider the functor @xmath358 defined as follows : for every @xmath354 and every open set @xmath289 , let @xmath359 . then @xmath360 is called the * canonical @xmath273-functored space associated to @xmath1*. + [ manifolds are functored ] obviously , for any smooth @xmath252-manifold @xmath1 , @xmath360 is a reduced @xmath273-functored space . furthermore , @xmath360 is , as a @xmath273-functored space , locally isomorphic to model @xmath273-functored spaces @xmath361 , where @xmath301 is an open subset of a space from the class @xmath252 . + let @xmath252 be a class of l.c a * structure functor - smooth @xmath252-manifold * is a reduced @xmath362-functored space @xmath363 which , as a @xmath272-functored space , is locally isomorphic to model functor spaces @xmath364 , where @xmath301 is an open subset of a space from the class @xmath252 . + by remark [ manifolds are functored ] , for @xmath365 , every smooth @xmath252-manifold is , in a natural way , a structure functor - smooth @xmath252-manifold . + let @xmath252 be a class of l.c . models . if @xmath269 and @xmath274 are smooth @xmath252-manifolds , and @xmath275 a continuous map , then @xmath17 is smooth if and only if @xmath366 is a morphism of @xmath272-functored spaces . * proof :* note that a morphism @xmath367 between _ reduced _ functored spaces is completely determined by the underlying continuous map @xmath321 , the information contained in the natural transformation @xmath315 being that of all possible pullbacks by @xmath321 . consequently , as in the case of _ reduced _ locally ringed spaces , we write @xmath17 instead of @xmath367 . suppose now that @xmath368 is smooth . then , for any open subset @xmath183 of a locally convex space @xmath12 and for any open subset @xmath369 of @xmath370 , the pullback @xmath371 is well - defined by the chain rule . we claim that @xmath372 , which assigns to every @xmath183 the sheaf map @xmath373 given by the pullbacks @xmath374 , is a natural transformation . indeed , we have to show that given @xmath375 ( resp . @xmath376 ) open subset of a locally convex space @xmath377 ( resp . @xmath378 ) , and given a smooth map @xmath379 , we have @xmath380 for every open set @xmath369 in @xmath370 . but this is true since both lhs and rhs , when evaluated at an element @xmath381 , are equal to @xmath382 . thus , @xmath17 is a morphism of @xmath272-functored spaces . now we prove the converse of the theorem . assume that @xmath17 induces , via the pullbacks @xmath374 , a morphism of @xmath272-functored spaces . this means that for a germ of a smooth map defined on @xmath370 having values in an open subset of an arbitrary locally convex space , the pullback is the germ of a smooth map on @xmath255 . since this property as well as the condition of smoothness are local , we replace w.l.o.g . @xmath255 and @xmath370 by open subsets @xmath15 resp . @xmath383 in locally convex spaces @xmath384 in @xmath252 . taking now @xmath385 , the pullback @xmath386 equals @xmath17 and thus @xmath17 is smooth . @xmath44 + the above theorem has the following important consequence . + let @xmath252 be a class of l.c . models . for every structure functor - smooth @xmath252-manifold @xmath387 , there is a canonical maximal atlas @xmath264 on @xmath255 such that @xmath269 is a smooth @xmath252-manifold , and @xmath388 . * proof :* mutatis mutandis the proof of corollary [ manifolds = structure sheaf - smooth manifolds ] shows this corollary as well . @xmath44 + one could also take @xmath252 to be the class of all complex locally convex spaces . the notion of smooth @xmath252-manifold is then replaced by that of _ complex - analytic locally convex manifold_. to such a manifold @xmath269 , one can then associate a _ complex - analytic locally convex manifold in the structure - sheaf sense _ @xmath389 , where @xmath390 is the sheaf of complex - analytic functions on @xmath1 . one has then the following result . + let @xmath269 and @xmath274 be complex - analytic locally convex manifolds , and @xmath275 a continuous map . then @xmath17 is complex - analytic if and only if @xmath391 is a morphism of locally ringed spaces . + * proof :* by the chain rule , analytic maps are morphisms of locally ringed spaces . assume now that @xmath17 is a morphism of locally ringed spaces . since the condition of analyticity is local , we can proceed as in the proof of theorem [ smooth equals ringed morphism ] , which amounts to replace @xmath255 and @xmath370 by open subsets @xmath392 and @xmath393 of complex locally convex spaces @xmath14 and @xmath12 respectively . this gives a map @xmath394 which is scalarwise analytic . in particular , viewed as a map from @xmath392 to @xmath12 , @xmath329 is weakly analytic . by @xcite ( part ii , proposition 1.6 ) , we deduce that @xmath329 is analytic . it follows that @xmath17 itself is analytic . @xmath44 + if , instead of complex - analytic locally convex manifolds , one considers more generally complex - analytic subsets of locally convex spaces ( and analytic spaces modelled on such analytic sets ) , then pathologies appear . there are examples of reduced analytic sets ( in the ringed space - sense ) with continuous maps into some complex banach space which are weakly analytic but not analytic ( see e.g. @xcite , pp . 73 - 80 ) . this phenomenon is avoided by defining analytic sets and spaces as functored spaces in @xcite . for the reader s convenience , we give a simple proof of the fact that the mackey - closure topology on a frchet space coincides with the given topology . + * proof :* we know that the @xmath120-topology is finer than the given topology . we need to show that conversely , any @xmath120-open subset of @xmath14 is open . let @xmath15 be an absolutely convex @xmath120-neighborhood of 0 . first we show that being @xmath120-open , @xmath15 must absorb every bounded set . + suppose that there exists a bounded set @xmath31 which is not absorbed by @xmath15 . then for every @xmath395 , we have that @xmath31 is not contained in @xmath396 . in particular , if we consider a sequence @xmath60 of positive real numbers converging to 0 , then for every @xmath0 , the set @xmath31 is not contained in @xmath397 . so for every @xmath0 , we can choose @xmath398 such that @xmath399 . now @xmath400 for every @xmath0 , therefore the sequence @xmath401 mackey - converges to 0 . since @xmath15 is @xmath120-open , this implies the existence of @xmath402 such that @xmath403 . on the other hand , since @xmath404 , there exists @xmath405 such that @xmath406 . for @xmath407 , we have @xmath408 , and so @xmath409 , contradiction . thus , @xmath15 absorbs every bounded set . + now assume @xmath15 is not a neighborhood of 0 . since @xmath14 is frchet , it is metrizable and therefore first - countable . let @xmath410 be a decreasing countable basis of neighborhoods of 0 in @xmath14 . we may assume each @xmath411 to be balanced , in which case @xmath412 is also a basis of neighborhoods of 0 . since @xmath15 is not a neighborhood of 0 , for every @xmath0 , the basic neighborhood @xmath413 is not contained in @xmath15 . for each @xmath0 , choose @xmath414 such that @xmath415 . it is easy to see that @xmath416 , and so the set @xmath417 is bounded . by what we have seen above , @xmath15 will absorb this set , and so there exists @xmath395 such that @xmath418 for all @xmath0 . for @xmath0 sufficiently large , we will then have @xmath419 , contradiction . @xmath44 +

as the nearest large spiral galaxy , m31 is a fundamental benchmark for studies of disk galaxy evolution , and our understanding of the evolution of m31 has changed profoundly during the past decade . there is a large body of evidence ( e.g. ibata et al . 2001 , 2007 ; hammer et al . 2007 ; tanaka et al . 2010 ) that interactions with companion galaxies , some of which may not have survived intact to the present day and remain only as debris trails , have occured throughout the history of m31 . the detection of similar tidal features close to other nearby galaxies ( e.g. martinez - delgado et al . 2010 ) indicates that galaxy - galaxy encounters in the local universe have not been rare , and have played a key role in sculpting the current appearance of many nearby galaxies . the current study focuses on investigating the star - forming history ( sfh ) of m31 during the past few hundred myr , using the brightest resolved stars as tracers . it has been suggested that m31 has interacted with some of its companions during this time , and a key element of our work is to search for observational signatures of this activity . a review that considers the entire history of m31 will guide our understanding of its present - day appearance , and allow a conceptual picture of the key events that shaped its evolution to be developed . to this end , in the remainder of this section we review previous work on the evolution of m31 . while current theories for the buildup of disk galaxies give primacy to the smooth accretion of hot and cold gas , the following focuses on the influence of major and minor mergers , as these imprint observable signatures on the kinematics , spatial distribution , and age distribution of stars . not every interaction can be traced in detail , especially those that happened more than a few dynamical times in the past , and the discussion of the earliest phases of the evolution of m31 are by necessity more speculative than the discussion of the most recent events . the ages listed in what follows should also be viewed as approximations . the initial assembly of m31 likely involved the merger of proto - galactic structures , which contained a mix of gas and stars that had formed _ in situ _ ( e.g. oser et al . 2010 ) . the gas in these structures probably contributed to the assembly of an m31 proto - disk , supplementing material accreted through the large - scale inflow of hot and cold gas . any early disk was likely short - lived due to the high frequency of major mergers . in addition to disrupting the early disk , mergers may also have spurred the formation of a classical metal - poor halo , through the displacement of stars out of the disk plane . stars ejected in such events are expected to dominate the central 20 kpc of halos ( zolotov et al . 2009 ) . at larger radii the majority of halo stars may not have had such a violent origin , and may instead have been accreted from companion galaxies during the first few gyr of galaxy assembly ( font et al . 2008 ; zolotov et al . 2009 ; cooper et al . 2010 ) . a classical metal - poor halo has been detected around m31 ( e.g. kalirai et al . 2006 ; chapman et al 2006 ; ibata et al . 2007 ; koch et al . 2008 ) . there is evidence that m31 accreted a number of satellites early - on . the integrated luminosities , masses ( @xmath6 m@xmath7 ) , and dynamical properties of the m31 and the galactic halos are similar ( chapman et al . 2006 ; ibata et al . 2007 ) , and the properties of galactic globular clusters suggest that 6 8 satellites large enough to form such large star clusters may have been accreted ( e.g. mackey & gilmore 2004 ; forbes & bridges 2010 ) . globular clusters in the m31 halo tend to be associated with tidal debris ( mackey et al . 2010 ) , and the range of ages , chemical mixtures , and overall metallicities amongst m31 clusters ( e.g. beasley et al . 2005 ) hints at diverse progenitors . some objects that were originally identified as globular clusters may not be simple stellar populations ( ssps ) ( e.g. davidge et al . 1990 ; meylan et al . 2001 , fuentes - carrera et al . 2008 ) , but instead may be the remnants of dwarf galaxies that were shredded early in the evolution of m31 ( fuentes - carrera et al . 2008 ) . the properties of classical globular clusters suggest that m31 and the galaxy may have experienced different chemical enrichment histories early - on . when compared with galactic globular clusters , the spectra of many m31 globular clusters have relatively strong cn absorption bands ( e.g. burstein et al . 1984 ; davidge 1990a ) , and this has been attributed to a comparatively large nitrogen abundance in m31 clusters ( burstein et al . such an abundance difference may have its origins in the numbers of present - day stars in globular clusters that formed in the diffuse proto - cluster environments , as opposed to those that formed in the more compact potential wells that existed after cluster formation ( carretta et al . 2010 ) , in the sense that m31 clusters are made up of a smaller fraction of progenitor stars . if correct , then the location of stars in m31 globular clusters on the [ na / fe ] versus [ o / fe ] diagram , which is a diagnostic of the rate of chemical enrichment , should differ from that defined by galactic objects . early merger activity likely also produced a pressure - supported bulge ( e.g. bekki & chiba 2001 ) . aside from the central few arcsec , the integrated spectrum of the m31 bulge at visible wavelengths originates predominantly from old stars ( e.g. davidge 1997 ; puzia et al . 2005 ; saglia et al . the brightest agb stars are well - mixed throughout the central @xmath8 kpc of the m31 bulge , as expected if they formed during a rapid , uniform star - forming episode ( davidge 2001b ) , such as would result from the violent merging of satellites . still , the structural ( e.g. beaton et al . 2007 ) and kinematic ( e.g. morrison et al . 2010 ) properties of the m31 bulge suggest that secular processes may have also contributed to its growth . the material from which stars in the galactic bulge formed may have been chemically enriched by the earliest generation of globular clusters ( davidge 2001a ) . if globular clusters did play a significant role in enriching the early ism of m31 then the chemical properties of stars in the bulge and globular clusters may show similarities . in fact , the strengths of cn absorption bands at visible wavelengths in the integrated spectrum of the inner bulge of m31 are reminiscent of those in m31 globular clusters ( davidge 1997 ) , suggesting a chemical kinship . a thick disk may also have been formed during the early assembly of m31 . using kinematic selection criteria , collins et al . ( 2011 ) detect a thick disk in m31 and find that the component stars are @xmath9 dex more metal - poor than thin disk stars at the same galactic radius . the estimated total mass of the thick disk suggests that it did not form from the destruction of a satellite that follows the trend between [ m / h ] and total mass defined by present - day systems . if the m31 thick disk formed from thin disk stars that were kinematically heated then its ( comparatively ) low metallicity suggests that this happened during early epochs , before the major merger that produced the extended extraplanar component discussed in the next section . the spatial extent of m31 at the end of its initial assembly was significantly smaller than the present - day galaxy . simulations suggest that a galaxy with a present - day mass like that of m31 probably accreted no more than one - half of this mass by moderate redshifts ( stewart et al . 2008 ) , while at @xmath10 the spatial extent of its disk may have been substantially ( by a factor of @xmath11 two - thirds ) smaller than its present - day value ( firmani & avila - reese 2009 ) . the modest size of the m31 progenitor aside , it was this system that served as the seed for the subsequent evolution that lead to the present - day galaxy . mergers were common events during intermediate epochs , and played a key role in sculpting the morphological properties of nearby disk galaxies ( hammer et al . 2007 ; 2009 ; van der kruit & freeman 2011 ) . simulations of a merger between a disk galaxy and a large companion find that the existing disk can be obliterated , with the orbits of disk stars thermalising to form a pressure - supported extraplanar population ( e.g. hopkins et al . a gas disk will re - form if the progenitors have sufficient quanities of gas ( robertson et al . 2006 ; governato et al . 2007 ) . there is evidence that a violent merger occured only a few gyr following the initial assembly of m31 . such an event would have profoundly affected the morphology of m31 , and may have produced features that are usually attributed to more recent interactions ( hammer et al . 2010 ) . the proposed merger heated the disk of m31 and produced a diffuse extraplanar component that is distinct from the classical metal - poor halo . it has long been known that the extraplanar regions of m31 are dominated by stars with metallicities that ( 1 ) are much higher than those in the galactic halo ( e.g. mould & kristian 1986 , pritchet & van den bergh 1988 , durrell et al . 2001 ) and ( 2 ) are consistent with an origin in either a large disk or an smc / lmc - like companion . more recently , a kinematically hot component ( e.g. chapman et al . 2006 ; gilbert et al . 2007 ) has been detected , that has been traced out to extraplanar distances of 60 kpc . this component is distinct from the thick disk , which has a vertical scale height of 3 kpc ( collins et al . the metallicities of stars in this component show little or no trend with distance from the galaxy ( chapman et al . 2006 ; koch et al . 2008 ; richardson et al . 2009 ) . a break in the surface brightness profile occurs near 30 kpc ( ibata et al . 2007 ) , possibly indicating the transition between a violently disturbed inner halo and a more metal - poor classical halo . the lack of sub - structure in the distribution of the moderately metal - rich extraplanar stars suggests that they are not the remnants of a galaxy that was slowly accreted by m31 , but instead are the result of a kinematically violent event . the extraplanar component must have an age of at least a few gyr , as the c / m5 + ratio is @xmath12 throughout the outer regions of m31 ( koch & rich 2010 ) . such a c star frequency amongst stars with a moderately sub - solar metallicity is suggestive of an age in excess of @xmath13 gyr ( e.g. cole & weinberg 2002 ) . brown et al . ( 2008 ) discuss observations of extraplanar fields with minor axis distances of 21 and 35 kpc , which are populated by stars with main sequence turn - off ( msto ) ages @xmath14 gyr . the age distribution is such that the majority of stars in the 35 kpc field are not associated with the early assembly of m31 , as only 10% are extremely old and metal - poor and so likely belong to the classical halo . that the bulk of the stars in the two brown et al . ( 2008 ) fields have ages @xmath15 gyr argues that ( 1 ) the merger occured a few gyr after the initial assembly of m31 , and ( 2 ) there has not been a subsequent merger of comparable magnitude while m31 has likely been subject to encounters with other galaxies in the more recent past ( see below ) , none of these have been as disruptive as the event described here . a sfr for m31 during this epoch can be estimated from the relation between the brightness of the brightest globular cluster , normalized to an age of 10 myr , and sfr . the most luminous globular cluster that is known to have formed during intermediate epochs is 292 - 010 , for which beasley et al . ( 2004 ) find a mass of @xmath16 m@xmath7 . the models of bruzual & charlot ( 2003 ) indicate that log(m / l)@xmath17 for a population with an age 10 myr , and so 292 - 010 would have had m@xmath18 when it was 10 myr old . applying relation 1 from bastian ( 2008 ) yields a sfr of @xmath19 m@xmath7 year@xmath20 . major mergers trigger large - scale star - forming activity that is not restricted to the central regions of the galaxy ( e.g. teyssier et al . 2010 ) , and m31 may have appeared as an ultra - luminous infrared galaxy ( ulirg ) for a period of time during intermediate epochs . star formation is seen throughout the disks of ulirgs , and is curtailed at progressively smaller radii as torques cause gas to move inwards ( e.g. soto & martin 2010 ) . if m31 was a typical ulirg then 5 10% of its current stellar mass may have been produced over a @xmath21 gyr ( marcillac et al . 2006 ) time span , and this is roughly consistent with the sfr estimated above from the brightest intermediate age globular cluster . the formation of globular clusters in the local universe is associated with the large - scale re - distribution of gas that occurs during major mergers ( e.g. whitmore et al . 1993 ) , and the concomitant violent turbulent stirring of the ism that produces dense star - forming clumps ( teyssier et al . globular clusters in m31 span a wide range of ages ( e.g. beasley et al . 2004 ; burstein et al . 2004 ; fusi pecci et al . 2005 ; caldwell et al . 2009 ) , and puzia et al . ( 2005 ) find peaks in the cluster age distribution at @xmath22 gyr , 8 gyr , and 1 gyr . the majority of globular clusters in m31 belong to the oldest age group , which is associated with the initial assembly of the galaxy . however , the number of clusters that belong to the 8 gyr peak and which likely formed during the major merger that produced the extraplanar component is a significant fraction of the older cluster system . these intermediate age clusters tend to be seen against the disk of the galaxy , and are not at large radii . to the extent that globular clusters are probes of a disturbed and turbulent ism , the 8 gyr cluster population is the remnant of an event that was almost as violent as the mergers associated with the initial assembly of m31 . a major merger would have imprinted signatures on the age distribution of stars throughout much of the disk . if the present - day disk of m31 formed during intermediate epochs following the obliteration of an earlier disk then there should be ( 1 ) an absence of disk stars with ages @xmath15 gyr , and ( 2 ) a large population of disk stars that formed @xmath23 gyr in the past - these are the field counterpart of the globular cluster population that formed at that time . in addition , starbursts deplete the gas supply at a rate that is an order of magnitude shorter than in a normal disk ( daddi et al . 2010 ) . therefore , the sfr would have plunged within a few disk rotation times of the merger . observations that fully test these predictions require photometric depths and angular resolutions that tax the capabilities of existing telescopes . still , olsen et al . ( 2006 ) investigate the sfh of bulge and disk fields in m31 , and find a preponderance of stars with ages @xmath14 gyr . this is consistent with the disk of m31 experiencing a large burst of star formation in the past that may not have been associated with the epoch of earliest disk assembly . m31 has not evolved in isolation since the major merger that spurred the formation of the present - day thin disk , and there is a high probability that there was a close encounter between m31 and m33 within the past few gyr . various studies have estimated when such an interaction may have occured . the sfh of m33 suggests that the encounter must have happened at least 0.5 gyr in the past ( davidge & puzia 2011 ) . using the locations and velocities of m31 and m33 , putman et al . ( 2009 ) argues that an encounter occured 1 3 gyr in the past , while bekki ( 2008 ) assigns an age of 4 8 gyr based on the properties of the hi bridge that may link the galaxies ( braun & thilker 2004 ) . simulations discussed by mcconnachie et al . ( 2009 ) that use currently available orbital constraints and the assumption that m33 was significantly perturbed by the encounter place this interaction a few billion years in the past . there is a high probability that the tidal radius of m33 was @xmath24 kpc during this encounter , and this would have disturbed the disk of m33 . the s - shaped hi distribution in the outer regions of m33 is testiment of such a disturbance , and this structure likely contains gas that was stripped from the m33 disk ( putman et al . the detection of extended stellar structures in the outer regions of the m33 disk , coupled with isophotal distortions at smaller radii , provide further evidence that the disk of m33 was disturbed ( mcconnachie et al . 2010 ; davidge & puzia 2011 ) . the disk of m31 would also have been affected by this event ( mcconnachie et al . 2009 ; davidge & puzia 2011 ) , with elevated levels of star - forming activity being sparked as tidal forces re - distributed gas . intermediate age globular clusters in m31 with ages @xmath25 gyr ( puzia et al . 2005 ) may have formed during this event . while the sfr of the m31 disk would have been greatly elevated immediately after the encounter , it would eventually drop as the ism is depleted and disrupted . there is evidence of a decrease in m31 star - forming activity @xmath8 gyr ago ( williams 2002 ) . given that the damping timescale for elevated sfrs is on the order of a few disk crossing times , then the drop in the sfr found by williams ( 2002 ) is consistent with an interaction that had peak star - forming activity @xmath25 gyr ago . if gas was re - distributed throughout m31 then intense star - forming activity would likely have been triggered in the galaxy center , and this part of m31 _ does _ contain stars that formed at a time that coincides with the proposed interaction with m33 . the integrated spectrum of the central few arcsec of m31 at visible wavelengths reveals a large population of objects with an age of a few gyr ( davidge 1997 ; silchenko et al . saglia et al . ( 2010 ) conclude from integrated spectra that the luminosity - weighted age of stars near the center of m31 is 4 8 gyr . mergers and interactions with companions that have @xmath27 the mass of the primary galaxy are not expected to permanently disrupt disks ( e.g. stewart et al . 2008 ) , and there is evidence that m31 has interacted with companions in this mass range during recent epochs . the most overt signature of such an encounter may be the giant stellar stream ( gss ; ibata et al . the gss is a substantial structure , with the northern and southern ends separated by at least 130 kpc along the line of sight ( mcconnachie et al . 2003 ) . the stellar content of the gss suggests that the progenitor was a massive dwarf galaxy , and ibata et al . ( 2001 ) discuss ngc 205 and m32 as possible sources . the photometric properties of rgb stars in the gss are indicative of a moderately metal - rich system , with a 0.5 dex dispersion in [ m / h ] ( mcconnachie et al . 2003 ) . using the red clump of core helium - burning stars as an age indicator , tanaka et al . ( 2010 ) conclude that stars in the gss have an age near 7 gyr , and a peak [ m / h ] near @xmath28 . this metallicity is comparable to that of stars in the lmc , and suggests a progenitor mass @xmath29 m@xmath7 . fardal et al . ( 2007 ) and ibata et al . ( 2007 ) arrive at similar progenitor mass estimates . upper limits to the mass of the progenitor can also be placed . assuming that the gss progenitor passed through the disk of m31 , then mori & rich ( 2008 ) conclude that it must have had a mass @xmath30 m@xmath7 so as not to widen the disk of m31 more than is observed . font et al . ( 2006 ) conclude that the encounter that formed the gss occured only a few hundred myr in the past , while mori & rich ( 2008 ) assign an age @xmath8 gyr . the age predicted for the gss by font et al . ( 2006 ) is similar to that modelled by block et al . ( 2006 ) for the supposed passage of a satellite ( presumably m32 ) through the m31 disk . the mean metallicity of stars in the outer regions of m32 is @xmath31 ( rose et al . 2005 ) , which is consistent with the metallicity of stars in the gss . models that attempt to reproduce the orbital properties of the gss predict that the encounter occured much more recently than might be inferred from its stellar content . the timings predicted by kinematic studies and the ages of stars in the gss are only in conflict if the progenitor contained a reservoir of cool gas that could have fueled star formation up to the time of its disruption . if the progenitor was depleted of gas a few gyr in the past and most present - day companions of m31 , such as m32 , are gas - depleted ( e.g. welch , sage , & mitchell 1998 ; sage , welch , & mitchell 1998 ; grcevich & putman 2009 ) then there is a clear physical explanation for the apparent conflict between the kinematic and stellar content age estimates . the extraplanar regions of m31 are laced with stellar streams and tidal structures , some of which may or may not be related to the gss . the completeness of the catalog of these structures is uncertain , as there is a bias against the detection of metal - poor and/or old features . this bias occurs because higher surface brightness structures tend to have higher metallicities and/or formed more recently ( e.g. gilbert et al . 2009 ) . the diverse stellar contents of streamlets along the minor axis of m31 led ibata et al . ( 2007 ) to suggest that these structures have multiple progenitors , all of which based on metal content would have initially been larger than the fornax dwarf spheroidal . mori & rich ( 2008 ) find that a single minor merger can pollute a large fraction of the extraplanar regions of m31 within @xmath25 gyr of an encounter . still , they also find that some structures in the outer regions of m31 likely formed more than 3 4 gyr in the past , and so are not related in a direct way to the gss , to which they assign an age of 1 gyr . fardal et al . ( 2008 ) , tanaka et al . ( 2010 ) , and hammer et al . ( 2010 ) argue that many features in the outer regions of m31 could result from an interaction between m31 and a single galaxy . the stream - to - stream diversity in stellar contents indicates that a progenitor that produces multiple observed streams would likely have had a metallicity gradient , and such gradients are seen in nearby local group dwarf galaxies ( e.g. harbeck et al . 2001 ) . clues of a significant accretion event within the past few hundred myr also lurk in the central regions of m31 . the nuclear star cluster p3 has an age of 200 myr ( bender et al . 2005 ) , and hence formed at the same time as the events modelled by block et al . ( 2006 ) and font et al . davidge et al . ( 2006 ) find a bright object that is a few arcsec from the nuclear star cluster p3 . this source dominates the central light output of m31 at @xmath32 m and appears to be a massive dust - enshrouded agb star that based on its luminosity may have an age of a few hundred myr . saglia et al . ( 2010 ) find a counter - rotating ring of gas in the central regions of m31 , which they interpret as the remnant of a merger @xmath21 gyr in the past . they suggest that @xmath33 m@xmath7 of stars formed in the central parts of m31 following this merger . the disk of m31 also contains signatures of an event ( or events ) that disrupted the ism . the hi disk of m31 is twisted and warped ( e.g. corbelli et al . 2010 ) , and there is a significant non - rotational kinematic component ( e.g. unwin 1983 ) . yin et al . ( 2009 ) model the chemical properties and the distribution of gas and stars in m31 assuming evolution in isolation . the model does not match the observed gas content at large radii or the radial distribution of star - forming activity throughout the galaxy . yin et al . ( 2009 ) attribute these discrepancies to interactions . an alternative is to accept that the star - forming efficiency in m31 varies with radius , but stays constant throughout the galaxy ( e.g. marcon - uchida et al . 2010 ) . the location and distribution of star - forming regions in m31 also differ systematically from those in the galaxy and m33 . while star formation in the galaxy and m33 tends to be distributed along well - defined spiral arms , a large fraction of the star - forming activity in m31 occurs in rings at r@xmath34 and 14 kpc ( thilker et al . 2005 ; gordon et al . 2006 ; barmby et al . 2006 ) , in arcs , and in spiral arm fragments ( e.g. discussion in efremov 2010 ) . much of the star - forming activity is ( and has been for at least the past 100 myr - 6 ) concentrated in the ring at r@xmath34 kpc , which contributes substantially to the total h@xmath35 emission from m31 ( devereux et al . 1994 ) . despite the evidence that the ism of m31 has been disrupted , the star - forming environment in these rings is probably not too different from that in the galactic disk . indeed , gmcs in m31 fall along the same size versus line width relation as galactic gmcs ( sheth et al . 2008 ) , indicating a structural kinship . the depletion timescale of molecular gas indicates that m31 is much less efficient at forming massive stars than m33 ( tabatabaei & berkhuijsen 2010 ) . the strength of the uv radiation field throughout the m31 disk is only 40% that in the solar neighborhood ( montalto et al . 2009 ) , and the sfr of m31 is roughly one half that expected for a galaxy of similar mass and morphology that has evolved in isolation ( yin et al . williams ( 2003 ) infers a sfr of @xmath8 m@xmath7 year@xmath20 for m31 , based on studies of resolved stars over a 1.4 degree@xmath36 area . integrated light diagnostics suggest present - day sfrs for m31 of 0.3 ( tabatabaei & berkhuijsen 2010 ) , 0.4 ( barmby et al . 2006 ) and 0.6 ( kang et al . 2009 ) m@xmath7 year@xmath20 . these are lower than the sfr of the galaxy ( e.g. boissier & prantzos 1999 ; robitaille & whitney 2010 ) and are on par with the much less massive galaxy m33 ( hippelein et al . 2003 ; verley et al . 2007 ; verley et al . 2009 ) . in the present paper , deep megacam observations are used to study the locations and photometric properties of bright resolved stars throughout the disk of m31 . massive main sequence and evolved stars are detected , and these are used to characterize the sfh during the past few hundred myr and search for signatures of recent interactions . atmospheric seeing and sky emission restrict the photometric depth that can be achieved with ground - based datasets when compared with space - based observations . however , ground- and space - based datasets of nearby galaxies provide complementary information because they usually sample very different spatial scales . the spatial coverage of space - based datasets of m31 at visible and near - infrared wavelengths tend to be limited . the most extensive such dataset is that discussed by dalcanton et al . ( 2011 ) , which is restricted to the north east quadrant of the galaxy . in contrast , our megacam data cover almost the entire star - forming disk of m31 out to major axis distances in excess of 30 kpc ( @xmath37 scale lengths ) . such comprehensive spatial coverage is of interest since the sfh can vary with location over large spatial scales in disks ( e.g. davidge 2010 ; davidge & puzia 2011 ) . spatial variations of this nature can complicate efforts ( 1 ) to determine a representative global sfh if only part of the disk is sampled , and ( 2 ) to assess the large scale radial trends that are probes of galaxy evolution . two important considerations are the distance to m31 and the reddening model . an absolute distance modulus @xmath38 , computed by vilardell et al . ( 2010 ) from two double - lined spectroscopic eclipsing binaries , is adopted for this study . such systems are primary distance indicators ( de vaucouleurs 1978 ) , that rely on geometric parameters directly measured from the application of basic physics . distances measured with these objects should be less susceptible to calibration uncertainties than those computed from secondary distance indicators . the distance modulus adopted here is in excellent agreement with that computed from cepheids by riess et al . ( 2012 ) . the baseline reddening model consists of a linear combination of foreground ( a@xmath39 ; schlegel et al . 1998 ) and internal ( a@xmath40 ; pierce & tully 1992 ) components . the total extinction is thus a@xmath41 magnitudes . this corresponds to a@xmath42 , a@xmath43 , and a@xmath44 magnitudes , based on the relations listed in table 6 of schlegel et al . it is demonstrated in 4 that this total extinction holds for a ` typical ' main sequence star in our sample . of course , the reddening towards any given object depends on factors such as its age , location , and evolutionary state ( e.g. massey et al . 2009 ) , and it is not surprising that there is a substantial spread in extinction throughout the m31 disk . for example , hodge et al . ( 2010 ) find a dispersion of @xmath45 mag in @xmath46 measurements among young clusters . there is almost certainly an environmental component , as hodge et al . ( 2010 ) and perina et al . ( 2010 ) find @xmath47 ( a@xmath48 ) for young star clusters , whereas massey et al . ( 2007 ) measure @xmath49 ( a@xmath50 ) for young field stars in the m31 disk . the paper is structured as follows . the observations and the reduction of the data are discussed in 2 , while details of the photometric measurements , including their calibration and characterization , are discussed in 3 . color - magnitude diagrams ( cmds ) are presented in 4 , while the sfh of the m31 disk is investigated in 5 using the lfs of main sequence stars . the spatial distribution of main sequence and evolved stars is examined in 6 . the paper closes with a summary and discussion of the results in 7 . the data were recorded on the 3.6 metre cfht as part of the 2010b megacam ( boulade et al . 2003 ) observing queue . the detector in megacam is a mosaic of thirty six @xmath51 e2v ccds . these are deployed in a @xmath52 format , and cover roughly one degree@xmath36 exposure@xmath20 with 0.185 arcsec pixel@xmath20 . four fields that cover almost all of the star - forming disk of m31 were imaged through @xmath53 , and @xmath54 filters . the locations of the megacam pointings are shown in figure 1 . the observations for each field were obtained during a single continuous block of time , that typically spanned 2 hours . while archival @xmath55 and @xmath54 megacam data were available of the disk of m31 , images in these filters were still recorded for our program to secure a set of single epoch measurements , and thereby suppress scatter in colors that might arise from photometric variability over timescales @xmath56 hours . a log of the observations , showing dates of observations and exposure times for each field , is shown in table 1 . the @xmath57 images have markedly longer exposure times than the @xmath55 and @xmath54 images because of the comparatively low system throughput at shorter wavelengths . initial processing of the data was done with the cfht elixir pipeline , and this included bias subtraction and flat - fielding . each pipeline - processed image was divided into six @xmath58 ccd sub - mosaics to produce datasets of a manageable size for subsequent processing and photometric analysis . sub - mosaic images were spatially registered to correct for offsets introduced during acquisition , and the results were stacked according to field and filter . the megacam optics produce modest chromatic distortions across the imaged field , which manifest as location - dependent offsets between the centroids of stars imaged in different filters . the offsets between @xmath57 and @xmath55 images taken at moderate airmass and aligned near the optical axis of the instrument can be up to an arcsec near the edge of the megacam science field . to correct for these , the @xmath57 and @xmath54 data were mapped into the @xmath55 reference frame using the iraf geomap / geotran tasks . stars in the final distortion - corrected images typically have fwhm @xmath59 arcsec . photometric measurements were made with the point spread function ( psf)-fitting program allstar ( stetson & harris 1988 ) . each psf was constructed from between 50 to 100 bright , unsaturated stars using routines in the daophot package ( stetson 1987 ) . in addition to brightness , psf stars were selected using star - like appearance and the absence of bright neighbors as criteria . faint companions were subtracted from the wings of psf stars in an iterative manner , using progressively improved psfs . following the procedure described by davidge ( 2010 ) , the raw photometric catalogues produced by allstar were culled of extended and very faint sources using the error in magnitude that is computed by allstar , @xmath60 . this quantity measures the quality of the psf fit , but does not take into account other sources of error , such as those introduced by crowding . while @xmath60 is only a lower limit to the total uncertainty in source brightness , it provides a means of identifying objects for which photometry may be problematic . two rejection criteria were applied . first , all sources with @xmath61 mag were deleted . obtaining meaningful photometric measurements is problematic for objects with such large uncertainties , the majority of which are near the faint limit of the data . second , point sources define a sequence in the magnitude _ vs. _ @xmath60 plane that is narrow at the bright end and fans out towards fainter magnitudes . objects that are outliers from this trend were removed . this step tends to remove extended sources ( e.g. davidge 2010 ) and cosmetic defects . standard stars are observed during each megacam observing block . these observations are used to determine photometric zeropoints , and the results are placed in megacam image headers during elixir processing . we used these header entries to transform instrumental magnitudes into the sloan digital sky survey @xmath62 system ( fukugita et al . we note that a comparison of photometric measurements in overlapping sections of fields indicates that the internal field - to - field calibration consistency is a few hundredths of a magnitude . there are significant differences between the @xmath57 and @xmath63 bandpasses , with the central wavelength of @xmath57 falling 200 redward of @xmath63 . our @xmath63 magnitudes were checked using the @xmath64 measurements of bright stars in m31 published by massey et al . the magnitudes from that study were transformed into the sloan system using equations from smith et al . ( 2002 ) , and the results were compared for stars in common with our dataset . the comparisons were restricted to bright stars in uncrowded parts of m31 to suppress the influence of image quality differences between the kpno and cfht observations . one consequence of restricting the areal coverage in this way is that the majority of the stars used in the comparison have red colors , as the main body of the star - forming disk ( where bright blue stars tend to be found ) is intentionally avoided . this caveat notwithstanding , there is excellent agreement between the two sets of measurements , with @xmath65 and @xmath66 , in the sense cfht kpno . the standard deviations about the mean are @xmath67 and @xmath68 . this agreement is comparable to that found by davidge & puzia ( 2011 ) in their study of m33 . sample completeness and uncertainties in the photometry were assessed through artificial star experiments . artificial stars were assigned magnitudes and colors that hold for main sequence stars , which are the primary probes of the sfh in this paper . as with observations of real objects on the sky , an artificial star was considered to be recovered only if ( 1 ) it was detected in two filters ( either @xmath69 or @xmath70 ) , and ( 2 ) it survived the @xmath60based culling criteria described in 3.1 . the uncertainties and systematic effects in the photometry that arise due to crowding and/or statistical fluctuations in the noise increase substantially with magnitude when sample completeness drops below 50% . the magnitude at which 50% completeness occurs is thus one estimate of the faint limit . the artificial star experiments indicate that 50% completeness occurs near @xmath71 and @xmath72 in the outer regions of the disk , and @xmath73 and @xmath74 at intermediate radii along the major axis . these values are roughly consistent with the observed faint limits of the cmds ( 4 ) . the impact of crowding on photometry can be investigated by stacking images to simulate higher stellar densities ( e.g. davidge 2001b ) . a shortcoming is that foreground stars and background galaxies , which are more - or - less uniformly distributed on the sky over angular scales of a few degrees , are over - represented in the stacked images . this can be mitigated by selecting fields for stacking that have a moderately high initial stellar density , with the aim of minimizing the number of frames that are co - added to simulate a given stellar density . four @xmath75 pixel@xmath36 areas from field 1 were selected as the building blocks for the stacking analysis . these are @xmath76 arcmin from the center of the galaxy , where the average surface brightness is @xmath77 magnitudes arcsec@xmath78 ( walterbos & kennicutt 1987 ) . the faint limit of the cmds in this part of the disk is almost the same as the outermost parts of the disk ( 4 ) . two sub - fields were stacked to simulate a field with @xmath79 mag arcsec@xmath78 , while three were combined to simulate @xmath80 mag arcsec@xmath78 . all four sub - fields were combined to simulate @xmath81 mag arcsec@xmath78 . the magnitudes of stars in the stacked images were measured using the procedures discussed in 3.1 , and the resulting cmds are shown in figures 2 and 3 . the left hand column of each figure shows the cmds from the unstacked data , while the right hand column shows the cmds of the stacked images . the simulated surface brightness of each stacked field is also listed , along with the corresponding major axis radius from the walterbos & kennicutt ( 1987 ) light profile . as expected , crowding elevates the faint limit of the cmds as one moves to progressively higher surface brightnesses . an interesting practical result is that the photometry of sources with @xmath82 and @xmath83 appears not to be sensitive to surface brightness in the density regime investigated here . the brightest main sequence stars and agb stars in these data can thus be used to trace the distribution of young stars to within a few kpc of the galaxy center . we define six regions in the m31 disk to facilitate analysis and subsequent discussion . two regions are associated with the rings of uv and fir emission at r@xmath34 and 14 kpc , and these will be refered to as ring 10 and ring 14 . the inner disk is defined to be the region that is interior to ring 10 . the image stacking experiments in 3.4 indicate that crowding is an issue for all but the brightest stars with r@xmath84 kpc , and so the minimum radius for the inner disk is set at r@xmath85 kpc . the inner disk thus probes a region where the disk light dominates over that of the bulge ( e.g. figure a1 of tempel et al . the middle disk is located between ring 10 and 14 , while the outer disk is external to ring 14 . the outer disk is split into two radial intervals . this is done in part because the outer radius of the m31 disk is not clearly defined . in addition , there is evidence that the properties of star formation in low density environments may differ from those in the inner regions of disks ( bigiel et al . 2010 ) , while secular processes may play a key role in defining the stellar content at large radii ( e.g. roskar et al . 2008 ; sanchez - blazquez et al . the outermost regions of disks may then be an environment that is distinct from the inner disk , containing population gradients with physical drivers that differ from those at smaller radii . the @xmath86 cmds of roughly 2.1 million sources in various radial intervals are shown in figure 4 . hess diagrams are shown in figure 5 to allow the relative properties of objects in the faint portions of the cmds to be examined . the radial distances listed in each panel of figures 4 and 5 are in the plane of the disk and assume 75@xmath87 inclination . this inclination produces roughly circular rings when observations of warm dust emission are de - projected to simulate the face - on appearance of m31 ( gordon et al . 2006 ) . while the assumption of a single inclination is an obvious pragmatic choice , it is a simplification given that the gas disk of m31 is warped ( e.g. corbelli et al . this warping , coupled with the orientation of m31 on the sky , frustrates efforts to extract clean radial samples that are free of annulus - to - annulus contamination . young main sequence stars form a prominent plume in the cmds , with @xmath88 at the bright end , and @xmath89 near the faint end . bright blue stars are not restricted to ring 10 and ring 14 , and are seen in all cmds with r@xmath90 kpc . recent star formation has thus not been restricted exclusively to ring 10 and ring 14 . many of the blue objects with @xmath91 ( m@xmath92 ) are bright blue supergiants ( bsgs ) , although some of these may be young compact star clusters . the diffuse sequence of objects that is displaced @xmath93 magnitude in @xmath94 redward of the main sequence , and is most pronounced in the inner disk , ring 10 , and middle disk cmds with @xmath55 between 20 and 22 , is due to helium burning stars that are at the blue extrema of their evolution . this same feature is seen in figure 2 of mcquinn et al . ( 2011 ) . while the primary probe of evolved stars in our data is the @xmath95 cmd ( 4.2 ) , evolved red stars are also present in the @xmath86 cmds . a population of red objects with @xmath96 and @xmath97 is seen in the hess diagram of the inner disk region in figure 5 . isochrones ( see below ) indicate that evolved stars with ages of @xmath98 gyr pass through this part of the @xmath86 cmd . that these red sources occur in large numbers in the inner disk @xmath99 cmd when compared with the ring 10 and the middle disk cmds suggests that the sfr in the inner disk was elevated with respect to regions of m31 at larger radii within the past 200 myr ( 6 ) . background galaxies contribute significantly to the source counts near the faint limit of these data . the majority of background objects have red colors , and populate the jumble of red sources with @xmath97 that is most obvious in the outer disk cmds . not all background objects have red colors , and davidge & puzia ( 2011 ) found that a significant fraction of the objects with blue colors at large radii in m33 are star - forming regions in moderately distant background galaxies . this being said , source counts in areas where the number of m31 disk stars is expected to be small indicate that many of the blue objects with @xmath100 in the outer disk 2 cmd are _ bona fide _ main sequence stars in m31 . m31 is viewed at intermediate galactic latitudes , and foreground stars dominate the diffuse spray of objects with @xmath101 and @xmath94 between 1 and 3 that grows more prominent towards larger r@xmath2 . the blue edge of this sequence is defined by the msto of stars in the outer disk and halo of the galaxy , while the red envelope reflects the properties of low mass main sequence stars . not all of the sources with @xmath94 between 1 and 3 are in the foreground , and bsgs with @xmath102 and @xmath101 ( i.e. to the immediate right of the m31 main sequence ) are evident in the cmds of sources with r@xmath103 kpc . the @xmath86 cmds of sources in ring 10 and ring 14 are compared with solar metallicity isochrones from girardi et al . ( 2004 ) in figure 6 . metallicity gradients are present in many spiral galaxies , and so a comparison with isochrones having only one metallicity may seem suspect . however , the metallicities of bsgs in m31 with r@xmath2 between 5 and 12 kpc do not vary with radius ( trundle et al . 2002 ) , suggesting that the interstellar medium of m31 is well mixed , and that the metallicity gradient among moderately young stars is modest . therefore , with the exception of a single z = 0.004 sequence to demonstrate metallicity effects , the comparisons with isochrones are restricted to a single metallicity . the image stacking experiments discussed in 3 indicate that crowding does not affect greatly the photometric properties of objects with m@xmath104 , the majority of which are main sequence stars with ages @xmath105 myr . it is thus significant that the zero age main sequence ( zams ) defined by the isochrones falls along the blue envelope of main sequence stars , while the terminal age main sequence ( tams ) , defined by the red extent of main sequence evolution , follows the red envelope of the blue plume . that the observed blue plume is bracketed by these two evolutionary phases suggests that age is the dominant driver of main sequence width in our data , rather than photometric errors or differential reddening . ring 10 and ring 14 both contain very young stars . the brightest blue objects in ring 10 and ring 14 fall more - or - less along the 5 myr isochrone in figure 6 . this is not unexpected given that these annuli are known to contain areas of very recent star formation . it is also evident from the isochrones that the part of the cmd with m@xmath106 and @xmath107 contains evolved stars with ages @xmath108 gyr . hence , the red concentration of objects near the faint limit of the @xmath86 cmd contains intermediate age stars . the blue core helium - burning ( bheb ) sequence in m31 is not well - matched by the locus of blue loops in the z = 0.019 isochrones , in the sense that the locus of blue loops during supergiant evolution is almost 1 magnitude in @xmath94 redward of the observed sequence . mcquinn et al . ( 2011 ) also found differences between the observed and predicted colors of the bheb sequence in their sample of dwarf galaxies . such disagreements with observations may be due to for example errors in the physics used to generate the models , uncertainies in circumstellar extinction , or uncertainties in the metallicities adopted for the stars . an interesting result from figure 4 of mcquinn et al . ( 2011 ) is that the colors predicted for bheb stars do not change greatly with metallicity when z @xmath109 . the blue loop of the z = 0.004 isochrone in figure 6 overlaps the m31 bheb sequence , and one _ possible _ explanation for the poor agreement between the observed and predicted color of the bheb sequence in figure 6 is that the stars in m31 have a metallicity that is no more than one half solar ( i.e. [ m / h ] @xmath110 ) . however , we reiterate that metallicity is just one of the factors that may affect the color predicted for the bheb sequence . the comparisons in figure 6 assume the same reddening for each star . reddening maps have been constructed for m31 using the distribution of warm and cool dust , and these could be used to estimate reddening for individual sources . however , our megacam data are not consistent with the radial extinction trends predicted from dust emission . this is demonstrated in table 2 , where the mean @xmath94 color , @xmath111 , of sources with @xmath55 between 21.75 and 22.25 is given for 5 radial intervals . an iterative @xmath112 rejection filter was applied to suppress outliers . the standard deviation about the mean , @xmath113 , is also shown , along with the mean optical depth at each r@xmath2 , @xmath114 , taken from figure 4 of tempel et al . ( 2011 ) . the entries in table 2 indicate that the plume containing main sequence stars and bsgs has a dispersion @xmath115 magnitudes at @xmath116 , supporting the notion that photometric errors make only a modest contribution to the width of the main sequence . the entries in table 2 further indicate that neither the mean color of the main sequence nor the dispersion about the mean changes significantly with radius between r@xmath117 kpc and 15 kpc . indeed , the mean @xmath94 color of the main sequence in this large range of r@xmath2 varies by only a few millimagnitude . to be sure , there are area - to - area variations in extinction within each annulus . however , the radial consistency of the mean @xmath94 color indicates that these variations average out to a remarkable degree when stellar measurements are combined azimuthally . the uniform color of the main sequence ridgeline in table 2 is contrary to what might be expected based on the optical depth measurements from tempel et al . ( 2011 ) , which predict systematic radial variations in total extinction , with @xmath114 peaking in ring 10 . such an inconsistency would occur if the radial @xmath114 measurements obtained from dust emission do not apply to the majority of stars in each angular resolution element . indeed , a core assumption made by tempel et al . ( 2011 ) is that cool dust is uniformly distributed . there almost certainly @xmath118 a uniformly distributed cool dust component throughout the m31 disk ; however , there are also areas of cool dust emission that are concentrated on scales that are smaller than the angular resolution of the facilities that have probed fir emission , and such a component will skew @xmath119 measurements . the tempel et al . ( 2011 ) optical depth measurements are based on data from the spitzer telescope , the angular resolution of which is roughly 40 arcsec , or @xmath120 kpc in m31 , at @xmath121 m . each resolution element then has a spatial extent that is roughly one order of magnitude larger than that of a typical giant molecular cloud ( gmc ) . if a significant part of the cool dust emission within a resolution element originates from gmcs then the extinction of objects that are exterior to the gmcs , but still within the same observed angular resolution element , will be over - estimated . the stars that are obscured in gmcs by the cool dust may even be so heavily extincted that they are not detected at visible wavelengths , and if this is the case then the dust and observed stars are spatially de - coupled . in summary , if there is a cool dust component in m31 that is concentrated on gmc - like spatial scales , then the stars within at least some spitzer resolution elements will not be subject to the level of dust extinction that is predicted from dust emission . the @xmath95 cmds of roughly 3.6 million sources in the m31 disk are shown in figure 7 , while the corresponding hess diagrams are shown in figure 8 . foreground stars form a spray of objects at the bright end of the cmds in figure 7 that has a well - defined blue and red cut - off ; this sequence falls between @xmath122 and 2 , and is most obvious when @xmath123 . there is also a prominent blue plume with @xmath124 that contains main sequence stars . the blue plume is less well - defined in the @xmath95 cmds than in the @xmath86 cmds , and it is demonstrated below that this is probably a consequence of the trajectories followed by stars on this cmd as they evolve off of the zams . a collection of objects with @xmath125 and @xmath126 between 0 and 2 forms a prominent feature in the @xmath95 cmds . at small and intermediate r@xmath2 this part of the cmd is dominated by a mix of red core helium - burning ( rheb ) giants / supergiants and stars that are evolving on the agb . background galaxies make a progressively larger contribution to the faint red end of the cmds as one moves to larger r@xmath2 . the @xmath95 cmds of ring 10 and ring 14 are compared with solar metallicity isochrones from girardi et al . ( 2004 ) in figure 9 . the isochrones follow a near - vertical trajectory on the @xmath95 cmd after leaving the zams ; this continues past the tams until the onset of red loops , smearing the main sequence in a vertical direction on the cmds , and making the blue plume in the @xmath95 cmd near @xmath124 more diffuse than in the @xmath86 cmd . it is encouraging that the blue plume in the @xmath95 cmds is bracketed by the zams and the inflexion points in the isochrones that mark the onset of red loops . the isochrones indicate that the red jumble is made up of stars in a mix of evolutionary stages , some of which are rheb and others evolving on the agb . given that the timescale for core helium burning is much longer than the timescale to evolve on the agb then stars in the former stage of evolution are the dominant population . the 160 and 240 myr isochrones pass through the main concentration of red sources that have m@xmath127 , indicating that the majority of these stars have progenitor masses @xmath128m@xmath7 . line blanketing affects the photometric properties of very red evolved stars at visible and red wavelengths , causing the most luminous portions of the agb at moderate and higher metallicities to bow over on the @xmath95 cmds ( e.g. bica et al . this is evident in figure 9 , as the 160 myr isochrone plunges to fainter m@xmath129 when @xmath130 . the most luminous portions of the agb are almost certainly missing from these cmds , and so identifying the agb - tip is problematic . the color distributions of stars can be used to investigate field - to - field variations in the recent sfh in a purely empirical manner . the @xmath126 histogram distributions of sources in three r@xmath2 intervals with m@xmath129 between 4 and 3 , which corresponds to @xmath54 between 20.9 and 21.9 based on the apparent @xmath54 distance modulus , are compared in figure 10 . comparisons with isochrones in figure 9 indicate that the blue peak is populated by stars with ages @xmath105 myr , while the red peak contains stars with ages between 60 and 200 myr . the number counts in each radial interval have been corrected statistically for foreground and background contamination using source counts in areas that are external to outer disk 2 . these control fields , the location of which are indicated in figure 1 , were examined by eye to ensure that they do not contain obvious concentrations of young stars . the control fields almost certainly contain some intermediate age stars that belong to m31 , and so the number counts in these areas may ( slightly ) overestimate the level of foreground and background contamination at the faint end . that ring 10 has been an area of star - forming activity during the past 100 myr is clearly evident from the number of blue stars in the top panel of figure 10 . the mix of young stars ( i.e. those in the blue color peak ) to intermediate age stars ( i.e. those in the red color peak ) changes with radius , and these variations are explored in the lower panel of figure 10 , where the color distributions have been normalized according to the number of objects with @xmath126 between 0.25 and 0.25 . there is a wide dispersion in the relative amplitudes of the blue and red peaks . the ratio of red to blue objects in the inner disk is substantially larger than in either ring 10 or the middle disk , indicating that the sfh in this part of m31 was skewed to higher levels of activity 60 200 myr in the past . this is another manifestation of the pronounced red stellar concentrations that are seen in the hess diagrams of the inner disk in figures 5 and 8 . the sfh of the inner disk during intermediate epochs is re - visited in 6 . the blue sequence in the middle disk peaks at a bluer @xmath126 color than at smaller radii . a comparatively low reddening for the middle disk is probably not driving the peak color of blue stars , as there is not a corresponding shift in the color of the main sequence on the @xmath86 cmd ( 4.1 ) . rather , the trajectory of the isochrones on the @xmath95 plane makes the @xmath126 color distribution of blue stars more susceptible to variations in the recent sfh . the blue sequence will narrow as stars at the older end of the @xmath131 myr age range are removed , as these stars have redder colors than younger stars in a given magnitude interval this may not be accompanied by a skewing of the color distribution of red stars , as these objects span a much larger range of ages than main sequence stars with the same m@xmath129 , and so their mean colors are less susceptible to variations in the sfh . as in our study of m33 ( davidge & puzia 2011 ) , we use the @xmath63 lf of blue stars to probe the sfh of the m31 disk . our @xmath63 photometry includes main sequence stars throughout much of the m31 disk with ages @xmath132 myr , allowing the sfh in this timeframe to be directly probed . the damping time for large scale star - forming activity in disks is @xmath133 years ( e.g. leitherer 2001 ) , and so elevated sfrs may linger for this time after they are initiated . therefore , inferences can be made into the more distant past than would otherwise be permitted from resolved stars . following davidge & puzia ( 2011 ) , lfs were constructed of sources with @xmath134 between 0.5 and 0.5 . the majority of objects in this color range are main sequence stars . while most foreground and background objects in the cmds have red colors , there are modest numbers of contaminants that fall within the color interval considered here . these include blue horizontal branch stars in the galactic halo , and star - forming regions in moderately distant spiral galaxies . we account for these statistically using number counts in portions of our megacam fields that are offset from the disk of m31 ( 4.3 ) . the mean lf in these control regions follows a power - law , and a characteristic exponent and zeropoint were determined by using the method of least squares to fit a power - law to this lf . this fitted relation was then subtracted from each of the observed lfs , after scaling to account for differences in angular coverage . the lfs of various radial intervals are compared in figure 11 . the number counts in the top panel are as observed on the sky , and these are given in units of counts arcmin@xmath78 per 0.5 magnitude interval in m@xmath135 . the @xmath8 dex offset between the middle and outer disk lfs is consistent with the walterbos & kennicutt ( 1987 ) surface brightness profile , as is the 0.5 dex ( i.e. 1.25 magnitude arcsec@xmath78 ) dispersion between the lfs with r@xmath2 in the range 5 - 15 kpc . this broad consistency between star counts and surface photometry is encouraging , and suggests that the bright blue stars detected in the megacam data are not affected by crowding . the lfs in the bottom panel of figure 11 have been normalized to match the ring 10 lf between m@xmath136 and 5.25 , thereby allowing the shapes of the lfs to be compared . the lfs follow power - laws , and the outer disk 2 lf is steeper than the rest . the mix of massive and intermediate mass stars in the outer disk thus differs from that in the inner disk . the lfs of the inner disk , ring 10 , and the middle disk in the lower panel of figure 11 are similar . this is perhaps surprising given the concentration of star - forming pockets in ring 10 ( 6 ) . however , it should be kept in mind that the lfs are azimuthal averages , and isolated star - forming pockets will only influence the mean lf if they occur in large numbers . in fact , while ring 10 and ring 14 contain areas of intense recent star formation , they do not stand out in the uv color profile measured by thilker et al . ( 2005 ) ; rather , the uv color stays roughly constant between 10 and 15 kpc . the comparisons in figure 11 indicate that the mix of young to intermediate age stars in ring 10 is not different from that in the inner or middle disks when averaged over large areas , the sfhs of these regions have been broadly similar during the past 100 200 myr . a comparison of the properties of bright main sequence stars in m31 and m33 provides a purely empirical means of assessing the relative star - forming activities of these galaxies during recent epochs . to be sure , star - forming activity in disk galaxies depends on a number of factors , such as environment and mass , and it would be of interest to compare the stellar content of m31 with that of other large galaxies of similar morphological type ; however , such datasets do not yet exist . differences in size and morphological type aside , m33 is an important comparison object as it is at roughly the same distance of m31 , and has been the subject of a number of investigations . the sfhs of m31 and m33 have probably also been coupled at some point in the past ( 1.3 ) . finally , the specific sfr ( ssfr ) of m33 is near the mid - point of spiral galaxies in general ( munoz - mateos et al . 2007 ) , making it a benchmark for activity in a ` typical ' star - forming galaxy . the m33 megacam dataset examined by davidge & puzia ( 2011 ) was recorded during observing conditions that were similar to those of the present m31 observations , and are adopted here for comparisons with m31 . main sequence stars with ages @xmath137 myr are traced throughout much of m33 ( davidge & puzia 2011 ) and m31 ( 3 ) , and these stars provide a direct means of comparing the sfrs of these galaxies . davidge & puzia ( 2011 ) found 3000 massive main sequence stars and bsgs in m33 with ages @xmath138 myr . applying the m33 selection criteria of m@xmath139 to the current data , we find that there are @xmath140 similar objects in m31 . the sfr in m31 during the past 10 myr has thus been one third that of m33 . for comparison , the integrated h@xmath35 luminosities of m31 and m33 are comparable ( e.g. kennicutt et al . 2008 ) , while m31 has a slightly higher fir flux than m33 ( rice et al . 1988 ) . the ssfr is a measure of the rate at which stellar mass grows due to star formation . the bulk of the @xmath141band light from most galaxies has its origins in old and intermediate age stars , and so integrated @xmath141band brightness serves as a crude proxy for total stellar mass . adopting the apparent @xmath142 magnitudes of m31 and m33 from jarrett et al . ( 2003 ) and a distance modulus of 24.92 for m33 ( bonanos et al . 2006 ) , then the integrated total brightnesses of these galaxies are m@xmath143 ( m31 ) and m@xmath144 ( m33 ) . based on the numbers of blue stars with ages @xmath138 myr ( see above ) and their integrated @xmath141band brightnesses , then the ssfr of m31 is only @xmath145 that of m33 . the difference between the ssfrs of m31 and m33 is almost certainly greater than this , as it has been assumed that m31 and m33 have identical @xmath141band mass - to - light ratios ( m / l@xmath146 ) . in actual fact , given the higher relative fraction of young stars in m33 , then the m / l@xmath146 of m33 is probably lower than that of m31 . m / l ratios at a given age depend on metallicity , although for effective ages @xmath147 years the differences between the m / l ratios predicted by z = 0.02 and z = 0.008 models are miniscule ( e.g. figure 4 of mouhcine & lancon 2003 ) . minor differences in the relative m / l ratios of these galaxies notwithstanding , it is clear that m31 is practically dead in terms of recent star - forming activity when compared with m33 . the m31 and m33 lfs , normalized in the interval m@xmath148 4 to 5 , are compared in figure 12 . the m33 lf is that of objects with r@xmath2 between 6 and 8 kpc from davidge & puzia ( 2011 ) . the lf of sources in this part of the galaxy is representative of the main body of the m33 disk . there are conspicuous differences between the m31 and m33 lfs in figure 12 . these differences are likely not a result of differences in metallicity between stars in the disks of m31 and m33 . while model lfs of blue sources with z = 0.019 and z = 0.008 do differ , the differences are largely restricted to m@xmath149 . this magnitude regime is dominated by bsgs , the properties of which are affected by metallicity - dependent mass loss rates . when m@xmath150 main sequence stars dominate , and the model lfs of blue sources show only slight differences . when compared with the m33 disk , outer disk 1 has a lower number of stars with ages of a few tens of myr with respect to stars that formed a few hundred myr in the past . when compared with the m31 lfs , the m33 lf is also deficient in stars with m@xmath135 between 2 and 4 , which is a m@xmath135 range that corresponds to msto ages of 40 100 myr . davidge & puzia ( 2011 ) found that the lf of bright main sequence stars that formed during the past few hundred myr in m33 is well matched by a constant sfr ( csfr ) model , and the comparisons in figure 12 indicate that such a sfh does not hold for m31 . the sfh of the m31 disk is examined in more detail in the next section . we use a forward modelling approach , in which model lfs are constructed for pre - defined sfhs and are then compared with the observations . models were generated from the z = 0.019 girardi et al . ( 2004 ) isochrones using routines in the starfish package ( harris & zaritsky 2001 ) . the models assume an imf power - law index @xmath151 , which kroupa et al . ( 2003 ) find holds for massive stars . models with a salpeter ( 1955 ) imf differ only slightly from those used here ( e.g. figure 11 of davidge & puzia 2011 ) . the @xmath63 lf of main sequence stars provides a relatively robust means of investigating sfhs , as the main sequence is probably the best understood phase of stellar evolution . still , there is an incomplete understanding of the physical processes that affect the structure and evolution of massive main sequence stars . these include uncertainties in convection , which is the dominant mode of energy transport in the central regions of massive stars , mass loss , and rotation . while these uncertainties affect the post - zams models of massive stars , it is encouraging that the zams and tams loci predicted by the isochrones match the blue and red envelopes of the main sequence on the cmds ( 4 ) . statistical studies of main sequence stars are susceptible to uncertainties in the properties of close companions , which are not resolved at the distance of m31 . these uncertainties are not an issue for statistical studies of evolved stars , as the light from the most evolved star in a close stellar system swamps that from less evolved companions at visible wavelengths . however , if all the stars in a close system are on the main sequence then depending on the mass ratio companions may contribute a significant fraction of the total system light . while many of the sources in our data are unresolved systems , if the statistical properties of these systems ( e.g. the frequency of binary or higher - order systems , the mass ratio of components , etc ) do not change with the mass of the primary then the main sequence lf will still yield useful information for a differential examination of the sfh . the megacam data sample stars in a wide range of evolutionary states , all of which provide information about the sfh of m31 . however , combining information from different evolutionary states to investigate the sfh may lead to complications when interpreting the results , as the dominant sources of uncertainties in the input model physics vary with evolutionary phase . comparisons of the sfhs of galaxies at different distances may be most prone to uncertainties related to evolutionary state , as the sfhs may be based on objects in very different stages of evolution . gogarten et al . ( 2010 ) suggest that this may be a factor when comparing the sfhs of m33 and ngc 300 . the impact of uncertainties in models of the most advanced stages of stellar evolution on sfhs has been investigated by melbourne et al . ( 2010 ) , who compute separate sfhs for a dwarf galaxy using main sequence stars and agb stars . the two sfhs differ substantially , and melbourne et al . ( 2010 ) attribute this difference to an incomplete empirical calibration of the agb models , which rely heavily on the lmc and smc as benchmarks . the models used here assume a fixed metallicity , and so do not account for metallicity evolution . while the mean metallicity of a disk grows with time as material that is processed in stars is recycled into the ism , a fixed metallicity is appropriate given that our models cover a time span that is short when compared with the enrichment timescale of a disk . indeed , the age - metallicity relation of the solar neighborhood during recent epochs has a shallow slope ( e.g. edvardsson et al . 1993 ) , and the rate of metallicity evolution in m31 will be even more subdued than this given the lower sfr . the most basic sfh model is that of a simple stellar population ( ssp ) , in which all stars have the same age and metallicity ; ssp lfs are also the building blocks from which lfs that track more elaborate sfhs are constructed . because galaxies are composite stellar systems , it is unlikely that their lfs will be well matched by ssp models . still , if a galaxy experiences a single large episode of star formation that is caught early in its evolution then the lf of the brightest stars may follow that of an ssp . the @xmath63 lfs of main sequence stars in three radial intervals are compared with a log(t@xmath152 ssp model lf in figure 13 . the ring 10 lf is similar to that of the inner and middle disk regions ( figure 11 ) , and so comparisons with models are restricted to the four intervals shown in figure 13 . m31 is a composite stellar system , and so it is not surprising that there is poor agreement between the model ssp lf and the observations . still , the ring 10 lf , which samples a region with a high concentration of recent star formation , has an exponent that comes closest to matching that of the ssp model . the sfhs of isolated , passively evolving disks may be affected by stochastic effects , with spikes in star - forming activity due to events such as the passage of spiral arms and the propagation of density waves from sne in young clusters and associations . evidence for variations in the sfh of the solar neighborhood due to the passage of spiral arms has been found by hernandez et al . nevertheless , the impact of stochastic events will average out when considered over long timespans and large areas , and the sfhs of passively evolving disks might be expected to come close to that of a csfr . the @xmath63 lfs are compared with csfr models in figure 14 , and there is much better agreement with the observations than in figure 13 . in fact , the csfr model provides a reasonable match to the outer disk 2 lf , and is a fair representation of the outer disk 1 lf . still , with the exception of the outer disk 2 lf , the csfr model lfs in figure 14 tend to be steeper than the observed lfs . indeed , while the csfr model is a reasonable match to the outer disk 1 observations with m@xmath153 , which corresponds to msto ages @xmath154 myr , it underestimates the number of stars at the bright end . these differences are even more evident when the ring 10 and ring 14 lfs are compared with the csfr model . given the differences between the observations and models in figure 14 , we investigated sfhs in which the sfr during the past 10 myr has increased with respect to that expected for a csfr at earlier epochs . models were generated in which the recent sfr was increased by various amounts , and the results were compared with the observed lfs ; the best matches , selected by eye , are shown in figure 15 . also shown in each panel is b@xmath155 , which is the ratio of the number of stars that formed within the past 10 myr to the number that formed in the interval 10 100 myr . the outer disk 2 lf is not shown in figure 15 , as it is represented adequately by the csfr model . models with elevated recent star - forming activity better match the lfs of the three regions shown in figure 15 than the csfr models . higher recent sfrs are required to match the ring 10 and ring 14 lfs than is required to match the outer disk 1 lf . while this is consistent with ring 10 and 14 being areas of comparatively intense recent star - forming activity , if elevated levels of recent star formation occur in only part of outer disk 1 ( i.e. at the smallest r@xmath2 in this region ) then the effect of any recent increase in the sfr will be diluted . the differences in the amplitude of recent star - forming activity aside , it is clear that the recent upswing in star - forming activity has not been restricted to ring 10 and 14 , but occured over a large fraction of the disk the change in the recent sfr is not a localized phenomenon that is restricted to only a small part of the m31 disk , but is more global in nature . the spatial distribution of stars in a known age range is an important probe of galaxy evolution . samples of young main sequence stars ( yms ; ages @xmath137 myr ) and intermediate - age main sequence stars ( ims ; ages @xmath5 myr ) were selected from the megacam data based on their location in the @xmath86 cmd . the boundaries of the yms and ims regions in figure 16 were defined using isochrones as guides , while also keeping in mind the results of the artificial star and stacking experiments discussed in 3 . the distributions of objects in the yms and ims samples are shown in figures 17 ( as observed on - sky ) and 18 ( as would be observed face - on ) . the face - on distributions assume that the stars are in an infinitely thin , unwarped disk , and departures from these assumptions will cause blurring of structures in the de - projected distributions . some blurring is expected , as the gas disk of m31 is warped ( corbelli et al . 2010 ) . it is evident from figures 17 and 18 that ( 1 ) recent star - forming activity in m31 has not been uniformly distributed with azimuthal angle , and ( 2 ) knots of stars in both samples are seen throughout the disk . the stars in the yms sample tend to be tightly clustered and define partial arcs . the distribution of yms stars indicates that the most intense area of recent star formation was in the north east quadrant of the disk . there is broad agreement between the yms distribution and images of uv ( figure 1 of thilker et al . 2005 ) and mir ( figure 1 of gordon et al . 2006 ) emission . while peaks in hi emission shown in figure 1 of unwin ( 1983 ) coincide with areas of high yms density in ring 10 , at radii @xmath156 arcmin ( r@xmath157 kpc ) there are collections of yms stars with no corresponding hi feature . in contrast , there is much better agreement between the location of yms concentrations and the peaks in co ( 10 ) emission mapped by nieten et al . ( 2006 ) . stars in the yms sample are seen at moderately small r@xmath2 . large - scale , well - defined structures that contain bright blue stars are seen near 20 arcmin ( i.e. r@xmath158 kpc ) , while individual objects in the yms sample are detected to within 10 arcmin ( r@xmath84 kpc ) of the nucleus . this r@xmath2 is where the bulge and disk contribute comparable amounts of light at visible wavelengths ( e.g. tempel et al . 2011 ) , and is where the fuv nuv color in the main body of the galaxy is reddest ( thilker et al . 2005 ) . ring structures are much more pronounced in the ims distribution than in the yms distribution . the ims sample contains stars that formed over a wider range of ages than the yms stars , and so the effects of spatial fluctuations in star - forming activity are expected to average out . the more - or - less uniform azimuthal distribution of ims objects in the rings suggests that the location of star formation in the immediate past differed from the present day , and that areas of active star formation in the rings have moved through a wide range of position angles within m31 . that ring 10 is seen in samples of objects that are as old as 100 myr indicates that the event that spurred its formation must have occured at least 100 myr ago . in fact , it is somewhat surprising that rings and arcs are present at all in the ims sample , given the rate at which stars in the disks of nearby star - forming galaxies acquire random velocities with time as they interact with giant molecular clouds ( gmcs ) . the velocities acquired in this manner can cause disk stars to move substantial distances over timespans of 100 myr . for example , spiral structure in m33 is well - defined by stars with ages @xmath138 myr , but is not apparent in the distribution of stars with ages of 100 myr ( davidge & puzia 2011 ) . the presence of organized structures in the disk of m31 that are defined by objects in the ims sample suggests that stars in these parts of m31 do not attain the same random motions as their counterparts in m33 . the narrow distribution of orbital angular momentum that is expected for star - forming material in rings undoubtedly contributes to restricting the time evolution of random velocities among stars in these structures . a sample of rheb@xmath159agb stars has been selected using the boundaries indicated on the @xmath95 cmd in figure 16 . the objects in this part of the cmd formed 60 200 myr in the past , and so probe earlier epochs than the ims sample . the faint limit of the extraction region in figure 16 reflects the results of the data characterization experiments discussed in 3 . the face - on distribution of objects in the rheb@xmath159agb sample is shown in figure 19 . the distribution observed on the sky is not shown , as the density of red sources makes it difficult to detect structure because of the orientation of m31 on the sky . as in figure 18 , the major axis of m31 points upwards . the concentration near ( dx , dy ) = ( 60,20 ) is centered on m32 ; the dense central regions of m32 where stars are not resolved appears as a hole in the rheb@xmath159agb distribution . the concentration near ( dx , dy ) = ( 100 , 0 ) is probably associated with ngc 205 ( see figure 1 ) . the distribution of rheb@xmath159agb stars in figure 19 is asymmetric about the horizontal axis , in the sense that the stellar density above and to the left of the galaxy center ( i.e. along the north east arm of the semi - major axis ) is higher than that below the galaxy center . there are also asymmetries to the left and right of the galaxy center , which fall along the minor axis . however , the orientation of m31 on the sky is such that the distribution of objects along the minor axis is more prone to uncertainties arising from for example the inherent thickness of the disk and the presence of stars in the bulge of m31 , both of which may affect the de - projected stellar distribution . we thus focus on the asymmetry along the major axis , where the de - projected distribution is most secure . number counts in an 8 arcmin wide strip along the north east and south west segments of the major axis are shown in figure 20 . the density of rheb@xmath159agb stars along the ne axis within 50 arcmin ( @xmath3 kpc ) of the galaxy center is higher than along the sw axis . at @xmath160 arcmin these stars have a density that is 0.4 dex higher on the ne axis than at the same point on the sw axis . this suggests that the north east part of the galaxy was an area of more intense star - forming activity @xmath161 myr in the past than the south west part . persistent large scale concentrations of elevated star - forming activity are seen in other nearby disk galaxies , such as ngc 253 ( davidge 2010 ) . the number counts of objects in the ims sample are also shown in figure 20 , and it is evident that the ims and rheb@xmath159agb samples have very different radial distributions . star - forming rings form a 5 kpc periodicity in the distribution of ims stars , and these structures are most pronounced along the ne axis . the rings have an amplitude of @xmath8 dex in the ims counts . the red stars show more gradual variations with radius , and there is little or no correlation with features in the ims distribution . departures from large scale trends in the rheb@xmath159agb distribution are substantially smaller than those in the ims distribution , amounting to @xmath98 dex . in addition to the periodic nature of structure in the radial distribution of ims stars , there is also a systematic gradient in the relative numbers of rheb@xmath159agb and ims stars . this is investigated in the lower panel of figure 20 , where the ratio of objects in the two samples is shown . there is a tendency for the numbers of rheb@xmath159agb and ims sources to become more equal towards larger r@xmath2 , indicating that star formation was more centrally concentrated in m31 during intermediate epochs than at present . this trend is most noticeable when considering sources along the ne axis ; while a gradient is seen out to r@xmath162 kpc along the sw axis , it may reverse at larger radii . we note that the fuv nuv color in m31 decreases ( i.e. becomes bluer ) towards larger radii in the inner disk when r@xmath163 kpc ( thilker et al . 2005 ) , and this is broadly consistent with the relative radial distributions of the rheb@xmath159agb and ims objects in our data . the change in uv color in the interval r@xmath164 kpc corresponds to a change in luminosity - weighted ssp age from @xmath165 myr to 290 myr ( thilker et al . 2005 ) . ferguson et al . ( 2002 ) investigate the distribution of rgb and agb stars in the outer regions of the m31 disk . their study intentionally avoids the main body of the disk where crowding is an issue for their aperture photometry measurements , and so there is only modest spatial overlap between our megacam and the ferguson et al . ( 2002 ) datasets . still , we note that figure 2 of ferguson et al . ( 2002 ) shows filamentary structure at large radii . while contamination from background galaxies accounts for a progressively larger fraction of the objects in figure 19 as one moves to larger r@xmath2 , there are hints of structures at large r@xmath2 in figure 19 . these will prove to be promising targets for future deep photometric studies . images obtained with the cfht megacam of the local group galaxy m31 have been used to examine the spatial distribution and photometric properties of the brightest stars in this system . m31 is an important target for studies of galaxy evolution not only because it is the closest external large disk galaxy , but also because it has anemic star - forming activity at the present day , and is the nearest example of a red disk that is evolving in the so - called ` green valley ' . with four megacam pointings we cover most of the m31 disk out to 100 arcmin ( r@xmath166 kpc ) along the major axis , allowing us to investigate large - scale trends in stellar content . the primary goals of our study are ( 1 ) to probe the sfh of m31 during the past few hundred myr , and ( 2 ) to search for evidence of recent interactions . when considered alongside studies of the outer regions and bulge of m31 , which have yielded insights into the evolution of the galaxy during early and intermediate epochs , the results of the current work help to develop an understanding of the evolution of m31 in the context of other nearby galaxies . m31 has been subjected to interactions with companions ( e.g. 1 ) , and this activity has continued to cosmologically recent epochs . these interactions have influenced the large scale re - distribution of stars and gas , and have had a major influence on the sfh of the galaxy . in the remainder of this section we discuss our four principal results : ( 1 ) the nature of the sfh during the past 100 myr , focusing on the apparent rise in the sfr over the last @xmath3 myr that has occured throughout much of the galaxy , ( 2 ) the spatial distribution of stars that formed 10 100 myr in the past , ( 3 ) the distribution of evolved red stars that formed @xmath167 myr ago , and ( 4 ) the subdued level of star - forming activity at the present day , and the evolutionary status of m31 . stellar density changes with location in m31 , and this affects the spatial coverage of the stellar samples used here to probe the sfh . because they are among the brightest stars in the galaxy , and have relatively blue colors when compared with the underlying disk , main sequence stars with ages @xmath3 myr are detected over a large fraction of the m31 disk . while resolving fainter stars is more challenging , main sequence stars with ages @xmath5 myr are still detected at r@xmath168 a few kpc in our data . the recent ( i.e. within a hundred myr ) sfh can thus be investigated over much of the m31 disk . the @xmath63 lf of main sequence stars are indicative of a constant sfr during the past 10 100 myr , and so suggest that any major disturbance to the disk of m31 during intermediate epochs probably occured more than @xmath169 myr in the past . however , the number counts of the brightest main sequence stars indicate that the sfr has increased by a factor of @xmath170 during the past @xmath3 myr . that this upturn in the sfr was not restricted to ring 10 and ring 14 , which are the areas of m31 that contain a large fraction of its young stars and star - forming material , indicates that the trigger was an event that spurred activity throughout the disk , and is probably not stochastic in nature . a rise in star - forming activity during the past 10 myr is not seen at r@xmath0 kpc , and this provides clues into the nature of any event that may have triggered the change in the sfr . if the sfr was elevated by an interaction with a companion , then the point of closest encounter between m31 and this companion would have been r@xmath90 kpc ( e.g. hopkins et al . 2009 ) . if the companion passed through the disk plane then the point of impact must then have been in the present - day star - forming disk . gordon et al . ( 2006 ) models an interaction with m32 , in which that galaxy passed through the inner disk of m31 roughly 20 myr ago . the timing of that event is not very different from our estimate of a recent upswing in the disk sfr . large scale spiral structure is often associated with recent interactions . if m32 ( or another companion ) passed through the disk of m31 in the recent past then why is there no large scale spiral structure ? the amplitude of any spiral pattern depends on the mass of the perturber and the cold gas content of the larger galaxy . in addition , interactions may spur spiral structure only over limited areas , and even then these will survive only over dynamical time scales ( revaz et al . 2009 ) . if m31 was depleted of cold gas or the cold gas distribution was disrupted prior to the most recent interaction with a companion , then prominent spiral structure may not be expected . the sfh measured from the megacam images is broadly consistent with that presented by williams ( 2002 ; 2003 ) . williams ( 2002 ) examined a suite of hst / wfpc2 images to investigate the sfhs of the m31 disk and its environs . given the widely distributed locations of these fields , coupled with hst / wfpc2 s modest science field , it is perhaps not surprising that there is a substantial field - to - field dispersion in the sfhs found from these data . still , the sfhs for the majority of his fields are consistent with a csfr during the past few hundred myr , in agreement with the sfh deduced from the megacam @xmath63 lfs of main sequence stars during the time interval 10 100 myr . not all of the hst / wfpc2 fields show evidence of an upturn in the sfr within the past 10 20 myr . this is perhaps not surprising , as the distribution of young stars , uv light , mir emission , and molecular material all indicate that the recent rise in star - forming activity occured in isolated areas that are distributed throughout the disk . the sfhs deduced from the megacam data are based on stellar content that covers substantial swaths of the m31 disk . the sampling of small star - forming areas by the substantially smaller hst / wfpc2 field will be less complete . williams ( 2003 ) used wide - field ground - based images that cover a total area of 1.4 degrees@xmath36 to probe the recent sfh of the m31 disk . the level of star - forming activity was found to decline by 50% from @xmath171 myr to @xmath172 myr in the past , although the sfr during the past 64 myr is steady to within @xmath173 m@xmath7 year@xmath20 , thereby approximating a csfr to within @xmath174 . the sfr was also found to increase during the past 25 myr , in broad agreement with the results found here . the qualitative agreement with the williams ( 2003 ) results are significant , as williams examined a much smaller portion of the disk , and used a cmd - fitting technique that relies on stars spanning a range of evolutionary states , in contrast to our study which is based on main sequence stars . [ [ signatures - of - recent - interactions - characterizing - coherent - structures - in - the - disk - of - m31 ] ] signatures of recent interactions : characterizing coherent structures in the disk of m31 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ many bright main sequence stars in m31 are found in ring- and shell - like structures . these rings have a radial periodicity of @xmath175 kpc and based on main sequence stars that formed within the past 100 myr a stellar density that is an order of magnitude higher than in the inter - ring regions ( figure 20 ) . such a distribution of young stars indicates that much of the cool gas in m31 has been displaced systematically throughout m31 , as is evident in the distribution of hi and molecular material ( e.g. emerson 1974 ; nieten et al . 2006 ) . the distribution of objects in the ims sample indicates that the rings have been in place ( and forming stars ) for at least @xmath5 myr . gmcs in the disks of ` normal ' star - forming galaxies play a key role in kinematically heating stars over time spans of a few tens of myr . the modest physical width of the rings in m31 ( @xmath176 kpc ; figure 20 ) suggests that the orbital angular momenta of gmcs in these rings must have a compact distribution , reducing the incidence of collisions and thus contributing to the longevity of these structures . the ims sample contains main - sequence stars that formed over the entirety of the past @xmath5 myr , and this might skew age estimates . still , the majority of objects in the ims sample likely have ages that are not much younger than 100 myr . indeed , there has been on - going star formation throughout the disk of m31 during the past 100 myr ( 5 ) , which proceeded at a pace that is consistent with a csfr for @xmath177 myr . given this sfh and the relation between age and msto magnitude , then the majority of objects in the ims sample will likely have ages that are weighted towards the older , as opposed to younger , end of the age range covered by the part of the @xmath178 cmd that contains ims stars . interactions can produce the ring - like distribution of young stars in the m31 disk . hammer et al . ( 2010 ) find that a gas ring forms in models in which m31 merges with a large companion on a polar orbit . block et al . ( 2006 ) discuss a model in which a companion passed through the central regions of m31 @xmath179 myr ago . a propagating density wave , similar to that in the cartwheel galaxy , then triggers star formation as it works its way outwards . block et al ( 2006 ) suggest that the companion that spurred this event was m32 , accompanied by a massive halo to enhance the strength of the interaction . however , the orbit that they assign to m32 , together with its velocity , places it behind m31 , whereas there is strong evidence that m32 lies in front ( ford et al . 1978 ) . in any event , the spatial pattern of the ims sample argues against this model . rather than commencing at its current location only within the last 10 myr or so , we find that the star formation in ring 10 has existed for at least 100 myr . the existence of similar long - lived star - forming structures outside ring 10 , such as ring 14 , is further evidence against a propagating wave origin for ring 10 . in order to have a csfr during the past @xmath5 myr , it is likely that the last major interaction between m31 and a companion occured at least @xmath180 myr in the past . this leads us to speculate that the displacement of cool gas into rings may be an artifact of the event that triggered the last major upswing in star - forming activity 1 2 gyr ago . the event that produced this rise in star - forming activity undoubtedly affected the ism of m31 given the large population of globular clusters that formed at this time ( puzia et al . 2005 ) . hammer et al . ( 2010 ) argue that the 10 kpc ring may have resulted from a merger between m31 and a large companion more than 5 gyr ago . there are other possible perturbers : m33 is one , while another is m32 ( or its progenitor ) . m32 is of interest as it contains a substantial population of stars with an age of a few billion years ( e.g. davidge 1990b ; davidge 2000 ; del burgo et al . 2001 ; worthey 2004 ; rose et al . 2005 ) that are uniformly distributed throughout the galaxy ( davidge & jensen 2007 ) . the properties of m32 are also consistent with it having been sculpted by interactions . compact elliptical galaxies like m32 are associated with dense environments , and there is evidence that the morphology of m32 may have changed dramatically with time . indeed , remnants of a fossil disk have been detected around m32 ( graham 2002 ; choi et al . 2002 ) , suggesting that the present - day galaxy started out as a much larger disk system . m32 also lacks an entourage of globular clusters , as well as interstellar gas ( sage et al . 1998 ) and dust ( gordon et al . 2006 ) . these components may have been removed if m32 was subjected to tidal stripping . the total mass of m32 is only @xmath181 that of m31 , raising the question if it would be large enough to influence star formation in m31 , although if it interacted with m31 then its initial mass was probably larger than at the present day . simulations suggest that satellites with masses that are 10% of the larger system can significantly affect disk properties ( mori & rich 2008 ; qu et al . 2011 ) . the events that re - distributed gas and spurred elevated levels of star formation may leave signatures near the disk boundary . mergers are expected to move the boundaries of stellar disks outwards ( e.g. younger et al . 2007 ; cook et al . 2009 ) , as gas that loses angular momentum and moves inwards scatters stars to larger radii , pushing out the radial extent of the disk . in fact , an extended disk is observed in m31 ( e.g. ferguson & johnson 2001 ) , and is traced out to @xmath182 kpc ( ibata et al . the outer regions of the disk account for 10% of the total m31 disk luminosity ( ibata et al . 2005 ) , indicating that it is a significant structural component . ferguson & johnson ( 2001 ) also conclude that the majority of stars in the outer disk have ages in excess of a gyr , which is consistent with a population of stars that were displaced outward during the event that produced the population of intermediate age globular clusters . in contrast , ibata et al . ( 2005 ) and pearrubia et al . ( 2006 ) argue that this part of the disk is the product of satellite accretion , rather than the re - configuring of the orbits of stars that originated at smaller radii . [ [ signatures - of - recent - interactions - the - asymmetric - distribution - of - evolved - red - stars ] ] signatures of recent interactions : the asymmetric distribution of evolved red stars ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ lop - sided structure in galaxies is not rare ( e.g. zaritsky & rix 1997 ) , and may be a signature of the large - scale radial movement of gas in disks ( e.g. reichard et al . 2009 ) , or the non - uniform accretion of cosmic gas ( bournaud et al . 2005 ) . the projected density of rheb@xmath159agb stars with r@xmath157 kpc along the ne segment of the major axis is roughly two times that along the sw segment . given that these stars have ages 100 200 myr , then the lop - sided distribution was likely spurred by an event that occured within the past @xmath11 0.5 gyr . such a time scale would allow an elevated sfr to die down so that a csfr could be seen in the time interval probed by the @xmath63 lf of main sequence stars in the inner disk . the asymmetric distribution of rheb@xmath159agb stars is consistent with other observations . williams ( 2003 ) found that the region @xmath183 arcmin of the north east of the galaxy center was an area of enhanced star - forming activity 64 256 myr in the past . in addition , circumstellar envelopes around massive agb stars contribute significantly to the mir light from systems containing stars that formed within the past few gyr . the [ 5.8 ] and [ 8.0 ] major axis profiles of m31 shown in figure 2 of barmby et al . ( 2006 ) have higher levels of emission between @xmath184 and 1800 arcsec ( 13 30 arcmin ) along the ne major axis segment than the corresponding section of the sw segment . finally , davidge ( 2012 ) examined 2mass images and discovered a substantial concentration of agb stars @xmath185 kpc ne of the center of m31 . this structure is populated by stars with ages in excess of 100 myr . the presence of a large - scale asymmetry in the distribution of stars with ages 100 - 200 myr is perhaps surprising given that rotational shear might be expected to blur such structures . however , shear will not be an issue for stars at radii where the disk kinematics follow solid body rotation . while rotation curves derived from hi suggest that significant shear might be expected even in the inner regions of m31 ( e.g. braun 1991 ) , the gas dynamics at small radii may not reflect those of objects in the stellar disk ( e.g. kent 1989a ) . still , there is agreement between the rotational properties of m31 defined by hi and pne at r@xmath186 kpc ( halliday et al . 2006 ) , and the rotation velocity measured from hii regions drops near 25 30 armin ( r@xmath187 kpc ; kent 1989b ) . for comparison , the asymmetry in the rheb@xmath159agb number counts disappears when r@xmath188 kpc . kinematic measurements of bright agb stars in the inner disk of m31 will help establish if they are in the disk plane . such measurements could be made in the @xmath141band , where the brightest agb stars in m31 have @xmath189 and significant gains in angular resolution can be realised with adaptive optics systems . the event that produced the asymmetric distribution of rheb@xmath159agb stars may have left signatures in the extraplanar regions if it involved an encounter with a satellite . the gss , which mori & rich ( 2008 ) model as the result of a moderate mass companion passing through the m31 disk , is one candidate signature of such an event . fardal et al . ( 2007 ) find a kinematic age for the gss of 750 myr , which is a few hundred myr older than the rheb@xmath159agb stars examined here . the dwarf elliptical galaxy ngc 205 is one of the largest companions of m31 , and has a projected distance of only @xmath3 kpc from the center of m31 . twisting of the outer isophotes of ngc 205 is consistent with tidal disruption ( choi et al . there is evidence in the resolved stellar content of ngc 205 for a large episode of star formation @xmath5 myr in the past , and another a few hundred myr before that ( davidge 2003 ) . the timing of the first event coincides with the last m31 disk - crossing predicted by cepa & beckman ( 1988 ) , while the second is consistent with their period of ngc 205 about m31 . the second star - forming episode is also consistent with the timing of the event that may have produced the lop - sided distribution of rheb@xmath159agb stars in m31 . howley et al . ( 2008 ) have re - examined the orbital properties of ngc 205 and suggest that it might be on its first passage by m31 , counter to the orbit proposed by cepa & beckman ( 1988 ) . if ngc 205 has interacted with m31 then a tidal tail would be expected , and mcconnachie et al . ( 2005 ) report the possible detection of such a structure . given that there is no consensus or firm evidence on the orbit of ngc 205 about m31 then the connection between this satellite and the larger galaxy must remain speculative for now . photometric surveys of galaxies reveal distinct red and blue sequences on cmds ( e.g. strateva et al . 2001 ; bell et al . the majority of objects in the blue sequence are disk - dominated , while the majority of objects in the red sequence are spheroid - dominated ( strateva et al . mutch et al . ( 2011 ) examine the morphology and integrated colors of m31 , and conclude that it is currently in the ` green valley ' that falls between the blue and red sequences . they also conclude that at least one in six nearby disk galaxies , including the galaxy , lie in the green valley or on the red sequence . if the sfr of m31 has recently increased by a factor of @xmath170 , then the appearance of m31 @xmath5 myr in the past m31 would have been different from the present day . if the current sfr is one third that of m33 , for which the sfr is 0.5 0.7 m@xmath7 year@xmath20 ( hippelein et al . 2003 ; verley et al . 2009 ) , then the sfr in m31 would have been only 0.1 m@xmath7 year@xmath20 a hundred myr ago . when considered in the context of star - forming activity , m31 at that time would have appeared as a largely dead system . the ssfr in m31 is similar to that in nearby red disk galaxies , such as the lenticular galaxy ngc 5102 . using the integrated @xmath141band brightnesses of m31 and ngc 5102 from jarrett et al . ( 2003 ) as proxies for computing relative stellar masses , and assuming a sfr of @xmath190 m@xmath7 year@xmath20 for ngc 5102 ( davidge 2008 ) , then the ssfr of m31 100 myr ago would have been comparable to that of ngc 5102 at the present day . it is perhaps worth noting that m31 and ngc 5102 share common morphological properties . young stars in ngc 5102 have a ring - like distribution and the lop - sided distribution of agb stars in ngc 5102 ( figure 5 of davidge 2010 ) is reminiscent of the distribution of rheb@xmath159agb stars in m31 . bundy et al . ( 2010 ) examine the morphologies of red sequence galaxies , and find that a significant number are passively evolving disks where large - scale star formation has ceased . the red disk galaxies examined by bundy et al . ( 2010 ) tend to have more prominent bulges than blue , star - forming disk galaxies . red disk galaxies are found in a range of environments , suggesting that they are not the exclusive product of evolution in high density regions . the stellar archeology of m31 may provide clues into how red disk galaxies evolve in environments like the local group . star formation in galaxies is influenced by a number of factors , one of which is environment , and galaxies with companions tend to have high ssfrs ( li et al . the ssfr also depends on galaxy mass , in the sense that the ssfr at the present day rises as one moves to lower masses ( e.g. kauffmann et al . 2003 ) . bundy et al . ( 2010 ) suggest that star formation in red disks may have been halted by mergers and subsequent interactions . these events would have driven the rapid consumption of star - forming material , and ultimately contributed to the disruption of the cold ism . red sequence disks are expected to evolve into pressure - supported systems if they experience a major merger , as there is no cool gas from which a disk could re - form . to the extent that m31 can be considered to be a typical red disk galaxy , its past history appears to favor the ` two stage ' model forwarded by bundy et al . ( 2010 ) , in which a disk is first stripped of gas , and is then pummeled by other systems to produce a red galaxy that is supported mainly by random motions , rather than rotation . the final pressure - supported configuration of m31 may be realised a few gyr in the future , when it is projected to merge with the galaxy ( e.g. cox & abraham 2008 ) . barring an injection of cool gas into the m31 disk during the next few gyr , this merger will occur when m31 is gas poor and based on the present - day modest sfr contains a preponderance of stars with ages in excess of a few gyr . lcc field # & date observed & exposure times + & & ( seconds ) + 1 & sept 13 , 2010 & @xmath191 + & & @xmath192 + & & @xmath193 + & & + 2 & oct 31 , 2010 & @xmath194 + & & @xmath192 + & & @xmath193 + & & + 3 & nov 1 , 2010 & @xmath194 + & & @xmath192 + & & @xmath193 + & & + 4 & oct 30 , 2010 & @xmath191 + & & @xmath192 + & & @xmath193 + & & + rccc r@xmath2 & @xmath111 & @xmath113 & @xmath114 + 6 kpc & @xmath195 & @xmath196 & 0.6 + 8 kpc & @xmath197 & @xmath196 & 0.6 + 10 kpc & @xmath198 & @xmath199 & 1.0 + 12 kpc & @xmath200 & @xmath201 & 0.8 + 14 kpc & @xmath202 & @xmath203 & 0.4 +

the study of topological phases of matter is one of the most vibrant directions of research in contemporary condensed matter physics . one core accomplishment has been the theoretical modeling and experimental discovery of two - dimensional topological insulators . @xcite the integer quantum hall effect ( iqhe ) was an early example of how states could be classified into distinct topological classes using an integer , the chern number , to express the quantized hall conductivity . @xcite in the iqhe , the number of delocalized edge channels is directly tied to the quantized hall conductivity through the chern number . more recently , it has been found that the symmetry under reversal of time acts as a protective symmetry for edge modes in ( bulk ) insulators with strong spin - orbit interactions in two and three dimensions , @xcite and that these systems are characterized by a @xmath0 topological invariant . the discovery of @xmath0 topological insulators has triggered a search for a classification of phases of fermionic matter that are distinct by some topological attribute . for non - interacting electrons , a complete classification , the tenfold way , has been accomplished in arbitrary dimensions . @xcite in this scheme , three discrete symmetries that act locally in position space time - reversal symmetry ( trs ) , particle - hole symmetry ( phs ) , and chiral or sublattice symmetry ( sls ) play a central role when defining the quantum numbers that identify the topological insulating fermionic phases of matter within one of the ten symmetry classes ( see columns 1 - 3 from table [ table : main table ] ) . the tenfold way is believed to be robust to a perturbative treatment of short - ranged electron - electron interactions for the following reasons . first , the unperturbed ground state in the clean limit and in a closed geometry is non - degenerate and given by the filled bands of a band insulator . the band gap provides a small expansion parameter , namely the ratio of the characteristic interacting energy scale to the band gap . second , the quantized topological invariant that characterizes the filled bands , provided its definition and topological character survives the presence of electron - electron interactions as is the case for the symmetry class a in two spatial dimensions , can not change in a perturbative treatment of short - range electron - electron interactions . @xcite on the other hand , the fate of the tenfold way when electron - electron interactions are strong is rather subtle . @xcite for example , short - range interactions can drive the system through a topological phase transition at which the energy gap closes , @xcite or they may spontaneously break a defining symmetry of the topological phase . even when short - range interactions neither spontaneously break the symmetries nor close the gap , it may be that two phases from the non - interacting tenfold way cease to be distinguishable in the presence of interactions . in fact , it was shown for the symmetry class bdi in one dimension by fidkowski and kitaev that the non - interacting @xmath1 classification was too fine in that it must be replaced by a @xmath2 classification when generic short - range interactions are allowed . how to construct a counterpart to the tenfold way for interacting fermion ( and boson ) systems has thus attracted a lot of interest . @xcite the fractional quantum hall effect ( fqhe ) is the paradigm for a situation by which interactions select topologically ordered ground states of a very different kind than the non - degenerate ground states from the tenfold way . on a closed two - dimensional manifold of genus @xmath3 , interactions can stabilize incompressible many - body ground states with a @xmath3-dependent degeneracy . excited states in the bulk must then carry fractional quantum numbers ( see ref . and references therein ) . such phases of matter , that follow the fqhe paradigm , appear in the literature under different names : fractional topological insulators , long - range entangled phases , topologically ordered phases , or symmetry enriched topological phases . in this paper we use the terminology long - range entangled ( lre ) phase for all phases with nontrivial @xmath3-dependent ground state degeneracy . all other phases , i.e. , those that follow the iqhe paradigm , are called short - range entangled ( sre ) phases . ( in doing so , we follow the terminology of ref . , that differs slightly from the one used in ref . the latter counts all chiral phases irrespective of their ground state degeneracy as lre . ) while there are nontrivial sre and lre phases in the absence of any symmetry constraint , many sre and lre phases are defined by some protecting symmetry they obey . if this protecting symmetry is broken , the topological attribute of the phase is not well defined any more . however , there is a sense in which lre phases are more robust than sre phases against a weak breaking of the defining symmetry . the topological attributes of lre phases are not confined to the boundary in space between two distinct topological realizations of these phases , as they are for sre phases . they also characterize intrinsic bulk properties such as the existence of gapped deconfined fractionalized excitations . hence , whereas gapless edge states are gapped by any breaking of the defining symmetry , topological bulk properties are robust to a weak breaking of the defining symmetry as long as the characteristic energy scale for this symmetry breaking is small compared to the bulk gap in the lre phase , for a small breaking of the protecting symmetry does not wipe out the gapped deconfined fractionalized bulk excitations . the purpose of this paper is to implement a classification scheme for interacting electronic systems in two spatial dimensions that treats sre and lre phases on equal footing . to this end , we use a coupled wire construction for each of the symmetry classes from the tenfold way . this approach has been pioneered in refs . and for the iqhe and in refs . and for the fqhe ( see also related work in refs . ) . to begin with , non - chiral luttinger liquids are placed in a periodic array of coupled wires . in doing so , forward - scattering two - body interactions are naturally accounted for within each wire . we then assume that the back - scattering ( i.e. , tunneling ) within a given wire or between neighboring wires are the dominant energy scales . imposing symmetries constrains these allowed tunnelings . whether a given arrangement of tunnelings truly gaps out all bulk modes , except for some ungapped edge states on the first and last wire , is verified with the help of a condition that applies to the limit of strong tunneling . we name this condition the haldane criterion , as it was introduced by haldane in his study of the stability of non - maximally chiral edge states in the quantum hall effect . @xcite we show that , for a proper choice of the tunnelings , all bulk modes are gapped . moreover , in five out of the ten symmetry classes of the tenfold way , there remain gapless edge states in agreement with the tenfold way . it is the character of the tunnelings that determines if this wire construction selects a sre or a lre phase . hence , this construction , predicated as it is on the strong tunneling limit , generalizes the tenfold way for sre phases to lre phases . it thereby delivers lre phases that have not yet appeared in the literature before . evidently , this edge - centered classification scheme does not distinguish between lre phases of matter that do not carry protected gapless edge modes at their interfaces . for example , some fractional , time - reversal - symmetric , incompressible and topological phases of matter can have fractionalized excitations in the bulk , while not supporting protected gapless modes at their boundaries . @xcite stated in a slightly more constructive way , we can think of our approach as ( 1 ) fixing , in a first step , a given desired edge theory at the boundary , and ( 2 ) continue , in a second step , by asking whether such an edge can be consistently defined with a set of symmetry - allowed periodic tunneling terms between wires which manage to gap out all other modes . alluding to a related strategy in philosophy , this is what we call wire deconstructionism of topological phases . the paper is organized as follows . we define the array of luttinger liquids in sec . [ sec : definitions ] . the haldane criterion , which plays an essential role for the stability analysis of the edge theory , is reviewed in sec . [ sec : conditions for a spectral gap ] . all five sre entries of table [ table : main table ] are derived in sec . [ sec : reproducing the tenfold way ] , while all five lre entries of table [ table : main table ] are derived in sec . [ sec : fractionalized phases ] . we conclude with sec . [ sec : discussion ] , where we allude to the generalization of our approach to additional symmetries , bosonic systems , and higher spatial dimensions . and @xmath4 , the scattering between them is forbidden by imposing locality in the limit @xmath5 . ( b ) periodic boundary conditions allow the scattering vector @xmath6 that gaps modes which were protected by locality before . , width=317 ] & @xmath7 & @xmath8&@xmath9 & & short - range entangled ( sre ) topological phase & long - range entangled ( lre ) topological phase + @xmath10 a & @xmath11 & @xmath11&@xmath11&@xmath12 & [ cols="^ " , ] + cii & @xmath13 & @xmath13&@xmath14 & & none & + [ table : main table ] we consider an array of @xmath15 parallel wires that stretch along the @xmath16 direction of the two - dimensional embedding euclidean space ( see fig . [ fig : boundaryconditions ] ) . we label a wire by the latin letter @xmath17 . each wire supports fermions that carry an even integer number @xmath18 of internal degrees of freedom that discriminate between left- and right - movers , the projection along the spin-@xmath19 quantization axis , and particle - hole quantum numbers , among others ( e.g. , flavors ) . we label these internal degrees of freedom by the greek letter @xmath20 . we combine those two indices in a collective index @xmath21 . correspondingly , we introduce the @xmath22 pairs of creation @xmath23 and annihilation @xmath24 field operators obeying the fermionic equal - time algebra [ eq : def hat psi s ] @xmath25 with all other anticommutators vanishing and the collective labels @xmath26 . the notation @xmath27 is used for the operator - valued row ( @xmath28 ) and column ( @xmath29 ) vector fields . we assume that the many - body quantum dynamics of the fermions supported by this array of wires is governed by the hamiltonian @xmath30 , whereby interactions within each wire are dominant over interactions between wires so that we may represent @xmath30 as @xmath15 coupled luttinger liquids , each one of which is composed of @xmath18 interacting fermionic channels . by assumption , we may thus bosonize the @xmath22 fermionic channels making up the array . to this end , we follow ref . . within abelian bosonization , this is done by postulating first the @xmath31 matrix [ eq : def mathcal k ] @xmath32 to be symmetric with integer - valued entries . because we are after an array of identical wires , each of which having its quantum dynamics governed by that of a luttinger liquid , it is natural to assume that @xmath33 is reducible , @xmath34 a second @xmath31 matrix is then defined by [ eq : def mathcal l ] @xmath35 where @xmath36 third , one verifies that , for any pair @xmath37 , the hermitian fields @xmath38 and @xmath39 , defined by the mandelstam formula [ eq : def hat phi s ] @xmath40 as they are , obey the bosonic equal - time algebra @xmath41= -\mathrm{i}\pi \left ( \mathcal{k}^{-1}_{\mathsf{a}\mathsf{a}'}\ , \mathrm{sgn}(x - x ' ) + \mathcal{k}^{-1}_{\mathsf{a}\mathsf{b}}\ , \mathcal{l}^{\,}_{\mathsf{b}\mathsf{c}}\ , \mathcal{k}^{-1}_{\mathsf{c}\mathsf{a } ' } \right ) . \label{eq : def hat phi 's b}\ ] ] here , the notation @xmath42 stands for normal ordering of the argument @xmath43 and the summation convention over repeated indices is implied . in line with eq . ( [ eq : def hat psi s b ] ) , we use the notation @xmath44 for the operator - valued row ( @xmath45 ) and column ( @xmath46 ) vector fields . periodic boundary conditions along the @xmath16 direction parallel to the wires are imposed by demanding that @xmath47 equipped with eqs . ( [ eq : def mathcal k])([eq : def hat phi s ] ) , we decompose additively the many - body hamiltonian @xmath30 for the @xmath48 interacting fermions propagating on the array of wires into [ eq : hat h bosonized ] @xmath49 hamiltonian @xmath50 even though quadratic in the bosonic field , encodes both local one - body terms as well as contact many - body interactions between the @xmath18 fermionic channels in any given wire from the array through the block - diagonal , real - valued , and symmetric @xmath31 matrix @xmath51 hamiltonian @xmath52 is not quadratic in the bosonic fields . with the understanding that the operator - multiplication of identical fermion fields at the same point @xmath16 along the wire requires point splitting , and with the short - hand notation @xmath53 , we interpret @xmath54 as ( possibly many - body ) tunnelings between the fermionic channels . here , we introduced the set @xmath55 comprised of _ all _ integer - valued tunneling vectors @xmath56 obeying the condition @xmath57 and we assigned to each @xmath58 from the set @xmath55 the real - valued functions @xmath59 and @xmath60 the condition ( [ eq : def mathcal t b ] ) ensures that these tunneling events preserve the parity of the total fermion number for the superconducting symmetry classes ( symmetry classes d , diii , c , and ci in table [ table : main table ] ) , while they preserve the total fermion number for the non - superconducting symmetry classes ( symmetry classes a , aiii , ai , aii , bdi , and cii in table [ table : main table ] ) . we emphasize that the integer @xmath61 dictates that @xmath58 encodes a @xmath62-body interaction in the fermion representation . our strategy consists in choosing the many - body hamiltonian @xmath63 defined in eq . ( [ eq : hat h bosonized ] ) so that ( i ) it belongs to any one of the ten symmetry classes from the tenfold way ( with the action of symmetries defined in sec . [ sec : representation of symmetries ] ) and ( ii ) all excitations in the bulk are gapped by a _ specific _ choice of the tunneling vectors @xmath55 entering @xmath54 ( with the condition for a spectral gap given in sec . [ sec : conditions for a spectral gap ] ) the energy scales in @xmath54 are assumed sufficiently large compared to those in @xmath64 so that it is @xmath64 that may be thought of as a perturbation of @xmath54 and not the converse . we anticipate that for five of the ten symmetry classes there can be protected gapless edge states because of locality and symmetry . step ( ii ) for each of the five symmetry classes supporting gapless edge states is represented pictorially as is shown in the fourth column of table [ table : main table ] . in each symmetry class , topologically trivial states that do not support protected gapless edge states in the tenfold classification can be constructed by gapping all states in each individual wire from the array . the classification is based on the presence or the absence of the trs and the phs that are represented by the antiunitary many - body operator @xmath65 and the unitary many - body operator @xmath66 , respectively . each of @xmath65 and @xmath66 can exist in two varieties such that their single - particle representations @xmath67 and @xmath68 square to the identity operator up to the multiplicative factor @xmath69 , @xmath70 respectively . by assumption , the set of all degrees of freedom in each given wire is invariant under the actions of @xmath65 and @xmath66 . if so , we can represent the actions of @xmath65 and @xmath66 on the fermionic fields in two steps . first , we introduce two @xmath71-dimensional matrix representations @xmath72 and @xmath73 of the permutation group of @xmath18 elements , which we combine into the block - diagonal @xmath31 real - valued and orthogonal matrices [ eq : def reps widehat theta and widehat pi ] @xmath74 where @xmath75 is the @xmath76 unit matrix and we make sure that @xmath72 and @xmath73 represent products of transpositions so that @xmath77 second , we introduce two column vectors @xmath78 and @xmath79 , which we combine into the two column vectors @xmath80 and the @xmath31 diagonal matrices @xmath81 with the components of the vectors @xmath82 and @xmath83 as diagonal matrix elements . the vectors @xmath84 and @xmath85 are not chosen arbitrarily . we demand that the vectors @xmath86 and @xmath87 are made of even [ for the @xmath88 in eq . ( [ eq : widehat theta and widehat phi square to pm1 ] ) ] and odd [ for the @xmath89 in eq . ( [ eq : widehat theta and widehat phi square to pm1 ] ) ] integer entries only , while @xmath90 and @xmath91 in order to meet @xmath92 and @xmath93 , respectively . the operations of reversal of time and interchanges of particles and holes are then represented by @xmath94 for the fermions and @xmath95 for the bosons . one verifies that eq . ( [ eq : widehat theta and widehat phi square to pm1 ] ) is fulfilled . hamiltonian ( [ eq : hat h bosonized ] ) is trs if [ eq : trs h ] @xmath96 this condition is met if @xmath97 the hamiltonian ( [ eq : hat h bosonized ] ) is phs if [ eq : phs h ] @xmath98 this condition is met if ( see appendix [ appendixsec : cond trs and phs ] ) @xmath99 the total number of fermions is a good quantum number in any metallic or insulating phase of fermionic matter . this is not true anymore in the mean - field treatment of superconductivity . in a superconductor , within a mean - field approximation , charge is conserved modulo two as cooper pairs can be created and annihilated . the existence of superconductors and the phenomenological success of the mean - field approximation suggest that the conservation of the total fermion number operator should be relaxed down to its parity in a superconducting phase of matter . if we only demand that the parity of the total fermion number is conserved , we may then decompose any fermionic creation operator in the position basis into its real and imaginary parts , thereby obtaining two hermitean operators called majorana operators . any hermitean hamiltonian that is build out of even powers of majorana operators necessarily preserves the parity of the total fermion number operator , but it might break the conservation of the total fermion number . by definition , any such hamiltonian belongs to the symmetry class d. the tool of abelian bosonization allows to represent a fermion operator as a single exponential of a bose field . in abelian bosonization , a majorana operator is the sum of two exponentials , and this fact makes it cumbersome to apply abelian bosonization for majorana operators . it is possible to circumvent this difficulty by representing any hamiltonian from the symmetry class d in terms of the components of nambu spinors obeying a reality condition . indeed , we may double the dimensionality of the single - particle hilbert space by introducing nambu spinors with the understanding that ( i ) a reality condition on the nambu spinors must hold within the physical subspace of the enlarged single - particle hilbert space and ( ii ) the dynamics dictated by the many - body hamiltonian must be compatible with this reality condition . @xcite the reality condition keeps track of the fact that there are many ways to express an even polynomial of majorana operators in terms of the components of a nambu spinor . the complication brought about by this redundency is compensated by the fact that it is straightforward to implement abelian bosonization in the nambu representation . we implement this particle - hole doubling by assigning to every pair of fermionic operators @xmath100 and @xmath101 ( whose indices we have been omitted for simplicity ) related to each other by the reality condition @xmath102 the pair of bosonic field operators @xmath103 and @xmath104 related by the reality condition @xmath105 invariance under this transformation has to be imposed on the ( interacting ) hamiltonian in the doubled ( nambu ) representation . in addition to the phs , we also demand , when describing the superconducting symmetry classes , that the parity of the total fermion number is conserved . this discrete global symmetry , the symmetry of the hamiltonian under the reversal of sign of all fermion operators , becomes a continuous @xmath106 global symmetry that is responsible for the conservation of the electric charge in all non - superconducting symmetry classes . in this way , all 9 symmetry classes from the tenfold way descend from the symmetry class d by imposing a composition of trs , @xmath106 charge conservation , and the chiral ( sublattice ) symmetry . hamiltonian @xmath64 in the decomposition ( [ eq : hat h bosonized ] ) has @xmath48 gapless modes . however , @xmath64 does not commute with @xmath54 and the competition between @xmath64 and @xmath54 can gap some , if not all , the gapless modes of @xmath64 . for example , a tunneling amplitude that scatters the right mover into the left mover of each flavor in each wire will gap out the spectrum of @xmath64 . a term in @xmath107 has the potential to gap out a gapless mode of @xmath64 if the condition ( in the heisenberg representation ) @xcite @xmath108= c^{\,}_{\mathcal{t}}(x ) \label{eq : locking condition}\ ] ] holds for some time - independent real - valued functions @xmath109 on the canonical momentum @xmath110 that is conjugate to @xmath111 , when applied to the ground state . the locking condition ( [ eq : locking condition ] ) removes a pair of chiral bosonic modes with opposite chiralities from the gapless degrees of freedom of the theory . however , not all scattering vectors @xmath58 can simultaneously lead to such a locking due to quantum fluctuations . the set of linear combinations @xmath112 that can satisfy the locking condition ( [ eq : locking condition ] ) simultaneously is labeled by the subset @xmath113 of all tunneling matrices @xmath55 defined by eqs . ( [ eq : def mathcal t a ] ) and ( [ eq : def mathcal t b ] ) obeying the haldane criterion ( [ eq : haldane conditions ] ) @xcite [ eq : haldane conditions ] @xmath114 for any @xmath115 and @xmath116 pairwise for any @xmath117 . our first goal is to apply the wire construction in order to reproduce the classification of non - interacting topological insulators ( symmetry classes a , aiii , ai , aii , bdi , and cii in table [ table : main table ] ) and superconductors ( symmetry classes d , diii , c , and ci in table [ table : main table ] ) in ( @xmath118 ) dimensions ( see table [ table : main table ] ) . @xcite in this section , we will carry out the classification scheme within the bosonized description of quantum wires . here , we will restrict the classification to one - body tunneling terms , i.e. , @xmath119 in eq . ( [ eq : q - def ] ) , for the non - superconducting symmetry classes , and to two - body tunneling terms , i.e. , @xmath120 in eq . ( [ eq : q - def ] ) , for the superconducting symmetry classes . in sec . [ sec : fractionalized phases ] , we generalize this construction to the cases @xmath121 and @xmath122 of multi - particle tunnelings in the non - superconducting and superconducting symmetry classes , respectively . the topological stability of edge modes will be an immediate consequence of the observation that no symmetry - respecting local terms can be added to the models that we are going to construct . within the classification of non - interacting hamiltonians , superconductors are nothing but fermionic bilinears with a particle - hole symmetry . the physical interpretation of the degrees of freedom as bogoliubov quasiparticles is of no consequence to the analysis . in particular , they still carry an effective conserved @xmath106 charge in the non - interacting description . topological insulators in symmetry class a can be realized without any symmetry aside from the @xmath106 charge conservation . the wire construction starts from wires supporting spinless fermions , so that the minimal choice @xmath123 only counts left- and right - moving degrees of freedom . the @xmath124-matrix reads [ eq : def h for class a ] @xmath125 the entry @xmath88 of the @xmath124-matrix corresponds to a right mover . it is depicted by the symbol @xmath126 in the first line of table [ table : main table ] . the entry @xmath89 of the @xmath124-matrix corresponds to a left mover . it is depicted by the symbol @xmath127 in the first line of table [ table : main table ] . the operation for reversal of time in any one of the @xmath15 wires is represented by [ one verifies that eq . ( [ eq : trs on p and i ] ) holds ] @xmath128 we define @xmath54 by choosing @xmath129 scattering vectors , whereby , for any @xmath130 , [ eq : def h of t for class a ] @xmath131 with @xmath17 and @xmath132 . in other words , @xmath133 for @xmath134 . intent on helping with the interpretation of the tunneling vectors , we use the @xmath135 s in eq . ( [ eq : def h of t for class a e ] ) to compartmentalize the elements within a given wire . henceforth , there are @xmath123 vector components within each pair of @xmath135 s that encode the @xmath123 degrees of freedom within a given wire . the @xmath136th scattering vector ( [ eq : def h of t for class a e ] ) labels a one - body interaction in the fermion representation that fulfills eq . ( [ eq : def mathcal t b ] ) and breaks trs , since the scattering vector @xmath137 is mapped into the scattering vector @xmath138 by the permutation @xmath72 that represents reversal of time in a wire by exchanging right- with left - movers . for any @xmath130 , we also introduce the amplitude @xmath139 and the phase @xmath140 according to eqs . ( [ eq : trs h d ] ) and ( [ eq : trs h e ] ) , respectively . the choices for the amplitude ( [ eq : def h of t for class a f ] ) and the phase ( [ eq : def h of t for class a g ] ) are arbitrary . in particular the amplitude ( [ eq : def h of t for class a f ] ) can be chosen to be sufficiently large so that it is @xmath64 that may be thought of as a perturbation of @xmath54 and not the converse . one verifies that all @xmath129 scattering vectors ( [ eq : def h of t for class a d ] ) satisfy the haldane criterion ( [ eq : haldane conditions ] ) , i.e. , @xmath141 correspondingly , the term @xmath54 gaps out @xmath142 of the @xmath143 gapless modes of @xmath64 . two modes of opposite chirality that propagate along the first and last wire , respectively , remain in the low energy sector of the theory . these edge states are localized on wire @xmath144 and @xmath145 , respectively , for their overlaps with the gapped states from the bulk decay exponentially fast as a function of the distance away from the first and end wires . the energy splitting between the edge state localized on wire @xmath144 and the one localized on wire @xmath145 that is brought about by the bulk states vanishes exponentially fast with increasing @xmath15 . two gapless edge states with opposite chiralities emerge in the two - dimensional limit @xmath5 . at energies much lower than the bulk gap , the effective @xmath33-matrix for the edge modes is @xmath146 here , @xmath147 follows from replacing the entries in the @xmath148 @xmath33 matrix for all gapped modes by 0 . the pictorial representation of the topological phase in the symmetry class a with one chiral edge state per end wire through the wire construction is shown on the first row and fourth column of table [ table : main table ] . the generalization to an arbitrary number @xmath149 of gapless edge states sharing a given chirality on the first wire that is opposite to that of the last wire is the following . we enlarge @xmath123 to @xmath150 by making @xmath149 identical copies of the model depicted in the first row and fourth column of table [ table : main table ] . the stability of the @xmath149 chiral gapless edge states in wire @xmath151 and wire @xmath15 is guaranteed because back - scattering among these gapless edges state is not allowed kinematically within wire @xmath151 or within wire @xmath15 , while back - scattering across the bulk is exponentially suppressed for @xmath15 large by locality and the gap in the bulk . the number of robust gapless edge states of a given chirality is thus integer . this is the reason why @xmath1 is found in the third column on the first line of table [ table : main table ] . it is imperative to ask whether the phases that we constructed so far exhaust all possible sre phases in the symmetry class a. by demanding that one - body interactions are dominant over many - body interactions , we have constructed all phases from the ( exhaustive ) classification for non - interacting fermions in class a and only those . in these phases , the same topological invariant controls the hall and the thermal conductivities . however , it was observed that interacting fermion systems can host additional sre phases in the symmetry class a where this connection is lost . @xcite these phases are characterized by an edge that includes charge - neutral chiral modes . while such modes contribute to the quantized energy transport ( i.e. , the thermal hall conductivity ) , they do not contribute to the quantized charge transport ( i.e. , the charge hall conductivity ) . by considering the thermal and charge hall conductivity as two independent quantized topological responses , this enlarges the classification of spt phases in the symmetry class a to @xmath152 . starting from identical fermions of charge @xmath153 , we now construct an explicit wire model that stabilizes a sre phase of matter in the symmetry class a carrying a non - vanishing hall conductivity but a vanishing thermal hall conductivity . in order to build a wire - construction of such a strongly interacting sre phase in the symmetry class a , we group three spinless electronic wires into one unit cell , i.e. , @xmath154 it will be useful to arrange the charges @xmath155 measured in units of the electron charge @xmath153 for each of the modes @xmath156 , @xmath20 , into a vector @xmath157 the physical meaning of the tunneling vectors ( interactions ) that we define below is most transparent , if we employ the following linear transformation on the bosonic field variables [ eq : trasf for fields phi a ] @xmath158 where @xmath159 is a @xmath31 block - diagonal matrix with the block @xmath160 having integer entries and unit determinant . the transformation @xmath160 that we employ is given by @xmath161 it brings @xmath124 to the form @xmath162 as we can read off from eq . , the parity of @xmath163 determines the self - statistics of particles of type @xmath164 . as eq is form invariant under the transformation ( [ eq : trasf for fields phi a ] ) , we conclude that , with the choice , the transformed modes @xmath132 as well as the modes @xmath165 are pairs of bosonic degrees of freedom , while the third and fourth mode remain fermonic . furthermore , the charges transported by the transformed modes @xmath166 are given by @xmath167 we may then define the charge - conserving tunneling vectors @xmath168 using eq . ( [ eq : t tilde if one wants to get bosons out of fermions ] ) , these tunneling vectors can readily be rewritten in the original electronic degrees of freedom . these tunneling vectors gap all modes in the bulk and the remaining gapless edge modes on the left edge are @xmath169 the only charge - conserving tunneling vector that could gap out this effective edge theory , @xmath170 , is not compatible with haldane s criterion . we conclude that the edge theory is stable against charge conserving perturbations . the hall conductivity supported by this edge theory is given by @xmath171 in units of @xmath172 . this is the minimal hall conductivity of a sre phase of bosons , if each boson is interpreted as a pair of electrons carrying the electronic charge @xmath173 . @xcite on the other hand , the edge theory supports two modes with opposite chiralities , for the symmetric matrix @xmath174 has the pair of eigenvalues @xmath175 . thus , the net energy transported along the left edge , and with it the thermal hall conductivity , vanishes . topological insulators in symmetry class aii can be realized by demanding that @xmath106 charge conservation holds and that trs with @xmath176 holds . the wire construction starts from wires supporting spin-@xmath19 fermions because @xmath176 , so that the minimal choice @xmath177 counts two pairs of kramers degenerate left- and right - moving degrees of freedom carrying opposite spin projections on the spin quantization axis , i.e. , two pairs of kramers degenerate helical modes . the @xmath124-matrix reads [ eq : def h of v for class aii ] @xmath178 the entries in the @xmath124-matrix represent , from left to right , a right - moving particle with spin up , a left - moving particle with spin down , a left - moving particle with spin up , and a right - moving particle with spin down . the operation for reversal of time in any one of the @xmath15 wires is represented by [ one verifies that eq . ( [ eq : trs on p and i ] ) holds ] @xmath179 we define @xmath64 by choosing any symmetric @xmath180 matrix @xmath181 that obeys @xmath182 we define @xmath183 by choosing @xmath142 scattering vectors as follows . for any @xmath130 , we introduce the pair of scattering vectors @xmath184 and @xmath185 the scattering vector ( [ eq : def h of t for class aii d ] ) labels a one - body interaction in the fermion representation that fulfills eq . ( [ eq : def mathcal t b ] ) . it scatters a left mover with spin up from wire @xmath136 into a right mover with spin up in wire @xmath186 . for any @xmath130 , we also introduce the pair of amplitudes @xmath187 and the pair of phases @xmath188 according to eqs . ( [ eq : trs h d ] ) and ( [ eq : trs h e ] ) , respectively . the choices for the amplitude ( [ eq : def h of t for class aii f ] ) and the phase ( [ eq : def h of t for class aii g ] ) are arbitrary . the subscript so refers to the intrinsic spin - orbit coupling . the rational for using it shall be shortly explained . one verifies that all @xmath142 scattering vectors ( [ eq : def h of t for class aii c ] ) and ( [ eq : def h of t for class aii d ] ) satisfy the haldane criterion ( [ eq : haldane conditions ] ) , i.e. , @xmath189 for @xmath190 . correspondingly , the term @xmath183 gaps out @xmath191 of the @xmath192 gapless modes of @xmath64 . two pairs of kramers degenerate helical edge states that propagate along the first and last wire , respectively , remain in the low energy sector of the theory . these edge states are localized on wire @xmath144 and @xmath145 , respectively , for their overlaps with the gapped states from the bulk decay exponentially fast as a function of the distance away from the first and end wires . the energy splitting between the edge state localized on wire @xmath144 and wire @xmath145 brought about by the bulk states vanishes exponentially fast with increasing @xmath15 . two pairs of gapless kramers degenerate helical edge states emerge in the two - dimensional limit @xmath5 . at energies much lower than the bulk gap , the effective @xmath33-matrix for the two pairs of helical edge modes is @xmath193 here , @xmath147 follows from replacing the entries in the @xmath194 @xmath33 matrix for all gapped modes by 0 . we are going to show that the effective scattering vector @xmath195 with the potential to gap out the pair of kramers degenerate helical edge modes on wire @xmath144 since it fulfills the haldane criterion ( [ eq : haldane conditions ] ) , is not allowed by trs . @xcite on the one hand , @xmath196 maps to itself under reversal of time , @xmath197 on the other hand , @xmath198 therefore , the condition ( [ eq : trs h e ] ) for @xmath196 to be a trs perturbation is not met , for the phase @xmath199 associated to @xmath196 then obeys @xmath200 a condition that can not be satisfied . had we imposed a trs with @xmath201 instead of @xmath202 as is suited for the symmetry class ai that describes spinless fermions with trs , we would only need to replace @xmath84 in eq . ( [ eq : def h of v for class aii b ] ) by the null vector . if so , the scattering vector ( [ eq : def t eff aii ] ) is compatible with trs since the condition ( [ eq : trs h e ] ) for trs then becomes @xmath203 instead of eq . ( [ eq : trs condition on alpha teff class aii ] ) . this is the reason why symmetry class ai is always topologically trivial in two - dimensional space from the point of view of the wire construction . note also that if we would not insist on the condition of charge neutrality ( [ eq : def mathcal t b ] ) , the tunneling vector @xmath204 that satisfies the haldane criterion and is compatible with trs could gap out the kramers degenerate pair of helical edge states . to address the question of what happens if we change @xmath177 to @xmath205 with @xmath149 any strictly positive integer in each wire from the array , we consider , without loss of generality as we shall see , the case of @xmath206 . to this end , it suffices to repeat all the steps that lead to eq . ( [ eq : def t eff aii ] ) , except for the change @xmath207 and @xmath208 are compatible with the condition that trs holds in that the pair is a closed set under reversal of time , @xmath209 one verifies that these scattering vectors fulfill the haldane criterion ( [ eq : haldane conditions ] ) . consequently , inclusion in @xmath183 of the two cosine potentials with @xmath210 and @xmath211 entering in their arguments , respectively , gaps out the pair of kramers degenerate helical modes on wire @xmath144 . the same treatment of the wire @xmath145 leads to the conclusion that trs does not protect the gapless pairs of kramers degenerate edge states from perturbations when @xmath206 . the generalization to @xmath205 channels is that it is only when @xmath149 is odd that a pair of kramers degenerate helical edge modes is robust to the most generic @xmath183 of the form depicted in the fourth column on line 3 of table [ table : main table ] . since it is the parity of @xmath149 in the number @xmath205 of channels per wire that matters for the stability of the kramers degenerate helical edge states , we use the group of two integers @xmath0 under addition modulo 2 in the third column on line 3 of table [ table : main table ] . if we were to impose conservation of the projection of the spin-@xmath19 quantum number on the quantization axis , we must then preclude from all scattering vectors processes by which a spin is flipped . in particular , the scattering vectors ( [ eq : def mathcal teff aii ] ) and ( [ eq : def mathcal teff aii ] ) are not admissible anymore . by imposing the @xmath106 residual symmetry of the full @xmath212 symmetry group for a spin-@xmath19 degree of freedom , we recover the group of integers @xmath1 under the addition that encodes the topological stability in the quantum spin hall effect ( qshe ) . we close the discussion of the symmetry class aii by justifying the interpretation of the index so as an abbreviation for the intrinsic spin - orbit coupling . to this end , we introduce a set of @xmath129 pairs of scattering vectors [ eq : def mathcal t r aii ] @xmath213 and @xmath214 for @xmath215 . the scattering vector ( [ eq : def mathcal t r aii a ] ) labels a one - body interaction in the fermion representation that fulfills eq . ( [ eq : def mathcal t b ] ) . the index r is an acronym for rashba as it describes a backward scattering process by which a left mover with spin down from wire @xmath136 is scattered into a right mover with spin up on wire @xmath186 and conversely . for any @xmath130 , we also introduce the pair of amplitudes @xmath216 and the pair of phases @xmath217 according to eqs . ( [ eq : trs h d ] ) and ( [ eq : trs h e ] ) , respectively . in contrast to the intrinsic spin - orbit scattering vectors , the rashba scattering vectors ( [ eq : def mathcal t r aii a ] ) fail to meet the haldane criterion ( [ eq : haldane conditions ] ) as @xmath218 hence , the rashba scattering processes fail to open a gap in the bulk , as is expected of a rashba coupling in a two - dimensional electron gas . on the other hand , the intrinsic spin - orbit coupling can lead to a phase with a gap in the bulk that supports the spin quantum hall effect in a two - dimensional electron gas . the simplest example among the topological superconductors can be found in the symmetry class d that is defined by the presence of a phs with @xmath219 and the absence of trs . with the understanding of phs as discussed in sec . [ sec : phs in superconductors ] , we construct a representative phase in class d from identical wires supporting right- and left - moving spinless fermions each of which carry a particle or a hole label , i.e. , @xmath177 . the @xmath124-matrix reads [ eq : def h of v for class d ] @xmath220 the entries in the @xmath124-matrix represent , from left to right , a right - moving particle , a left - moving particle , a left - moving hole , and a right - moving hole . the operation for the exchange of particles and holes in any one of the @xmath15 wires is represented by [ one verifies that eq . ( [ eq : phs on p and i ] ) holds ] @xmath221 we define @xmath64 by choosing any symmetric @xmath222 matrix @xmath181 that obeys @xmath223 we define @xmath54 by choosing @xmath224 scattering vectors as follows . for any wire @xmath225 , we introduce the scattering vector [ eq : def h of t for class d ] @xmath226 between any pair of neighboring wires we introduce the scattering vector @xmath227 for @xmath130 . we observe that both @xmath228 and @xmath229 are eigenvectors of the particle - hole transformation in that @xmath230 thus , to comply with phs , we have to demand that the phases @xmath231 while @xmath232 are unrestricted . similarly , the amplitudes @xmath233 and @xmath234 can take arbitrary real values . one verifies that the set of scattering vectors defined by eqs . ( [ eq : def h of t for class d d ] ) and ( [ eq : def h of t for class d e ] ) satisfies the haldane criterion . correspondingly , the term @xmath54 gaps out @xmath235 of the @xmath192 gapless modes of @xmath64 . furthermore , one identifies with @xmath236 a unique ( up to an integer multiplicative factor ) scattering vector that satisfies the haldane criterion with all existing scattering vectors eqs . ( [ eq : def h of t for class d d ] ) and ( [ eq : def h of t for class d e ] ) and could thus potentially gap out the remaining pair of modes . however , the tunneling @xmath237 is non - local for it connects the two edges of the system when open boundary conditions are chosen . we thus conclude that the two remaining modes are exponentially localized near wire @xmath144 and wire @xmath145 , respectively , and propagate with opposite chirality . to give a physical interpretation of the resulting topological ( edge ) theory in this wire construction , one has to keep in mind that the degrees of freedom were artificially doubled . we found , in this doubled theory , a single chiral boson ( with chiral central charge @xmath238 ) . to interpret it as the edge of a chiral @xmath239 superconductor , we impose the reality condition to obtain a single chiral majorana mode with chiral central charge @xmath240 . the pictorial representation of the topological phase in the symmetry class d through the wire construction is shown on the fifth row of table [ table : main table ] . the generalization to an arbitrary number @xmath149 of gapless chiral edge modes is analogous to the case discussed in symmetry class a. the number of robust gapless chiral edge states of a given chirality is thus integer . this is the reason why the group of integers @xmath1 is found in the third column on the fifth line of table [ table : main table ] . the remaining two topological nontrivial superconducting classes diii ( trs with @xmath176 and phs with @xmath219 ) and c ( phs with @xmath241 ) involve spin-@xmath19 fermions . each wire thus features no less than @xmath242 internal degrees of freedom corresponding to the spin-@xmath19 , chirality , and particle / hole indices . the construction is very similar to the cases we already presented . we relegate details to the appendices [ appendixsec : symmetry class c ] and [ appendixsec : symmetry class diii ] . the scattering vectors that are needed to gap out the bulk for each class of class diii and c are represented pictorially in the fourth column on lines 4 and 9 of table [ table : main table ] . we have provided an explicit construction by way of an array of wires supporting fermions that realizes all five insulating and superconducting topological phases of matter with a nondegenerate ground state in two - dimensional space according to the tenfold classification of band insulators and superconductors . the topological protection of edge modes in the bosonic formulation follows from imposing the haldane criterion ( [ eq : haldane conditions ] ) along with the appropriate symmetry constraints . in the next section we shall extend the wire construction to allow many - body tunneling processes that delivers fractionalized phases with degenerate ground states . the power of the wire construction goes much beyond what we have used in sec . [ sec : reproducing the tenfold way ] to reproduce the classification of the sre phases . in this section we describe how to construct models for interacting phases of matter with intrinsic topological order and fractionalized excitations by relaxing the condition on the tunnelings between wires that they be of the one - body type . while these phases are more complex , the principles for constructing the models and proving the stability of edge modes remain the same : all allowed tunneling vectors have to obey the haldane criterion ( [ eq : haldane conditions ] ) and the respective symmetries . first , we review the models of quantum wires that are topologically equivalent to the laughlin state in the fqhe , @xcite following the construction in ref . for abelian fractional quantum hall states . here , we want to emphasize that the choice of scattering vectors is determined by the haldane criterion ( [ eq : haldane conditions ] ) and at the same time prepare the grounds for the construction of fractional topological insulators with trs in sec . [ subsec : symmetry class aii : fractional topological insulators ] . we want to construct the fermionic laughlin series of states indexed by the positive odd integer @xmath243 . @xcite ( by the same method , other fractional quantum hall phases from the abelian hierarchy could be constructed.@xcite ) the elementary degrees of freedom in each wire are spinless right- and left - moving fermions with the @xmath124-matrix [ eq : def array quantum wires tse a ] @xmath244 as is done in eq . ( [ eq : def h for class a a ] ) . reversal of time is defined through @xmath72 and @xmath84 given in eq ( [ eq : def h for class a b ] ) . instead of eq ( [ eq : def h of t for class a ] ) , the scattering vectors that describe the interactions between the wires are now defined by @xmath245 for any @xmath215 , where @xmath246 [ see table [ table : main table ] for an illustration of the scattering process ] . for any @xmath215 , the scattering ( tunneling ) vectors ( [ eq : def array quantum wires tse a b ] ) preserve the conservation of the total fermion number in that they obey eq . , and they encode a tunneling interaction of order @xmath247 , with @xmath62 defined in eq . . as a set , all tunneling interactions satisfy the haldane criterion ( [ eq : haldane conditions ] ) , for @xmath248 we note that the choice of tunneling vector in eq . is unique ( up to an integer multiplicative factor ) if one insists on charge conservation , compliance with the haldane criterion ( [ eq : haldane conditions ] ) , and only includes scattering between neighboring wires . the bare counting of tunneling vectors shows that the wire model gaps out all but two modes . however , we still have to convince ourself that the remaining two modes ( i ) live on the edge , ( ii ) can not be gapped out by other ( local ) scattering vectors and ( iii ) are made out of fractionalized quasiparticles . to address ( i ) and ( ii ) , we note that the remaining two modes can be gapped out by a unique ( up to an integer multiplicative factor ) charge - conserving scattering vector that satisfies the haldane criterion ( [ eq : haldane conditions ] ) with all existing scatterings , namely @xmath249 connecting the opposite ends of the array of wires through the tunneling @xmath6 is not an admissible perturbation , for it violates locality in the two - dimensional thermodynamic limit @xmath5 . had we chosen periodic boundary conditions corresponding to a cylinder geometry ( i.e. , a tube as in fig . [ fig : boundaryconditions ] ) by which the first and last wire are nearest neighbors , @xmath6 would be admissible . hence , the gapless nature of the remaining modes when open boundary conditions are chosen depends on the boundary conditions . these gapless modes have support near the boundary only and are topologically protected . applying the transformation with @xmath250 where @xmath251 transforms the @xmath124-matrix into @xmath252 as its determinant is not unity , the linear transformation ( [ eq : trasf for fields phi a b r ] ) changes the compactification radius of the new field @xmath253 relative to the compactification radius of the old field @xmath254 accordingly . finally , the transformed tunneling vectors are given by @xmath255 where @xmath256 and @xmath215 . in view of eqs . and , the remaining effective edge theory is described by @xmath257 this is a chiral theory at each edge that can not be gapped by local perturbations . equation is precisely the edge theory for anyons with statistical angle @xmath258 and charge @xmath259 , @xcite where @xmath153 is the charge of the original fermions . having understood how fractionalized quasiparticles emerge out of a wire construction , it is imperative to ask what other phases can be obtained when symmetries are imposed on the topologically ordered phase . such symmetry enriched topological phases have been classified by methods of group cohomology . @xcite here , we shall exemplify for the case of trs with @xmath176 how the wire construction can be used to build up an intuition for these phases and to study the stability of their edge theory . the elementary degrees of freedom in each wire are spin-@xmath19 right- and left - moving fermions with the @xmath124-matrix [ eq : def array quantum wires tse aii ] @xmath260 as is done in eq . ( [ eq : def h of v for class aii a ] ) . reversal of time is defined through @xmath72 and @xmath84 given in eq ( [ eq : def h of v for class aii b ] ) . instead of eq ( [ eq : def h of t for class aii d ] ) , the scattering vectors that describe the interactions between the wires are now defined by @xmath261 and @xmath262 for any @xmath215 , @xmath243 a positive odd integer , and @xmath263 . for any @xmath215 , the scattering ( tunneling ) vectors ( [ eq : def array quantum wires tse aii b ] ) preserve conservation of the total fermion number in that they obey eq . , and they encode a tunneling interaction of order @xmath247 with @xmath62 defined in eq . . they also satisfy the haldane criterion ( [ eq : haldane conditions ] ) as a set [ see table [ table : main table ] for an illustration of the scattering process ] . applying the transformation with @xmath264 to the bosonic fields , leaves the representation of time - reversal invariant @xmath265 while casting the theory in a new form with the transformed @xmath266-matrix given by @xmath267 and , for any @xmath215 , with the transformed pair of scattering vectors @xmath268 given by @xmath269 and @xmath270 when these scattering vectors have gapped out all modes in the bulk , the effective edge theory is described by @xmath271 this effective @xmath124-matrix describes a single kramers degenerate pair of @xmath258 anyons propagating along the first wire and another single kramers degenerate pair of @xmath258 anyons propagating along the last wire . their robustness to local perturbations is guaranteed by trs . unlike in the tenfold way , the correspondence between the bulk topological phase and the edge theories of lre phases is not one - to - one . for example , while a bulk topological lre phase supports fractionalized topological excitations in the bulk , its edge modes may be gapped out by symmetry - allowed perturbations . for the phases discussed in this section , namely the abelian and trs fractional topological insulators , it was shown in refs . and that the edge , consisting of kramers degenerate pairs of edge modes , supports at most one stable kramers degenerate pair of delocalized quasiparticles that are stable against disorder . ( note that this does not preclude the richer edge physics of non - abelian trs fractional topological insulators . @xcite ) we will now argue that the wire constructions with edge modes given by eq . exhaust all stable edge theories of abelian topological phases which are protected by trs with @xmath176 alone . let the single protected kramers degenerate pair be characterized by the linear combination of bosonic fields @xmath272 and its time - reversed partner @xmath273 where the tunneling vector @xmath58 was constructed from the microscopic information from the theory in ref . and @xmath274 is the @xmath124-matrix of a trs bulk chern - simons theory from the theory in ref . . [ in other words , the theory encoded by @xmath274 has nothing to do a priori with the array of quantum wires defined by eq . ( [ eq : def array quantum wires tse aii ] ) . ] the kramers degenerate pair of modes @xmath275 is stable against trs perturbations supported on a single edge if and only if @xmath276 is an odd number . here , @xmath277 is the charge vector with integer entries that determines the coupling of the different modes to the electromagnetic field . provided @xmath275 is stable , its equal - time commutation relations follow from eq . ( [ eq : def hat phi s b ] ) as @xmath278 = & \ , -\mathrm{i}\pi\ , \left ( \mathcal{t}^{\mathsf{t}}\ , \mathcal{k}'\ , \mathcal{t}\ , \mathrm{sgn}(x - x ' ) + \mathcal{t}^{\mathsf{t}}\ , \mathcal{l}\ , \mathcal{t } \right ) , \\ \left[\hat{\bar{\varphi}}(x),\hat{\bar{\varphi}}(x')\right ] = & \ , -\mathrm{i}\pi\ , \left ( -\mathcal{t}^{\mathsf{t}}\ , \mathcal{k}'\ , \mathcal{t}\ , \mathrm{sgn}(x - x ' ) + \overline{\mathcal{t}}^{\mathsf{t}}\ , \mathcal{l}\ , \overline{\mathcal{t } } \right),\end{aligned}\ ] ] where we used that @xmath274 anticommutes with @xmath279 according to eq . ( [ eq : trs h c ] ) . by the same token , we can show that the fields @xmath280 and @xmath281 commute , for @xmath282 we conclude that the effective edge theory for _ any _ abelian trs fractional topological insulator build from fermions has the effective form of one kramers degenerate pairs @xmath283 and is thus entirely defined by the single integer @xmath284 with the scattering vectors ( [ eq : def array quantum wires tse aii c ] ) we have given an explicit wire construction for each of these cases , thus exhausting all possible stable edge theories for abelian fractional topological insulators . for each positive odd integer @xmath243 , we can thus say that the fractionalized mode has a @xmath0 character : it can have either one or none stable kramers degenerate pair of @xmath243 quasiparticles . in sec . [ subsec : symmetry class aii : fractional topological insulators ] we have imposed trs on the wire construction of fractional quantum hall states and obtained the fractional topological insulator in symmetry class aii . in complete analogy , we can impose phs with @xmath219 on the wire construction of a fractional quantum hall state , thereby promoting it to symmetry class d. physically , there follows a model for a superconductor with fractionalized " majorana fermions or bogoliubov quasiparticles . lately , interest in this direction has been revived by the investigation of exotic quantum dimensions of twist defects embedded in an abelian fractional quantum hall liquid , @xcite along with heterostructures of superconductors combined with fractional quantum hall effect , @xcite or fractional topological insulators . @xcite furthermore , the kitaev quantum wire has been generalized to @xmath285 clock models hosting parafermionic edge modes , @xcite along with efforts to transcend the read - rezayi quantum hall state @xcite to spin liquids @xcite and superconductors , @xcite all of which exhibit parafermionic quasiparticles . as in the classification of non - interacting insulators , we treat the bogoliubov quasiparticles under bosonization as if they were dirac fermions . the fractional phase is driven by interactions among the bogoliubov quasiparticles . the elementary degrees of freedom in each wire are spinless right- and left - moving fermions and holes as was defined for symmetry class d in eqs . - . we construct the fractional topological insulator using the set of phs scattering vectors @xmath228 , for @xmath225 with @xmath228 as defined in eq . in each wire and the phs as defined in eq . . we complement them with the set of phs scattering vectors @xmath229 , for @xmath215 defined by @xmath286 with @xmath243 an odd positive integer . notice that @xmath287 so that we have to demand that @xmath288 has to comply with phs . thus , together the @xmath228 and @xmath229 gap out @xmath235 of the @xmath192 chiral modes in the wire . we can identify a unique ( up to an integer multiplicative factor ) scattering vector @xmath289 with @xmath243 the ame odd positive integer as in eq . ( [ eq : def overline t for class d j=1 ... n-1 ] ) that satisfies the haldane criterion with all @xmath228 and @xmath229 and thus can potentially gap out the 2 remaining modes . however , it is physically forbidden for it represents a non - local scattering from one edge to the other . we conclude that each boundary supports a single remaining chiral mode that is an eigenstate of phs . to understand the nature of the single remaining chiral mode on each boundary , we use the local linear transformation @xmath160 of the bosonic fields @xmath290 with determinant @xmath291 . when applied to the non - local scattering vector @xmath292 that connects the two remaining chiral edge modes , @xmath293 while the @xmath124 matrix changes under this transformation to @xmath294 noting that the representation of phs is unchanged @xmath295 we can interpret the remaining chiral edge mode as a phs superposition of a laughlin quasiparticle and a laughlin quasihole . it thus describes a fractional chiral edge mode on either side of the two - dimensional array of quantum wires . the definite chirality is an important difference to the case of the fractional @xmath0 topological insulator discussed in sec . [ subsec : symmetry class aii : fractional topological insulators ] . it guarantees that any integer number @xmath296 layers of this theory is stable , for no tunneling vector that acts locally on one edge can satisfy the haldane criterion ( [ eq : haldane conditions ] ) . for each @xmath243 , we can thus say that the parafermion mode has a @xmath1 character , as does the sre phase in symmetry class d. needed are the many - body tunneling matrices for class diii and c. we refer the reader to the appendices [ appendixsec : symmetry class c ] and [ appendixsec : symmetry class diii ] for their definitions . for class diii , the edge excitations ( and bulk quasiparticles ) of the phase are trs fractionalized bogoliubov quasiparticles that have also been discussed in one - dimensional realizations . ( in the latter context , these trs fractionalized bogoliubov quasiparticles are rather susceptible to perturbations . in this work , we have developed a wire construction to build models of short - range entangled and long - range entangled topological phases in two spatial dimensions , so as to yield immediate information about the topological stability of their edge modes . as such , we have promoted the periodic table of integer topological phases to its fractional counterpart . the following paradigms were applied . + ( 1 ) each luttinger liquid wire describes ( spinfull or spinless ) electrons . we rely on a bosonized description . + ( 2 ) back - scattering and short - range interactions within and between wires are added . modes are gapped out if these terms acquire a finite expectation value . + ( 3 ) a mutual compatibility condition , the haldane criterion , is imposed among the terms that acquire an expectation value . it is an incarnation of the statement that the operators have to commute if they are to be replaced simultaneously by their expectation values . + ( 4 ) a set of discrete and local symmetries are imposed on all terms in the hamiltonian . when modes become massive , they may not break these symmetries . + ( 5 ) we do not study the renormalization group flow of the interaction and back - scattering terms , but analyze the model in a strong - coupling limit . using this strategy , the following directions present themselves for future work . first , for symmetry class a , we have shown that sufficiently strong interactions among identical electrons can turn any topological phase with the same topological number controlling both the hall and thermal conductivities into a sre topological phase with independent quantized values of the hall and thermal conductivities . [ we only need to make the interaction encoded by eq . ( [ eq : definition of interactions that produce bosons out of fermions ] ) dominant . ] hence , it is natural to seek a putative breakdown of the topological counterpart to the wiedemann - franz law for metals in the symmetry class aii and for the lre phases in the symmetry classes a and aii . second , we can impose on our wire construction additional , albeit less generic , symmetries such as a non - local inversion symmetry or such as a residual @xmath106 spin symmetry . third , our construction can be extended to topological phases of systems that have bosons as their elementary degrees of freedom . for bosons , no analogue of the tenfold way exists to provide guidance . however , several works are dedicated to the classification of sre and lre phases of bosons , which might provide a helpful starting point . @xcite fourth , extensions to higher dimensions could be considered . @xcite this would , however , entail leaving the comfort zone of one - dimensional bosonization , with a necessary generalization of the haldane criterion in a layer construction . this work was supported by the european research council through the grant topolectrics , erc - stg - thomale-336012 , by darpa spawarsyscen pacific n66001 - 11 - 1 - 4110 ( t.n . ) , and by doe grant def-06er46316 ( c.c . ) . the conditions ( [ eq : trs h ] ) and ( [ eq : phs h ] ) for trs and phs can be derived by adapting the derivations @xmath297 and @xmath298 of eqs . ( [ eq : trs h d ] ) and ( [ eq : trs h e ] ) , respectively . class c is defined on line 9 of table [ table : main table ] by the operator @xmath66 for the phs obeying @xmath241 with neither trs nor chiral symmetry present ( as is implied by the entries 0 for @xmath7 and @xmath9 in table [ table : main table ] ) . in physical terms , class c describes a generic superconductor for which full spin @xmath212 symmetry is retained but trs is broken . the only difference to the case of class d considered in the main text is that the number of degrees of freedom is doubled . we postulate that under phs the following transformation rules hold @xmath299 for the creation operators of bogoliubov - degennes quasiparticles that are right ( r ) movers at the fermi energy and carry the spin quantum numbers @xmath300 . we apply the same transformation law to the creation operators of bogoliubov - degennes quasiparticles that are left ( l ) movers at the fermi energy and carry the spin quantum numbers @xmath300 . we consider identical wires with quasiparticles of type 1 and 2 , `` particles '' and `` holes '' , as well as left- and right - moving degrees of freedom . for any given wire with the basis ( @xmath301 , @xmath302 , @xmath303 , @xmath304 , @xmath305 , @xmath306 , @xmath307 , @xmath308 ) , the @xmath124-matrix reads @xmath309 where phs has the representation @xmath310 to complete the definition of an array of quantum wires realizing a sre phase in the symmetry class c , we specify the @xmath235 scattering vectors @xmath311 [ eq : tunneling vectors for c sre ] for @xmath225 , and @xmath312 . these scattering vectors gap out all modes in the bulk and comply both with phs and with the haldane criterion ( [ eq : haldane conditions ] ) . of the remaining four modes , two are localized at either edge of the system . the remaining two modes on either edge share the same chirality , for they could be gapped out by the non - local scattering vectors @xmath313 which act on modes with @xmath14 chirality on the left and and of @xmath13 chirality on the right edge only . we conclude that the pair of chiral modes on either edge is protected from backscattering . extending this construction to any integer number of layers yields the @xmath1 classification of class c. to complete the definition of an array of quantum wires realizing a lre phase in the symmetry class c , we use the @xmath143 scattering vectors @xmath314 and @xmath315 , @xmath316 defined in eq . and supplement them with the @xmath142 scattering vectors latexmath:[\[\begin{aligned } \mathcal{t}^{(j)}_{3,\mathrm{lre}}:=&\ , ( 0,0,0,0,0,0,0,0|\cdots| 0,-m^{\,}_{-},m^{\,}_{+},0,-m^{\,}_{+},0,0,m^{\,}_{-}|0,-m^{\,}_{+},m^{\,}_{-},0,-m^{\,}_{-},0,0,m^{\,}_{+ } \\ \mathcal{t}^{(j)}_{4,\mathrm{lre}}:=&\ , ( 0,0,0,0,0,0,0,0|\cdots| m^{\,}_{-},0,0,-m^{\,}_{+},0,m^{\,}_{+},-m^{\,}_{-},0|m^{\,}_{+},0,0,-m^{\,}_{-},0,m^{\,}_{-},-m^{\,}_{+},0 for @xmath134 , and @xmath263 as well as @xmath243 an odd positive integer . these tunneling vectors gap out all modes in the bulk and comply both with phs and with the haldane criterion ( [ eq : haldane conditions ] ) . one verifies that there exists a linear transformation with integer entires @xmath160 and @xmath318 such that @xmath319 the @xmath124-matrix transforms according to eq . , leaving the effective effective edge theory with two chiral modes of phs symmetric superpositions of laughlin quasiparticles with laughlin quasiholes on either edge of the system . as with the sre phase of symmetry class c , this is a completely chiral theory and no back - scattering mechanism can gap out modes by the haldane criterion ( [ eq : haldane conditions ] ) . extending this construction to any integer number of layers yields a @xmath1 classification of the lre phase in symmetry class c for every positive odd integer @xmath243 . class diii is defined on line 4 of table [ table : main table ] by the operator @xmath66 for the phs obeying @xmath219 and with the trs @xmath176 . in physical terms , class diii describes a generic superconductor for which full spin @xmath212 symmetry is broken but trs is retained . we use the same basis and @xmath124-matrix as specified for class c in eq . , namely for @xmath225 , and @xmath312 . these tunnelings gap out all the bulk modes . here , @xmath314 and @xmath315 as well as @xmath324 and @xmath325 are pairwise related by trs , while each of the tunneling vectors is in itself phs . the phases of the corresponding cosine terms in hamiltonian comply with both trs and phs as if @xmath326 on both wire @xmath327 and wire @xmath4 , there remains a single kramers degenerate pair of propagating modes . now , the tunneling vector @xmath328 acts locally on the left edge , satisfies the haldane criterion with all existing scattering vectors , and is unique up to an integer multiplicative factor . it might thus be concluded that the left pair of kramers degenerate edge states can be gapped by the tunneling @xmath329 . this is not so however . indeed , while @xmath329 itself is both compliant with phs and trs , its contribution to @xmath54 induces another expectation value that breaks trs spontaneously . to see this , we note that , by the particle - hole redundancy , the bosonic fields @xmath330 and @xmath331 as well as @xmath332 and @xmath333 have to be identified . thus , it is @xmath334 that acquires an expectation value . now , the term @xmath335 is more relevant from the point of view of the renormalization group than @xmath336 . if @xmath336 acquires an expectation value , so does @xmath335 . however , @xmath335 corresponds to @xmath337 ( and scattering vectors related by phs ) must then break trs spontaneously , for the resulting condition @xmath338 on the phase of its cosine term can not be met . if we rule out the spontaneous breaking of trs on the left edge , we must rule out the tunnelings @xmath339 for any integer @xmath149 . under this assumption , there remains a single gapless left pair of kramers degenerate edge states . we conclude that there is no possibility to localize the remaining edge modes with perturbations that comply with both trs and phs . had we considered two layers of this wire model , edge modes in both layers can be gapped out pairwise , similar to the case of class aii that we discussed in the main text . we conclude that the sre phase of symmetry class diii features a @xmath0 topological classification . to complete the definition of an array of quantum wires realizing a lre phase in the symmetry class d iii , we use the @xmath143 scattering vectors @xmath314 and @xmath315 , @xmath316 defined in eq . and supplement them with the @xmath142 scattering vectors latexmath:[\[\begin{aligned } \mathcal{t}^{(j)}_{3,\mathrm{lre}}:=&\ , ( 0,0,0,0,0,0,0,0|\cdots| -m^{\,}_{-},0,m^{\,}_{+},0,-m^{\,}_{+},0,m^{\,}_{-},0|-m^{\,}_{+},0,m^{\,}_{-},0,-m^{\,}_{-},0,m^{\,}_{+},0 for @xmath134 , and @xmath263 as well as @xmath243 an odd positive integer . these tunneling vectors gap out all modes in the bulk and comply both with phs and with the haldane criterion ( [ eq : haldane conditions ] ) . one verifies that there exists a linear transformation with integer entires @xmath160 and @xmath318 such that @xmath341 the @xmath124-matrix transforms according to eq . , leaving the effective effective edge theory with one kramers degenerate pair of phs symmetric superpositions of laughlin quasiparticles with laughlin quasi - holes on either edge of the system . as with the sre phase of symmetry class diii , this edge theory is protected by phs and trs . two copies of it , however , can be fully gapped out while preserving phs and trs . this yields a @xmath0 classification of the lre phase in symmetry class diii for every positive odd integer @xmath243 . a chiral symmetry is preseent if there exists a chiral operator @xmath342 that is antiunitary and commutes with the hamiltonian . the single - particle representation @xmath343 of @xmath342 is a unitary operator that anticommutes with the single - particle hamiltonian . in a basis in which @xmath343 is strictly block off diagonal , @xmath343 reverses the chirality . this chirality is unrelated to the direction of propagation of left and right movers which is also called chirality in this paper .

with ages of the order of @xmath2-@xmath3 gyr , globular clusters ( gcs ) are thought to be among the first stellar systems formed at early epochs in the local group . thanks to the general homogeneity in terms of age and chemical composition of their stars , gcs have been used for decades as ideal tracers of the chemistry of their environments , allowing us to reconstruct the chemical enrichment history and the age - metallicity relation of their host galaxies . the current picture of gc formation and evolution moves away from the traditional paradigm of gcs as simple stellar population ( see for instance the seminal paper by * ? ? ? * ) , according to which all the stars in a gc share the same initial chemical abundances for all the chemical elements . in fact , the recent discoveries of multiple sequences in gc colour - magnitude diagrams ( cmds , see e.g. @xcite ) and of star - to - star variations in the chemical abundances of some light elements ( like c , n , na , o , mg , al , see e.g. * ? ? ? * and references therein ) demonstrate that such systems are indeed much more complex . multiple populations ( mps ) are ubiquitous in all gcs studied so far , both in the galactic ( see for example @xcite ) and extra - galactic ( @xcite ) environment . since this chemical pattern has been observed in stars at all the evolutionary stages ( @xcite ) , it can not be explained in terms of internal mixing but it should have been imprinted in stars at formation . several scenarios interpret the photometric and spectroscopic evidence in terms of different generations of stars , with a first generation polluting the gas out of which second generation of stars formed with the products of their internal evolution . a number of candidate polluters have been proposed ( see @xcite for a comprehensive review ) ; namely asymptotic giant branch stars ( @xcite ) , fast - rotating massive stars ( @xcite ) , massive binaries ( @xcite ) and supermassive stars ( @xcite ) . alternative scenarios have been proposed as well , where low - mass stars accrete the polluted material during the pre - main sequence phase to give rise to a chemically peculiar population of stars that is coeval to that with pristine composition ( @xcite ) . the abundance spreads typical of mps involve only a few light elements . for most of the elements ( in particular @xmath1-elements like si , ca and ti , iron - peak and neutron - capture elements ) the stars in a typical gc exhibit a remarkable level of internal homogeneity , thus suggesting that the abundances of these elements in gcs can be used to trace the chemical composition of the gas from which the cluster formed . therefore , the investigation through high - resolution spectroscopy of the chemistry of poorly ( or not yet ) studied clusters is crucial to understand the chemical evolution of the galactic halo ( see e.g. * ? ? ? * ; * ? ? ? * ) and to identify , through the chemical tagging , clusters that likely originated in extragalactic environments @xcite . in this paper , we present the first detailed chemical study of ngc 6362 , an intermediate metallicity clusters which belongs to the low - mass tail of the gc mass distribution , with a mass of only @xmath4 @xcite . we list cluster fundamental properties in tab . [ tab0 ] , along with other useful informations . @xcite firstly measured its iron content ( [ fe / h]= 1.09 @xmath0 0.01 dex ) by analysing 160 giant stars observed with flames at the very large telescope . they also found a bimodal [ na / fe ] distribution , that makes ngc 6362 one of the least massive clusters where mps have been detected . the peculiar radial distribution of its stars , with both first and second populations being completely mixed out to several half light radii , has been interpreted by @xcite as the result of heavy mass - loss due to long - term dynamical evolution . such a claim is also supported by the quite shallow present - day mass function observed for this cluster in @xcite . thus , it would be extremely interesting to investigate whether it also shows other exceptional chemical features or it follows the general trends observed for standard gcs in the milky way with mps . to this end , we here present a detailed chemical analysis of 11 members in ngc 6362 . the paper is organized as follows . the dataset analysed in this work is described in section [ data ] , while the details of the chemical analysis are discussed in section [ analysis ] . we present the results of this study in section [ res ] and we finally discuss them in section [ discuss ] . the dataset analysed in this work was acquired under the programs 093.d-0618(a ) and 097.d-0325(a ) , pi : dalessandro , with the flames spectrograph ( @xcite ) at the eso very large telescope . we used the combined medusa+uves configuration , which allowed the simultaneous allocation of eight uves high - resolution fibres and 132 medusa mid - resolution fibres per exposure . while the stars observed in the medusa mode have been discussed in @xcite , in this work we focus on the eleven stars observed with the uves 580 red arm spectral configuration , with a resolution of r@xmath545000 in the spectral range @xmath6 . targets have been selected from the wide field imager ( wfi ) photometric catalogue presented in @xcite . for this study , only red giant branch ( rgb ) stars brighter than v@xmath7 mag have been targeted ( see figure [ cmd_targ ] ) . also , to avoid contamination from neighbours , only stars with no close ( i.e. within 2 ) sources brighter than v@xmath8 v@xmath9 have been selected . four exposures of 45 min for each target have been secured . two uves fibres have been used to sample the sky background , thus allowing a proper sky subtraction for each individual exposure . cc right ascension ( h : m : s ) & @xmath10:@xmath11:@xmath12 + declination ( @xmath13 : : ) & -@xmath14:@xmath15:@xmath16 + distance ( kpc ) & @xmath17 + r@xmath18 ( ) & @xmath19 + mass ( @xmath20 ) & @xmath21 + @xmath22fe / h@xmath23 $ ] ( dex ) & @xmath24 + [ tab0 ] to reduce the acquired data , we used the last version of the flames - uves common pipeline libraries based eso pipeline , which includes bias - subtraction , flat - field correction , wavelength calibration with a standard th - ar lamp , extraction of one - dimensional spectra and order merging . the accuracy of the dispersion solution has been checked by comparing the observed position of several sky emission lines with their rest - frame position as reported in the sky lines atlas by @xcite . no significant wavelength shifts have been found . once extracted , all the individual ( sky - subtracted ) exposures of each target have been brought to the same reference by correcting for the corresponding heliocentric radial velocity ( see sect.[vrad ] for the details ) , and finally combined together to obtain a median spectrum . the final reduced spectra have signal - to - noise ratio ( snr ) larger than @xmath530 at any wavelength . radial velocities ( v@xmath25 ) have been measured using the wrapper 4dao ( @xcite ) , which allows to run daospec @xcite for large sets of spectra , tuning automatically its main input parameters . for all the eleven targets , we treated the two uves chips of each single exposure independently . for the lower ( l ) chip , v@xmath25 have been computed by using @xmath26 absorption lines , while for the upper ( u ) chip the lines used were @xmath27 . for each target we obtained a remarkably good agreement between the measurements of the two independent chips , finding an average difference of only v@xmath28-v@xmath29 km@xmath30s@xmath31 ( @xmath32 km@xmath30s@xmath31 ) . moreover , we did not find any significant difference among the v@xmath25 measured for the four exposures of each target . this indicates that none of the observed targets are binary systems . we then computed the final values of v@xmath25 as the average value of the eight single exposure measurements ( two chips for each of the four exposures ) and we adopted the dispersion around the mean divided by @xmath33 as uncertainty . heliocentric velocities and related errors are listed in table [ tab1 ] . we measured an average v@xmath25 of v@xmath34 km@xmath30s@xmath31 ( @xmath35 km@xmath30s@xmath31 ) , which is in agreement with the value derived by @xcite ( v@xmath36 km@xmath30s@xmath31 , @xmath37 km@xmath30s@xmath31 ) . according to their v@xmath25 distribution all the targeted stars are members , having v@xmath25 values that lie well within 2@xmath38 from the systemic velocity . effective temperatures ( t@xmath39 ) and surface gravities ( @xmath40 @xmath41 ) for target stars have been derived from their the b and v magnitudes , in the same way as described in in @xcite . we correct magnitudes and colours for differential reddening using the procedure outlined in @xcite ( see also * ? ? ? differential reddening corrections across the whole wfi field of view range from @xmath42=-0.03 $ ] mag to + 0.03 mag around the adopted average colour excess e(b - v)@xmath43 mag ( @xcite ) . errors on the parameters affecting the determination of the absolute colour for the analysed targets , i.e photometric errors or uncertainty on the absolute and differential reddening ( @xmath44}}$ ] and @xmath45}}$ ] , respectively ) could potentially affect our t@xmath39 estimates . thus , to evaluate the uncertainties on @xmath46 we re - determined temperatures assuming typical errors on colors and extinction of @xmath47 mag , @xmath44}}=0.04 $ ] mag and @xmath45}}=0.02 $ ] mag , respectively . we underline that the quoted values are conservative upper limits , since the nominal photometric errors of such bright and well exposed stars are @xmath48 mag , while the errors on the absolute and differential reddening estimates correspond to about the 50% of their value ( see @xcite ) . finally , we measure a typical uncertainty on t@xmath39 of @xmath49 k. stars located in the brightest portion of the rgb , especially those approaching the rgb tip , can be significantly be affected by non - lte effects that spuriously decrease the iron abundances from fei lines leaving those from feii lines unaltered ( see @xcite ) . this causes the spectroscopic derivation of @xmath40 @xmath41 through ionization equilibrium ( i.e. @xmath40 @xmath41 is constrained by imposing that both neutral and ionized iron lines give the same abundance ) to be systematically biased towards lower gravities . thus , we prefer to rely on photometric gravities , derived by using the stefan - boltzmann relation . we adopted an absolute distance modulus of ( m - m)@xmath50 mag ( from @xcite ) , bolometric corrections from @xcite and a mass of @xmath51 . such a mass has been derived from the best fit isochrone taken from the basti dataset ( @xcite ) , with an age of 12 gyr , z=0.004 and @xmath1-enhanced chemical mixture ( corresponding to [ fe / h]@xmath52 dex ) . uncertainties in @xmath40 @xmath41 have been computed by taking into account the errors on t@xmath39 ( as described above ) , bolometric luminosity ( due to all the photometric uncertainties ) and mass ( we assumed @xmath53 , that corresponds to the range of masses that populate the entire rgb according to the best - fit isochrone ) . the final uncertainty on @xmath40 @xmath41 is about @xmath54 dex . finally , microturbulent velocities v@xmath55 have been derived spectroscopically , by requiring no trend between the measured iron abundances and the line strengths . typical uncertainties on this parameter are about @xmath56 km@xmath30s@xmath31 . the atmospheric parameters for each analysed target are shown in table [ tab1 ] . the adopted linelist has been compiled by selecting only transitions that are unblended at the temperatures , gravities , and metallicities of sampled stars . atomic data are from the lastest version of the kurucz - castelli database , improved for some specific cases with new and updated values . for the elements in our linelist with single and unblended lines we estimated the chemical abundances from the measured equivalent width ( ew ) , by using the package gala @xcite . we run gala keeping t@xmath39 and @xmath40 @xmath41 of the model fixed and allowing its metallicity to vary iteratively in order to match the iron abundance measured from ews . all the model atmospheres have been computed by means of the atlas9 code ( @xcite ) , while ews were measured by using daospec through the code 4dao ( see section [ vrad ] ) . ew uncertainties are estimated by daospec as the standard deviation of the local flux residuals ( see * ? ? ? all the lines with ew errors larger than 10% were excluded from the analysis . solar reference abundances are taken from @xcite . only for the lines with hyperfine structure and/or isotopic splitting ( cu , ba , nd , eu , la ) , abundances have been obtained by individually comparing the observed spectral lines with a grid of synthetic spectra computed with synthe ( @xcite ) , by running a @xmath57-minimization algorithm ( see the procedure described in * ? ? ? in particular , synthetic spectra computed by assuming for each star the appropriate atmospheric parameters derived as described in section [ atm ] are convolved at the uves resolution and finally re - sampled at the pixel size as the observed spectra . the total uncertainties on abundance measurements has been derived by considering two main sources of error : the internal error associated to the measurement procedure and the errors arising from the uncertainties in the atmospheric parameters . we defined as internal error the dispersion around the mean divided by the square root of the number of lines used to compute abundances . abundances of elements measured via spectral synthesis method ( namely cu , la and eu ) come from the measurement of only one line . in this case , the corresponding internal error has then been computed by means of monte carlo simulations . briefly , poissonian noise is added to the best - fit synthetic spectrum in order to reproduce the observed snr and then the procedure to derive the abundance is repeated . the dispersion of the abundances measured from 1000 monte carlo realizations has been adopted as the internal abundance uncertainty . depending on the snr around these lines ( ranging from @xmath58 to @xmath59 ) , the typical internal errors obtained in this way range from @xmath60 dex up to @xmath61 dex . in order to quantify the error coming from the uncertainties on the atmospheric parameters , we repeated the chemical analysis by varying each parameter for the corresponding uncertainty ( see section [ atm ] ) . the total uncertainty on the [ x / h ] abundances has been computed by summing in quadrature this contribution and the intrinsic error described above . as discussed by in @xcite this kind of uncertainty , when related to abundance ratios as [ x / y]=[x / h]-[y / h ] , partially cancel out because lines of the same ionization stage tend to react in a similar way to changes in the stellar parameters . therefore , we followed their prescription to calculate our final [ x / fe ] abundance uncertainties . in this section the results of the chemical analysis are described for each elemental group . in particular , elemental abundances as found for ngc 6362 ( always shown as a red empty star symbol in figures 3 - 8 ) are directly compared to abundances taken from the literature for other gcs ( values from @xcite are shown as green filled circles , while data from @xcite are plotted as blue filled circles , if not stated otherwise ) and galactic field stars ( grey dots , data from from @xcite ) . all the measured values and the corresponding uncertainties are listed in tables [ tab2 ] , [ tab3 ] and [ tab4 ] , while the average abundance ratios are listed in table [ tab5 ] . * _ iron abundances_. + by using the maximum likelihood ( ml ) algorithm described in @xcite , we found that the iron distribution for our spectroscopic sample is best described by a gaussian function with mean @xmath8[fei / h]@xmath62 and a null dispersion of @xmath63}}=0.00\pm0.01 $ ] . the derived fe abundance agrees well with the value provided by @xcite , i.e. [ fei / h]@xmath64 dex . a very similar value is obtained from the single ionized fe ii lines that provide an average abundance of @xmath8[feii / h]@xmath65 , with null dispersion . + * _ light elements : mg , al_. + as already demonstrated by @xcite for a larger sample of stars including the 11 targets studied here , ngc 6362 displays a broad [ na / fe ] distribution , with the presence of two ( equally populated ) stellar groups , peaked at [ na / fe]@xmath66 dex and [ na / fe]@xmath67 dex . another significant feature related to the existence of mps in the form of light element abundance spreads is the presence of star - to - star variations in mg and al abundances . this feature arises as an anti - correlation in a few clusters , whereas a large al variation is coupled with a small or null mg variation in most cases @xcite . + for the @xmath68 uves targets of ngc 6362 analysed in this work ( red stars ) , for 12 giants belonging to m 4 ( black triangles ) , and for @xmath10 galactic gcs ( grey dots , data from @xcite ) . typical errors are shown in the bottom - left region of the plot . ] + as shown in figure [ mgal ] , ngc 6362 does not display intrinsic variations in both mg and al . in fact ngc 6362 stars describe only a clump with very small dispersion compared to the whole extent of the anti - correlation covered by the uves targets analysed in @xcite in 17 gcs ( grey dots ) . according to a ml analysis , the mean abundances and intrinsic dispersions for these two elements are @xmath8[mg / fe]@xmath69 , @xmath70}}=0.00\pm0.01 $ ] and @xmath8[al / fe]@xmath71 , @xmath72}}=0.00\pm0.02 $ ] , thus confirming the lack for any abundance spread in both [ mg / fe ] and [ al / fe ] . + vs [ fe / h ] trends for ngc 6362 ( red star ) and m 4 ( black triangle ) , compared to a sample of field stars ( grey dots ) . _ lower panel : _ same comparison for [ al / fe ] . ] + in figure [ mgal_gc ] the average mg and al abundances of ngc 6362 stars are compared to those of galactic field stars . although ngc 6362 stars appear to be quite rich in both mg and al content with respect to stars at similar same metallicity ( upper and lower panel of figure [ mgal_gc ] , respectively ) , their mg and al abundances are still in agreement with the trends observed for field stars . + * _ @xmath1-elements : si , ca , ti_. + we adopt the @xmath1-elements produced by explosive nucleosynthesis ( si , ca and ti ) as tracers of the total @xmath1-element abundance for the cluster . mg ( that is produced by hydrostatic nucleosynthesis ) is excluded from this discussion because self - enrichment processes can , in principle , alter its initial abundance ( even if in ngc 6362 no evidence of a mg spread is found ) . + , [ ca / fe ] , [ ti / fe ] vs. [ fe / h ] trends for ngc 6362 , field stars ( same symbols of figure [ mgal_gc ] ) and galactic gcs ( green circles are taken from @xcite , blue circles from @xcite ) . _ lower panel : _ same comparison on the [ @xmath1/fe ] vs. [ fe / h ] trend . ngc 6362 is in very good agreement with what observed for the other gcs . ] + when analysed separately , si is the element with the highest average enhancement , having @xmath8[si / fe]@xmath73 , while ca and ti show lower values that are very similar each other , @xmath8[ca / fe]@xmath74 and @xmath8[ti / fe]@xmath75 , respectively . according to the ml analysis , none of the three elements show any hints of intrinsic dispersion . figure [ sicati ] shows the comparison among ngc 6362 , and the same objects as those shown in fig.[mgal_gc ] . also in terms of si , ca , and ti content , ngc 6362 does not behave differently from what commonly is observed for the other populations . + * _ iron - peak elements : sc , v , mn , cr , co , ni_. + we measured abundances for six iron - peak elements , namely sc , v , mn , cr , co and ni . + for cr and ni the abundances have been derived from the ew measurement , and both the abundances of these elements turn out to be scaled - solar , with average values of @xmath8[cr / fe]@xmath76 and @xmath8[ni / fe]@xmath77 . + for odd - z elements like sc and mn , we used the spectral synthesis method since their lines suffer for hyperfine splitting . in this case we found a super - solar [ sc / fe ] abundance ratio , with an average @xmath8[scii / fe]@xmath78 , while we found a significant deficiency in the mn content of the cluster ( @xmath8[mn / fe]@xmath79 ) . + finally , the presence of both isolated lines and lines split for hyperfine structure for v and co within the uves wavelength range allowed us to check for possible systematic effects arising from the different method of analysis used . for both the elements we found that the abundances measured with the ew method and those coming from the spectral synthesis agree very well , being coincident within a @xmath80 uncertainty . in particular we obtained slightly super - solar abundance ratios for both v ( @xmath8[v / fe]@xmath81 and @xmath8[v / fe]@xmath82 , respectively ) and co ( @xmath8[co / fe]@xmath83 and @xmath8[co / fe]@xmath84 ) . therefore we can safely claim that no systematic uncertainties are introduced by the use of two different abundance measurement methods . + as measured for ngc 6362 , m 4 and field stars . green circles indicate the [ mn / fe ] average abundances derived by @xcite for 21 galactic gcs . ] + the iron - peak elemental abundances for ngc 6362 are compared to the typical values found for field stars in fig [ ironpeaks ] . + * _ copper , s- and r- elements_. + the abundances for these elements have all been measured by means of spectral synthesis . + for cu , the only available transition is that at 5105 , since the other cu optical line ( at 5782 ) falls in the gap between the two uves chips . ngc 6362 behaves similarly to the other ( few ) gcs in the galaxy for which cu measurements exist ( see @xcite ) , with an average @xmath8[cu / fe]@xmath85 . this is also in good agreement with measurements for field stars at similar [ fe / h ] ( see the upper panel of figure [ copper ] ) . + we determined the abundance of the slow neutron - capture elements ba , nd and la . regarding the abundance of ba , we measured @xmath8[baii / fe]@xmath86 . by using only the la line at 6390.5 @xcite we derived @xmath8[laii / fe]@xmath87 , while for nd we found @xmath8[nd / fe]@xmath88 . + finally we estimated the abundance of the rapid neutron - capture element eu by using the transition at 6645.1 , finding @xmath8[eu / fe]@xmath89 . ngc 6362 matches well the typical behaviour of [ eu / fe ] measured for the other gcs in the galaxy ( see the bottom panel of figure [ copper ] , where data for eu abundances in 14 gcs have been taken from @xcite ) . + vs [ fe / h ] trends for a sample of galactic gcs ( green circles ) taken from @xcite , a sample of thick disk stars taken from @xcite and the cluster under analysis . _ central panel : _ observed ba trends for the same populations ( data for the 14 plotted gcs have been taken from @xcite ) . _ lower panel : _ same comparison but for eu abundances . ] in the previous paragraphs we compared the elemental abundances of ngc 6362 with those of field stars and other gcs available in the literature . however , such a comparison could be prone to systematic effects due for example to different used instruments or adopted linelists , atomic parameters , model atmospheres etc . to provide a further comparison free from all of these systematics , we decided to repeat our analysis on another gc with mass ( see @xcite ) and metallicity very similar to those of ngc 6362 : m 4 . this differential comparison between two clusters with similar properties and analysed with the same procedure will allow us to accurately highlight any intrinsic difference among their chemical composition . following the same approach described by @xcite , we thus re - analysed 12 rgb stars belonging to m 4 and observed with flames - uves red arm 580 ( i d 073.0211 , pi : carretta ) . b and v magnitudes for m 4 have been obtained from the analysis of a dataset of wfi2.2 m images ( program 68.d-0265(a ) , pi : ortolani ) , and have been corrected for differential reddening . the distance modulus ( @xmath90 mag ) and colour excess ( e(b - v)@xmath91 ) mag have been taken from @xcite . atmospheric parameters for the m4 targets have been obtained following the same approach used for the ngc 6362 stars . the results of such an investigation are shown in figures 3 - 8 , where abundance ratios for m 4 are always shown as black triangles . the average values are listed in tab.[tab5 ] , where the one - to - one comparison between the ngc 6362 and m 4 is displayed . as a first step , we checked that the abundance ratios we obtained as output of our analysis were in agreement with previous results available in the literature . the iron abundance we measured for m 4 is @xmath8[fe / h]@xmath92 . this value is in good agreement with previous metallicity determinations obtained from giant stars ( see @xcite . mg and al abundance ratios are shown in fig.[mgal ] as black triangles . like ngc 6362 stars , they describe a compact clump as well , with no dispersion in mg and only a small hint of dispersion in al , confirmed by the ml analysis which found @xmath72}}=0.05\pm0.03 $ ] ( a small al dispersion for the cluster has also been found by @xcite ) . their average values are shown in fig . [ mgal_gc ] , and agree well with the results in @xcite and @xcite , as @xmath1- and iron - peak elemental abundances do . it is worth noticing that for m 4 @xcite found @xmath8[cu / fe]@xmath93 dex by using both the cu lines previously described . these authors used the same solar values and oscillator strength for the line used by us , and they obtained agreement within @xmath56 dex between measurements coming from the two lines individually . therefore their estimate is compatible with ours to within @xmath94 . finally , our measurement of @xmath8[ba / fe]@xmath95 for m 4 is @xmath96 dex larger than that found in @xcite . however , these authors used a solar value a(ba)@xmath97 instead of our a(ba)@xmath98 . therefore such a discrepancy cancels out when the different adopted solar values are taken into account . at this point we performed the direct , one - to - one comparison between the chemistry of the two clusters , shown in fig . [ valmedi ] . in general , we found that the agreement between the elemental abundances of ngc 6362 and m 4 is remarkable . in fact , all the elemental abundance ratio match within a @xmath99 uncertainty , with the only exception of [ baii / fe ] and [ laii / fe ] , for which the agreement is only within a @xmath100 uncertainty . we measured abundance ratios for different elemental groups in the galactic gc ngc 6362 , which has never been investigated before using high - resolution spectroscopy , and we compared it with other gcs in the milky way . with the aim of anchoring our findings to a solid touchstone , we also repeated the same chemical analysis ( thus erasing any systematics ) on a sample of rgb stars in m 4 , a cluster that shares with ngc 6362 very similar mass and metallicity . [ valmedi ] and [ interq ] show the comparison of the measured abundance ratios in ngc 6362 ( red star symbols ) and m 4 ( black triangles ) , using the average values and the interquartile - range ( iqr ) , respectively . * _ light elements . _ + @xcite demonstrated that ngc 6362 hosts two distinct populations with different na abundances . while na star - to - star variations have been observed in most gcs , mg and al intrinsic spreads are not as universal . in particular , according to both observations ( see e.g. @xcite ) and theoretical predictions ( @xcite ) , the al production in gcs depends on two main parameters : ( @xmath101 ) metallicity , since a high metallicity causes lower al - yields from polluters ( ) and ( @xmath102 ) cluster mass , since ( under the hypothesis of the same initial mass function ) a more massive cluster would have more polluters and would retain more polluted gas within its potential well . however , several exceptions to these trends exist , such as 47 tucanae ( which shows a small al spread despite being quite massive , see @xcite ) or m71 ( which is an outlier in the carretta relation , see @xcite ) . therefore , the al abundances measured in this work for ngc 6362 and their comparison to m 4 , are of particular interest . + as already shown is sect . [ res ] , mg and al internal variations in ngc 6362 are compatible with being null ( see also fig.[interq ] ) , while m 4 stars show a possible small dispersion in al content . the corresponding iqrs are extremely small as well , being of few hundredths of dex for mg and only slightly larger for al ( iqr@xmath103 dex , iqr@xmath104 dex ) . therefore , we can conclude that in terms of light element abundance , ngc 6362 behaves as the other gcs with similar mass and metallicity , both according to the differential analysis with respect to m 4 and to the observational - based prediction of @xcite . * _ @xmath1- and iron - peak elements . _ + ngc 6362 is characterized by enhanced abundances for all the measured @xmath1-elements , with an average value of @xmath8[@xmath1/fe]@xmath105 . this value matches well those measured in m 4 and in the other galactic gcs ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? this finding confirms that ngc 6362 formed from gas enriched by core - collapse supernovae in a similar way to other galactic gcs ( excluding some gcs likely accreted by extragalactic galaxies and characterized by lower [ @xmath1/fe ] ratios ) . in particular , ngc 6362 and m 4 show very similar @xmath1 abundances , suggesting that they share a similar chemical enrichment by core - collapse supernovae . also for the iron - peak elements , the analysis shows that ngc 6362 follows the same abundance patterns of the other gcs and no remarkable ( element - by - element ) difference is found between ngc 6362 and m 4 . * _ s- and r - process elements _ + the cluster has a [ ba / fe ] abundance ratio compatible with those of gcs of similar metallicity , like m4 , ngc 288 and ngc 6171 , and , in general , agrees well with the run of [ s / fe ] abundance ratios with [ fe / h ] ( see fig.1 in * ? ? ? in fact , metal - poor gcs show roughly solar [ ba / fe ] ratios , as the production of ba at those metallicities is dominated by r - processes . on the other hand [ ba / eu ] increases with increasing the metallicity , because of the higher efficiency of the s - process . in the case of ngc 6362 , [ ba / eu ] reaches @xmath106 dex , compatible with the value measured in m 4 ( [ ba / eu]@xmath107 dex ) and the same is observed also for [ la / eu ] and [ nd / eu ] . even if ngc 6362 has a [ ba / fe ] compatible with that of m 4 at a level of 2@xmath38 ( at variance with the other abundance ratios that are compatible within 1@xmath38 ) , the two clusters show the same relative efficiency of s- to r - process therefore , the general enhancement of s - process element abundances measured in ngc 6362 suggests that the first generation of cluster stars @xmath101 ) formed from gas already enriched by low - mass ( @xmath108 ) agb stars , that are the main producers of s - process elements , the so - called main - component ( see e.g * ? ? ? * ; * ? ? ? * ) , and @xmath102 ) did not introduce any spread the s - process abundance of second generation stars . summing all up , the differential comparison between ngc 6362 and m 4 revealed that all the elemental abundances measured for the two clusters match within 1-@xmath38 ( with the marginal exception of [ ba / fe ] ) . none of the elements analysed in this work show any internal spread in ngc 6362 . it is worth noting that ngc 6362 and m 4 also display the same extent of na variation ( see * ? ? ? according to the compilation of gc masses in @xcite , and of gc metallicities in @xcite , the only clusters with properties similar to ngc 6362 and m 4 ( and not associated with the sagittarius dwarf spheroidal galaxy , see @xcite ) which have been chemically investigated in detail are ngc 288 and ngc 6171 . they both belong to the sample of clusters studied by @xcite , and as such they are plotted in fig.[sicati ] , in the same metallicity range as that of ngc6362 . their chemical composition is consistent with that of the two clusters analysed in this work . therefore , we conclude that ngc 6362 is a regular gc that shows the chemical composition representative of the milky way gcs with similar mass and metallicity , with the signatures of chemical enrichment by core - collapse supernovae and agb stars . we thank the anonymous referee for her / his comments and suggestions which improved the presentation of our results . dm and gf has been supported by the firb 2013 ( miur grant rbfr13j716 ) . sc acknowledges financial support from prin - inaf 2014 ( pi : s.cassisi ) . this research is part of the project cosmic - lab ( web site : http://www.cosmic-lab.eu ) funded by the european research council ( under contract erc-2010-adg-267675 ) . m.b . acknowledges financial support from prin miur 2010 - 2011 project `` the chemical and dynamical evolution of the milky way and local group galaxies '' , prot . 2010ly5n2 t . cccccccccc & & + 601063 & 262.8751296 & -67.0970861 & 15.023 & 13.789 & 4308 & 1.33 & 1.5 & -14.39 & 0.06 + 602339 & 262.8462133 & -67.0100153 & 14.919 & 13.674 & 4292 & 1.27 & 1.5 & -14.23 & 0.12 + 709358 & 263.0091350 & -67.0564120 & 14.744 & 13.393 & 4147 & 1.05 & 1.5 & -15.24 & 0.07 + 710376 & 262.9964883 & -67.0425252 & 15.099 & 13.916 & 4382 & 1.43 & 1.4 & -17.39 & 0.13 + 711565 & 262.9825550 & -67.0373852 & 15.058 & 13.845 & 4338 & 1.37 & 1.5 & -10.75 & 0.07 + 716150 & 262.9265237 & -67.0608503 & 15.086 & 13.902 & 4381 & 1.42 & 1.5 & -16.01 & 0.06 + 601269 & 262.8706968 & -67.1041887 & 14.706 & 13.355 & 4147 & 1.04 & 1.6 & -13.42 & 0.09 + 604027 & 262.7885538 & -67.1157141 & 14.848 & 13.579 & 4259 & 1.21 & 1.5 & -14.98 & 0.08 + 703323 & 263.1077119 & -67.0442269 & 14.658 & 13.282 & 4114 & 0.98 & 1.5 & -17.33 & 0.09 + 714494 & 262.9480400 & -67.0372456 & 14.644 & 13.296 & 4151 & 1.02 & 1.5 & -14.02 & 0.07 + 716237 & 262.9253519 & -67.0381319 & 14.610 & 13.185 & 4051 & 0.90 & 1.5 & -17.93 & 0.11 + . [ tab1 ] ccccccc & & + 601063 & @xmath109 & @xmath110 & @xmath111 & @xmath112 & @xmath113 & @xmath114 + 602339 & @xmath115 & @xmath116 & @xmath117 & @xmath118 & @xmath119 & @xmath120 + 709358 & @xmath121 & @xmath122 & @xmath123 & @xmath124 & @xmath125 & @xmath126 + 710376 & @xmath127 & @xmath128 & @xmath123 & @xmath129 & @xmath113 & @xmath130 + 711565 & @xmath131 & @xmath132 & @xmath117 & @xmath133 & @xmath134 & @xmath130 + 716150 & @xmath135 & @xmath136 & @xmath137 & @xmath138 & @xmath139 & @xmath140 + 601269 & @xmath141 & @xmath142 & @xmath143 & @xmath118 & @xmath144 & @xmath145 + 604027 & @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath151 + 703323 & @xmath152 & @xmath153 & @xmath154 & @xmath133 & @xmath119 & @xmath155 + 714494 & @xmath156 & @xmath110 & @xmath157 & @xmath112 & @xmath144 & @xmath158 + 716237 & @xmath159 & @xmath160 & @xmath161 & @xmath162 & @xmath163 & @xmath164 + [ tab2 ] cccccccc & & + 601063 & @xmath165 & @xmath166 & @xmath167 & @xmath168 & @xmath169 & @xmath170 & @xmath171 + 602339 & @xmath172 & @xmath173 & @xmath174 & @xmath175 & @xmath165 & @xmath176 & @xmath177 + 709358 & @xmath178 & @xmath179 & @xmath180 & @xmath181 & @xmath182 & @xmath183 & @xmath184 + 710376 & @xmath185 & @xmath186 & @xmath187 & @xmath188 & @xmath189 & @xmath190 & @xmath191 + 711565 & @xmath192 & @xmath193 & @xmath180 & @xmath194 & @xmath195 & @xmath196 & @xmath197 + 716150 & @xmath198 & @xmath199 & @xmath200 & @xmath201 & @xmath202 & @xmath203 & @xmath204 + 601269 & @xmath205 & @xmath206 & @xmath207 & @xmath208 & @xmath209 & @xmath210 & @xmath211 + 604027 & @xmath212 & @xmath199 & @xmath213 & @xmath194 & @xmath214 & @xmath215 & @xmath216 + 703323 & @xmath172 & @xmath217 & @xmath218 & @xmath194 & @xmath209 & @xmath219 & @xmath220 + 714494 & @xmath221 & @xmath222 & @xmath223 & @xmath201 & @xmath224 & @xmath225 & @xmath226 + 716237 & @xmath227 & @xmath228 & @xmath213 & @xmath168 & @xmath229 & @xmath219 & @xmath230 + [ tab3 ] ccccc 601063 & @xmath231 & @xmath232 & @xmath233 & @xmath117 + 602339 & @xmath234 & @xmath235 & @xmath236 & @xmath237 + 709358 & @xmath238 & @xmath239 & @xmath240 & @xmath241 + 710376 & @xmath242 & @xmath113 & @xmath243 & @xmath244 + 711565 & @xmath245 & @xmath246 & @xmath247 & @xmath248 + 716150 & @xmath249 & @xmath250 & @xmath139 & @xmath251 + 601269 & @xmath252 & @xmath253 & @xmath254 & @xmath255 + 604027 & @xmath256 & @xmath257 & @xmath258 & @xmath259 + 703323 & @xmath260 & @xmath250 & @xmath139 & @xmath261 + 714494 & @xmath262 & @xmath263 & @xmath119 & @xmath259 + 716237 & @xmath264 & @xmath233 & @xmath265 & @xmath266 + [ tab4 ] ccc @xmath8[fe / h]@xmath267 & @xmath268 & @xmath269 + @xmath8[mg / fe]@xmath267 & @xmath270 & @xmath271 + @xmath8[al / fe]@xmath267 & @xmath272 & @xmath273 + @xmath8[si / fe]@xmath267 & @xmath274 & @xmath271 + @xmath8[ca / fe]@xmath267 & @xmath275 & @xmath275 + @xmath8[sc / fe]@xmath267 & @xmath276 & @xmath277 + @xmath8[ti / fe]@xmath267 & @xmath278 & @xmath279 + @xmath8[v / fe]@xmath267 & @xmath280 & @xmath281 + @xmath8[cr / fe]@xmath267 & @xmath282 & @xmath283 + @xmath8[mn / fe]@xmath267 & @xmath284 & @xmath285 + @xmath8[co / fe]@xmath267 & @xmath286 & @xmath287 + @xmath8[ni / fe]@xmath267 & @xmath288 & @xmath289 + @xmath8[cu / fe]@xmath267 & @xmath290 & @xmath291 + @xmath8[ba / fe]@xmath267 & @xmath292 & @xmath293 + @xmath8[la / fe]@xmath267 & @xmath294 & @xmath272 + @xmath8[nd / fe]@xmath267 & @xmath295 & @xmath296 + @xmath8[eu / fe]@xmath267 & @xmath297 & @xmath298 + [ tab5 ]

duality between gravity and gauge theory in spacetime with different dimensionalities has been discovered by maldacena in 1998 @xcite . the type - iib string theory in ads@xmath0 is conjectured to be dual to a gauge theory in four dimensional minkowski spacetime ( @xmath1 ) at the boundary of the ads space . the correspondence can be used as a complementary method to study the strongly coupled gauge theory in four dimensional minkowski spacetime , a cousin of quantum chromodynamics , by avoiding the uncontrollable non - perturbative calculation via the application of weak - strong duality . we can deal with this problem by alternatively performing calculations in the tractable weakly interacting string theory in five ( plus five compact dimensions which provide details that are not relevant here ) dimensional anti de sitter space ( ads@xmath2 ) . the duality is extended to a finite temperature situation by adding a horizon in the radial coordinate @xcite . the string theory in an ads space with black hole horizon in the radial direction is proposed to be dual to a gauge theory at finite temperature . the duality is made quantitative in the sense that the hawking temperature in the bulk theory corresponds to the temperature of the gauge theory on the boundary . the ads / cft correspondence provides the first string - theoretic example of the underlying generic principle of the holographic duality ( i.e. the holographic principle ) . the idea of holographic duality was originally proposed by t hooft @xcite in a generic quantum gravity situation involving a gravitational horizon . the precise string theoretic version was given by susskind @xcite . when an object falls into a black hole , it will be stretched , torn apart into bits and eventually the bits will be smeared out over the horizon . consequently , all of the bulk information is spread over the horizonal surface resulting in an effective boundary description of the bulk theory . the bulk world is holographically encoded on the boundary . connection between ads space and holography was further clarified by witten @xcite after discovery of the ads / cft correspondence . given an ads space , the weakly - coupled bulk gravity theory corresponds to a strongly - coupled boundary gauge theory . adding a black hole to the ads space , the dual gauge theory on the boundary becomes thermal with the temperature equal to the corresponding hawking - page temperature of the background @xcite . it is thus interesting to investigate the intermediate situation where there exists a massive object before gravitational collapse into a black hole in the ads space and search for the dual description in the gauge theory side . it is argued in ref . @xcite that the degenerate fermions in the ads correspond to the composite multitrace operator constructed from product of single trace operators in the large central charge limit on the boundary . it is not unreasonable to think of this free " fermionic operator as the conformal cousin of a qcd nucleon such as neutron and proton . these free fermions " , however , still interact with each other by the colour - singlet interaction of order @xmath3 assuming negligible in the large @xmath4 limit . the colour - singlet ( glueball ) exchange on the boundary corresponds holographically to the gravitational interaction in the bulk . while gravity pulls the bulk mass together causing the gravitational collapse , the colour - singlet interaction should be responsible for the deconfinement phase transition of the injected mass in the dual picture . arguably , the gravitational collapse of the star in the ads would correspond to a thermalization process of the dual gauge matter on the boundary @xcite . consideration of the mass limit of the fermionic star in the ads bulk could reveal certain details of the pre - thermalization process in the dual gauge picture . the mass limit of the ads star corresponds to the minimum amount of injected mass required in order for the boundary gauge matter to start the thermalization process ( since the bulk gravitational collapse starts when injected mass exceeds the mass limit ) . specifically , it is also interesting to ask what the dual object of the bulk temperature is on the boundary world before black hole formation ? should it correspond to some parameter characterizing the superheated phase of gauge matter before the start of the thermalization ? moreover , what is the exact nature of the colour - singlet ( glueball exchange ) interaction responsible for the deconfinement of the dual gauge matter into the thermalized deconfined plasma ( which is the dual picture of the gravitational collapse caused by gravity ) ? the heavy - ion collision experiments at rhic and cern s lhc ( large hadron collider ) smash two charged ions at extreme energies , producing dense and hot nuclear matter with properties of the strongly coupled plasma . in the vicinity of the collision point , the induced magnetic field could be enormous @xcite . understanding the physics of dense hot nuclear plasma under such circumstances requires nonperturbative treatments of the strong interaction and the holographic method is one option . one holographic dual of the magnetized nuclear matter at finite temperature is proposed to be a magnetized black brane in the ads space @xcite . it was found that the entropy density of the magnetized brane in the ads obeys the third law of thermodynamics with entropy @xmath5 ( temperature ) for small temperature . in this article , we consider a fermionic star in the holographic ads@xmath2 background in the presence of external magnetic field at finite bulk temperature . the mass limit and other properties of the star is studied with respect to the changes in the magnetic field and bulk temperature . even though there is no complete understanding of the dual description in the gauge theory side of this situation , we argue certain aspects of the duality . in section ii , the tolman - oppenheimer - volkoff ( tov ) equation @xcite in the background ads@xmath2 is calculated starting from the general dimensionality . the energy levels of the bulk charged fermions in the presence of the magnetic field are calculated in the flat space approximation . the pressure and density of the bulk fermions at finite field and temperature are subsequently derived . section iii presents analytic and numerical results for each case of finite temperature and field . the mass limits depend crucially on the field and bulk temperature . the mass - radius relations for each case are discussed in section iv . the bulk adiabatic index and sound speed of the fermions inside the ads star are discussed in section v. the entropy density and total entropy in the bulk are also computed . section vi investigates the dependence of mass limit on the ads radius . section vii contains further discussions and summary of our results the study of the magnetized star in the ads space consists of two main calculations . first , the pressure and energy density need to be calculated for the system of charged fermions in the magnetic field at arbitrary temperature . the star will be assumed electrically neutral and we will focus only on the effect of magnetic field to the charged particles . at zero temperature , the energy states of the charged fermions in the magnetic field are separated naturally into landau levels . the partition function in the macrocanonical ensemble of these energy levels will provide the generic expression for the pressure and energy density of the fermionic system at finite temperature . the pressure and energy density are subsequently used in the equation of state required by the tov equation in the 5-dimensional ads spacetime . even though we will focus on interpreting the results of the bulk ads star in terms of the dual gauge theory , the calculations in the bulk picture are self - consistent and satisfactorily describe a real magnetized fermionic star in the 5-dimensional ads spacetime . in order to study the behaviour of a degenerate star in @xmath6-dimensional ads spacetime , we derive the spherical symmetric tov equation in @xmath6 dimensions as given in appendix [ appa ] . in the presence of external magnetic field , the pressure of the fermionic matter in the star is actually anisotropic due to the quantization of the energy levels . however , in the classical limit where the momentum in the direction of the magnetic field is much larger than the square root of the magnetic field , @xmath7 , the pressure becomes isotropic @xcite and the spherical symmetric tov equation is applicable . the resulting equations of motion describing the ads star in the spherical symmetric approximation are @xmath8 for the temperature @xmath9 , and @xmath10 where @xmath11 is the ads radius , @xmath12 is the area of @xmath13 and @xmath14 . to solve the equations of motion , we need the equation of state or the explicit expression of @xmath15 in terms of the chemical potential @xmath16 . standard evaluation of the partition function requires the layout of energy states of the fermionic system which can be obtained in the following subsection . we now solve the dirac equation to find the relativistic energy level of a charged fermion in the presence of external magnetic field in the 5 dimensional spacetime . as an approximation , we will ignore the effect of curvature on the energy levels of the fermions . the effects of gravity and the ads curvature will be considered only through the einstein equations stated in the previous subsection . starting from the dirac equation in flat space @xmath17 where @xmath18 is the mass of the fermion . the gamma matrices are chosen to be in the dirac representation as the following @xmath19 where @xmath20 and @xmath21 are @xmath22 identity matrix and pauli matrices respectively . we will consider only the positive energy solution since we are interested in the particle not the antiparticle . the positive energy solution @xmath23 satisfies the equation @xmath24 . let @xmath25 and consider a particle in an external magnetic field , the effect of the magnetic field can be taken into account by adding the field momentum , @xmath26 . we will choose the magnetic field to point in the @xmath27 direction and uniformly distributed over the entire @xmath28 space . the equation of motion of the fermion in 5 dimensional space becomes @xmath29 the momentum component in the extra dimension is represented by @xmath30 corresponding to the coordinate @xmath31 . we have assumed the solution in the form @xmath32 and neglect the effect of the ads curvature to the momentum component @xmath30 . this is a good approximation as long as the ads radius of curvature is large compared to the wavelength of the bulk fermions . the energy condition from the equation of motion is given by @xmath33 if we let @xmath34 , then we have @xmath35 from equation ( [ landau energy level 1 ] ) and ( [ landau energy level 2 ] ) , energy is quantized in the @xmath36 plane and contains certain degeneracy of states , _ i.e. _ , there are several states with the same one - particle energy . the number of states @xmath37 of a discrete energy level @xmath38 is @xmath39 where @xmath40(@xmath41 ) is a spin degeneracy independent of @xmath38 . the degeneracy is proportional to the field and vanishes for @xmath42 . the discrete energies from the degrees of freedom of the plane perpendicular to the magnetic field is called the landau levels , characterizing the statistical properties of the fermionic system . extension to finite temperature situation can be done by considering the corresponding partition function . thermodynamical pressure and energy density of the magnetized fermion gas can be calculated from the grand canonical partition function given by @xmath43 use the euler - maclaurin formula ( see appendix [ appb ] ) and certain tricks of integration , we finally have the pressure in the asymptotically approximated form @xmath44 - \left(\frac{2\pi^{2}m^{2}\mu_{b}^{2}b^{2}}{3h^{4}}\right)\bigg[\left(\mu - mc^{2}\right ) \nonumber \\ & & -~k_{b}t\int_{0}^{\frac{\mu - mc^{2}}{k_{b}t}}\frac{dy}{e^{y } + 1 } + k_{b}t\int_{0}^{\infty}\frac{dy}{e^{y } + 1}\bigg ] . \label{pressure in ads space under external magnetic field at finite temperature}\end{aligned}\ ] ] likewise the energy density is given by @xmath45 again , use the euler - maclaurin formula and tricks of integration , so that @xmath46 becomes @xmath47 + \left(\frac{2g_{s}\pi^{2}m^{2}\mu_{b}^{2}b^{2}}{3h^{4}}\right)\bigg [ ( \mu - mc^{2 } ) \nonumber \\ & & - k_{b}t\int_{0}^{\frac{\mu - mc^{2}}{k_{b}t } } \frac{dy}{\left(e^{y } + 1\right ) } + k_{b}t\int_{0}^{\infty } \frac{dy}{\left(e^{y } + 1\right ) } - \int_{mc^{2}}^{\mu}\frac{\epsilon e^{\frac{\epsilon-\mu}{k_{b}t}}}{k_{b}t}d\epsilon - 2\int_{0}^{\frac{\mu - mc^{2}}{k_{b}t}}\frac{\left ( \mu - k_{b}t y \right)e^{-y}}{\left ( e^{y } + 1\right)}dy \nonumber \\ & & + \int_{0}^{\frac{\mu - mc^{2}}{k_{b}t}}\frac{\left(\mu - k_{b}ty\right)e^{-y}}{\left(e^{y } + 1\right)^{2}}dy + \int_{0}^{\infty}\frac{\left(\mu + k_{b}ty\right)e^{y}}{\left(e^{y } + 1\right)^{2}}dy \bigg ] . \label{energy density in ads space under external magnetic field at finite temperature}\end{aligned}\ ] ] both expressions for the pressure and energy density are in the remarkable form with the dependence on @xmath48 separated out in simple quadratic functions . the integrations can be cast into logarithmic and polylogarithmic functions depending only on the temperature ( and not the field ) as are shown in the next section . in this section , the equations of motion , eqn . ( [ eom of m]),([eom of mu ] ) will be solved numerically . to emphasize effects of both temperature and external magnetic field , the physical properties of the degenerate star in the ads@xmath2 under the influence of both temperature and external magnetic field are investigated by dividing into 4 cases ; 1 . ) @xmath49 , 2 . ) @xmath50 , 3 . ) @xmath51 , and 4 . ) @xmath52 . before going into the details of each case , we integrate equations ( [ pressure in ads space under external magnetic field at finite temperature ] ) and ( [ energy density in ads space under external magnetic field at finite temperature ] ) to obtain @xmath53 , \label{pressure in simulation } \end{aligned}\ ] ] @xmath54 + \left(\frac{4m^{2}\pi^{2}\mu_{b}^{2}b^{2}}{3h^{4}}\right ) \nonumber \\ & & \bigg[\frac{mc^{2}}{1 + e^{\frac{\mu(r)-mc^{2}}{k_{b}t } } } - \mu(r ) - k_{b}t\ln\left(1 + e^{\frac{mc^{2}-\mu(r)}{k_{b}t}}\right ) + k_{b}t\ln\left(1 + e^{\frac{\mu(r)-mc^{2}}{k_{b}t}}\right ) \bigg ] , \label{energy density in simulation } \end{aligned}\ ] ] where @xmath55 is a polylogarithm function . for numerical analysis , we set @xmath56 , @xmath57 , @xmath58 . we can transform the numerical results to the si unit by using the table of dimensional translation given in appendix [ appc ] . the coupled equations of motion between mass and chemical potential ( eqn . ( [ eom of m ] ) , ( [ eom of mu ] ) ) are numerically solved to find the chemical potential and the accumulated mass within the star . the density and pressure profiles can be subsequently obtained . the boundary conditions at the center of star are chosen to be @xmath59 and @xmath60 for every case . this is the condition of degenerate star in ads@xmath2 considered in ref . the fermions degenerate into the lowest possible energy states filling the energy levels up until the fermi energy in 5 dimensions . in this limit , the pressure and the energy density , eqn . ( [ pressure in simulation ] ) , ( [ energy density in simulation ] ) , reduce to @xmath61 first , the surface of the star can be defined at the radial distance , @xmath62 , where the pressure becomes zero . apparently from eqn . ( [ pressure in simulation at t=0 , b=0 ] ) , the pressure is zero when @xmath63 . on the other hand , from eqn . ( [ energy density in simulation at t=0 , b=0 ] ) , the density vanishes when @xmath64 . therefore in this case , both the pressure and energy density become zero at the radius @xmath62 where @xmath65 . the accumulated mass , the chemical potential , the density and the pressure distribution of the star versus the radius are presented in figure [ t=0,b=0mmuvsr ] and [ t=0,b=0rhopvsr ] . relations between the total mass and the central chemical potential / density of the degenerate star are shown in figure [ t=0,b=0masslimitmu ] . + + + from the numerical solution , the edge of the degenerate star is at @xmath66 where the pressure drops to zero . in figure [ t=0,b=0mmuvsr](a ) , the accumulated mass grows rapidly , in particular for the interval between @xmath67 and @xmath68 . beyond the central region , the accumulated mass increases less rapidly and becomes steady . the behavior of the accumulated mass is determined by the density and the pressure distribution within the star . initially , both the energy density and pressure in figure [ t=0,b=0rhopvsr ] , decrease rapidly then they drop to zero more gradually at larger distance . the chemical potential also behaves similarly(figure [ t=0,b=0mmuvsr](b ) ) . it is clear that the matter in the star becomes extremely dense in the region near the core . figure [ t=0,b=0masslimitmu ] shows the mass curve of the degenerate star as a function of the central chemical potential and density . from numerical analysis , the maximum mass is found to be @xmath69 for the central chemical potential equal to @xmath70 or at the central energy density equal to @xmath71 . this maximal mass can be interpreted to be the mass limit above which gravitational collapse occurs . a mass injection into an empty ads space until the accumulated mass exceeds the mass limit would result in a gravitational collapse in the bulk . the collapse corresponds to a thermalization process to finite temperature of the dual gauge matter . therefore , the mass limit corresponds to the minimum injected mass required by the dual gauge matter to start the thermalization process into the thermal equilibrium . after deconfinement thermalization , the dual gauge matter is in thermal equilibrium at the hawking temperature at this mass limit , i.e. @xmath72 with @xcite @xmath73 where the horizon radius @xmath74 for ads@xmath2 . note that the mass dependence of the hawking - page temperature in the limit of large ( @xmath75 ) and small ( @xmath76 ) black hole in the ads is @xmath77 respectively . it is interesting to note that for @xmath78 ( small black hole with negative specific heat after the gravitational collapse ) , the higher the mass limit , the smaller temperature the dual gauge matter would thermalize to . this corresponds to @xmath79(for @xmath80 , approximately 0.8836 ) . the mass limit of our ads star for @xmath81 is roughly @xmath82 and therefore the black hole at the end of gravitational collapse for ads star at this mass limit is a small black hole with small negative specific heat . for this case , the magnetic field is turned on and the mass limit and other properties at zero temperature are studied by comparing to the results of case i. since the changes from case i is small , we will present the results using the numerical differences between the two cases . starting from the pressure and energy density for nonzero magnetic field @xmath83 observe that the pressure of the star has almost the same form as the pressure in case i. the correction term to the pressure from the magnetic field contains the factor @xmath84 . the density appears to be smaller due to the contribution from the term @xmath85 . since the pressure vanishes at @xmath86 as in case i , the surface of the star is defined in the similar way , at @xmath65 . interestingly at this radius , the density becomes negative @xmath87 due to the interaction energy between the fermion s magnetic moment and the external field . interestingly , there is a critical field strength where the density becomes zero , @xmath88 where @xmath89 is a rescaled chemical potential . for magnetic field stronger than this critical value , the energy density becomes negative and there is no star formation or black hole in the bulk . since there is no horizon in the bulk , the dual gauge matter is at zero temperature under extremely strong magnetic field . likewise , there is a critical field where the pressure becomes zero , @xmath90 for @xmath91 , @xmath92 is always smaller than @xmath93 , therefore the pressure becomes negative before the density as the field is increased . at @xmath94 , both @xmath93 and @xmath92 are zero . + for numerical study , the magnetic field strength is chosen to be @xmath95 and @xmath96 for our consideration . fig . [ mixedatt=0masslimitmu ] show that the mass limit , comparing to case i , decreases when the magnetic field increases . the maximum mass for @xmath97 is appreciably smaller than the maximum mass at @xmath98 . consider the equation of state in the energy density part ( eqn . ( [ energy density in simulation at t=0 ] ) ) . since the coupled equations of motion between mass and chemical potential of the star ( eqn . ( [ eom of m ] ) and ( [ eom of mu ] ) ) involve the energy density , decreasing the energy density leads to the decrease of mass and the chemical potential of the star comparing to case i. the increase of the chemical potential subsequently leads to the decrease in the pressure of the star . numerical analysis confirms these behaviour as are shown in fig . [ otherpropertiesatt=0mvsr ] . note that in the core region ( @xmath99 ) , the density increases due to the increase of the chemical potential . however , in the outer region of the star , the effect of the magnetic field becomes dominant resulting in the decrease of the density . accumulated mass eventually becomes smaller than the mass in case i. the maximal mass or the mass limit of the ads star when the magnetic field is turned on is smaller than the mass limit in case i. therefore the dual gauge matter under magnetic field thermalizes to larger temperature when the accumulated mass exceeds the mass limit even though it requires smaller injected mass in order to start the thermalization . gravitational collapse of an ads star under strong magnetic field corresponds to thermalization of the magnetized gauge matter from zero to finite temperature . remarkably , the thermalized temperature ( at the mass limit ) is _ larger _ than when the field is absent previously discussed in case i. the magnetized gauge matter thermalizes more easily by requiring smaller injected mass , and also becomes hotter after the deconfinement thermalization . for finite bulk temperature , the bulk fermions become thermal in the ads space . since the kinetic energy of the particles increases , the pressure becomes larger and the star grows bigger . again , we study the small changes in the mass limit and other properties of the star by comparing the results to the zero temperature case . the pressure and energy density , eqn . ( [ pressure in simulation]),([energy density in simulation ] ) in this case reduce to @xmath100 , \label{pressure in simulation at b=0 } \\ \rho c^{2 } & = & \left(\frac{g_{s}2\pi^{2}}{15c^{4}h^{4}}\right)\bigg[3\mu(r)^{5 } - 5m^{2}c^{4}\mu(r)^{3 } + 2m^{5}c^{10 } - 5k_{b}^{2}t^{2}m^{2}c^{4}\pi^{2}\mu(r ) + 7k_{b}^{4}t^{4}\pi^{4}\mu(r ) \nonumber \\ & & + ~10k_{b}^{2}t^{2}\pi^{2}\mu(r)^{3 } + 30k_{b}^{2}t^{2}m^{3}c^{6}li_{2}\left(-e^{\frac{mc^{2}-\mu(r)}{k_{b}t}}\right ) - 150k_{b}^{3}t^{3}m^{2}c^{4}li_{3}\left(-e^{\frac{mc^{2}-\mu(r)}{k_{b}t}}\right ) \nonumber \\ & & + ~360k_{b}^{4}t^{4}mc^{2}li_{4}\left(-e^{\frac{mc^{2}-\mu(r)}{k_{b}t}}\right ) - 360k_{b}^{5}t^{5}li_{5}\left(-e^{\frac{mc^{2}-\mu(r)}{k_{b}t}}\right)\bigg ] . \label{energy density in simulation at b=0}\end{aligned}\ ] ] it is interesting to investigate the large temperature limit , @xmath101 . in this limit , the polylogarithmic function becomes a zeta function @xmath102 and thus @xmath103 if we assume the star to be in a uniform temperature , this implies that the thermal fermions are not confined within a finite - size star when the temperature is sufficiently large , i.e. @xmath101 . the result is not surprising , any particles with sufficiently large kinetic energy will escape the gravitational influence of the star . we set temperature values in the simulation unit to be @xmath104 . figure [ mixedatb=0masslimitmu ] show that temperature increasing hardly affects the mass limit . for this case , the surface of star is defined at @xmath105 since the density and pressure do not necessarily reduce to zero . the maximum masses increase with the bulk temperature . this is because the small increase in the temperature affects the fermi - dirac distribution very slightly . most particles are still in the same quantum states , mostly degenerate , and a very small part of the particles occupy higher energies than the fermi energy and exert more pressure . increasing temperature thus results in a small increase of pressure and energy density . consequently , when temperature increases , the maximum mass also grows . for @xmath106 , the energy density and chemical potential reduce to zero at much larger radii as shown in fig . [ otherpropertiesatb=0rhovsr ] . for sufficiently large temperature , even though the chemical potential reduces to @xmath107 at smaller radii , the pressure does not reduce to zero . in other words , the thermal bulk fermions refuse to be confined within a finite - size star above a critical temperature . + to interpret the results in the dual gauge picture , caution has to be made regarding the bulk temperature . during the thermalization process corresponding to the gravitational collapse in the gravity picture , the gauge matter is not in thermal equilibrium until a black hole is formed when the mass injection exceeds the mass limit . a zero - bulk - temperature ads star collapsing into a black hole becomes thermal at nonzero hawking temperature due to the emergence of a horizon . therefore , the bulk temperature does not correspond to any sort of temperature of the gauge matter on the boundary world . one of the effects of the bulk temperature of the fermions in the ads star is the increase of mass limit . once a black hole is formed from gravitational collapse of the warm ads star , the corresponding hawking temperature is always smaller than the the zero bulk temperature case . after thermalization process , the dual gauge matter will be in thermal equilibrium at _ lower _ temperature than the case of zero bulk temperature collapse . however , the total injected energy is larger than the zero bulk temperature case . the bulk temperature thus serves as a parameter which delays the onset of the thermalization process as well as reducing the temperature of the resulting thermal equilibrium . certainly , the dual gauge matter at exactly the same temperature can be alternatively achieved by injecting mass into a black hole in ads space , increasing its mass and reducing the corresponding hawking temperature ( however , if we keep increasing the black hole mass , it will finally become large black hole with positive specific heat and the temperature will start to increase with the mass ) . this choice would correspond to in - equilibrium thermalization where the gauge matter is always kept at thermal equilibrium as temperature decreases . the final thermal equilibrium at certain temperature can always be achieved by infinitely many different thermalization processes . we now consider effects from both the finite bulk temperature and nonzero magnetic field to the mass limit and other properties of the star . the equations of state have the full form according to eqn . ( [ pressure in simulation ] ) and ( [ energy density in simulation ] ) . and @xmath108 in comparison to the mass limit curve at @xmath109.,scaledwidth=60.0% ] again , it is interesting to consider the extreme limit of large temperature in the presence of the magnetic field . for nonzero field , the pressure in this limit becomes @xmath110 provided that the field @xmath48 is also comparably large . from eqn . ( [ largebtp ] ) , the star will have definite surface at finite radius when @xmath111 . sufficiently hot star requires sufficiently strong field to confine its fermionic content . we see the similar behaviour as in case ii and iii , temperature increase leads to the increase of the mass limit whereas the effect of the magnetic field is the opposite . in fig . [ mixed ] , when we set the field @xmath112 , the temperature @xmath113 has stronger effect on the profile of the star . the mass limit becomes larger than the mass limit in the case of the zero temperature and magnetic field . similar to case iii , when the temperature increases , the mass limit grows larger ( the upper line in the fig . [ mixed ] ) . however , if the magnetic field is enhanced further to @xmath114 , the mass limit becomes smaller than the zero - field zero - temperature mass limit . namely , the influence of the magnetic field has overcome those of the temperature when it is sufficiently large . let us summarize implications for the thermalization of the dual gauge matter from the results in this mixed situation with @xmath115 . generically , turning on the bulk temperature results in a larger mass limit in the ads space while finite magnetic field leads to a smaller mass limit . if the injected mass exceeds the mass limit , gravitational collapse will occur and we end up with a black hole . the injected mass at the mass limit is also the minimum mass required for the dual gauge matter to start the thermalization . the hawking temperature of the black hole can be identified with the temperature of the dual gauge matter at thermal equilibrium after the non - equilibrium thermalization process corresponding to the collapse , it is larger ( smaller ) for finite field ( bulk temperature ) than the collapse with @xmath116 . the field and the bulk temperature compete with opposite effects . for zero - field finite temperature collapse , the final black hole has higher mass and thus corresponds to small temperature of the gauge matter . the final equilibrium at the same temperature can be achieved via in - equilibrium process by injecting mass into a black hole resulted from gravitational collapse of an ads star with @xmath116 ( case i ) . on the contrary , when the field is turned on , we need to extract mass from a magnetized black hole , reducing its mass and increasing its hawking temperature in order to achieve the thermal equilibrium at the same temperature and magnetic field . the black hole immersed in the constant magnetic field in 4 dimensions was originally investigated in ref . extension to the magnetized black hole in ads@xmath2 spacetime is required to fully understand the holographic description of the strongly coupled magnetized gauge matter , one such solution ( magnetic brane ) is discussed in ref . it is found that the entropy density of the black brane in ads@xmath2 is proportional to @xmath117 for small @xmath117 and has a @xmath118 dependence for higher temperatures . we will calculate the entropy density of the ads star and compare to the case of magnetic brane in section v. however , as stated above , we have assumed the field is not sufficiently strong that it affects the spacetime of the background and our analyses are thus limited to the moderate magnetic field situation . the mass sequences diagram of the ads star for each case can be presented by the mass - radius plot of the star as shown in fig . [ mrrelation ] . figure [ mrrelation](a ) is the mass - radius sequence for case i with zero temperature and zero magnetic field . the stars in this case have larger radius than case ii ( zero temperature , finite field ) in fig . [ mrrelation](b ) but smaller radius than case iii ( zero field , finite temperature ) in fig . [ mrrelation](c ) . interesting competition between temperature and magnetic field can be seen in fig . [ mrrelation](d ) , a sufficiently large field helps to confine the fermions within a finite - size star even for relatively higher temperatures comparing to case iii . for sufficiently high temperature , the mass - radius curve can change the way it spirals to the attractor fixed point at @xmath119 . for @xmath120 in fig . [ mrrelation](c ) , the curve with @xmath121 oscillates down " to the fixed point from the small radii instead of the typical anticlockwise spiralling . this is because at this temperature the radius of the star is an increasing function of @xmath122 with no oscillation as we can see from fig . [ mixedatb=0masslimitmu](b ) . for @xmath98 in fig . [ mrrelation](d ) , the curve with @xmath121 , oscillates down " to the fixed point from the large radii without spiralling . it should be remarked that for case i and iii ( zero field ) , the mass at the attractor fixed point for @xmath119 is around 0.7 . for case ii and iv at @xmath98 , the fixed point mass for @xmath119 is around 0.68 . the radius of the ads star at the fixed point decreases with the field but does not depend very sensitively on the temperature . many interesting physical properties of the fermions squeezed within the ads star by its own gravity can be illustrated by certain thermodynamic and transport quantities . in this section , we consider two transport coefficients , the adiabatic index and sound speed of the ads fermionic matter for each limiting case . the entropy density and total entropy of the ads star are discussed subsequently . generically the adiabatic index , @xmath123 , and the sound speed , @xmath124 , of a medium are defined as @xmath125 which can be calculated through the dependence on the chemical potential @xmath126 of both @xmath127 and @xmath128 . the general expressions for both quantities are very lengthy but they are simplified for the zero - temperature limit . @xmath129 , @xmath130 where @xmath89 is the rescaled chemical potential and @xmath131 is the rescaled magnetic energy of the fermions . @xmath132 , @xmath133 a number of remarks are in order for the zero - temperature limit . from eqn . ( [ cs ] ) and ( [ csb0 ] ) , the sound speed for the nonzero field case ( @xmath134 ) is shown to be larger than the case with @xmath120 . for @xmath120 since @xmath135 , the sound speed is always real and the upper limit of @xmath124 is always smaller than @xmath136 or half the speed of light . for nonzero field , reality condition of @xmath124 leads to the constraint @xmath137 . namely , for a given @xmath138 , the upper limit on the magnetic field for ordinarily - compressible fermionic matter is @xmath139 on the other hand , the upper limit from the light speed @xmath140 is satisfied trivially for any value of @xmath48 . numerical results for each case are presented in fig . the @xmath141 and @xmath142 label represents the curve with @xmath143 and @xmath144 respectively . the @xmath115 label represents the curve with @xmath145 . when a thermodynamical system is injected with energy until it reaches a thermal equilibrium , the total energy density , pressure and number density are related to the entropy density by the relation @xmath146 where the entropy density in our case can be computed via @xmath147 from the gibbs - duhem relation . using eqn . ( [ pressure in simulation ] ) , the entropy density of the fermionic content of the ads star at finite temperature can be calculated to be @xmath148 . \label{sd}\end{aligned}\ ] ] the entropy density of the fermion gas approaches zero as @xmath149 , a typical behaviour from a quantum ensemble satisfying the third law of thermodynamics . in the low temperature limit , the last two terms of eqn . ( [ sd ] ) remain dominant and thus @xmath150.\end{aligned}\ ] ] it is interesting to compare the @xmath117-dependence of our entropy density with the magnetized black hole studied in ref . @xcite where @xmath151 for small temperatures and @xmath152 for larger temperatures . in our case of the fermions in the ads star , the origin of the temperature dependence is the typical behaviour of free relativistic fermi gas persisting in any dimensions . for the magnetized ads black hole , the entropy is determined from the central charge of the ads@xmath153 subspace of ads@xmath154 interpolating with the ads@xmath2 . however , it must be aware that the bulk temperature of the ads star and the hawking temperature of the black hole are two distinct kinds of temperature . only the latter corresponds to the temperature of dual gauge matter at a thermal equilibrium . the entropy density of the magnetized fermion gas given by eqn . ( [ sd ] ) also depends on the magnetic field @xmath155 . the dependence nevertheless vanishes in the @xmath149 limit . however , this formula is the result of the euler - maclaurin formula which is a good approximation for @xmath156 , i.e. sufficiently high temperature . for smaller temperatures , starting with eqn . ( [ parf ] ) , the zeroth mode becomes dominant and the field - dependence becomes @xmath157 . this is also similar to the behaviour of the magnetized black brane @xcite . the entropy density from two numerical solutions are shown in fig . [ figsd ] . since @xmath158 is an increasing function of @xmath117 , the star at relatively small temperatures will have smaller entropy density . for small temperatures or small entropy density , we can approximate @xmath159 leading to @xmath160 . label represents @xmath145 curve and the @xmath161 label represents @xmath162 curve respectively.,scaledwidth=60.0% ] and @xmath163.,title="fig:",scaledwidth=45.0% ] and @xmath163.,title="fig:",scaledwidth=45.0% ] ( from left to right ) and @xmath163 ( from right to left).,title="fig:",scaledwidth=45.0% ] ( from left to right ) and @xmath163 ( from right to left).,title="fig:",scaledwidth=45.0% ] next we calculate the total entropy of the ads star which should be equivalent to the entropy of the dual gauge matter before the thermalization . the total entropy of the star should be the lower bound of the total entropy of the black hole at the end of gravitational collapse when the mass of the ads star exceeds the mass limit . this black hole entropy in turn corresponds to the total entropy of the dual gauge matter at the end of thermalization . in @xmath6 dimensions , the total entropy is given by @xmath164 where the volume factor @xmath165 becomes @xmath166 for @xmath167 . figure [ figs ] shows the total entropy of the ads star for @xmath168 at temperature @xmath169 and for @xmath170 under field @xmath171 . the total entropy is an increasing function of the temperature and a decreasing function of the magnetic field . from small radii , the total entropy is an increasing function of the star radius . this is a similar behaviour to the accumulated mass which is also a global quantity . remarkably , the total entropy converges to zero in the attractor fixed point @xmath119 limit . as the central density grows , the content of the ads star concentrates more in the central region resulting in the decrease of total entropy towards zero ( the volume weighing factor @xmath172 enhances contribution in the outer region in contrast to the core ) . the black hole at the end of gravitational collapse should possess at least the same amount of total entropy as the initial ads star above the mass limit . the second law of thermodynamic demands that the entropy of the ads star above the mass limit is always less than the black hole entropy after the collapse @xcite . the entropy increase could continue until it reaches the maximum when a black hole is formed @xcite . unfortunately , the time evolution of the entropy during the gravitational collapse is not completely known . partially because the thermal entropy is ill - defined during off - equilibrium processes and partially due to the geometric nature of black hole entropy at the end of the collapse . there are other kinds of entropy that can be assigned to the ads star and the black hole . the entanglement entropy quantifies how much we do not know about the region behind the horizon and it is consistent with the geometric nature of the bekenstein - hawking entropy of the black hole . entanglement entropy is found to increase in a different manner from the kolmogorov - sinai entropy @xcite during the collapse . however , all kinds of entropy are found to increase approximately linearly during the initial state of the gravitational collapse and saturate to a constant value at the end . each maximum of the @xmath173-@xmath174 curve in fig . [ figs1 ] corresponds holographically to the entropy of the dual gauge matter at the beginning of the thermalization process . it is also proportional to entropy of the black hole after gravitational collapse assuming the linear progression to the saturated bekenstein - hawking entropy mentioned above . arguably , the increase of entropy of the dual gauge matter from the injected mass state to the thermal equilibrium should also be the linear progression following by saturation as well . note that the entropy is not maximal at the maximal mass nor the maximal radius as we can see from fig . [ figs ] , [ figs1 ] . we vary the curvature radius of the ads space , @xmath175 , and study the changes in the profile of the star in this section . for simplicity , we will set the temperature and the external magnetic field to be zero . we let the curvature radius to be @xmath176 and @xmath177 , and observe considerable changes in the mass limit of the star as are shown in fig . [ mixed3 ] . .,scaledwidth=50.0% ] the mass limit of the degenerate star increases evidently when we raise the curvature radius of the ads space . moreover , the peak of the mass limit curve shifts to the lower central density side . note that increasing @xmath175 corresponds to decreasing the bulk cosmological constant @xmath178 . for @xmath179 , the maximum mass is @xmath180 for the central chemical potential @xmath181 or the central energy density @xmath182 . for @xmath183 , the maximum mass is @xmath184 for the central chemical potential @xmath185 or the central energy density @xmath186 . for @xmath187 , the maximum mass is @xmath188 for the central chemical potential @xmath189 or the central energy density @xmath190 . in this work we have found that both temperature and external magnetic field affect the mass limit and other physical properties of the fermionic ads star . the increase of bulk temperature enables the pressure and energy density of the star to increase . consequently , the mass limit becomes slightly greater due to the larger pressure . this is the typical behavior of the fermi gas at finite temperature . too large temperature results in the the fermions refusing to be confined within a finite - size star , they will leak to the space inevitably . in the presence of external magnetic field , the mass limit decreases when the magnetic field increases . as we can see from eqn . ( [ pressure in simulation ] ) and ( [ energy density in simulation ] ) , an increase in the magnetic field results in a smaller energy and pressure density as well as a smaller chemical potential . the mass limit becomes smaller naturally . there is an interesting competition between the temperature and the magnetic field to the density profile and mass limit of the star . extremely strong magnetic field tends to make the bulk fermions stay in the landau states with lower energies whilst the temperature causes the particles to flee the star . the radius of curvature of the ads space also affects the mass limit evidently . when the radius of curvature increases , the mass limit increases substantially as are shown in fig . [ mixed3 ] . interestingly , the peak of the mass limit curve shifts to the lower central density side . gravitational collapse in the ads space has holographic dual in terms of the non - equilibrium thermalization of the gauge matter on the boundary . even though the hawking temperature of the black hole at the end of the gravitational collapse can be matched with the temperature at thermal equilibrium of the gauge matter at the end of thermalization , the bulk temperature of the ads star does not seem to have such a straightforward relationship with the dual gauge matter . the hawking temperature of the resulting black hole is not directly related to the temperature of the fermionic star before the collapse but inversely proportional to the mass of the star . therefore it is the mass limit studied in our work which corresponds to the temperature of the gauge matter at the thermal equilibrium after thermalization , i.e. @xmath191 ( the black hole formed at our mass limit is small ads black hole with negative specific heat , the precise relationship is given in eqn . ( [ thm ] ) ) , for a given mass injection @xmath192 in the dual gauge picture . the mass limit also plays the role of the minimum injected mass required for the dual gauge matter to undergo the thermalization into the thermal equilibrium . larger mass limit means that it requires more injected energy to thermalize , and once it thermalizes , the gauge matter will be at lower hawking temperature . it should be remarked that the ads black holes formed at the end of the gravitational collapse of the ads stars at our mass limits are small black holes with negative specific heat . they are previously thought to be less thermodynamically preferred than the ads vacuum in the context of the ads / cft correspondence and only the large ads black hole with positive specific heat was considered relevant for the dual of the thermal gauge matter . however , inevitable collapse at the mass limits corresponding to small ads black holes suggest that there might exist the phase of thermal gauge matter with negative specific heat dual to these black holes at the end of the gravitational collapse . injecting more mass would make these ads black holes and their gauge duals eventually become thermodynamically stable with positive specific heat . entropy density of the ads star under uniform magnetic field is found to show interesting behaviour ; @xmath151 for small and @xmath152 for higher _ bulk _ temperatures . such @xmath117-dependence is typical for free fermion gas ( modulo the magnetic field existence ) and it is amusingly similar to the @xmath117-dependence of the magnetic black brane entropy in the ads @xcite even though the latter is the hawking - page temperature of the brane , not the bulk temperature of the material making up the brane itself . nevertheless , the correspondence between the bulk and boundary exists throughout the gravitational collapse as long as the background is the ads . the holographic duality suggests that the hilbert spaces of both the gravity and gauge theory as well as their partition functions are equivalent . a global probe for the number of degrees of freedom on both sides of the duality is the entropy . the total entropy of the ads star above the mass limit , which indicates the lower bound of the black hole entropy at the end of gravitational collapse , should also be the lower bound of the total entropy of the gauge matter at the end of thermalization in the dual picture , @xmath193 . we found that the entropy ( at the mass limit ) of the ads star is an increasing ( decreasing ) function of the temperature ( magnetic field ) , similar behaviour to the mass limit . the remaining unanswered question is the exact correspondence between the gravitational collapse in the bulk and non - equilibrium deconfinement thermalization of the dual gauge matter . if bulk gravity is dual to colour - singlet glueball interaction and it causes the gravitational collapse in the ads , how could the glueball exchange describe the deconfinement thermalization in the dual gauge picture ? should there exist the critical glueball density corresponding to the mass limit in the bulk which determines the deconfinement phase transition on the gauge theory side ? what is the boundary ( cft ) gauge description of the tov equation and more generically the einstein equation in the ( ads ) bulk ? what are the duals of bulk temperature and other thermodynamic and transport quantities such as the adiabatic index and sound speed of the ads star in the gauge theory side ? t.c . would like to thank sirachak panpanich for valuable helps . is supported in part by the thailand research fund ( trf ) and commission on higher education ( che ) under grant rmu5380048 and also by the thailand center of excellence in physics ( thep ) . is supported in part by the thailand research fund ( trf ) and commission on higher education ( che ) under grant rmu5380048 . we solve the einstein equation in @xmath6-dimensional spacetime in this section . starting from the einstein equation , @xmath194 where @xmath195 , @xmath196 , @xmath62 , @xmath197 , @xmath12 , @xmath198 are ricci tensor , metric tensor , ricci scalar , energy - momentum tensor , the area of @xmath13 and constant @xmath199 respectively . assuming a perfect fluid , the energy - momentum tensor is given by @xmath200 where we use a spherically symmetric metric in @xmath6 dimensions in the polar coordinates @xcite @xmath201 the lagrangian of this metric is then given by @xmath202 we will use the euler - lagrange equation to find the equations of motion and read off the connections , @xmath203 consider @xmath204 component , the equation of motion is @xmath205 and the connections are @xmath206 the equation of motion in the @xmath207 component reads @xmath208 and the connections in the @xmath207 component are @xmath209 likewise , the equation of motion in the @xmath210 component is @xmath211 and the connections in the @xmath210 component are @xmath212 where @xmath213 . similarly , the equation of motion in the @xmath214 component is @xmath215 and the relevant connections are @xmath216 where @xmath217 . the equation of motion in the @xmath218 component is @xmath219 and the connections in @xmath220 component are @xmath221 where @xmath222 and @xmath223 . the ricci tensor and ricci scalar can be calculated from @xmath224 after some calculations , we have @xmath225 consider @xmath226 , then @xmath227 @xmath228 if we consider an ads space(with a negative cosmological constant , @xmath178 ) , then the einstein equation reads @xmath229 and equation ( [ result of gtt ] ) becomes @xmath230 change @xmath231 , so that @xmath232 the solution to this equation is @xmath233 let @xmath234 , then @xmath235 also the accumulated mass can be defined to be @xmath236 consider @xmath237 , then @xmath238 use equation ( [ 1 st result of gtt ] ) from @xmath239 and multiply by @xmath240 , @xmath241 change @xmath242 , @xmath231 , equation ( [ gtt grr 2 ] ) becomes @xmath243 solve this equation to find relations between @xmath244 and @xmath48 , @xmath245 where @xmath246 finally we obtain the coupled equations of motion from equation ( [ equation of motion of mass ] ) and ( [ equation of motion of chi ] ) @xmath247 moreover , when we consider the energy momentum conservation @xmath248 by letting @xmath249 and @xmath250 , @xmath242 , it leads to the tov equation in @xmath6-dimension , @xmath251 next we want to rewrite this equation in the form containing thermodynamic quantities such as the chemical potential , the entropy , and the temperature of matter within the spherically symmetric star . from thermodynamic relations involving the entropy density @xmath158 ; @xmath252 the tov equation can be rewritten as @xmath253 implying two equations to be satisfied simultaneously @xmath254 the temperature equation can be solved to obtain the redshifted temperature profile within the star @xmath255 where @xmath256 and @xmath257 is the temperature at the star center . the chemical potential equation similarly gives @xmath258 the coupled equations of motion can then be written in terms of the accumulated mass and chemical potential as the following @xmath259 a slowly converging series can be evaluated effectively by using an integral as in the euler - maclaurin formula @xmath260 in this article , the partition function sum over landau states is approximated using this conventional method by letting @xmath261 where @xmath262 . [ cols="^,^,^",options="header " , ] j. m. maldacena , `` the large n limit of superconformal field theories and supergravity , '' _ adv . phys . _ * 2 * , 231 ( 1998 ) [ _ int . j. theor . phys . _ * 38 * , 1113 ( 1999 ) ] [ arxiv : hep - th/9711200 ] . edward witten , anti - de sitter space , thermal phase transition , and confinement in gauge theories , " _ adv . * 2 * ( 1998 ) 505 , [ arxiv : hep - th/9803131 ] . g. t hooft , `` dimensional reduction in quantum gravity , '' arxiv : gr - qc/9310026 . l. susskind , `` the world as a hologram , '' _ j. math . phys . _ * 36 * , 6377 ( 1995 ) [ arxiv : hep - th/9409089 ] . edward witten , anti - de sitter space and holography , " _ adv . _ * 2 * ( 1998 ) 253 - 291 , [ arxiv : hep - th:9802150 ] . j. de boer , k. papadodimas and e. verlinde , `` holographic neutron stars , '' _ jhep _ * 1010 * , 020 ( 2010 ) [ arxiv:0907.2695 [ hep - th ] ] . x. arsiwalla , j. de boer , k. papadodimas and e. verlinde , `` degenerate stars and gravitational collapse in ads / cft , '' _ jhep _ * 1101 * , 144 ( 2011 ) [ arxiv:1010.5784 [ hep - th ] ] . u. h. danielsson , e. keski - vakkuri and m. kruczenski , spherically collapsin matter in ads , holography , and shellons , " _ nucl . b _ * 563 * , 279 ( 1999 ) [ arxiv : hep - th/9905227 ] . s. lin and e. shuryak , toward the ads / cft gravity dual for high energy collisions . 3 . gravitationally collapsing shell and quasiequilibrium , " _ phys . d _ * 78 * , 125018 ( 2008 ) [ arxiv:0808.0910[hep - th ] ] . v. balasubramanian , a. bernamonti , j. de boer , n. copland , b. craps , e. keski - vakkuri , b. muller , a. schafer , m. shigemori , and w. staessens , thermalization of strongly coupled field theories , " _ phys . lett . _ * 106 * , 191601 ( 2011 ) [ arxiv:1012.4753[hep - th ] ] . v. balasubramanian , a. bernamonti , j. de boer , n. copland , b. craps , e. keski - vakkuri , b. muller , a. schafer , m. shigemori , and w. staessens , holographic thermalization , " _ phys . rev . _ * d84 * , 026010 ( 2011 ) [ arxiv:1103.2683[hep - th ] ] . kharzeev , l.d . mclerran , and h.j . warringa , the effects of topological charge change in heavy ion collisions:event by event p and cp violation , " _ nucl . phys . a _ * 803 * ( 2008 ) 227 , [ arxiv:0711.0950 , hep - th ] . r. c. tolman , `` static solutions of einstein s field equations for spheres of fluid , '' _ phys . rev . _ * 55 * , 364 ( 1939 ) . j. r. oppenheimer and g. m. volkoff , `` on massive neutron cores , '' _ phys _ * 55 * , 374 ( 1939 ) . v. canuto and h. y. chiu , quantum theory of an electron had in intense magnetic fields , " _ phys . rev . _ * 173 * , 1210 ( 1968 ) . l. e. blumenson , `` a derivation of n - dimensional spherical coordinates , '' _ the american mathematical monthly _ * 67 * , 1 ( jan 1960 ) . jacob d. bekenstein , entropy bounds and black hole remnants , " _ phys . rev . _ * d49 * , 1912 - 1921 ( 1994 ) [ arxiv : gr - qc/9307035 ] .

consistent unification of relativity and quantum mechanics remains an unsolved theoretical problem in spite of many efforts applied to its solution in the 20th century . the fundamental difference between relativistic and non - relativistic physics follows from the famous einstein s formula @xmath1 . this formula , in particular , implies that if a system of particles has sufficient energy @xmath2 of their relative motion , then this energy may be converted to the mass @xmath3 of newly created particles . generally , there is no limit on how many particles can be created in collisions , so any realistic quantum mechanical description of high - energy systems should involve states with any number of particles from zero to infinity . the number of particles is not conserved during time evolution . the most familiar example of such a behavior is the emission and absorption of light ( photons ) in electrodynamics . first attempts to describe relativistic quantum systems were undertaken immediately after creation of the formalism of quantum mechanics in 1920 s . a quantum theory of the electromagnetic field was constructed by quantization of the classical maxwell electrodynamics . in lowest perturbation orders , this theory agreed well with experimental results . however , perturbative calculations for the @xmath0-matrix did not work in higher orders , in particular , due to ultraviolet divergences . the way to calculate the @xmath0-matrix in qft accurately in all orders was provided by the renormalized qed formulated by feynman , schwinger , and tomonaga in the late 1940 s . however this approach created a host of other problems . according to the prevailing interpretation , the creation and annihilation operators present in the hamiltonian and lagrangian of qed correspond to _ bare _ particles having infinite masses and charges . however , the bare particles have never been directly observed in experiments . they are believed to be surrounded by clouds of virtual photons and electron - positron pairs , thus forming complex objects called _ dressed _ particles . the dressed particles are supposed to be the eigenstates of the full hamiltonian . they have finite experimentally observable masses and charges . the problem is that the bare particle hamiltonian of qft is formally infinite . although , these infinities cancel out when the @xmath0-matrix is calculated , the hamiltonian is useless if one wants to calculate the time evolution or to find wavefunctions of bound states via diagonalization procedure . two lines of research were initiated to cope with these problems . the _ dressed particle _ approach was suggested by greenberg and schweber @xcite . their goal was to get rid of the bare particles and to express the entire formalism of qft through observable dressed particles only . the operators for dressed particles were sought as unitary transforms of bare particle operators . from another direction , gazek and wilson @xcite introduced the _ similarity renormalization _ formalism . the idea was to apply a unitary transformation to the hamiltonian in order to ensure that interaction potentials rapidly fall off as functions of energy differences , and to guarantee that all loop integrals are convergent . the similarity and dressing transformations are strikingly similar . they can be combined into one unitary _ similarity - dressing _ transformation which achieves two goals at once : the theory can be expressed in terms of real dressed particles and the hamiltonian can be made finite in all perturbation orders . at the same time , the accurate and well - tested @xmath0-matrix of the renormalized theory remains intact . this combined approach was developed in refs . @xcite and dubbed the relativistic quantum dynamics , or rqd . in this paper we will discuss the reasons why the renormalization difficulties ( e.g. , the absence of a well - defined hamiltonian and unsatisfactory treatment of the time evolution ) persist in current relativistic quantum field theories . to avoid unnecessary mathematical complications , we will be working with a simple model theory which , nevertheless , shares some important features with qed . in section 9 we will explain how the similarity - dressing transformation of this model theory leads to a well defined ( perturbatively ) finite hamiltonian that can be used for all kinds of quantum mechanical calculations ( @xmath0-matrix , bound states , time evolution , etc . ) without the need for renormalization . in addition , the rqd formalism does not use the dubious notions of bare and virtual particles . in this paper we will be concerned with three types of phenomena that are normally studied in physical experiments : bound states , time evolution , and scattering . these three areas account for the most of experimental information available about fundamental particles and their interactions . the key theoretical quantity involved in quantum mechanical description of these phenomena is the hamilton operator @xmath4 . the energies @xmath5 and state vectors @xmath6 of bound states can be found as eigenvalues and eigenvectors of the hamiltonian @xmath7 the development of the state vector @xmath8 from time @xmath9 to time @xmath10 is described by the _ time evolution operator _ @xmath11 @xmath12 if the eigenvalues @xmath5 and eigenvectors @xmath13 of the hamiltonian are known and the initial state is represented as a sum ( and/or integral ) over the basis states @xmath14 unfortunately , in most cases , the full spectrum of the hamiltonian is not known , and the time evolution is difficult to predict . this difficulty , however , is not that disappointing , because experimental observations of the time evolution in subatomic world are even more difficult than calculations . most experiments in high energy physics are performed by preparing free particles or their bound states ( like hydrogen atoms or deuterons ) , bringing them into collision , and studying the properties of free particles or bound states leaving the region of collision . in these experiments , it is not possible to observe the time evolution during interaction : particle reactions occur almost instantaneously and one can only register the reactants and products which move freely before and after the collision . this gives a lucky break for theoreticians : in such situations the theory is not required to describe the detailed dynamics of particles during the short interval of collision . it is sufficient to provide a mapping of free states before interaction onto the free states after the interaction . to describe scattering experiments , one needs only the formula for the time evolution from the remote past @xmath15 to the distant future @xmath16 @xmath17 where the @xmath0__-operator _ _ [ s - oper ] is defined by @xmath18 one can read the right hand side of eq . ( [ eq : s - oper ] ) from right to left as a sequence of three steps : ( i ) the non - interaction evolution of the system from time @xmath9 in the past to 0 ; ( ii ) the sudden jump at @xmath19 described by the @xmath0-operator ; ( iii ) the free evolution from @xmath19 to the future time @xmath10 . the most effective technique available for calculations of the @xmath0-operator is the perturbation theory which can be written in many equivalent forms . the dyson time - ordered expansion provides the most economical expressions that can be encoded in familiar feynman diagrams . however , for the discussion in this paper , we found more useful two other perturbative expressions which differ from the dyson s formula only by re - shuffling the terms . the `` old - fashioned '' formula for the @xmath0-operator is @xmath20 where @xmath21 is the interaction part of the total hamiltonian @xmath22 , @xmath23 and the factor latexmath:[$e^{-\epsilon switching the interaction on and off . operators with @xmath10-dependence determined by the free hamiltonian @xmath25 as in eq . ( [ eq:8.67 ] ) will be called _ regular_. using convenient symbols for @xmath10-integrals [ underline ] @xmath26 formula ( [ eq:8.68 ] ) can be written compactly as @xmath27 where @xmath28 another equivalent expression for @xmath0 was suggested by magnus @xcite @xmath29 here the hermitian operator @xmath30 can be represented as a series of multiple commutators with @xmath10-integrations @xmath31 -\frac{1}{6}[\underline{\underline{v(t)},[v(t)},v(t ) ] ] -\frac{1}{6}[\underline{[\underline{v(t)},v(t)]},v(t ) ] \nonumber \\ & \mbox { } & + \frac{i}{12}[\underline{\underline{\underline{v(t)},[[v(t)},v(t)]},v(t ) ] ] \nonumber \\ & + & \frac{i}{12}[\underline{[\underline{\underline{v(t)},[v(t)},v(t)]]},v(t ) ] + \frac{i}{12}[\underline{\underline{[\underline{v(t)},v(t)]},[v(t)},v(t ) ] ] + \ldots \label{eq:8.72}\end{aligned}\ ] ] one important advantage of this representation is that the @xmath0-operator ( [ eq:8.71 ] ) is manifestly unitary . it follows from equations ( [ eq:8.69 ] ) and ( [ eq:8.71 ] ) that operators @xmath32 and @xmath30 are related to each other @xmath33 so finding @xmath30 or @xmath32 are equivalent tasks . the @xmath0-operator and the hamiltonian provide two different ways to describe dynamics . the @xmath0-operator represents only `` integrated '' time evolution from the remote past to the distant future . the knowledge of the @xmath0-operator is sufficient to calculate the scattering cross - sections as well as energies and lifetimes of stable and metastable bound states .- operator on the complex energy plane . ] however , in order to describe the time evolution and the wavefunctions of bound states the full interacting hamiltonian @xmath4 is required . it can be shown @xcite that two hamiltonians @xmath34 and @xmath35 related to each other by a unitary transformation @xmath36 @xmath37 yield the same scattering @xmath38 as long as condition @xmath39 is satisfied . such hamiltonians @xmath4 and @xmath40 are called _ scattering - equivalent_. the energy spectra of two scattering equivalent hamiltonians are identical . however , the eigenvectors are different and , according to eq . ( [ eq:8.61a ] ) , the corresponding descriptions of dynamics are different as well . therefore scattering - equivalent theories may be not _ physically _ equivalent . calculations of bound states , time evolution and scattering is a routine practice in non - relativistic quantum mechanics . however , the situation is less certain in relativistic quantum field theories . as mentioned in introduction , the hamiltonian of renormalized qft is infinite . therefore , it is not immediately clear if the above formulas ( [ eq:1 ] ) - ( [ eq:8.74 ] ) remain valid for the high energy relativistic phenomena , and what modifications , if any , should be introduced in quantum theory to take into account the variable number of particles . in this paper we illustrate the difficulties encountered in quantum field theories by analyzing a simple model theory with variable number of particles . we will demonstrate that the hamiltonian can be redefined so that there is no need for renormalization and usual quantum mechanical techniques remain applicable even in the relativistic case . our model theory describes two kinds of particles . these are massive spinless fermions which will be called _ electrons _ and massless bosons with zero helicity , which will be called _ photons_. here we disregard the spin and polarization degrees of freedom as they are not so important for the discussion of renormalization . to allow for creation and annihilation of particles , the system is described in the fock space which is built as a direct sum of sectors with various numbers of particles . for example , if we denote @xmath41 the no - particle vacuum state , @xmath42 the one - electron hilbert space and @xmath43 the one - photon hilbert space , then the fock space can be written as an infinite direct sum @xmath44 the anticommutation and commutation relations for particle creation and annihilation operators ( @xmath45 , @xmath46 for electrons and @xmath47 , @xmath48 for photons , respectively ) are , as usual , @xmath49 @xmath50 & = & \delta ( \mathbf{p}-\mathbf{p } ' ) \label{eq : cc}\\ \{a_{\mathbf{p}},a_{\mathbf{p}'}\ } & = & \{a^{\dag}_{\mathbf{p}},a^{\dag}_{\mathbf{p}'}\ } = 0 \label{eq : aa2}\end{aligned}\ ] ] @xmath51 = [ c^{\dag}_{\mathbf{k}},c^{\dag}_{\mathbf{k } ' } ] = 0 \label{eq : cc2}\end{aligned}\ ] ] @xmath52&= & [ a^{\dag}_{\mathbf{p}},c_{\mathbf{k } } ] = [ a_{\mathbf{p}},c^{\dag}_{\mathbf{k } } ] = [ a_{\mathbf{p}},c_{\mathbf{k } } ] = 0.\end{aligned}\ ] ] the full hamiltonian @xmath53 is the sum of the free hamiltonian @xmath54 ( where @xmath55 and @xmath56 are one - particle energies of electrons and photons , respectively ) and the interaction , which we choose in the following form @xmath57 the coupling constant @xmath58 is the absolute value of the electron charge . here and in what follows the _ perturbation order _ [ pert - order ] of an operator (= the power of the coupling constant @xmath58 ) is shown by the subscript . for example , the free hamiltonian @xmath25 does not depend on @xmath58 , so it is of zero perturbation order ; @xmath59 is of the first perturbation order , etc . the number of electrons is conserved by the interaction ( [ eq:9.69 ] ) , but the number of photons is not conserved . so , this theory is capable of describing important processes of the emission and absorption of photons . in this paper we will discuss ultraviolet divergences associated with the interaction ( [ eq:9.69 ] ) . however , we will skip completely the discussion of `` infrared divergences '' which are related to the zero mass of photons and singularities @xmath60 in ( [ eq:9.69 ] ) . the easiest way to avoid infrared problems in practical calculations is to assign a small non - zero mass to photons . in section [ ss : e - e - scatt ] we are going to calculate the scattering operator ( [ eq:8.69 ] ) with the above hamiltonian . before doing that , some remarks are in order . the operator @xmath0 is obtained as a sum of products ( [ eq:8.70 ] ) ( or commutators ( [ eq:8.72 ] ) ) of interactions @xmath61 with @xmath10-integrations . each term in these expressions can be written as a normally ordered product of @xmath62 creation operators @xmath63 and @xmath64 annihilation operators @xmath65 . and @xmath65 refer to generic creation and annihilation operators without specifying the type of the particle . the pair of integers @xmath66 will be referred to as the _ index _ [ index ] of the term @xmath67 . ] @xmath68 d_{nm}[\mathbf{q } ] e^{ite_{nm}[\mathbf{q } ] } \delta ( \mathbf{p}_{nm}[\mathbf{q } ] ) \alpha^{\dag}_{\mathbf{q}'_1 } \ldots \alpha^{\dag}_{\mathbf{q}'_n } \alpha _ { \mathbf{q}_1 } \ldots \alpha_{\mathbf{q}_m } \label{eq:9.49}\end{aligned}\ ] ] where integration is carried over momenta of all created and annihilated particles @xmath69 $ ] . the momentum conservation law is guaranteed by the delta function in ( [ eq:9.49 ] ) whose argument is the sum of momenta of created particles minus the sum of momenta of annihilated particles @xmath70 usually , we will perform explicit integration over one momentum @xmath71 which removes the delta function and expresses @xmath71 as a linear function of other momenta . the argument of the exponent in ( [ eq:9.49 ] ) contains the _ energy function _ @xmath72 which is the difference of energies of particles created and annihilated by @xmath67 . @xmath73 is a numerical _ coefficient function_. suppose that a term @xmath67 has coefficient function @xmath73 , then we introduce a useful notation [ circ ] @xmath74 for the operator whose coefficient function @xmath75 is a product of @xmath73 and a function @xmath76 of the same arguments @xmath77 & = & d_{nm}[\mathbf{q } ] \zeta [ \mathbf{q}]\end{aligned}\ ] ] then , a @xmath10-dependent regular term @xmath78 can be written as @xmath79 and its definite @xmath10-integral is @xmath80 eq . ( [ eq:9.52 ] ) means that each term in @xmath81 is non - zero only on the hypersurface of solutions of the equation @xmath82 if such solutions exist . this hypersurface in the momentum space is called the _ energy shell _ [ energy - shell ] of the term @xmath78 . note that the scattering operator ( [ eq:8.69 ] ) is non - trivial only on the energy shell , i.e. , where the energy conservation condition holds . in this section we would like to get a further insight into the nature of operators in the fock space by dividing them into three groups depending on their index @xmath66 . we will call these types of operators _ renorm _ , _ phys _ , and _ unphys_. * renorm operators * [ oper - renorm ] have either index ( 0,0 ) ( a numerical constant @xmath83 ) or index ( 1,1 ) in which case the same type of particle is created and annihilated . the most general form of a renorm operator in our theory is omitting all numerical factors , indices , integration and summation signs . note also that the interaction operator ( [ eq:9.69 ] ) commutes with the operator of charge @xmath84 , so any product or commutator of terms derived from @xmath59 should also commute with @xmath85 . this is not true for terms like @xmath86 and @xmath87 , so they are not allowed in the theory . ] @xmath88 the free hamiltonian ( [ eq : free - h0 ] ) is an example of a renorm operator . the class of renorm operators is characterized by the property that the energy function ( [ eq:9.51 ] ) is identically zero . so , renorm operators always have an non - empty energy shell where they do not vanish . regular renorm operators do not depend on @xmath10 . * phys operators * [ oper - phys ] have at least two creation operators and at least two destruction operators ( index @xmath66 with @xmath89 and @xmath90 ) . in this case the energy shell is non - empty . for example , the energy shell for phys operator @xmath91 is given by the set of solutions of equation @xmath92 which is not empty . * unphys operators * [ oper - unph ] have index @xmath93 or @xmath94 . it can be shown that the energy shell is empty for unphys operators . for example , the interactions ( [ eq:9.69 ] ) are unphys . the energy shell equation for these two terms @xmath95 does not have a solution , because there are no photon states with zero momentum . renorm , phys , and unphys operators exhaust all possibilities in our theory , therefore any regular operator @xmath21 must have a unique decomposition @xmath96 the rules for calculations of commutators , derivatives and @xmath10-integrals with different operator types are summarized in table 9.2 . for example , the @xmath10-integrals of phys and unphys operators are given by formula @xmath97 @ccccccc type of operator & & & & & & @xmath98 & @xmath99 $ ] & @xmath100 $ ] & @xmath101 $ ] & @xmath102 & @xmath103 & @xmath104 & & & & & & p & p & p+u & p & p & p & p u & p+u & p+u+r & u & u & u & 0 r & p & u & r & 0 & nr & @xmath105 our goal in this section is to introduce the diagram technique which greatly facilitates perturbative calculations of scattering operators ( [ eq:8.70 ] ) and ( [ eq:8.72 ] ) . let us graphically represent each term in the interaction operator ( [ eq:9.69 ] ) as a _ ( see fig . 1 ) . [ diagram ] each particle operator in ( [ eq:9.69 ] ) is represented as an oriented _ line _ [ line ] or arrow . the line corresponding to the annihilation operator enters the vertex , and the line corresponding to the creation operator leaves the vertex . electron lines are shown by full arcs and photon lines are shown by broken arrows . each line is marked by the momentum label of the corresponding particle operator . free ends of the electron lines are attached to the vertical `` order bar '' on the left hand side of the diagram . the order of these _ external lines _ [ extern ] ( from bottom to top of the diagram ) corresponds to the order of particle operators in the interaction term ( from right to left ) . an additional numerical factor is indicated in the upper left corner of the diagram . the @xmath10-integral @xmath106 differs from @xmath107 only by the factor @xmath108 ( see eq . ( [ eq:9.56 ] ) ) which is represented in the diagram by drawing a box that crosses all external lines . a line entering ( leaving ) the box contributes its energy with the negative ( positive ) sign to the energy function @xmath109 . the diagram representation of the integral @xmath110 is shown in fig . the product of two operators @xmath111 is represented by simply placing diagram @xmath112 below diagram @xmath98 and attaching the external electron lines of both diagrams to the same order bar . for example , the diagram for the product of the second term in ( [ eq:9.69 ] ) ( fig . 1(b ) ) and the first term in ( [ eq:9.70 ] ) ( fig . 2(a ) ) @xmath113 is shown in fig . this product should be further converted to the normal form , i.e. , all incoming lines should be positioned below the outgoing lines . due to the relations ( [ eq : aa ] ) - ( [ eq : cc2 ] ) each exchange of positions of the electron particle operators ( full external lines on the diagram ) changes the total sign of the expression . each permutation of annihilation and creation operators ( incoming and outgoing lines ) of similar particles creates an additional expression and a new diagram in which the swapped lines are joined together . using these rules we first move the photon operators in ( [ eq:9.71 ] ) to the rightmost positions , move the operator @xmath114 to the leftmost position , and add another term due to the anticommutator @xmath115 . @xmath116 this expression is represented by two diagrams 3(b ) and 3(c ) . in the diagram 3(b ) the electron line marked @xmath117 has been moved to the top of the order bar . in the diagram 3(c ) the product @xmath118 and the integration by @xmath117 are represented by merging or _ pairing _ the incoming electron line carrying momentum @xmath119 with the outgoing electron line carrying momentum @xmath117 . this produces the _ internal electron line _ [ intern ] carrying momentum @xmath119 between two vertices . in the expression ( [ eq:9.72 ] ) , the electron operators are in the normal order , however , the photon operators are not . the next step is to bring the photon operators to the normal order @xmath120 according to equation ( [ eq : cc ] ) , the normal ordering of photon operators in 3(b ) yields diagrams 3(d ) and 3(e ) . diagrams 3(f ) and 3(g ) are obtained from 3(c ) in a similar manner . using diagrams , with some practice , one can perform calculations of scattering operators ( [ eq:8.70 ] ) and ( [ eq:8.72 ] ) much easier than in the usual algebraic way . during these diagram manipulations we , actually , do not need to keep momentum labels of lines . the algebraic expression of the result can be easily restored from an unlabeled diagram by following these steps : * assign a distinct momentum label to each external line , except one , whose momentum is obtained from the momentum conservation condition . * assign momentum labels to internal lines so that the momentum conservation law is satisfied at each vertex . if there are _ loops _ , [ loop ] one needs to introduce additional independent _ loop momenta_. is the loop momentum . ] * read external lines from top to bottom of the order bar and write corresponding particle operators from left to right . * for each box , write a factor @xmath121 , where @xmath122 is the sum of energies of particles going out of the box and @xmath123 is the sum of energies of particles coming into the box . * write a factor @xmath124 , where @xmath125 is the energy function of the diagram which is the sum of energies of all outgoing external lines minus the sum of energies of all incoming external lines . * for each vertex introduce a factor @xmath126 , where @xmath127 is the momentum of the photon line attached to this vertex . * integrate the obtained expression by all independent external and loop momenta . let us now try to extract some physical information from the above theory . we will calculate low order terms in the perturbation expansion ( [ eq:8.70 ] ) for the @xmath0-operator . @xmath128 to obtain the corresponding contributions to the @xmath0-operator we need to take @xmath10-integrals of these expressions @xmath129 note that the right hand side of ( [ eq:9.73 ] ) and the first term on the right hand side of ( [ eq:9.74 ] ) are unphys . their energy shell is empty , so , according to table 1 , they do not contribute to the @xmath0-operator . operator @xmath130 on the right hand side of eq . ( [ eq:9.74 ] ) has two terms corresponding to two types of scattering processes allowed in the 2nd perturbation order . the term of the type @xmath131 ( see , e.g. , fig . 3(f ) ) annihilates an electron and a photon in the initial state and recreates them ( with different momenta ) in the final state . so , this term describes the electron - photon ( compton ) scattering . [ compton2 ] in this paper we will focus on the other term which describes the electron - electron scattering . let us consider in more detail the second - order contribution to this process ( see fig . 3(e ) ) @xmath132 & = & 2 \pi \int d \mathbf{p } d \mathbf{q } d \mathbf{k } \delta ( \omega_{\mathbf{p - k } } + \omega_{\mathbf{q+k } } -\omega_{\mathbf{q}}- \omega_{\mathbf{p } } ) d_2(\mathbf{p},\mathbf{q } , \mathbf{k } ) a^{\dag}_{\mathbf{p - k } } a^{\dag}_{\mathbf{q+k } } a_{\mathbf{q}}a_{\mathbf{p } } \label{eq:9.76}\end{aligned}\ ] ] the coefficient function in ( [ eq:9.76 ] ) can be read from the diagram , according to the rules ( i ) - ( vii ) , @xmath133 in the non - relativistic approximation ( @xmath134 ) , the coefficient function @xmath135 has singularity @xmath136 which is characteristic for scattering of two electrons interacting via repulsive coulomb potential [ coulomb1 ] @xmath137 so , our model theory is quite realistic . next consider the third term on the right hand side of eq . ( [ eq:9.74 ] ) . it is given by the diagram in fig . according to rules ( i ) - ( vii ) this diagram is represented by the expression @xmath138 there are serious problems with this term . first , the loop integral by @xmath127 is divergent because the integrand in ( [ eq:9.88 ] ) has asymptotic behavior @xmath139 at large @xmath140 . however , even if the integral were convergent , the presence of a renorm contribution in the operator @xmath32 is unacceptable , because , according to table 1 , the @xmath10-integral of any renorm term is infinite . so , the scattering phase in the second order @xmath141 is infinite , which is absurd . moreover , if we continued calculations ( [ eq:9.73 ] ) - ( [ eq:9.74 ] ) to higher perturbation orders we would find out that even phys terms in @xmath32 become infinite due to divergent loop integrals , so the theory with the hamiltonian ( [ eq : free - h0 ] ) - ( [ eq:9.69 ] ) is seriously flawed . in a consistent theory we must require that @xmath142 therefore , operator @xmath143 must be purely phys @xmath144 it was shown in ref . @xcite that condition ( [ eq:11.43 ] ) is equivalent to requiring that the @xmath0-operator leaves the vacuum and one - particle states invariant @xmath145 this is the _ mass renormalization condition _ of the traditional renormalization theory . to satisfy this condition , we must modify the hamiltonian ( [ eq : free - h0 ] ) - ( [ eq:9.69 ] ) by adding certain unphys @xmath146 and renorm @xmath147 _ counterterms _ to the interaction operator @xmath59 . in other words , we are saying that the original hamiltonian @xmath148 is not correct , and the modified hamiltonian with renormalization counterterms @xmath149 better describes interactions between particles . to comply with eq . ( [ eq:11.43 ] ) , we must choose the counterterms in such a way that operator @xmath150 does not contain renorm terms . from table 1 , it is clear that renorm terms in @xmath151 may appear due to the presence of renorm and unphys terms in @xmath152 and their products . so , in order to satisfy eq . ( [ eq:11.42 ] ) , there should be such a balance between unphysical and renorm terms in @xmath153 that all renorm terms in @xmath151 cancel out in all orders of the perturbation theory . the mass renormalization is achieved by adding renorm counterterms in even orders : @xmath154 , @xmath155 , etc . the charge renormalization procedure will be discussed in section [ ss : charge - renorm ] . it requires addition of unphys counterterms in odd orders @xmath156 , @xmath157 , etc . we take these considerations into account by writing the general expression for the hamiltonian of the renormalized theory - independent hamiltonian @xmath158 . although , at intermediate calculation steps it is convenient to keep all operators @xmath10-dependent , as in eq . ( [ eq:8.67 ] ) , in the end we should set @xmath19 . ] @xmath159 where @xmath160 we obtain formulas for @xmath161 ( @xmath162 ) by inserting interaction ( [ eq : inter ] ) in ( [ eq:8.70 ] ) and collecting terms of equal order . @xmath163 where we denoted @xmath164 now we go order - by - order and choose counterterms @xmath165 so that renorm terms are eliminated from the left hand sides of eqs ( [ eq:11.47 ] ) - ( [ eq:11.50 ] ) . the first - order term ( [ eq:11.47 ] ) is unphys , so there is no need for renormalization in the first order . to ensure that @xmath166 does not have a renorm part we choose the counterterm @xmath167 ( see diagram 3(g ) ) . with this choice , we can rewrite the contributions to the @xmath0-operator in the 4 lowest orders @xmath168 where @xmath169 and the superscript @xmath170 denotes the sum of phys and unphys terms . the term on the right hand side of ( [ eq:11.51 ] ) has odd number of particle operators , hence it is free of renorm parts ( which have either zero or two particle operators ) . therefore , no renorm counterterms should be added there . just as we did in the 2nd order , we can choose a renorm counterterm @xmath155 which simply cancels all renorm terms which may be present in the first three terms on the right hand side of ( [ eq:11.52 ] ) . with the above choices , the operator @xmath151 does not contain renorm terms up to the 4th order , as required by the mass renormalization condition , ) and in more realistic theories , such as qed , qcd , and standard model . products or commutators of such terms in ( [ eq:8.70 ] ) or ( [ eq:8.72 ] ) give rise to renorm terms in the @xmath0-operator which should be compensated by adding renorm terms to the hamiltonian . the operator structure of these terms is the same as in @xmath25 ( [ eq : free - h0 ] ) . this results in the difference of masses of bare and dressed particles , as discussed at the end of section [ ss : charge - renorm ] . ] and the expression for @xmath171 simplifies @xmath172 using the diagram technique we find that the @xmath173 part of the first two terms on the right hand side of eq . ( [ eq:11.55 ] ) is represented by 7 diagrams shown in fig . 4 . from diagram rules ( i ) - ( vii ) , we obtain the contribution of diagrams 4(a ) - 4(b ) to the coefficient function @xmath174 for the electron - electron scattering on the energy shell . @xmath175 where @xmath176 , @xmath177 , @xmath178 , and @xmath179 . unfortunately , this contribution is infinite ( at large values of @xmath180 the integrand behaves as @xmath181 ) . thus we conclude that the mass renormalization procedure described above has not removed all divergences . to solve this problem we need to perform the second renormalization step known as the charge renormalization procedure . this step is explained in the next section . first note that the divergent terms ( [ eq:11.58 ] ) have a singularity @xmath182 at @xmath183 . as we know from section [ ss : e - e - scatt ] , such a singularity is responsible for the low - energy electron - electron scattering at large distances . from classical physics we also know that long - distance interactions between charged particles depend on @xmath184 ( in our language , they are of the second perturbation order ) and they are accurately described by the 2nd order term ( [ eq : d2 ] ) . non - zero terms @xmath185 mean that the charge of the electron is modified by the interaction . actually , this modification is infinite , but even finite values of the terms like ( [ eq:11.58 ] ) are inconsistent with the classical limit . so , we will postulate that in orders higher that 2nd , singular coefficient functions like ( [ eq:11.58 ] ) should not be present at all , whether they are infinite or finite . this is the _ charge renormalization condition_. to eliminate the divergent contribution ( [ eq:11.58 ] ) we can choose unphys 3rd order counterterms @xmath186 . two diagrams describing these counterterms are shown in fig . 5(a ) and 5(b ) . as expected , they are infinite , and expressions @xmath187 in ( [ eq:11.52 ] ) exactly cancel unwanted infinite diagrams 4(a ) and 4(b ) on the energy shell . surviving diagrams 4(c ) - 4(g ) are the 4th order _ radiative corrections _ [ rad - corr ] to the electron - electron scattering . on the energy shell they yield the following contribution to the coefficient function of @xmath32 @xmath188 where @xmath189 the integrand in ( [ eq:11.59 ] ) is proportional to @xmath190 , and the integral is convergent at large values of the loop momentum @xmath180 . therefore @xmath191 is finite on the energy shell . so , in the renormalized theory the electron - electron scattering is represented by a finite @xmath0-operator which up to the 4th order is described by the coefficient function @xmath192 where @xmath193 is given by eq . ( [ eq : d2 ] ) and @xmath194 is given by eq . ( [ eq:11.59 ] ) . [ charge - end ] the renormalization technique presented above has not been applied to realistic theories , such as qed , yet . however , the close analogy between renormalization issues encountered in our toy model and in qed allows us to speculate that similar renormalization steps can be repeated in all perturbation orders in qed , so that finite and accurate values for various scattering amplitudes can be obtained . note that in our approach the counterterms in the hamiltonian ( [ eq : ren - ham ] ) have the same operator structure as the terms in the original hamiltonian ( [ eq : org - ham ] ) . in relativistic renormalizable theories , like qed , even stronger statements can be made : the mass renormalization counterterms differ from the one - particle energy terms in @xmath25 only by a constant ( but infinite ) factor , and the charge renormalization counterterms are equal to the interaction term @xmath21 in the original hamiltonian multiplied by a constant infinite factor . this conclusion is true in all perturbation orders . thus , if the original hamiltonian depends on finite masses @xmath195 , @xmath196 and coupling constants ( charges ) @xmath197 , @xmath198 @xmath199 then the addition of renormalization counterterms is equivalent to simply modifying ( making them infinite ) the values of masses and charges . the hamiltonian @xmath158 after renormalization has the same functional form as @xmath4 where parameters @xmath195 , @xmath196 and @xmath197 , @xmath198 substituted by renormalized ( infinite ) values @xmath200 , @xmath201 and @xmath202 , @xmath203 @xmath204 the traditional approach offers the following physical interpretation of these results . the creation and annihilation operators present in the theory describe so - called bare particles having infinite masses and charges . the interaction between bare particles drastically change their properties . for example , bare electrons constantly emit and re - absorb virtual photons and electron - positron pairs . so , bare electrons are surrounded by a coat of virtual particles . the mass of this coat is infinite , thus compensating the infinite mass of the bare particle . the virtual particles in the coat shield the true ( infinite ) charge of the bare electron . the resulting mass and charge of the dressed particle is finite and exactly equal to the parameters measured in experiment . in experiment , we never see the bare particles and their virtual coats , we only see the dressed particles . the hamiltonian @xmath158 is formally infinite , but these infinities cancel when the @xmath0-operator is calculated by formula ( [ eq:8.68 ] ) , and accurate results are obtained for scattering amplitudes and energies of bound states . however , the infinities do not cancel when one tries to find the eigenvectors of the hamiltonian @xmath158 or calculate the time evolution operator @xmath205 . direct application of this formula would lead to unphysical results . for example , if we calculated the time evolution of the simplest one - electron state we would obtain that this states dissociates into a complex linear combination of states over time @xmath206 moreover the coefficients multiplying the terms in ( [ eq : ihct ] ) are given by divergent integrals . therefore , with the infinite hamiltonian @xmath158 the basic formulas of quantum theory ( [ eq:1 ] ) and ( [ eq:8.60 ] ) become useless for practical calculations . however , in most cases this is not a cause of trouble . as discussed in section [ sc : bound - states ] , most experiments in high energy physics are concerned either with bound state energies or with scattering cross - sections . to calculate these properties , the knowledge of the @xmath0-operator is sufficient and the renormalized theory works fine . for approximate description of the time evolution in low energy , e.g. , atomic , systems , one can use heuristic approaches , such as those based on the dirac - fock hamiltonian . we see two fundamental problems with the traditional renormalization theory presented above . first , from the practical point of view , the absence of a well - defined finite hamiltonian does not allow one to study the time evolution of interacting states . one can expect that with advancement of experimental tools , the time - dependent information from the region of interaction will soon become available . however , accurate analysis of this information is not possible with ill - defined hamiltonians of renormalized quantum field theories . ) . ] second , from the theoretical point of view , the traditional approach operates with notions of bare and virtual particles that are , in principle , non - observable . it is desirable to have a theory formulated directly in terms of dressed particles and their interactions without mentioning the non - observable bare and virtual particles at all . this idea was first realized in the dressed particle formalism by greenberg and schweber @xcite . in the rest of this paper we will discuss the rqd approach @xcite which combines the similarity renormalization and dressed particle ideas . rqd is capable to fix the both problems of the traditional renormalization formalism mentioned above . let us first focus on the dressing part of rqd . there are two approaches to dressing with different interpretations but equivalent physical results . one approach @xcite tries to find an explicit unitary ( dressing ) transformation connecting creation and annihilation operators of bare particles ( @xmath207 , and @xmath208 ) with creation and annihilation operators of dressed particles ( @xmath209 , and @xmath83 ) and find a function @xmath210 which expresses the hamiltonian @xmath158 of the renormalized theory through dressed operators @xmath211 another approach @xcite is to pretend that operators @xmath207 , and @xmath208 already describe the dressed particles and try to find a new finite hamiltonian @xmath212 whose functional dependence on bare particle operators is given by the same function @xmath210 which expresses the dependence of @xmath158 on the dressed operators @xmath213 we will stick to the second approach in this paper . there are two ways to proceed . first , we can try to find @xmath212 by applying a unitary dressing transformation to @xmath158 . this way was described in ref . @xcite . in this paper we are choosing another route : we will simply fit the hamiltonian @xmath212 to the finite scattering operator @xmath214 known from the renormalized theory . there are three requirements that we want to satisfy when looking for the `` dressed particle '' hamiltonian @xmath215 . * @xmath212 is _ scattering - equivalent _ to @xmath158 . * @xmath216 is _ finite_. * @xmath216 is _ phys_. the condition ( a ) is understandable as we know that the @xmath0-operator of renormalized theories ( qed , standard model ) agrees well with experiments , and we would like to preserve this agreement in a theory with the new hamiltonian @xmath212 . using eqs . ( [ eq:8.71 ] ) and ( [ eq:8.72 ] ) we can write @xmath217 } \ldots \nonumber\end{aligned}\ ] ] then we see that condition ( a ) implies the following infinite set of relations between @xmath218 and @xmath219 on the energy shell is preferable to the old - fashioned or dyson s formulas , because the right hand sides of eqs . ( [ eq:12.4 ] ) - ( [ eq:12.7 ] ) are expressed through commutators , so they are manifestly hermitian . in addition , this approach automatically generates cluster separable interactions @xmath220 ( see ref . @xcite ) . ] @xmath221 } \label{eq:12.6}\\ \underbrace{v_i^d(t ) } & = & \underbrace{f_i^c(t ) } + \underbrace{q_i(t ) } , \mbox { } i > 4 \label{eq:12.7}\end{aligned}\ ] ] where @xmath222 denotes a sum of multiple commutators of @xmath223 from lower orders ( @xmath224 ) with @xmath10-integrations . we have expressions for operators @xmath225 and @xmath191 in the renormalized theory ( eqs . ( [ eq : d2 ] ) and ( [ eq:11.59 ] ) ) . clearly , we can perform similar calculations for operators @xmath219 whose values on the energy shell are present on the right hand sides of eqs . ( [ eq:12.4 ] ) - ( [ eq:12.7 ] ) . alternatively , we can express @xmath226 through @xmath151 using relationship ( [ eq:8.74 ] ) . operators @xmath219 are , of course , finite . this immediately implies that @xmath227 and @xmath228 are finite on the energy shell . moreover , from eqs . ( [ eq:8.74 ] ) and ( [ eq:11.43 ] ) it follows that @xmath229 and @xmath230 are phys on the energy shell . therefore , @xmath231 and @xmath232 can be also chosen phys on the energy shell . this proves that conditions ( b ) and ( c ) are satisfied on the energy shell up to the 3rd order . however , eqs . ( [ eq:12.4 ] ) - ( [ eq:12.7 ] ) tell us nothing about the behavior of @xmath227 and @xmath228 off the energy shell . the same is true for interactions in higher orders : the interaction operators @xmath220 off the energy shell remain undetermined by our condition ( a ) . this freedom in choosing interactions simply means that different choices of the coefficient functions of @xmath220 off the energy shell result in scattering equivalent hamiltonians . since our experimental knowledge about interactions in high energy physics is limited to their effects on scattering , we are allowed to freely choose the behavior of @xmath220 off the energy shell without any danger to get into contradiction with experiment . this freedom is exactly what is needed to satisfy conditions ( b ) and ( c ) in all orders . as we will see shortly , it is important to choose the behavior of @xmath218 such that their coefficient functions fall off rapidly when the arguments move away from the energy shell . is diagonal , this condition is equivalent to bringing the matrix of the hamiltonian to the band - diagonal form , which is the central idea of the similarity renormalization method @xcite . ] for example , we can choose the coefficient functions being proportional to @xmath233 where @xmath234 is a positive constant and @xmath2 is the energy function . to infinity . in this case , the coefficient functions of interaction operators @xmath235 become non - differentiable . this means that interaction is not separable @xcite , and the scattering theory formalism from section [ sc : bound - states ] is no longer applicable . ] then the electron - electron interaction in the second order takes the form @xmath236^{ph } \circ \zeta_2 \nonumber \\ & = & \frac { e^2 } { ( 2 \pi ) ^3 } \int d \mathbf{p } d \mathbf{q } d \mathbf{k } \frac { e^{- \gamma e^2}e^{-it e } } { |\mathbf{k}|(|\mathbf{k}| + \omega_{\mathbf{p - k } } - \omega_{\mathbf{p } } ) } a^{\dag}_{\mathbf{p - k } } a^{\dag}_{\mathbf{q+k } } a_{\mathbf{q}}a_{\mathbf{p } } \label{eq : el - el}\end{aligned}\ ] ] where @xmath237 is the energy function . there are no terms of the type @xmath173 in the 3rd order term @xmath238 . therefore the lowest ( 4th ) order radiative correction to the electron - electron interaction ( [ eq : el - el ] ) should be obtained from eq . ( [ eq:12.6 ] ) . as discussed above , the value of @xmath239 on the energy shell is finite . let us now consider the term @xmath240 } \label{eq:12.10}\end{aligned}\ ] ] on the right hand side of eq . ( [ eq:12.6 ] ) . first , we note that , according to table 1 , this commutator is phys . second , as we agreed above , the coefficient functions of @xmath227 fall off rapidly outside the energy shell . without this condition , the loop integrals encountered in calculations of ( [ eq:12.10 ] ) could be divergent . in our case these integrals are convergent : when the loop integration momentum goes to infinity , the @xmath227 factors in the integrand go away from the energy shell , i.e. , rapidly fall off . this guarantees that interaction @xmath241 is phys and finite on the energy shell . its coefficient function off the energy shell should be again chosen to decay rapidly to ensure the convergence of loop integrals in higher order operators @xmath222 , where @xmath241 may contribute . these arguments can be repeated in higher orders , which proves that the dressed particle hamiltonian @xmath212 is free of ultraviolet divergences . the contribution @xmath242 on the right hand side of eq . ( [ eq:12.6 ] ) is well - defined near the energy shell , but this is not true for the contribution @xmath243 $ ] . this commutator depends on the behavior of @xmath244 everywhere in the momentum space . so , it substantially depends on our choice of @xmath245 off the energy shell . there is a great freedom in this choice which is reflected in the uncertainty of @xmath246 even on the energy shell . it is interesting to note that although the off - shell behavior of the 2nd order interaction and the on - shell behavior of the 4th order interaction can not be separately determined in our theory , they are connected to each other in such a way that the ambiguity of the interaction does not affect the @xmath0-matrix for the electron - electron scattering . the above construction does not allow us to obtain full information about @xmath216 : the off - shell behavior of interactions is rather arbitrary , and the on - shell behavior can be determined only for lowest order terms . however , this uncertainty is perfectly understandable : it simply reflects the one - to - many correspondence between the @xmath0-operator and hamiltonians ( see section [ sc : bound - states ] ) . it means that there is a class of finite phys interactions @xmath247 all of which satisfy our requirements ( a ) - ( c ) and can be used for @xmath0-matrix calculations without encountering divergent integrals . in order to find the unique hamiltonian correctly describing the dynamics of particles , the theoretical predictions should be compared with time - resolved experimental data . however , our model theory is not sufficiently accurate to be comparable with experiment , and further efforts in this direction require building the dressed particle formulation of the full - blown quantum electrodynamics @xcite . @lll operator & physical meaning & perturbation & & orders & * elastic potentials & @xmath173 & electron - electron & @xmath248 @xmath249 & electron - photon ( compton ) & @xmath250 @xmath251 & 3-electron potential & @xmath252 & * inelastic potentials & @xmath253 & bremsstrahlung in electron - electron collisions & @xmath254@xmath255 & photon absorption in electron - electron collisions & @xmath254 * * in contrast to the original hamiltonian ( [ eq : org - ham ] ) , there seems to be no way to write the dressed particle hamiltonian @xmath212 in a closed form . from the derivation outlined above it is clear that in higher perturbation orders there are more and more terms with increasing complexity in the interaction operator @xmath216 . @xmath256 some of them are shown in table 2 . however , all these high order terms directly reflect real interactions and processes observable in nature . for example , the term @xmath253 ( _ bremsstrahlung _ ) [ bremsstr ] describes creation of a photon in electron - electron collisions . in the language of classical electrodynamics , this can be interpreted as radiation due to the acceleration during interaction of charged particles and is often referred to as the _ radiation reaction _ [ rad - res ] force . the hermitian - conjugated term @xmath255 describes absorption of a photon by a colliding pair of charged particles . note that in contrast to the traditional renormalization approach in which the mass renormalization condition ( [ eq:11.42 ] ) was ensured by maintaining a balance between unphys and renorm terms in the interaction , in the rqd hamiltonian these terms are absent altogether . the interaction operator @xmath216 is purely phys which guarantees that scattering operators @xmath257 and @xmath258 are phys too . the phys character of the interaction @xmath216 means that it yields zero when acting on the vacuum and one - particle states . interaction @xmath216 acts only when there are two or more dressed particles present . this is equivalent to saying that self - interaction effects are not present in the dressed particle approach . naturally , there should be no dressing in our theory , because we are working with particles which are already fully dressed . in this paper , we considered a model theory which provides a simplified description of interactions between electrons and photons , similar to exact interactions in qed . we found that in order to get sensible results for the @xmath0-matrix , the theory should be renormalized , just as qed , by adding infinite counterterms to the hamiltonian . we also demonstrated that the renormalized version of the theory can be reformulated entirely in terms of dressed particles and their interactions without affecting predictions of the theory about scattering cross - sections or bound state energies . the dressed particle hamiltonian @xmath212 is finite , thus , the ultraviolet divergences are not present anymore . there are no bare and virtual particles in the dressed particle approach . this simplifies substantially the physical interpretation of the theory . from the operator form of the dressed particle hamiltonian ( [ eq : dress ] ) it is clear that dressed particles interact with each other via instantaneous forces that generally do not conserve the number of photons i am thankful to drs . igor khavkine , arnold neumaier , and dan solomon for helpful comments , discussions , and criticism and to dr . arnold neumaier for drawing my attention to ref . @xcite and related works . w. magnus , _ on the exponential solution of differential equations for a linear operator _ , commun . pure appl . math . * 7 * ( 1954 ) , 649 ; p. pechukas and j. c. light , _ on the exponential form of time - displacement operators in quantum mechanics _ , j. chem * 44 * ( 1966 ) , 3897 . a. krger and w. glckle , _ one - nucleon effective generators of the poncar group derived from a field theory : mass renormalization _ , phys . c * 60 * ( 1999 ) , 024004 ; m. i. shirokov , _ mass renormalization using noncovariant perturbation theory _ , preprint jinr p2 - 2000 - 277 , dubna ( 2000 ) . a. v. shebeko and m. i. shirokov , _ unitary transformations in quantum field theory and bound states _ , fiz . yadra * 32 * ( 2001 ) , 31 [ english translation in phys . part . * 32 * ( 2001 ) , 15 ] , nucl - th/0102037 .

the search for star clusters in the galaxy is of a great interest for investigators in the recent years . on the one hand , the usage of new methods and instruments in observations allows scientists to involve clusters in solving a larger number of astrophysical problems and raises many new ones , such as : the processes of formation and evolution of young massive clusters , the nature of nuclear star clusters , the presence of multiple populations in globular clusters . on the other hand , the availability of new multiband all - sky surveys stimulates us to search them for new clusters and build a homogeneous catalog of parameters of both newly discovered and already known clusters . in paper i ( koposov et al . 2008 ) , we described the new method of an automated search for star clusters as a density peaks in huge stellar catalogs . it is based on the convolution of density maps with a special 2-d filter , which is the difference between two 2-d gaussian profiles and has zero integral . if convolved with this special filter , the areas of flat or slowly changing background would produce zero signal , whereas the areas of star concentrations would exhibit a high signal . using this method , we analyzed the distribution of stars in two micron all sky survey ( 2mass ) in the field of @xmath3 degrees towards the galactic anticenter and found 15 new open clusters . to verify the reality of detected clusters , we developed a method , which is based on the assumption that probable cluster members lying along the same isochrone on the color - magnitude diagram ( cmd ) should also form the spatial density peak , whereas the background stars should have a flat distribution . the other benefit of this method is that it not only allows to verify a cluster candidate , but at the same time estimates the cluster s main parameters distance , age , and color excess . we employed this method to derive parameters of 12 new and 13 known , but poorly studied clusters , which were also detected in the field of interest , using the technique of @xmath4 and @xmath5 diagrams built with the data from 2mass catalog . later , @xmath6 magnitudes of the stars in three new clusters were measured using ccd images taken at 104-cm telescope of aryabhatta research institute of observational sciences ( aries , india ) . the isochrones fitted to these data using another color - magnitude diagrams , and derived clusters parameters ( glushkova et al . 2009 ) show good agreement with those obtained from @xmath4 and @xmath5 2mass diagrams and therefore independently confirm the reality of clusters found by koposov et al . ( 2008 ) and the correctness of the method of new cluster verification . the aims of this study are : ( 1 ) search for overdensities in the galactic plane in the range of the galactic latitude @xmath7 using 2mass data ; ( 2 ) verification of some of them as a real clusters ; ( 3 ) estimate of their physical parameters using proven technique we have developed earlier . for the new clusters search and verification procedures we made use of http://vo.astronet.ru , the virtual observatory ( vo ) resource of the sternberg astronomical institute ( koposov et al . 2007 ) , which provides fast remote access to 2mass catalog data through the standard vo - compliant interfaces . we detected 11186 overdensities with the significance level of more than @xmath8 in the region of the milky way within the interval of the galactic latitude @xmath1 . since the open clusters subsystem concentrates to the galactic plane where the interstellar extinction reaches maximum and the background changes rapidly , the most from 11186 overdensities should be attributed to the fluctuations of interstellar dust especially towards the galactic center , where the density fluctuations are maximal . we undertook a visual inspection of all detected overdensities by images from digitized sky survey ( dss ) and 2mass catalogs to recognize real density concentrations against these background fluctuations . we selected 962 candidates to clusters and matched them to the online catalogs by dias et al . ( 2002 ) and harris ( 2003 ) . this match demonstrated that 565 candidates are known open clusters , 114 are known globular clusters , whereas 283 overdensities should be examined in detail as a new cluster candidates , some of which may only be attributed to random star concentrations . we ran all these 283 overdensities through our verification procedure , which includes the analysis of their hess - diagrams , color - magnitude diagrams and radial density distributions in accordance with the technique by koposov et al . 149 candidates were decided to be the real star clusters . however , a part of non - selected overdensities may also be real clusters , which we were unable to verify , because they were detected at the sensitivity limit of 2mass catalog . if we extend our data pattern with the data from the sources with a higher limit in @xmath9 magnitudes , then some other overdensities may manifest themselves as real clusters , and their parameters can also be determined . we tested this idea by using galactic plane survey of ukirt infrared deep sky survey , data release 3 ( ukidss gps dr3 , warren et al . 2007 ) as an additional data source . the average 2mass sensitivity in @xmath10 is @xmath11 , whereas in ukidss gps , the magnitude limit for @xmath12-band is as high as @xmath13 . 22 cluster candidates found in 2mass were studied by @xmath4 and @xmath14 diagrams and radial density distributions built using data from ukidss gps . we verified 9 star clusters and determined their distance moduli , ages and color excesses . among these nine objects , four ( sai 50 , sai 131 , sai 133 , and sai 141 in table 1 ) did not reveal themselves as star clusters when only 2mass catalog data were employed , because the upper part of their main sequences and/or red giant branches is only seen on cmd and hess - diagrams . for example , left panel of fig . 1 displays @xmath4 diagram of cluster sai 50 built by data from ukidss gps . one can only see here the red - giant branch limited to the stars fainter than @xmath15 . if we build such a cmd for this cluster with 2mass data only , we would not detect the cluster at all , because the limiting magnitude for @xmath16-band in this catalog is exactly @xmath17 . in fig . 1 , the isochrone is fitted to give the lower estimate of the distance to the cluster . thus , after involving ukidss gps data , the number of confirmed clusters reached 153 . for 130 clusters we automatically obtained main physical parameters : ages , distances and color excesses using the data from 2mass or ukidss gps and isochrones of solar metallicity by girardi et al . @xmath9 magnitudes were taken from the deeper ukidss gps for the nine clusters , whose centers fall into this survey : sai 50 , sai 74 , sai 75 , sai 130 , sai 131 , sai 133 , sai 141 , sai 142 , sai 145 . the center coordinates , diameters , color excesses @xmath2 , distance moduli and ages as @xmath18 are listed in table 1 . we consider the position of the maximum in a density peak as the center of a cluster on the density map . the cluster radius is such a distance from its center , at which the star density becomes flat on the radial density distribution plot . the errors in color excesses , distance moduli and ages were estimated as described by koposov et al . ( 2008 ) : from the differences in the parameters derived by @xmath4 and @xmath5 diagrams . we used the relations @xmath19 , @xmath20 , @xmath21 from the paper by dutra et al . ( 2002 ) to derive the distance modulus and color excess @xmath2 , and the relation @xmath22 from the paper by he et al . ( 1995 ) for @xmath12 magnitudes from ukidss gps . for some clusters , we are able to estimate the upper limit of the age only , in particular , when their color - magnitude diagrams do not contain red - giant stars . besides the cluster name sai ( _ _ s__ternberg _ _ a__stronomical _ _ i__nstitute ) , other names from papers by kronberger et al . ( 2006 ) and by froebrich et al . ( 2007 ) are used in table 1 . as mentioned by koposov et al . ( 2008 ) , the authors of both papers published the list of probable cluster candidates where further investigation is necessary to clarify their nature . some of the overdensities independently found by us coincide with the cluster candidates from papers by kronberger et al . ( 2006 ) and froebrich et al . ( 2007 ) . we performed a thorough analysis of these matching candidates and confirmed some of them as the real clusters . however , remaining candidates turned to be mere random star concentrations or background fluctuations ( we do not publish this list here ) . all clusters from table 1 having the other name according to kronberger et al . ( 2006 ) or froebrich et al . ( 2007 ) , are listed neither in the database of open clusters ( webda ) by paunzen , mermilliod ( 2009 ) , nor in the catalog by dias et al . that is why we consider them as a new clusters . special attention should be given to the cluster sai 92 from table 1 . this cluster is situated @xmath23 to the south - east of ngc 2645 , and both clusters feature approximately the same values of parameters . according to webda , the distance from sun to ngc 2645 equals to 1668 pc ; we estimate the distance to the cluster sai 92 to be @xmath24 pc . color excess @xmath2 in the direction to ngc 2645 is @xmath25 , and to sai 92 , @xmath26 . both clusters are pretty young , but differ in @xmath18 by 0.72 : webda estimates the age of ngc 2645 as 7.283 , whereas the present study gives @xmath27 for sai 92 . however , 0.05 is the formal error , which , as mentioned earlier , should be attributed to the difference in the ages determined by @xmath4 and @xmath5 diagrams . a real error for age estimate by 2mass is not less than 0.3 for clusters which do not contain stars on the red - giant branch . if we build the @xmath28 diagram of ngc 2645 using data from webda and fit them by the isochrone of solar metallicity by girardi et al . ( 2002 ) using the distance and color excess also from webda , then the lower limit for the age is 7.6 in @xmath18 . therefore , the age of both clusters can be considered approximately the same taking into account the real errors of its estimate . apparently , ngc 2645 and sai 92 form a double cluster then . 2 shows the image of @xmath29 x @xmath29 from dss where both clusters fall into . diameters of clusters ( @xmath30 and @xmath31 ) are represented by circles . sai 50 looks like a globular cluster by its age and cmd built using data from ukidss gps , although the errors in parameters are large , because the isochrone was fitted to the red - giant branch only . the left panel of fig . 1 represents @xmath4 diagram with fitted isochrone ; the radial density distribution corresponding to this fitted isochrone is displayed in the right panel . solid circles denote the cluster members that are the stars deviating from the isochrone by less than @xmath32 in color ; open circles denote background stars , that are all other stars . it can clearly be seen that the members of the cluster are concentrated to its center , whereas field stars exhibit flat distribution with small density fluctuations . besides 130 clusters which are listed in table 1 , we found 23 new embedded clusters . they are detected as density peaks and are clearly visible on 2mass images and hess - diagrams . but such a clusters reveal themselves as a cloud on the cmd @xmath4 and @xmath5 ; therefore , it was impossible to fit isochrones and find their parameters . coordinates and diameters of these clusters are presented in table 2 . [ meansp ] l|l|c|c|c|c|c|c & other name & @xmath33 & @xmath34 & d & e(b - v ) & @xmath35 & age + & & @xmath36 @xmath37 @xmath38 & @xmath39 @xmath40 @xmath41 & @xmath40 & mag & mag & @xmath18 + sai 1 & & 00:08:20.4 & + 51:43:15 & 4 & 0.34@xmath420.14 & 11.68@xmath420.06 & 9.10@xmath420.05 + sai 2 & & 00:12:28.4 & + 76:14:06 & 5 & 0.81@xmath420.27 & 12.45@xmath420.26 & @xmath437.70 + sai 3 & fsr 480 & 00:14:49.6 & + 61:28:36 & 6 & 0.68@xmath420.02 & 14.18@xmath420.05 & 9.20@xmath42 0.05 + sai 4 & & 00:23:40.0 & + 62:42:14 & 4 & 0.27@xmath42 0.18 & 11.02@xmath42 0.36 & @xmath438.70 + sai 5 & fsr 494 & 00:25:37.8 & + 63:45:40 & 4 & 0.60@xmath42 0.08 & 12.70@xmath42 0.07 & 9.10@xmath42 0.05 + sai 6 & & 00:27:52.6 & + 60:41:21 & 5 & 0.89@xmath42 0.09 & 12.17@xmath42 0.01 & @xmath438.50 + sai 7 & fsr 503 & 00:29:11.3 & + 68:56:11 & 7 & 0.82@xmath42 0.12 & 11.88@xmath42 0.38 & 9.40@xmath42 0.05 + sai 8 & fsr 505 & 00:40:08.8 & + 61:34:03 & 3 & 0.30@xmath42 0.11 & 14.28@xmath42 0.20 & 9.15@xmath42 0.10 + sai 9 & fsr 508 & 00:41:51.9 & + 64:58:51 & 4 & 0.66@xmath42 0.17 & 14.42@xmath42 0.20 & 9.10@xmath42 0.10 + sai 11 & fsr 524 & 00:57:12.4 & + 62:06:18 & 8 & 0.51@xmath42 0.13 & 10.70@xmath42 0.29 & 9.10@xmath42 0.05 + sai 12 & & 01:04:24.1 & + 45:36:25 & 4 & 1.13@xmath42 0.30 & 12.06@xmath42 0.02 & @xmath438.15 + sai 13 & fsr 536 & 01:19:37.6 & + 63:03:20 & 8 & 1.27@xmath42 0.23 & 12.38@xmath42 0.17 & @xmath438.20 + sai 14 & & 01:26:00.3 & + 62:37:33 & 4 & 0.55@xmath42 0.03 & 14.05@xmath42 0.03 & 9.10@xmath42 0.05 + sai 16 & & 02:05:29.6 & + 62:15:54 & 5 & 0.70@xmath42 0.01 & 14.42@xmath42 0.20 & 9.15@xmath42 0.05 + sai 17 & & 02:20:48.3 & + 59:11:06 & 2 & 0.98@xmath42 0.17 & 12.88@xmath42 0.10 & 8.90 @xmath42 0.05 + sai 21 & & 02:34:45.4 & + 62:34:22 & 3 & 0.58@xmath42 0.01 & 11.35@xmath42 0.10 & @xmath438.50 + sai 25 & & 03:00:26.7 & + 57:16:02 & 7 & 0.84@xmath42 0.13 & 11.80@xmath42 0.09 & 9.15@xmath42 0.05 + sai 26 & & 03:09:30.8 & + 56:47:26 & 3 & 1.47@xmath42 0.29 & 11.79@xmath42 0.51 & @xmath438.20 + sai 27 & & 03:11:21.9 & + 68:54:01 & 5 & 0.62@xmath42 0.04 & 10.48@xmath42 0.28 & 9.25@xmath42 0.10 + sai 29 & fsr 634 & 03:40:21.5 & + 59:24:52 & 5 & 0.84@xmath42 0.01 & 12.24@xmath42 0.07 & 9.00@xmath42 0.05 + sai 30 & & 03:40:22.2 & + 82:13:60 & 3 & 1.25@xmath42 0.44 & 12.46@xmath42 0.05 & @xmath438.00 + sai 31 & & 03:51:15.4 & + 58:46:08 & 6 & 0.16@xmath42 0.22 & 12.12@xmath42 0.02 & 9.60@xmath42 0.05 + sai 33 & juchert 19 & 04:02:20.2 & + 52:26:39 & 2 & 0.28@xmath42 0.01 & 10.64@xmath42 0.02 & @xmath439.30 + sai 34 & fsr 660 & 04:02:37.8 & + 51:55:50 & 5 & 1.20@xmath42 0.12 & 13.78@xmath42 0.03 & 9.10@xmath42 0.05 + sai 35 & juchert 20 & 04:10:47.0 & + 46:52:01 & 5 & 0.70@xmath42 0.18 & 11.92@xmath42 0.02 & @xmath438.55 + sai 36 & & 04:16:50.4 & + 41:04:00 & 3 & 0.55@xmath42 0.05 & 12.36@xmath42 0.19 & 9.10@xmath42 0.05 + sai 37 & fsr 687 & 04:39:27.1 & + 48:12:23 & 4 & 0.77@xmath42 0.03 & 14.30@xmath42 0.04 & 9.15@xmath42 0.05 + sai 40 & & 05:00:56.9 & + 41:14:02 & 8 & 0.35@xmath42 0.12 & 13.29@xmath42 0.05 & 8.75@xmath42 0.05 + sai 41 & & 05:07:05.4 & + 38:25:05 & 2 & 0.51@xmath42 0.18 & 13.58@xmath42 0.23 & 9.30@xmath42 0.05 + sai 42 & juchert 23 & 05:07:39.7 & + 17:36:00 & 8 & 0.62@xmath42 0.18 & 11.67@xmath42 0.09 & 8.70@xmath42 0.05 + sai 43 & & 05:08:16.6 & + 49:52:08 & 2 & 0.18@xmath42 0.03 & 12.92@xmath42 0.07 & 8.95@xmath42 0.05 + sai 44 & fsr 716 & 05:11:07.4 & + 45:43:09 & 5 & 0.24@xmath42 0.01 & 12.64@xmath42 0.06 & 8.95@xmath42 0.05 + sai 45 & fsr 727 & 05:16:35.0 & + 45:34:56 & 6 & 0.25@xmath42 0.02 & 10.55@xmath42 0.17 & 9.20@xmath42 0.05 + sai 46 & & 05:19:37.0 & + 36:30:32 & 3 & 0.78@xmath42 0.01 & 13.16@xmath42 0.12 & 9.45@xmath42 0.05 + sai 47 & & 05:23:58.0 & + 42:18:52 & 3 & 0.60@xmath42 0.01 & 12.93@xmath42 0.05 & 8.50@xmath42 0.05 + sai 48 & & 05:24:16.8 & + 33:30:02 & 4 & 0.31@xmath42 0.07 & 11.30@xmath42 0.20 & 9.35@xmath42 0.05 + sai 49 & & 05:26:24.9 & + 50:47:03 & 6 & 0.47@xmath42 0.01 & 12.17@xmath42 0.17 & 9.05@xmath42 0.05 + sai 50 & & 05:28:38.9 & + 35:01:19 & 3 & 0.60@xmath42 0.30 & 16.70@xmath42 0.50 & 9.70@xmath42 0.20 + sai 51 & & 05:30:01.9 & + 33:25:31 & 3 & 0.62@xmath42 0.18 & 9.54@xmath42 0.20 & @xmath43 9.30 + sai 56 & fsr 852 & 05:53:30.6 & + 25:10:41 & 8 & 0.43@xmath42 0.07 & 11.82@xmath42 0.04 & 8.90@xmath42 0.05 + sai 57 & fsr 932 & 06:04:21.5 & + 14:34:22 & 5 & 0.39@xmath42 0.07 & 10.18@xmath42 0.10 & 8.90@xmath42 0.10 + sai 58 & fsr 921 & 06:05:09.4 & + 16:41:03 & 5 & 1.33@xmath42 0.33 & 11.80@xmath42 0.20 & 8.50@xmath42 0.05 + sai 59 & fsr 942 & 06:05:59.2 & + 13:39:52 & 8 & 0.59@xmath42 0.01 & 11.51@xmath42 0.25 & 9.00@xmath42 0.05 + sai 60 & & 06:06:38.6 & + 15:56:51 & 4 & 0.90@xmath42 0.01 & 14.50@xmath42 0.05 & 9.05@xmath42 0.05 + sai 61 & fsr 904 & 06:06:55.2 & + 19:00:26 & 8 & 0.14@xmath42 0.16 & 10.51@xmath42 0.02 & 8.80@xmath42 0.05 + sai 62 & fsr 985 & 06:11:49.0 & + 07:01:22 & 3 & 0.55@xmath42 0.02 & 12.10@xmath42 0.09 & @xmath43 8.60 + sai 63 & fsr 987 & 06:13:44.5 & + 06:56:58 & 4 & 0.37@xmath42 0.13 & 11.40@xmath42 0.09 & 8.65@xmath42 0.05 + sai 64 & fsr 948 & 06:25:55.9 & + 15:51:39 & 4 & 0.08@xmath42 0.13 & 11.84@xmath42 0.17 & 9.00@xmath42 0.05 + sai 65 & fsr 979 & 06:31:16.2 & + 11:04:38 & 3 & 1.17@xmath42 0.17 & 11.65@xmath42 0.05 & 8.45@xmath42 0.05 + sai 66 & fsr 1063 & 06:34:37.9 & -00:16:01 & 5 & 0.58@xmath42 0.01 & 11.47@xmath42 0.14&@xmath43 8.60 + sai 67 & teutsch59 & 06:43:49.0 & -00:52:60 & 6 & 0.57@xmath42 0.14 & 12.34@xmath42 0.04 & @xmath43 8.90 + sai 68 & patchick 90 & 06:44:42.7 & -00:32:22 & 3 & 0.41@xmath42 0.09 & 14.00@xmath42 0.20 & 8.95@xmath42 0.05 + sai 69 & & 06:45:27.8 & -11:47:40 & 3 & 0.35@xmath42 0.07 & 14.27@xmath42 0.10 & 9.10@xmath42 0.05 + sai 70 & fsr 1125 & 06:50:26.6 & -05:26:08 & 4 & 0.39@xmath42 0.01 & 13.99@xmath42 0.09 & @xmath439.10 + sai 71 & fsr 1085 & 06:53:11.2 & -00:11:31 & 4 & 0.08@xmath42 0.09 & 10.62@xmath42 0.19 & @xmath438.60 + sai 72 & & 06:55:48.4 & + 00:13:37 & 5 & 0.82@xmath42 0.06 & 12.49@xmath42 0.05 & 8.50@xmath42 0.20 + sai 73 & fsr 1171 & 07:00:13.1 & -10:19:14 & 2 & 0.65@xmath42 0.07 & 13.07@xmath42 0.29 & 9.50@xmath42 0.05 + sai 74 & fsr 1150 & 07:04:36.6 & -06:36:38 & 3 & 0.94@xmath42 0.06 & 13.60@xmath42 0.26 & @xmath438.30 + sai 75 & patchick 79 & 07:15:22.5 & -07:25:20 & 4 & 0.23@xmath42 0.03 & 12.24@xmath42 0.24 & @xmath438.60 + sai 76 & fsr 1253 & 07:26:17.4 & -18:25:55 & 6 & 0.62@xmath42 0.01 & 12.11@xmath42 0.24 & 8.40@xmath42 0.05 + sai 77 & fsr 1275 & 07:28:06.2 & -21:47:46 & 4 & 0.60@xmath42 0.10 & 12.87@xmath42 0.12 & 9.40@xmath42 0.05 + sai 78 & teutsch 61 & 07:34:39.8 & -19:47:23 & 3 & 0.55@xmath42 0.02 & 11.42@xmath42 0.10 & 6.95@xmath42 0.05 + sai 79 & fsr 1347 & 07:48:03.4 & -33:42:35 & 6 & 1.09@xmath42 0.15 & 11.95@xmath42 0.07 & @xmath438.80 + sai 80 & teutsch 25 & 07:48:26.7 & -27:54:51 & 5 & 0.57@xmath42 0.12 & 11.48@xmath42 0.16 & @xmath439.00 + + [ meansp ] l|l|c|c|c|c|c|c & other name & @xmath33 & @xmath34 & d & e(b - v ) & @xmath35 & age + & & @xmath36 @xmath37 @xmath38 & @xmath39 @xmath40 @xmath41 & @xmath40 & mag & mag & @xmath18 + sai 81 & & 07:52:07.2 & -28:07:21 & 7 & 0.40@xmath42 0.02 & 12.21@xmath42 0.19 & @xmath43 9.20 + sai 82 & & 07:52:14.4 & -33:02:27 & 3 & 0.39@xmath42 0.07 & 10.30@xmath42 0.05 & 8.65 @xmath42 0.05 + sai 83 & & 07:53:20.1 & -32:33:30 & 3 & 1.00@xmath42 0.15 & 12.96@xmath42 0.34 & 8.85@xmath42 0.05 + sai 84&kronberger 85 & 07:58:21.6 & -34:46:09 & 2 & 1.17@xmath42 0.19 & 14.65@xmath42 0.11 & 8.80@xmath42 0.05 + sai 85 & fsr 1378 & 08:01:11.5 & -40:40:41 & 6 & 1.21@xmath42 0.25 & 11.74@xmath42 0.13 & 8.80@xmath42 0.05 + sai 86 & & 08:08:15.0 & -36:36:33 & 6 & 0.69@xmath42 0.09 & 12.41@xmath42 0.16 & 8.60@xmath42 0.05 + sai 87 & fsr 1380 & 08:11:05.0 & -39:59:51 & 4 & 1.28@xmath42 0.30 & 11.80@xmath42 0.19 & @xmath438.65 + sai 88 & & 08:12:57.6 & -41:53:01 & 2 & 0.92@xmath42 0.02 & 14.58@xmath42 0.25 & 9.00@xmath42 0.05 + sai 89 & fsr 1387 & 08:24:25.3 & -39:56:11 & 3 & 1.19@xmath42 0.11 & 11.82@xmath42 0.18 & @xmath438.80 + sai 90 & & 08:27:58.9 & -41:46:10 & 7 & 1.02@xmath42 0.21 & 12.19@xmath42 0.11 & 9.05@xmath42 0.05 + sai 91 & & 08:37:03.4 & -50:03:52 & 4 & 0.51@xmath42 0.02 & 12.45@xmath42 0.16 & 8.80@xmath42 0.05 + sai 92 & fsr 1436 & 08:39:34.5 & -46:17:54 & 5 & 0.39@xmath42 0.10 & 10.99@xmath42 0.17 & 8.00@xmath42 0.05 + sai 93 & fsr 1415 & 08:40:21.9 & -44:44:01 & 3 & 1.87@xmath42 0.35 & 12.74@xmath42 0.06 & 9.25@xmath42 0.05 + sai 94 & & 08:44:39.8 & -46:17:46 & 5 & 0.48@xmath42 0.05 & 12.96@xmath42 0.29 & 9.10@xmath42 0.10 + sai 95 & fsr 1454 & 08:51:00.6 & -48:27:37 & 5 & 0.57@xmath42 0.03 & 13.96@xmath42 0.05 & 9.15@xmath42 0.05 + sai 96 & fsr 1430 & 08:51:49.4 & -44:15:47 & 4 & 2.50@xmath42 0.33 & 12.27@xmath42 0.33 & 8.85@xmath42 0.05 + sai 98 & fsr 1435 & 08:57:03.0 & -43:45:40 & 6 & 1.06@xmath42 0.06 & 10.27@xmath42 0.30 & 9.75@xmath42 0.10 + sai 99 & & 08:58:04.4 & -43:24:38 & 5 & 1.84@xmath42 0.49 & 12.60@xmath42 0.04 & 8.95@xmath42 0.05 + sai 100 & fsr 1450 & 08:58:12.2 & -46:17:19 & 3 & 1.62@xmath42 0.08 & 13.65@xmath42 0.08 & 9.10@xmath42 0.15 + sai 101 & & 08:58:22.4 & -43:07:35 & 5 & 1.56@xmath42 0.04 & 11.09@xmath42 0.05 & @xmath439.00 + sai 102 & fsr 1402 & 09:05:43.2 & -37:52:05 & 4 & 0.08@xmath42 0.04 & 12.03@xmath42 0.29 & 9.60@xmath42 0.05 + sai 103 & fsr 1460 & 09:07:38.5 & -47:50:01 & 7 & 1.31@xmath42 0.20 & 11.30@xmath42 0.70 & 9.30@xmath42 0.70 + sai 104 & & 09:09:54.8 & -48:51:03 & 3 & 1.17@xmath42 0.11 & 9.59@xmath42 0.14 & @xmath438.90 + sai 105 & & 09:10:24.5 & -49:54:46 & 4 & 1.91@xmath42 0.31 & 12.35@xmath42 0.13 & 9.05@xmath42 0.05 + sai 106 & & 09:17:59.4 & -51:01:46 & 5 & 1.02@xmath42 0.02 & 13.56@xmath42 0.10 & 9.35@xmath42 0.05 + sai 107 & teutsch 103 & 09:28:35.8 & -46:35:60 & 3 & 0.62@xmath42 0.06 & 12.23@xmath42 0.38&@xmath43 8.75 + sai 108 & & 09:38:52.9 & -52:57:28 & 6 & 0.66@xmath42 0.02 & 12.32@xmath42 0.08 & 8.50@xmath42 0.05 + sai 109 & & 09:43:34.4 & -50:47:33 & 5 & 0.23@xmath42 0.04 & 14.43@xmath42 0.50 & 9.20@xmath42 0.20 + sai 110 & fsr 1509 & 09:48:00.8 & -55:04:01 & 2 & 1.17@xmath42 0.13 & 11.59@xmath42 0.05 & 8.75@xmath42 0.05 + sai 111 & fsr 1521 & 09:55:22.6 & -56:36:15 & 8 & 0.90@xmath42 0.17 & 13.00@xmath42 0.30 & 9.45@xmath42 0.10 + sai 112 & & 10:21:24.9 & -59:05:34 & 2 & 1.00@xmath42 0.02 & 14.30@xmath42 0.15 & 9.10@xmath42 0.10 + sai 114 & fsr 1555 & 10:48:55.7 & -59:02:58 & 3 & 0.76@xmath42 0.05 & 13.40@xmath42 0.30 & 9.25@xmath42 0.10 + sai 115 & fsr 1586 & 11:22:44.9 & -62:19:06 & 5 & 1.60@xmath42 0.29 & 11.92@xmath42 0.20 & 8.65@xmath42 0.10 + sai 116 & & 11:49:18.0 & -62:13:47 & 5 & 0.98@xmath42 0.08 & 11.72@xmath42 0.17 & 8.60@xmath42 0.05 + sai 117 & fsr 1663 & 13:41:04.5 & -60:12:27 & 4 & 1.01@xmath42 0.16 & 11.66@xmath42 0.25 & 8.90@xmath42 0.05 + sai 118 & & 13:43:03.8 & -63:09:53 & 8 & 0.17@xmath42 0.21 & 10.28@xmath42 0.62 & 9.75@xmath42 0.25 + sai 120 & & 13:58:42.0 & -61:40:08 & 2 & 1.99@xmath42 0.30 & 12.03@xmath42 0.21 & 8.75@xmath42 0.10 + sai 121 & fsr 1686 & 14:40:06.8 & -60:23:13 & 5 & 1.48@xmath42 0.23 & 13.02@xmath42 0.16 & 8.95@xmath42 0.05 + sai 122 & & 15:00:03.9 & -58:48:13 & 8 & 2.26@xmath42 0.43 & 11.12@xmath42 0.04 & 8.25@xmath42 0.15 + sai 123 & & 16:08:17.4 & -50:32:06 & 7 & 1.35@xmath42 0.18 & 11.35@xmath42 0.18 & 9.20@xmath42 0.05 + sai 124 & fsr 1744 & 16:51:35.8 & -42:25:47 & 3 & 2.58@xmath42 0.34 & 12.48@xmath42 0.26 & 8.85@xmath42 0.20 + sai 127 & fsr 124 & 19:06:52.8 & + 13:14:44 & 4 & 0.98@xmath42 0.12 & 11.56@xmath42 0.41 & 9.10@xmath42 0.05 + sai 128 & fsr 133 & 19:29:46.9 & + 15:33:51 & 6 & 2.26@xmath42 0.45 & 11.59@xmath42 0.12 & 8.35@xmath42 0.05 + sai 129 & fsr 160 & 19:44:44.3 & + 26:54:08 & 3 & 1.42@xmath42 0.05 & 14.38@xmath42 0.30 & 8.95@xmath42 0.05 + sai 130 & fsr 158 & 19:47:58.1 & + 26:01:55 & 8 & 1.17@xmath42 0.04 & 12.78@xmath42 0.11 & @xmath43 8.30 + sai 131 & fsr 154 & 19:48:00.8 & + 23:20:53 & 3 & 0.70@xmath42 0.13 & 12.23@xmath42 0.05 & 9.05@xmath42 0.05 + sai 132 & & 19:57:01.5 & + 31:36:31 & 8 & 0.94@xmath42 0.09 & 12.28@xmath42 0.06 & 8.75@xmath42 0.05 + sai 133 & & 20:01:58.7 & + 31:26:18 & 4 & 1.80@xmath42 0.05 & 13.94@xmath42 0.16 & 7.85@xmath42 0.05 + sai 135 & & 20:02:55.9 & + 24:33:17 & 5 & 1.34@xmath42 0.29 & 12.89@xmath42 0.06 & 9.10@xmath42 0.05 + sai 136&kronberger 36&20:04:35.7 & + 35:13:03 & 6 & 0.76@xmath42 0.18 & 11.40@xmath42 0.26 & @xmath438.40 + sai 137 & & 20:49:52.6 & + 41:15:18 & 6 & 0.89@xmath42 0.08 & 10.12@xmath42 0.24 & @xmath438.70 + sai 138 & fsr 275 & 20:56:43.0 & + 46:53:15 & 10 & 0.76@xmath42 0.05 & 11.98@xmath42 0.11 & 9.25@xmath42 0.05 + sai 139 & teutsch 22 & 21:01:40.8 & + 44:06:43 & 2 & 0.86@xmath42 0.13 & 12.16@xmath42 0.33 & 8.55@xmath42 0.05 + sai 140 & fsr 282 & 21:02:20.1 & + 48:06:26 & 4 & 0.78@xmath42 0.01 & 11.79@xmath42 0.12 & 9.05@xmath42 0.05 + sai 141 & & 21:03:46.9 & + 46:58:50 & 3 & 0.94@xmath42 0.04 & 13.47@xmath42 0.16 & 8.60@xmath42 0.05 + sai 142 & teutsch 156 & 21:03:46.7 & + 47:12:54 & 3 & 0.94@xmath42 0.04 & 12.97@xmath42 0.22 & 8.95@xmath42 0.05 + sai 143 & & 21:10:51.4 & + 50:18:31 & 3 & 1.72@xmath42 0.24 & 14.07@xmath42 0.17 & 9.25@xmath42 0.05 + sai 144 & teutsch 74 & 21:45:43.6 & + 58:05:03 & 4 & 1.05@xmath42 0.17 & 12.12@xmath42 0.53 & 9.05@xmath42 0.05 + sai 145 & fsr 336 & 21:50:47.5 & + 55:16:31 & 8 & 0.55@xmath42 0.09 & 11.67@xmath42 0.37 & 8.90@xmath42 0.05 + sai 146 & fsr 342 & 22:07:38.6 & + 53:06:09 & 8 & 0.27@xmath42 0.01 & 11.52@xmath42 0.47 & 9.05@xmath42 0.05 + sai 147 & fsr 394 & 22:55:04.8 & + 58:42:53 & 5 & 0.70@xmath42 0.04 & 13.73@xmath42 0.21 & 9.40@xmath42 0.05 + sai 148 & fsr 430 & 23:15:29.5 & + 64:09:58 & 5 & 1.08@xmath42 0.20 & 11.93@xmath42 0.05 & 8.05@xmath42 0.05 + sai 149 & & 23:38:01.7 & + 60:32:57 & 7 & 1.29@xmath42 0.30 & 11.87@xmath42 0.07 & @xmath438.20 + sai 150 & fsr 441 & 23:41:58.8 & + 58:32:29 & 5 & 0.78@xmath42 0.14 & 11.84@xmath42 0.04 & 8.55@xmath42 0.05 + sai 151 & fsr 465 & 23:50:54.5 & + 66:10:49 & 5 & 0.78@xmath42 0.04 & 13.60@xmath42 0.10 & 9.20@xmath42 0.05 + sai 153 & fsr 460 & 23:59:07.0 & + 60:40:40 & 7 & 0.86@xmath42 0.19 & 13.45@xmath42 0.09 & 9.20@xmath42 0.05 + + [ meansp ] l|l|c|c|c & other name & @xmath33&@xmath34 & d + & & @xmath36 @xmath37 @xmath38 & @xmath39 @xmath40 @xmath41 & @xmath40 + sai 10 & & 00:43:03.2 & + 41:37:21 & 3 + sai 15 & & 01:49:52.6 & + 53:54:34 & 2 + sai 18 & & 02:21:55.9 & + 61:07:21 & 3 + sai 19 & & 02:26:31.9 & + 61:59:33 & 6 + sai 20 & fsr 584 & 02:27:04.8 & + 61:37:40 & 6 + sai 22 & & 02:48:58.1 & + 60:44:34 & 5 + sai 23 & & 02:54:05.0 & + 60:39:52 & 2 + sai 24 & & 02:59:25.4 & + 60:33:58 & 6 + sai 28 & & 03:39:16.7 & + 55:58:24 & 3 + sai 32 & fsr 655 & 03:56:11.8 & + 53:52:09 & 1 + sai 38 & fsr 699 & 04:56:05.6 & + 47:23:25 & 2 + sai 39 & fsr 696 & 04:58:31.6 & + 47:59:03 & 1 + sai 52 & fsr 812 & 05:38:09.1 & + 31:44:16 & 3 + sai 53 & fsr 874 & 05:40:58.3 & + 20:21:18 & 3 + sai 54 & fsr 855 & 05:42:22.4 & + 22:50:04 & 3 + sai 55 & fsr 826 & 05:42:51.2 & + 28:56:43 & 3 + sai 97 & fsr 1432 & 08:52:53.3 & -44:12:06 & 4 + sai 113 & & 10:22:43.6 & -59:30:20 & 2 + sai 119 & fsr 1662 & 13:44:14.2 & -62:04:03 & 2 + sai 125 & fsr 31 & 18:06:27.8 & -21:23:27 & 5 + sai 126 & & 18:13:26.7 & -17:52:49 & 3 + sai 134 & fsr 198 & 20:02:26.8 & + 35:40:34 & 4 + sai 152 & fsr 469 & 23:57:04.4 & + 65:24:52 & 1 + + table 3 lists the coordinates , diameters , color excesses @xmath2 , distance moduli , distances in pc , and @xmath18 of known , but yet - unstudied clusters . we detected and investigated all clusters by our methods , which simultaneously allowed us to improve the accuracy of the center coordinates . for instance , one of the overdensities was attributed to the open cluster berkeley 53 , although it is situated @xmath44 away to the west from the coordinates contained in the catalog by dias et al . webda gives the same position and the dss image shows no star concentration there . hence , be 53 was not studied previously . webda misses all @xmath45 clusters , whereas the dias catalog lists their coordinates and diameters only . @xmath45 311 - 21 is marked by dias et al . ( 2002 ) as `` doubtful '' , and @xmath45 211 - 03 , as `` not found '' . diameter of cluster @xmath45 368 - 11 is @xmath46 according to dias et al . ( 2002 ) , but we found it to be @xmath47 . diameters of other known clusters are also less then that in dias catalog as compared to the data from table 3 in this paper . this difference can be explained by the fact that all these clusters are barely perceptible on dss images , but well noticeable on images from 2mass . diameters presented in table 3 were derived from radial density distribution plots in the same manner as for newly discovered clusters . dias catalog and webda only give the coordinates and underestimated diameters for all @xmath48 clusters from table 3 . open cluster @xmath49 131 is marked as `` doubtful '' by dias et al . ( 2002 ) , @xmath50 19 has no data in the literature available . koposov et al . ( 2008 ) published the coordinates and diameters only for cluster @xmath51 7 , whereas in present work we determined its age , distance and color excess using the data from ukidss gps . [ meansp ] l|l|c|c|c|c|l|c & @xmath33 & @xmath34 & d & e(b - v ) & @xmath35 & distance & age + & @xmath36 @xmath37 @xmath38 & @xmath39 @xmath40 @xmath41 & @xmath40 & @xmath52 & @xmath52 & pc & @xmath18 + koposov 7 & 05:40:44.1 & + 35:55:25 & 6 & 3.75@xmath42 0.37 & 13.39@xmath42 0.35 & 4760@xmath42 830 & 8.20@xmath42 0.05 + eso 368 - 11&07:44:22.2 & -34:37:07 & 6 & 0.06@xmath42 0.10 & 10.86@xmath42 0.41 & 1490@xmath42 310 & 9.00@xmath42 0.05 + eso 311 - 21 & 08:01:41.6 & -41:35:57 & 5 & 0.49@xmath42 0.02 & 11.84@xmath42 0.08 & 2330@xmath42 90 & 10.10@xmath42 0.05 + ru 48 & 08:02:42.5 & -32:03:02 & 3 & 0.29@xmath42 0.04 & 12.41@xmath42 0.05 & 3030 @xmath42 70 & @xmath438.60 + ru 54 & 08:11:21.7 & -31:56:37 & 2 & 0.35@xmath42 0.04 & 12.71@xmath42 0.27 & 3480 @xmath42 460 & 8.45@xmath42 0.05 + ru 60 & 08:24:23.0 & -47:12:37 & 4 & 0.57@xmath42 0.06 & 12.84@xmath42 0.11 & 3700 @xmath42 190 & @xmath438.65 + eso 312 - 03 & 08:24:26.1 & -41:17:52 & 6 & 0.52@xmath42 0.10 & 11.68@xmath42 0.57 & 2170 @xmath42 650 & @xmath438.90 + eso 312 - 04 & 08:26:51.2 & -41:14:39 & 2 & 0.49@xmath42 0.01 & 11.91@xmath42 0.11 & 2410 @xmath42 130 & @xmath438.75 + ru 63 & 08:32:41.4 & -48:18:21 & 6 & 0.54@xmath42 0.04 & 12.13@xmath42 0.08 & 2670 @xmath42 100 & @xmath438.50 + ru 68 & 08:44:34.5 & -35:54:00 & 10 & 0.39@xmath42 0.05 & 11.49@xmath42 0.08 & 1990 @xmath42 80 & 9.15@xmath42 0.05 + eso 211 - 03 & 08:51:34.3 & -50:14:43 & 5 & 0.90@xmath42 0.07 & 12.88@xmath42 0.04 & 3770 @xmath42 70 & 9.05@xmath42 0.10 + tr 19 & 11:14:20.2 & -57:33:04 & 14 & 0.08@xmath42 0.01 & 11.40@xmath42 0.05 & 1900 @xmath42 50 & 9.65@xmath42 0.05 + bh 131 & 12:26:13.7 & -63:24:50 & 5 & 0.62@xmath42 0.04 & 13.98@xmath42 0.22 & 6250 @xmath42 670 & 9.10@xmath42 0.10 + be 53 & 20:55:55.3 & + 51:02:54 & 10 & 1.31@xmath42 0.20 & 12.74@xmath42 0.13 & 3530 @xmath42 220 & 9.15@xmath42 0.05 + + to facilitate dissemination and scientific usage of the results of the present study and data from koposov et al . ( 2008 ) , we have developed a standardized tool for continuous publication of the results of ongoing catalog compilation on the web , available as a dedicated web - site at http://ocl.sai.msu.ru . besides the standard access via a web browser to the catalog of individual cluster pages with hess - diagrams , color - magnitude diagrams , 2mass or dss cluster images , and catalogs of photometric data ( for clusters with ccd observations ) , the web - site offers the catalog exported to csv , dat , and votable formats as well and implements a standard virtual observatory programmatic access interface for positional queries called _ conesearch_. moreover , it is possible to start the vo client software , such as topcat or cds aladin , by a single click in a web browser with the catalog preloaded for quick - look analysis of the whole sample and all accompanying data we provide for individual clusters . main results are available as a single table at centre de donnees astronomiques de strasbourg at http://vizier.u-strasbg.fr/viz-bin/vizier?-source=v/132 and also within standalone vo client applications running on any workstation connected to the internet , e.g. mentioned topcat and cds aladin tools , using query to search for the catalogs by key words or by authors of this publication . using @xmath0 photometric data available in 2mass catalog , we have performed an automated search for the star overdensities in the galactic plane for the latitudes ranged within @xmath7 and detected 11186 density peaks . a short list of 283 new cluster candidates has been compiled after we excluded the overdensities attributed to the background fluctuations or star concentrations coinciding with already known clusters . employing our method described earlier ( koposov et al . , 2008 ) , we have studied these 283 candidates and verified 149 of them to be real clusters . having involved the ukidss gps survey in our investigation as an additional data source , we studied 22 more cluster candidates and found four new open clusters . therefore , the number of new clusters found in the frame of the present paper is 153 . for 130 clusters of them , we have determined their main physical parameters : distance , age , and color excess . together with the clusters published in koposov et al . ( 2008 ) , we have found 168 new open clusters in total , and determined parameters for 142 of them . besides that , we confirmed the existence of 14 known , but yet - unstudied clusters and determined their accurate coordinates and other parameters . 3 shows the distribution of 143 open clusters projected onto the galactic plane . the crosses denote all clusters , except for sai 50 , whose distances are determined in this paper ( 129 new and 14 known clusters ) . we excluded cluster sai 50 from this pattern because of large errors in its parameters . solid circle represents the sun . it can be clearly seen from this figure that no new clusters were discovered in the direction towards the galactic center , because this area features maximum extinction and level of background fluctuations , which makes the detection of clusters more difficult . the major part of new clusters have been found within the radius of 2.5 kpc from the sun , because of a relatively low limiting values of @xmath53 in the 2mass catalog . we have not discovered any new clusters closer than 500 pc ; however , we did not attempt to search for them specifically in this range of the distance , because the open cluster sample there is believed to be complete . for this reason , we used the filter allowing to detect clusters with the diameter of @xmath44 and less , whereas for the close clusters this value is typically greater . some new rich clusters can only be seen at the ir wavelengths . for example , sai 122 is well noticeable on the image from 2mass data ( fig . 4 ) and is fully invisible on the dss images , because the color excess in the direction toward this cluster amounts to @xmath54 . among the clusters discovered by us , there are few , which reveal themselves on the dss images as a very faint objects ; however , it is normally impossible to estimate their parameters . the majority of new objects are either poorly populated clusters , or rich ones , yet imperceptible against a dense background : for example , sai 132 ( see fig . 5 ) . in both these cases , an object can not be detected while visually inspecting 2mass images . thus , the application of our automated method of searching infrared data for density peaks has provided us with the tool capable of detecting these 168 open clusters . 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 . this publication has made use of the digitized sky surveys which 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 discovery of the planet around 51 pegasi ( mayor & queloz 1995 ) has spurred an avalanche of observations over the last 15 years , leading to a census of almost 700 confirmed planets to date . within this data we can find a variety of exotic and unexpected systems , such as planets with short orbital periods , extreme orbital eccentricities , or systems with strong resonant interactions . these observations have also inspired a wealth of theoretical work , on both the origin and structure of such planets . furthermore , planet searches have continued to push the observational thresholds towards lower masses , with current observations now probing below @xmath5 ( mayor et al . 2009a , b ; leger et al . 2009 ; howard et al . 2011 ; batalha et al . 2011 ) , with the hope of eventually finding planets that may convincingly be argued to be hospitable to life . despite the wealth of data and interpretation now available , it is still difficult to draw together the different elements into an overall synthesis , in large part because of the heterogeneous nature of the observations . however , in the next few years a variety of observations offer the possibility to provide new constraints , which may allow us to frame a rigorous context for planet formation and evolution . recently , howard et al . ( 2010 ) and mayor et al . ( 2011 ) published the results of volume limited radial velocity surveys of nearby sun - like stars . the importance of these studies is that one can now begin to compare the relative frequencies of different types of planets , illuminating which are the generic product of the planet formation process , and which are the exotica . the results suggest that lower mass planets ( @xmath6 ) are even more common than gas giants at short periods , and that they appear to form a distinct population for orbital periods @xmath7 days . particularly striking is the fact that the low - mass planet distribution appears to show an upper edge in a plot of mass versus orbital period ; planets are found with masses between the completeness limit of the survey and an upper limit ranging between @xmath8 at @xmath9 day and @xmath10 at @xmath11 days . this is in contrast to the results of the predominant planet population synthesis models ( e.g. ida & lin 2008 ; mordasini , alibert & benz 2009 ) , which predict that this mass range ( @xmath12 ) should be liberally populated with partially - accreted giant planet cores ( `` hot neptunes '' ) that have migrated in from larger radii . in addition to the radial velocity surveys , the kepler satellite has begun to release the contents of its large - scale transit catalogue ( borucki et al . 2011 , lissauer et al . the majority of the present sample are still unconfirmed ` candidates ' , but a variety of tests suggest that the true planetary yield from this sample is expected to be high ( borucki et al . 2011 ) . the kepler sample once again has the advantage of being drawn from a relatively homogeneous program , and the distribution of planetary radii supports the aforementioned claim that the dominant species of short period planets corresponds to radii and masses below that expected from gas giant planets ( howard et al . 2011 ; wolfgang & laughlin 2011 ; youdin 2011 ) . these new surveys highlight the importance of short - period low - mass planets , as the low - mass planets now appear to outnumber their more massive brethren . the origins of such hot neptunes and ` super - earths ' is still poorly understood . most models of their formation ( alibert et al . 2006 ; terquem & papaloizou 2007 ; raymond , barnes & mandell 2008 ; mordasini et al . 2009 ; ida & lin 2010 ; mcneil & nelson 2010 ) adopt the position that they form at larger radii , in a manner related ( if not similar ) to the planets uranus and neptune in our own solar system , and then migrate inwards , to be trapped closer to the star . however , quantitative comparisons with the properties of observed systems remain unsatisfactory ( mcneil & nelson 2010 ) . ultimately , many of the problems with these calculations stem from the properties of the migration models assumed . although the notion of gaseous disk migration has a long , physically motivated provenance ( e.g. goldreich & tremaine 1980 ; ward 1997 ) , the nature of the migration and when it occurs is still a subject of significant uncertainty ( see kley & nelson 2012 for a recent review ) . the principal reason for the revival of interest in migration models was the notion that hot jupiters were emplaced in this fashion ( e.g. lin , bodenheimer & richardson 1996 ) . however , it is becoming clear that these systems represent a small minority of all planetary systems ( e.g. howard et al . 2010 ; mayor et al . 2011 ) and may therefore reflect rather special circumstances . indeed , there is a growing accumulation of recent results demonstrating the misalignment of stellar spin and planetary orbits in many hot jupiter systems ( e.g. winn et al . 2011a and references therein ) . the simplest interpretation is that at least some hot jupiters did not migrate through a gas disk . furthermore , it is no longer clear that migration implies a monotonic inward motion ( e.g. paardekooper & mellema 2006 ; oishi , mac low & menou 2007 ; adams & bloch 2009 ) . with the large numbers of @xmath13 planets now seen at short periods , where the simplest migration models suggest they should not be , indicates that large radial movement need not be ubiquitous in planetary systems . motivated by this renewed skepticism , we have chosen to revisit the question of in situ accumulation of rocky planets on small scales , and to explore the consequences of planet assembly without significant late - time gaseous migration . however , we will consider the possibility that significant radial migration is possible for smaller bodies , before they accumulate into planets , and so we will allow for disks with a wide variety of masses , discussed in [ inventory ] . in [ accum ] we then present the results of a series of traditional in situ planetary accumulation models on scales of 0.051 au , from the resulting disks . in [ accretion ] we investigate the consequences of gas accretion onto the massive rocky cores that form . we address the role of jupiter mass planets in [ hotties ] and evaluate the possible role of photoevaporation in [ evap ] . the prototypical initial condition for assembly of a planetary system is the ` minimum - mass solar nebula ' ( weidenschilling 1977a ; hayashi 1981 ) , based on an estimate of the material required to form our own planetary system . the required surface density of solids ( based here on the hayashi estimate ) is @xmath14 with the gas surface density @xmath15 times larger . this applies interior to the ` snow line ' ( taken to be at @xmath16 au ) . exterior to the snow line the available material in solids is a factor of four higher , reflecting the fact that more volatile elements are able to condense and form solids as the nebular temperature drops . this enhancement brings the solid / gas ratio closer to the estimated metallicity of the sun , which presumably reflects the original composition of the nebular gas . if we integrate this surface density we infer a mass in solids of 3.3@xmath17 interior to 1 au , @xmath18 inside the snow line , and 25.5 @xmath17 within 10 au . this inventory is the minimum amount of mass required to match the solar system planets , but factors of two or three more are not unreasonable , if we assume a certain inefficiency in the process of converting nebular material into planets . in addition , many planet hosting stars are known to show super - solar atmospheric abundances ( gonzalez 1998 ; santos , israelian & mayor 2001 ; fischer & valenti 2005 ; udry & santos 2007 ) , often up to a factor of two . as such , we can reasonably argue for an increase in the above inventory of solids by a factor of five . thus , the simplest extension of the mmsn mass accounting could plausibly provide 17@xmath17 inside 1 au , 29@xmath17 inside the snow line , and 128@xmath17 inside 10 au . the above estimates are still very much in the spirit of the original mmsn model , in which solid material condenses out of a gaseous nebula and accumulates in situ to form planets from the local reservoir of material . however , the modern view of nebular and planetary evolution is considerably more dynamic . recent models have focused largely on the migration of planet - sized bodies , but it has been known for a long time that the dust and small rocks from which planets form are also potentially subject to significant radial migration ( weidenschilling 1977b ; nakagawa , sekiya & hayashi 1986 ; youdin & shu 2002 ; bai & stone 2010 ) . if a significant fraction of the solid material can migrate interior to 1 au while it is still in the form of small bodies , then the potential reservoir of material for the accumulation of terrestrial - class planets could be considerably larger . we will consider mass reservoirs up to 100 @xmath17 , corresponding to a radial concentration of mass by factors of at least 10 . the nature of the processes that produce such radial concentrations is still a subject of active research , and it is not guaranteed that the solids will preserve the same radial profile as the original gas disk . thus we will consider solid surface density profiles with a range of slopes . there is a long history of studies of terrestrial planet accumulation in our own solar system . we adopt the oligarchic model of kokubo & ida ( 1998 ) as our starting point , which provides an analytic prescription for a protoplanetary system consisting of a sequence of annuli , each dominated by a single planetary embryo . simulations of the last stages of gravitational interaction and accumulation of embryos into planets in this model , using n - body techniques ( chambers & wetherill 1998 , agnor , canup & levison 1999 ; chambers 2001 ; raymond , quinn & lunine 2006 ) , shows that configurations of three or four planets emerge naturally from a minimum mass solar nebula disk , although the low masses of mercury and mars in the solar system may require a more localised initial distribution of mass ( hansen 2009 ) . the production of multi - planet systems is fairly robust with respect to varying the amount of mass and the disk density profile ( raymond , quinn & lunine 2005 ; kokubo , kominami & ida 2006 ) , although the masses of the planets tend to scale with the reservoir mass . prior simulations of this process have generally restricted themselves to distances @xmath19 au because of the lack of any closer planet in the solar system ( and shorter timesteps add unecessarily to the computational burden ) . this is also often motivated on the basis of arguments regarding the chemistry of volatiles and a presumed reduction in the mass fraction that condenses out of the nebula . however , our knowledge of nebula dynamics and chemistry is sufficiently poorly constrained that it is prudent to examine the dynamical consequences of different disk masses regardless , and argue about the underpinnings later . thus , in the following , we will examine the assembly of terrestrial planet - class systems from power law surface density distributions of solid material extending down to 0.05 au and up to 1 au . we use the n - body code mercury ( chambers 1999 ) to simulate the final stages of accumulation of planetary embryos , born of the oligarchic model , into planets around a @xmath20 star . we have performed a series of simulations of disks that extend from 0.05 to 1 au , while varying the total mass and surface density profile , to assess whether such an initial distribution can produce the planetary systems observed in a homogenously selected sample , such as that by howard et al . ( 2010 ) . in order to resolve the orbital dynamics at the inner edge of the disk , we use timesteps of 12 hours ! fortunately , most of the interesting dynamics in these collisional accumulation simulations occurs within a fixed number of orbits , so that the compact systems simulated here evolve very rapidly in absolute terms . we evolve our systems to 10 myr , which is sufficient to produce dynamically stable final results . we have verified this by evolution of selected systems to 100 myr and 1 gyr , and have observed little significant change . results of these and subsequent simulations are collected in the table [ simparam ] . we consider three surface density profiles , all of the power law form @xmath21 , and examine models with @xmath22 and @xmath23 . this encompasses a wide range of behaviour and is taken to represent the range of uncertainty in the initial conditions , given the possibility that this might be sculpted by radial migration of solids from large scale to small . with each choice of @xmath24 , we normalise the surface density to have a particular mass of solids interior to 1 au . we choose three different representative normalisations , of 25 , 50 and 100 @xmath17 . in each case , we start with the surface density profile consisting of a series of protoplanetary embryos realised according to the oligarchic distribution described in equations 17 & 18 of kokubo & ida ( 2002 ) , spaced out between 0.05 au to 1 au . for the disk masses considered here , which are larger than usually assumed , the mass of the initial oligarchs can sometimes reach several @xmath17 , and one might be concerned that such bodies are massive enough that their mutual perturbations would invalidate the assumption of mutually independent radial annuli . in appendix [ democratic ] , we have also investigated models in which we restrict the mass of oligarchs to be less than @xmath25 , and we found no meaningful difference in the eventual outcomes . figures [ snap0][snap2.5 ] show the characteristic evolution of three different disks . all have the same total mass ( @xmath26 ) within 1 au , but distributed according to surface densities with different slopes . the disk with the flattest profile ( @xmath27 ) is shown in figure [ snap0 ] . in this case , the masses of the initial oligarchs increase rapidly with radius , scaling with the area of individual annuli . the consequence of this is that the most massive planets at the end are located from 0.51 au . this is somewhat too large a radius to match most of the observations , which , at the time of writing , are most complete for orbital periods of 100 days or less . figure [ snap1.5 ] shows the evolution for an intermediate slope ( @xmath28 ) . here we find that the most massive final planets are found between 0.2 and 0.5 au . although the original oligarchs still increase in mass with semi - major axis , there is sufficient mass on smaller scales so that the faster dynamical evolution results in the accumulation of mass into a few large , comparable - mass objects by 10 myr . figure [ snap2.5 ] shows the evolution of a disk with a steep density profile ( @xmath29 ) . in this case , the oligarch masses decrease with semi - major axis , and the final result retains many of the characteristics of the original mass profile , with the largest planets having the shortest orbital periods . the three histories shown here demonstrate the characteristic behaviour of the models steep density profiles result in a distribution of planet masses that decreases with increasing orbital period , while flatter profiles move the maximum planet mass to larger scales . the stochastic nature of n - body systems means that there is some variation in the total number and division of mass from one simulation to the next , but the general distribution of mass with semi - major axis is reasonably constant . for a fixed density profile , the final planet masses will also , obviously , scale with the total mass budget . figure [ param2_4 ] shows the mass and orbital period of the largest final planet for several simulations with @xmath28 ( the choice which comes closest to matching the observations ) . the open circles indicate a total disk mass of 100@xmath17 , the open triangles indicate @xmath26 and the crosses indicate a total mass of @xmath30 . the filled circles are the observed periods and masses of the largest bodies in systems detected by howard et al . ( 2010 ) ( jupiter mass planets will lie off the top of this plot ) . we see that the known super - earth candidates require total disk masses in the range @xmath31 . as noted in [ inventory ] , even the optimistic estimates of the static nebula solid inventory are not sufficient to meet this requirement of @xmath0@xmath1 needed to match the observed planets . such massive planetesimal disks are more plausible if we allow for the possibility of migration of solids from several au to small scales . another generic consequence of in situ accumulation of rocky planets is that the resulting systems are almost always multiple , and the degree of multiplicity and clustering can be characterised by a variety of statistics ( see chambers 2001 for a summary , and appendix [ stats ] ) that were initially developed to analyse the formation of the terrestrial planets of our own solar system . this proves a useful test of the in situ accumulation hypothesis , as we would then expect the observed systems to be primarily multiple , and to be consistent with the statistical distribution of the simulated systems . a complication for such a comparison is that the observations are likely to be still severely incomplete ( howard et al . 2010 ) and there may be many additional planets lurking in the observed systems , but we can hope for this to improve with time . however , there are some systems for which the noise levels are low enough to demonstrate the existence of multiple systems already . the howard et al sample contains several multiple systems , although some contain planets beyond 1 au , and so are not directly comparable to our calculations . however , there are two systems , 61 vir ( originally announced by vogt et al . 2009 ) and hd69830 ( originally announced by lovis et al . 2006 ) which contain multiple planets within the period and mass range under discussion .. ] figures [ param2 ] and [ param3 ] show the comparison of these two systems with the corresponding simulation results . in order to ensure a proper comparison , we do not count planets in the simulations whose radial velocity amplitude would be @xmath32 . figure [ param2 ] plots two statistical measures , @xmath33 and @xmath34 , of how the mass in the system is distributed with semi - major axis . the observed systems hd69830 and 61 vir are shown as well , where the range indicates an uncertainty in the global system mass because of the unknown inclination angle of the system to the line of sight , assuming coplanarity to within 5 degrees . again , we see that the @xmath35 model produces systems with similar values of both mass concentration and distribution in eccentricity and semi - major axis , whereas the shallower and steeper slopes can not do match all observed statistics simultaneously . to some extent this is not surprising , given the qualitative features of figures [ snap0][snap2.5 ] , although it is a more quantitative statement . figure [ param3 ] compares the measures of angular momentum deficit , amd , ( a measure of eccentricity and inclination spread ) and mass - weighted mean semi - major axis . the spread on the observed systems account for randomly varying the observed eccentricities according to the observed error bars , and allowing the unknown mutual inclinations to fluctuate randomly by 5 degrees . again , we see that the @xmath35 models bracket the observations . it is striking that both the 61 virginis and hd69830 parameters overlap with the simulated values , suggesting that it is possible to realise these observed systems within this model . the spread in the model statistics derives primarily from two sources . the variation in the mass concentration statistics ( figure [ param2 ] ) reflects the degree to which the mass is spread evenly amongst the surviving planets , or concentrated in one or two more massive bodies . the spread in the amd ( see appendix [ stats ] ) seen in figure [ param3 ] is caused primarily by one of the surviving planets having significant ( @xmath36 ) inclination with respect to the others . if such conditions apply to the observed systems , then their parameter range will be expanded from that shown here , in which we limited the amount of assumed mutual inclinations . the outlier systems in the simulation results usually result from an incomplete accounting due to one or more of the final bodies being too small to meet our radial velocity criterion . the result of these accumulation models suggests that both the 61 vir and hd69830 systems are consistent with in situ accumulation of rocky planets from disks of sufficient mass . it also suggests that many of the other systems in the howard et al sample should harbour planets not far below the current detection limits , as the ratio of the two most massive planets in the simulated systems ranges from 0.40.98 , with a median value of 0.78 . the ability of the in situ accumulation model to match the properties of the observed low mass planet population is encouraging , but the masses of the rocky bodies that result suggest that a reconsideration of the issue of gaseous accretion is warranted . in the traditional model of gaseous giant planet formation , the first stage is the collisional assembly of a rocky core , which eventually gravitationally captures gas from the surrounding nebula ( e.g. pollack et al . 1996 ) when it reaches as mass @xmath3 . the final planet masses in [ accum ] exceed this threshold , albeit in a much different region of the disk than the original cores of jupiter and saturn . therefore , the assumption that these planets are wholly rocky is likely to be in error . the question of gas giant formation on small scales was widely reconsidered upon the discovery of 51 pegasi and other ` hot jupiters ' . however , a consensus quickly emerged that these planets most likely formed on larger scales and migrated to their currently observed positions . part of this consensus was the finding , from detailed calculations of the core accretion process , that in situ formation was unlikely ( bodenheimer , hubickyj & lissauer 2000 ; ikoma , emori & nakazawa 2001 , rafikov 2006 ) . however , these findings rest largely on the assumed implausibility of assembling a massive enough rocky core on small scales , given traditional estimates of the rocky material reservoir . if the solid mass available is substantially enhanced by inward radial transport of material , then such massive cores are no longer an outlandish idea . furthermore , on these scales , the assembly timescales of the planets in [ accum ] is short , often @xmath37 years , and the system configurations are largely in place by @xmath38 years . thus , it is quite conceivable that cores of several earth masses are in place before the gas is removed from the system . as a demonstration of this , figure [ acc ] shows the temporal history of the accumulation in a representative simulation with total disk mass in solids of 50 @xmath17 interior to 1 au . the final configuration has 4 planets in the ` super - earth ' mass range interior to 1 au . the assembly of all four planets is largely completed within 1 myr and so it is quite likely that the planets achieve a significant mass while there is still gas present in the system . on scales @xmath39 au , detailed calculations suggest that the energy transport in a hydrostatic atmosphere above a rocky core is regulated by convection ( ikoma et al . 2001 ; rafikov 2006 ) . however , these authors also note that traditional estimates of the critical core mass on these scales become uncertain because the hill sphere radii of the resulting cores begin to approach the size of the disk scale height i.e. a core could potentially open a gap in the disk before the threshold mass is reached . for a disk temperature profile given by @xmath40 this is @xmath41 . the criterion for the planet to open a gap in the gas disk is to set the hill diameter of a planet of mass @xmath42 equal to the gas disk scale height @xmath43 , which yields @xmath44 this threshold mass is less than many of the simulated planets in [ accum ] and most of the observed planets in the howard et al thus , if the assembly of these planets takes place in the presence of gas , accretion is likely to occur . however , the final outcome will not be a traditional gas giant planet . the normal core accretion sequence of events proceeds through a sequence of hydrostatically supported atmospheres until runaway gas accretion occurs ( see rafikov 2006 for a review ) . however , for planets on these small scales , the hydrostatic atmosphere becomes tidally limited before instability sets in , and the opening of the gap limits the amount of gas accretion . therefore , tidally limited accretion from the disk will produce a different , gas - starved , branch of the core accretion family than is found on larger scales . for a gas disk of surface density @xmath45 , the gas mass within an annulus of half width equal to the hill sphere of a core of mass @xmath42 is @xmath46 where @xmath47 is the gas disk mass inside 1 au , and @xmath48 is the orbital semimajor axis of the core . the value @xmath49 corresponds to the gas mass in the standard hayashi minimum mass solar nebula . the relation between this mass and the solid mass will depend on whether or not solids migrate radially . if the gas is eventually accreted by the planet , then the final mass of the planet will result from the combination of the two estimates . in the case where both the gas and solids follow a profile with @xmath50 , this yields a constant core mass fraction for the final planet , @xmath51 suggesting core mass fractions @xmath52 , for ranges of disk mass from 13 the mmsn value . if the gas disk had a different profile , the core mass fraction will vary with orbital period . for a gas disk profile @xmath53 , the mass - period relation would be @xmath54 it is striking that this estimate seems to naturally produce planets in the right range of masses and orbital periods . for example , at the fiducial period of 50 days , we derive a gap opening core mass of 7 @xmath17 which results in a total mass of 2149@xmath17 , for gas disks within a factor of three of the mmsn . this naturally covers the range of the observed masses , with most of the currently observed planets falling in this intermediate class with comparable amounts of rock and gas ( see figure [ pm ] ) . there is potentially also an intermediate regime , where the planet core is large enough to maintain a gravitationally bound atmosphere but not big enough to open a gap in the gas disk . as the disk gas dissipates , the planet may retain enough of this gas to have a significant observational effect . this is the case treated by rafikov ( 2006 ) , whose estimate of the core instability threshold is really a criterion for the gravitionally bound gaseous atmosphere to be some fraction of the core mass . if we adopt an atmosphere fraction of 10% , rafikov s estimate ( in the case of the inner disk his equation 60 ) , amounts to @xmath55 this turns out to be very similar to equation ( [ gapmaker ] ) , so that there are probably few planets in the intermediate regime . if a planet accretes a non - negligible mass fraction of gas , the accretion is likely limited by the opening of a gap . thus , we envision two different outcomes for in situ rocky accretion , depending on the amount of mass available . traditional terrestrial class planets , such as those in the solar system , emerge from disks below some mass threshold so that the time to assemble bodies of several earth masses is longer than the gas dispersal timescale ( @xmath56 years ) . for larger mass disks , cores form that are massive enough to bind gaseous atmospheres of significant mass before the dispersal of the disk . this process is also self - limiting , as they will accrete enough gas to actually open a gap in the disk , which limits their masses to the observed values . it does not depend sensitively on the evolution of the background gas disk a difference from the traditional ` failed core ' model , often invoked to explain these objects ( e.g. mordasini et al . 2009 ; ida & lin 2010 ) . to determine the threshold of disk mass that divides the gas - poor and gas - rich systems , we have simulated the assembly of disks of various masses for an @xmath35 profile , and find that the threshold disk mass is @xmath57 of solids inside 1 au . for disk masses above this value , one or more planets can assemble , within 1 myr , to masses that satisfy equation ( [ gapmaker ] ) , and therefore should accrete gas . for lower mass disks , planets either never reach the threshold value or do so when the gas has dispersed , leaving behind rocky cores . figure [ pm ] shows the overall picture that emerges , along with the observed planets in the howard et al . sample . the solid and open points indicate planets and planet candidates . the dotted line indicates the threshold mass for gas accretion ( eqn . [ gapmaker ] ) . planets above this dotted line will frequently harbour non - negligible gaseous envelopes . the exact size of the envelope depends on the mass in the gaseous disk . the long - dashed line indicates the final mass if the gas accreted is given by equation ( [ gaplimit ] ) , and assuming a gas disk that contains @xmath58 inside 1 au ( equivalent to four times the mmsn value ) . the region between these two lines is approximately the domain of the neptune - like planets born in situ . the short - dashed line running diagonally from upper left to lower right divides the parameter space into regions where accretion of solid bodies is efficient ( to the left of the line ) and not efficient ( to the right ) . the in situ solid body accumulation process is largely conservative of mass on scales smaller than this , whereas planets that lie above the line are more likely to eject smaller bodies than to accrete them . these planets will not grow significantly , but rather will migrate inwards until they reach the accretion / ejection boundary given by the short - dashed line in figure [ pm ] ( murray et al . further accretion will drive planets inwards approximately along the boundary ( i.e. they will grow and migrate inwards ) . the fact that there are few objects located near the intersection of these lines suggests that concurrent migration and accretion in this fashion is not a significant contributor to the demographics of the observed populations . the accretion of gas will change the assembly history , because one or more bodies will increase in mass and therefore the gravitational perturbations on other bodies in the system will be stronger . to account for the effects of this , we repeated the in situ assembly simulations using the @xmath35 profile , but we stop the simulation at 0.1 , 0.3 and 1 myr to compare the masses of all bodies to the threshold value of equation ( [ gapmaker ] ) . if any body has crossed the threshold in the interim , we instantaneously increase the mass according to a model for gas accretion , and then continue the simulation . this model is not fully self - consistent as it does not treat the gravitational influence of the gas disk before it is accreted , nor does it account for any difference in the specific angular momentum of the accreted gas , but it does address the issue of how the rapid mass growth of some bodies affects the gravitational interactions that regulate the assembly of the n - body system into a final system of planets . after the last gas accretion episode the systems are run to ages of 10 myr . we examined two models for gas accretion . in the first , we started with a rocky disk of mass @xmath26 , @xmath35 and assumed gas was accreted according to equation ( [ gaplimit ] ) , i.e. , from a gas disk that also obeys @xmath59 , and with a mass three times the mmsn value . this gas disk mass is chosen to produce planets that trace the upper edge of the observed lower mass planet population . in the second model , we start with a lower mass rocky disk ( @xmath60 , @xmath35 ) and assume gas is accreted according to equation ( [ gasscale ] ) , i.e. , from a gas profile @xmath53 , with a mmsn normalisation . this is chosen to capture the median parameters of the observed low mass population . note that the gas and solid disks have different power law slopes , which is possible if the solid inventory was assembled by radial inward migration , and not in situ deposition . figure [ param4 ] shows the statistical comparison of these simulations with the observed parameters of 61 vir and hd69830 . both sets of simulations show an inward shift in terms of the mass - weighted semi - major axis . this is a result of the tendency of planets with shorter orbital periods to assemble more rapidly and therefore they are more likely accrete gas . the end result is that including gas accretion in the model tends to steepen the mass profile . we have therefore repeated the experiments using a slightly shallower rocky profile of @xmath61 , using two different normalisations ( total masses of 50 @xmath17 and 80 @xmath17 ) . in both cases , the gas accretion is given by equation ( [ gasscale ] ) . the results are shown in figure [ param7 ] . with the use of a steeper profile , both 61 virginis and hd69830 remain within the parameter space of the simulations , although the simulations are now biased to larger semi - major axis ( because there is a larger fraction of mass at large radii and so it is easier to assemble the critical core mass ) . the overall conclusion to draw is that simulations can produce a wide range of systems , depending on the original slope and the amount of gas accreted , bracketing those observed . the dynamical effect of the gas accretion is to accelerate the orbital instabilities and dynamical interactions that characterise the final stages of terrestrial planet assembly . figure [ ta22 ] shows an example of one system ( drawn from the @xmath35 case ) in which three different planets cross the threshold given by equation ( [ gapmaker ] ) , two at 0.1 myr , and a third at 0.3 myr . close examination of the various curves shows that a rapid increase in mass leads to orbital instability in neighbouring , less massive bodies . the ultimate result is frequent collisions , and occasionally accretion onto the star or ejection from the system . this example also shows a common occurrence in which two of the planets which accreted gas eventually merge . the end result of this simulation is that we are left with three planets , two of which possess substantial gas fractions , although somewhat less than indicated by equation ( [ gasscale ] ) because of the accretion of substantial amounts of rocky material during the final clearing stages . this plot also shows why the mass - weighted semi - major axis @xmath62 moves inwards for these simulations , relative to the purely rocky ones the innermost planets grow the fastest and accrete gas earlier . this shifts the mass distribution to shorter radii . nevertheless , accretion occurs for planets at larger radii in some simulations as well , and we also find collisions between planets on larger scales too . this offers an interesting avenue to get the upper envelope of the masses in figure [ pm ] even if the gas disk is not quite as massive as assumed in equation ( [ gaplimit ] ) . our model thus far has addressed the nature of the low mass population at short periods . the howard et al . ( 2010 ) sample also contains a handful of genuine jovian mass planets with orbital periods @xmath6350 days . these must arise in some other fashion . the original model for such planets was migration through a gas disk , but recent observations suggest the situation may be different . for planets with orbital periods of a few days , tidal capture from eccentric orbits might be an alternative model ( rasio & ford 1996 ; matsumura , peale & rasio 2010 ) . however , this seems less likely to produce a planet with orbital periods @xmath64 days , because tidal forces weaken rapidly with distance . the planetary system around the star 55 cancri may offer insights because it appears to contain members of both the low mass and high mass population . we have investigated several hypotheses to see which are consistent with the existence of the 55 cancri system . * can the disk of rocky material we invoke play the role of a dissipative medium in capturing an eccentric jupiter ? the model in which a high eccentricity planet is tidally captured was originally motivated by the onset of dynamical instability , which can potentially occur late in a planetary system s lifetime ( rasio & ford 1996 ) after the removal of the gas disk . we have therefore examined whether a rocky planet system as described in previous sections can capture and circularise a jovian mass planet injected into the system on a highly eccentric orbit . the short answer is that this model is wildly unsuccessful , because the mass of the injected planet is comparable to , or larger than , the entire mass budget of the disks we have considered . thus , the end result of this model is that the injected planet remains on an orbit with a slightly decreased semi - major axis , and partially circularised eccentricity , having ejected essentially all of the mass of the initial disk . to reduce the semi - major axis and eccentricity to the level required by observation would require more mass , and would leave very little surviving mass to make the other planets in the 55 cancri system . + * it is possible to generate rocky planets of jovian mass , given a large enough mass reservoir . in a model where the solid material is laid down from a marginally gravitationally stable @xmath65 surface density profile ( appendix [ gi ] ) , the resulting innermost planet is quite massive , @xmath66 , with orbital periods @xmath64 days ] . this opens the possibility for the formation of rocky planets of saturn mass in short period orbits . at present there is little evidence for the existence of such objects in the sample discovered in transit surveys , although that sample is heavily biased to shorter orbital periods , and is possibly dominated by the tidal capture of genuine gas giants from eccentric orbits . these simulations do not , however , faithfully reproduce the lower mass planets found in the 55 cancri system . nevertheless , if some fraction of protoplanetary disks are significantly sculpted by gravitational instability , this class of planets may eventually emerge in a sample like that of the kepler mission . * a hybrid version of the model in [ accum ] is to simply assume the giant planet was in place in the beginning through some other , unspecified , process ) in the case of the q=1.5 gas disk , which is much more massive . however , the rest of the 55 cancri system is consistent with a less massive disk . ] , and then to repeat our calculation of in situ accumulation with this additional component . the end result of such a simulation actually produces systems qualitatively similar to that of 55 cancri , as shown in figure [ 55match ] . we are able to produce both an inner planet and a system of planets further out . in the event of no gas accretion , the resulting planets are too small compared to the observed planets , but they do cross the threshold of equation ( [ gapmaker ] ) and should grow to appropriate mass . figure [ 55match ] shows this for the analogue of 55 cancri c. the conclusion is that the presence of a jovian mass planet in the disk does not impede the in situ assembly of planets along the lines we have investigated , and that the resulting dynamical environment can produce systems with similar organisation to 55 cancri . jovian mass planets can also potentially influence the assembly on sub - au scales even if they orbit farther out ( such perturbations are often advanced as potential causes for the low mass of the asteroid belt , and possibly mars e.g. chambers 2001 ) . there are several systems in the howard et al . sample which contain a gas giant at long orbital periods and a lower mass planet at shorter orbital periods . to investigate the degree to which an external jupiter could influence the assembly histories considered here , we have repeated the @xmath35 , 50 @xmath17 runs in the presence of a planet of mass 351 @xmath17 , orbital period 493 days , and @xmath67 . this is chosen based on the planet hd114783b , as an example . we find that planet assembly is largely unaffected interior to @xmath68 au ( orbital periods @xmath69 days ) . the bulk of our analysis applies to smaller scales and shorter periods , so that the general character of the solutions is largely unaffected by the presence or absence of jupiters on scales @xmath70 au . this is consistent with the approximate separation between jupiter and the asteroid belt in the solar system and the rule - of - thumb that a gas giant does not significantly reduce planet assembly interior to it s 3:1 mean motion resonance . it has been suggested by numerous authors that short period planets may experience significant atmospheric mass loss driven by irradiation from the central star ( e.g. lammer et al . 2003 ; yelle 2004 ; lecavelier des etangs et al . 2004 ; tian et al . 2005 ; murray - clay , chiang & murray 2009 ) . as a result , we need to consider how much of a limit this will place on the accretion or retention of gas by planets at short orbital periods . empirical estimates of the evaporation rate for exoplanets are based on the so - called ` energy diagram ' , in which the gravitational binding energy of the planet ( @xmath71 ) is compared to the energy input ( @xmath72 ) from x - ray and euv radiation ( lecavelier des etangs 2007 , ehrenreich & desert 2011 ) , after accounting for various uncertain efficiency factors . using the estimates of ehrenreich & desert for the life - time averaged euv flux for the host star , we can rewrite the formula for the characteristic evaporation time , @xmath73 , as a mass - period relation @xmath74 where @xmath75 is the efficiency factor for converting incident radiation into mass loss ( set equal to unity here , so that this is a very generous estimate ) . the meaning of this equation is that planets with masses above this threshold are too large to experience significant dimunition through the evaporation process , under even very optimistic assumptions . we see that this relation has the wrong slope to explain the edge in figure [ pm ] , and will likely affect only the planets with periods @xmath76 days . the slope may be changed if we postulate a particular relationship between mass and radius , but not by much , as the mass - radius relationship for planets with gaseous envelopes is relatively flat in this mass range . therefore , evaporation is unlikely to be responsible for the shape of the mass - period relation in figure [ pm ] . nevertheless , it is likely to be important for the lowest mass , shortest period planets . in cases where the mass given by equation ( [ evaplim ] ) is greater than that given by equation ( [ gaplimit ] ) , the planet may very well accrete some gas , but it will likely lose it again over the course of time . thus , we may expect to see a transition from significant gaseous envelopes to thin , remnant envelopes below some period . estimating the period is somewhat uncertain because of the rather unconstrained nature of @xmath75 , but it likely lies in the range of 3.7 days ( @xmath77 ) to 8.5 days ( @xmath78 ) . motivated by recent observational challenges to the prevailing paradigm of exoplanet migration , we have re - examined the role that in situ accumulation might play in the assembly of planets on small scales , with masses ranging up to several tens of earth masses . we have demonstrated that the observed properties of planets in the hot neptune``/super - earth '' mass range , orbiting solar mass stars with periods of days to several months , can be reproduced with a simple model of in - situ accumulation of rocky planets from traditional oligarchic conditions , without any additional radial migration . however , in order for this to be the case , the original amount of rocky material required is @xmath79 within 1 au . this is somewhat in excess of that expected from traditional estimates of the mass profiles of protoplanetary disks , but could be achieved if there was significant radial migration of solid material from the outer parts of disks to the scales of interest , before the migrating bodies achieved the mass of planetary embryos . thus , our calculations argue that migration of solid material did indeed occur , but that it served to rearrange the mass budget of the nebula before the final stages of planet assembly . such a position is not without physical foundation , as the timescale for small bodies to migrate over au scales through the action of aerodynamic drag forces can be as small as @xmath80 years ( weidenschilling 1977b ; nakagawa et al . 1986 ; youdin & shu 2002 ; bai & stone 2010 ) . a further consequence of such a mass concentration is that the collisional assembly of planets is fast enough that bodies of several earth masses can form within 1 myr of the original material being emplaced . if this growth occurs in the vicinity of a gas disk , accretion and gravitational retention of gas from the nebula is possible . the accretion of this gas is limited , however , by the fact that it is easier to open a gap in the gas disk on smaller scales , and the accretion may be quenched before the accretion flow enters the traditional runaway regime in the core instability scenario ( e.g. rafikov 2006 ) . this process provides a natural mechanism for the production of planets with comparable amounts of gas and rock , intermediate between the traditional classifications of rocky terrestrial planets , and gas - dominated giants . figure [ mx ] shows the heavy element fraction of planets of various mass formed in our simulations , compared to published estimates for observed hot neptunes around solar mass stars . it should be noted that many of the published systems do not strictly fall within the bounds of our model ( e.g. hd149026b ) in terms of the parameter space of figure [ pm ] . furthermore , our crude estimate of a tidally - limited gas accretion ignores a lot of potentially interesting physics that determines the final mass of a planet embedded in a gas disk . in particular , planets that open gaps in disks can continue to accrete gas if the disk viscosity is high enough to allow flow into the gap ( e.g. artymowicz & lubow 1996 ; bryden et al . 1999 ; kley 1999 ) . based on the formulae of lin & papaloizou ( 1979 ) or taniguchi & ikoma ( 2007 ) , we estimate that disks with @xmath81 ( shakura & sunyaev 1973 ) may result in further gas accretion across the gaps we have estimated here . nevertheless , it is interesting that the simple considerations outlined above provide a natural framework for producing planets of this observed class . it is particularly encouraging that the objects on this plot most likely to fall within our model are the kepler-9 planets , which match the results of our simulations quite well . the quantitative agreement between observations and model is in striking contrast to the failures of traditional migration - based models ( alibert et al . 2006 , terquem & papaloizou 2007 ; ida & lin 2010 ) to match the same observations ( mcneil & nelson 2010 ) . the neptune mass analogs in these calculations formed further out and represent ` failed cores ' of giant planets , whose inward migration due to disk torques quenches the gas accretion that would produce a traditional gas giant . multiplicity is often important in the survival of such objects , as strong mutual gravitational interaction can serve to slow the migration and avoid spiralling into the central star . however , the effects of such strong interaction imprints itself on the final system , even after the loss of gas and further dynamical evolution ( ida & lin 2010 ; mcneil & nelson 2010 ) . recent observations of hot jupiters have cast doubt on the migration of gas giants as well , suggesting that perhaps the very idea of planetary migration needs to be extensively reviewed ( although we stress that the inventory of solids invoked in our model most likely still requires significant radial migration of solid material ) . we should be clear , however , that the model presented here is not complete , as we have not addressed the details of how these systems are related to those with hot jupiters or jupiters on more traditional scales . our aim is to better understand the individual physical processes at work before attempting a complete synthesis of all known planetary systems . furthermore , it is now evident that the lower planet mass systems are the most generic kind , and that stars which host hot jupiters are a small minority . it may thus be useful to consider our models independently of those developed for hot jupiters , which could represent only one very special case amongst the pathways to planetary system formation . our original discussion was purposely limited to the sample of howard et al . ( 2010 ) because of the homogeneity of selection . mayor et al . ( 2011 ) recently released a catalog of detections from a homogeneously selected sample of their own . in addition to 61 vir and hd 69830 , their sample contains seven more stars with at least three sub - jupiter mass companions . figure [ param6 ] shows a comparison of these systems with the same models as discussed before . the model range encompasses nearly all of these , with hd 10180 ( lovis et al . 2011 ) , hd 31527 and hd 136352 located closest to the bulk of the simulation points . several systems , such as hd 40307 and hd 39194 , could be brought closer to the main locus of points if additional planets are discovered in these systems with longer orbital period ( which seems likely in the case of hd 40307 , since mayor et al . 2009a also report the detection of a linear trend ) . conversely , hd 134606 has an anomalously large @xmath82 , which could be made consistent if another planet is found in the system with orbital period @xmath83 days . as a check on the howard et al . sample , figure [ pmall ] shows the harps ( mayor et al . 2011 ) and kepler - based exoplanet samples in the orbital period - mass plane , along with the proposed physical classification based on our model . at the lower right , at masses below those given by equation ( [ gapmaker ] ) , planets are not large enough to accrete significant gas and thus are likely to be truly rocky planets . this is a regime that is poorly explored as yet , although current radial velocity limits are slowly inching lower . above this limit , and at orbital periods @xmath84 days , we expect to find cores which have accreted substantial , but not overwhelming , gaseous envelopes . within our model , the bulk of the planets in this period and mass range are expected to be more like neptunes than scaled - up rocky planets ( a.k.a . super - earths ) . this is consistent with the newly emerging constraints from kepler , which suggests planetary radii for this class of object that require substantial gaseous envelopes ( holman et al . 2010 ; lissauer et al . further encouragment comes from the analysis of howard et al . ( 2011 ) , who estimate that planets with masses @xmath85 and periods @xmath7 days have substantial gas fractions on average , while those at lower masses are less gas - rich . evaluating our equation ( [ gapmaker ] ) at a mean period for the howard sample @xmath86 days , yields a mass of 5@xmath17 . as the observations and analysis improve , we expect that this division will be found to rise to higher masses at longer orbital periods . although the criterion is based on the ability of a core to gravitationally retain gas ( e.g. ikoma et al . 2001 ; rafikov 2006 ) , if the ordering of mass with semi - major axis is preserved to the present day , it would represent strong evidence that there was little orbital evolution post assembly . for orbital periods @xmath4 days , the division between the two classes is less uncertain because of the possibility that a gaseous envelope could be accreted but then removed over time . the fact that the observations tend to group near the line in figure [ pmall ] is a tantalising suggestion that evaporation could be very efficient . if true , it would suggest that most of the planets with @xmath87 days , and @xmath88 are neptune - like planets that have experienced significant erosion over the course of their lifetime . the essential point of these calculations is that many of the quantitative features of the observations can be modelled within a relatively simple in situ accumulation formalism . as such , several aspects of the more general astrophysical context need to be addressed , but the agreement with the observations argues that the basic model is correct . the mechanism(s ) by which heavy elements are concentrated in the inner parts of the nebula remain to be properly quantified . although the mobility of small bodies in protoplanetary nebulae is well established , the precise nature of the material rearrangement will depend on the properties of the background nebula . similarly , the effects of a residual gas disk on the damping of planetary eccentricities and inclinations is something to be considered . however , moeckel & armitage ( 2012 ) have shown that the statistical outcome of planetary scattering remains largely unaffected by the presence of gas . they do identify a subpopulation of low eccentricity orbits in their population , but presently the observations are still in excellent agreement with the outcomes of pure n - body calculations ( e.g. figure [ pmall ] ) . the model also predicts a period - dependent threshold mass that separates true rocky planets from those with comparable amounts of rock and gas . this will be quantitatively tested as more low mass planets are found , but already finds some statistical support in the planet candidates uncovered by kepler ( howard et al . 2011 ; wolfgang & laughlin 2011 ; youdin 2011 ) . the authors thank phil armitage , ravit helled , kristen menou , hilke schlichting and chris thompson for discussions and comments . is supported by the nasa atp program . is supported in part by the canada research chair program and by nserc of canada . adams , f. c. & bloch , a. m. , 2009 , apj , 701 , 1381 agnor , c. b. , canup , r. m. & levison , h. f. , 1999 , icarus 142 , 219 alibert , y. et al . , 2006 , a&a , 455 , l25 artymowicz , p. & lubow , s. h. , 1996 , apj , 467 , l77 bai , x .- & stone , j. m. , 2010 , apj , 722 , 1437 batalha , m. , et al . , 2011 , apj , 729 , 27 bodenheimer , p. , hubickyj , o. & lissauer , j. j. , 2000 , icarus , 143 , 2 borucki , w. b. et al . , 2011 , apj , 736 , 19 bryden , g. , chen , x. , lin , d. n. c. , nelson , r. p. & papaloizou , j. c. b. , 1999 , apj , 514 , 344 burrows , a. et al . , 2007 , apj , 661 , 502 carter , j. a. , winn , j. n , gilliland , r. & holman , m. j. , 2009 , apj , 696 , 241 chambers , j. e. , 1999 , mnras , 304 , 793 chambers , j. e. , 2001 , icarus , 152 , 205 chambers , j. e. & wetherill , g. w. , 1998 , icarus , 136 , 304 chambers , j. e. , wetherill , g. w. & boss , a. p. , 1996 , icarus , 119 , 261 dawson , r. i. & fabrycky , d. c. , 2010 , apj , 722 , 937 ehrenreich , d. , & dsert , j .- m . , 2011 , arxiv:1103.0011 fischer , d. a. & valenti , j. , 2005 , apj , 622 , 1102 gammie , c. , 2001 , apj , 553 , 174 goldreich , p. & tremaine , s. , 1980 , apj , 241 , 425 gonzalez , g. , 1998 , a&a , 334 , 221 hansen , b. m. s. , 2009 , apj , 703 , 1131 hartman , j. d. , et al . , 2011 , apj , 728 , 138 havel , m. , guillot , t. , valencia , d. & crida , a. , 2011 , arxiv:1103.6020 hayashi , c. , 1981 , prog . suppl . , 70 , 35 holman , m. et al . , 2010 , science , 330 , 51 howard , a. w. et al . , 2010 , science , 330 , 653 howard , a. w. et al . , 2011 , apj , 726 , 73 ida , s. & lin , d. n. c. , 2008 , apj , 685 , 84 ida , s. & lin , d. n. c. , 2010 , apj , 719 , 810 ikoma , m. , emori , h. & nakazawa , k , 2001 , apj , 553 , 999 kley , w. , 1999 , mnras , 303 , 696 kley , w. & nelson , r. p. , 2012 , arxiv:1203.1184 kokubo , e. , kominami , j. & ida , s. , 2006 , apj , 642 , 1131 kokubo , e. & ida , s. , 1998 , icarus , 131 , 171 kokubo , e. & ida , s. , 2002 , apj , 581 , 666 lammer , h. , selsis , f. , ribas , i. , guinan , e. , bauer , s. j. , & weiss , w. , 2003 , apj , 598 , l121 laughlin , g. , & rozyczka , m. , 1996 , apj , 456 , 279 leger , a. , et al . , 2009 , a&a , 506 , 287 lecavelier des etangs , a. , vidal - madjar , a. , mcconnell , j. c. & hebrard , g. , 2004 , apj , 621 , 1049 lecavelier des etangs , a. , 2007 , a&a , 461 , 1185 lin , d. n. c. , bodenheimer , p. & richardson , d. c. , 1996 , nature , 380 , 606 lin , d. n. c. & papaloizou , j. , 1979 , mnras , 186 , 799 lissauer , j. j. , et al . , 2011b , nature , 470 , 53 lissauer , j. j. et al . , 2011a , apj , 197 , 8 lovis , c. et al . , 2006 , nature , 441 , 305 lovis , c. et al . , 2011 , a&a , 528 , a112 matsumura , s. , peale , s. j. & rasio , f. a. , 2010 , apj , 725 , 1995 marcy , g. , butler , p. , fischer , d. , vogt , s. , lissuaer , j. & rivera , e. , 2001 , apj , 556 , 296 mayor , m. & queloz , d. , 1995 , nature , 378 , 355 mayor , m. et al . , 2009a , a&a , 493 , 639 mayor , m. et al . , 2009b , a&a , 507 , 487 mayor , m. et al . , 2011 , arxiv:1109.2497 mcneil , d. s. & nelson , r. p. , 2010 , mnras , 401 , 1691 moeckel , n. & armitage , p. j. , 2012 , mnras , 419 , 366 mordasini , c. , alibert , y. & benz , w. , 2009 , a&a , 501 , 1139 murray , n. , b. hansen , m. holman & s. tremaine , 1998 , science , 279 , 69 murray - clay , r. a. , chiang , e. i. & murray , n. , 2009 , apj , 693 , 23 nakagawa , y. , sekiya , m. & hayashi , c. , 1986 , icarus , 67 , 375 oishi , j. s. , mac low , m .- & menou , k. , 2007 , apj , 670 , 805 paardekooper , s. j. & mellema , g. , 2006 , a&a , 459 , l17 pollack , j. b. , hubickyj , o. , bodenheimer , p. , lissauer , j. j. , podolak , m. & greenzweig , y. , 1996 , icarus , 124 , 62 rafikov , r. , 2006 , apj , 648 , 666 rasio , f. a. & ford , e. b. , 1996 , science , 274 , 954 raymond , s. n. , barnes , r. & mandell , a. , 2008 , mnras , 384 , 663 raymond , s. n. , quinn , t. & lunine , j. i. , 2004 , icarus , 168 , 1 raymond , s. n. , quinn , t. & lunine , j. i. , 2005 , apj , 632 , 670 rice , w. k. m. & armitage , p. j. , 2009 , mnras , 396 , 2228 rogers , l. a. & seager , s. , 2010 , apj , 712 , 974 santos , n. c. , israelian , g. & mayor , m. , 2001 , a&a , 373 , 1019 shakura , n. i. & sunyaev , r. a. , 1973 , a&a , 24 , 337 taniguchi , t. & ikoma , m. , 2007 , apj , 667 , 55 terquem , c. & papaloizou , j. c. b. , 2007 , apj , 654 , 1110 tian , f. , toon , o. b. , pavlov , a. a. & de sterck , h. , 2005 , apj , 621 , 1049 udry , s. & santos , n. c. , 2007 , ara&a , 45 , 397 vogt , s. et al . , 2009 , apj , 708 , 1366 ward , w. r. , 1997 , icarus , 126 , 261 weidenschilling , s. j. , 1977a , ap&ss , 51 , 153 weidenschilling , s. j. , 1977b , mnras , 180 , 57 winn , j.n . , et al . , 2011a , aj , 141 , 63 winn , j. n. et al . , 2011b , apj , 737 , l18 wolfgang , a. & laughlin , g. , 2011 , arxiv:1108.5842 yelle , r. v. , 2004 , icarus , 170 , 167 youdin , a. & shu , f. , 2002 , apj , 580 , 494 youdin , a. , 2011 , apj , 742 , 38 for the kinds of disk masses considered in this paper , the oligarchic formalism of kokubo & ida ( 1998 ) can produce bodies with masses @xmath89 or more . with masses this large , one might wonder whether the long - range gravitational perturbations between neighbouring annuli become important sooner , and perhaps the assumption of initially independent annuli no longer holds . to examine this , we have repeated our calculations but have restricted the maximum size of an oligarch to @xmath25 . the bodies are still spaced by annuli with width ten times the hill radius , and so the overall surface density distribution is preserved . in table [ democheck ] we show the statistics for 5 realisations of this model , for a disk with power law @xmath28 and total mass @xmath26 . these can be compared with the results of the runs run@xmath9012 in table [ simparam ] . we see that there is little difference in the final statistical measures . the only noticeable difference is that the runs for this section lose a few percent of material in scattered and ejected bodies , which leads to a slightly higher probability that one of the surviving bodies will have a radial velocity signature @xmath32 . as a result we quote two sets of statistics for some of the runs in table [ democheck ] . the first line represents the direct comparison to other simulations . in those cases where we show a second set of values for a simulation , it means that one of the surviving planets would not have been detected with a 1 m / s radial velocity threshold . the second line shows the statistics calculated if we include those planets that would likely have been missed . the equations collected here offer a brief description of each of the statistical measures used in the text . further information can be found in chambers ( 2001 ) and references therein . the angular momentum deficit is defined as @xmath91 } { \sum_j m_j \sqrt{a_j}}\ ] ] where @xmath92 , @xmath93 , @xmath94 and @xmath95 are the mass , semi - major axis , eccentricity and inclination of the j - th particle . it is a measure of how much the orbits in the system deviate from the case of circular , coplanar orbits , which represent the state of maximal angular momentum per unit energy . the mass - weighted semi - major axis is @xmath96 and is a measure of where the mass is eventually accumulated . the mass concentration statistic is given as the maximum value of the function @xmath97 ^ 2 } \right)\ ] ] as the semi - major axis @xmath48 is varied . the orbital spacing statistic is @xmath98 where @xmath99 is the mean mass of the planets , and @xmath100(@xmath101 ) is the semi - major axis of the shortest ( longest ) period of the n planets in the system . this quantity is often considered a measure of the long - term dynamical stability of the system , as it bears a qualitative similarity to the average spacing of the planets measured in terms of their hill radii , @xmath102 . dynamical simulations can determine the threshold value of these parameters for long term stability ( e.g. chambers , wetherill & boss 1996 ) , but require some choices with regards to the masses of the various components . thus , we have repeated this calculation for the specific case of interest here . we consider three - planet systems , with the middle planet at 0.35 au . the planet masses are given by equation ( [ gasscale ] ) , and two other planets in the system were placed at equal separation @xmath103 on either side . the results suggest that stability is achieved for @xmath104 au , corresponding to @xmath105 . comparison to figure [ param6 ] shows that nearly all the observed systems satisfy this criterion . the two that do not are gliese 876 ( which corresponds to a pair of jupiter mass planets orbiting an m - dwarf in resonance ) and kepler-11 ( which satisfies this criterion if the mass of the outermost planet , kepler-11 g , lies somewhat below the currently quoted upper limit ) . in the early stages of star formation , the mass fraction of material in the disk may be substantial and the disk may exhibit a variety of large scale gravitational instabilities . this can also lead to significant radial transport of gas and dust ( e.g. laughlin & rozyczka 1996 ; gammie 2001 ) and may result ina somewhat different radial profile than those considered above . thus , we will also consider a model in which the solid distribution is deposited in situ , but from a gas disk which follows a disk profile dictated by setting the toomre parameter to @xmath65 , which we take as the marginally stable end state of a gravitationally unstable disk . the temperature profile is also somewhat different because the disk is heated by the dissipation inher ent in the mass transport . based on the results of rice & armitage ( 2009 ) , we assume a constant temperature @xmath106 k interior to 1 au , and derive @xmath107 where we have used a conversion factor of @xmath108 between gas and solid material . this is a small fraction of the gas metallicity , but captures the likely inefficiency of sedimentation at these temperatures . the assembly of embryos , drawn from this distribution , into planets proceeds in the same way as in the other simulations , except for the fact that there is considerably more mass available . the final planets produced have masses that range up to @xmath1 . this particular scenario may never be realised in practice , but is worth calculating if only to demonstrate that the gap in the observed masses is the result of the available reservoir of material , not of the physics of assembly itself . it is only when the masses begin to approach the ejection / accretion division shown in figure [ pm ] that the process of collisional accumulation becomes self - limiting . lccccc|cccccc run@xmath109 & 230 & 1.75 & @xmath110 & @xmath110 & 57 & 2 & 99.3 & 36.5 & 0.073 & 8.5 & 0.168 + run@xmath109 & 230 & 1.75 & @xmath110 & @xmath110 & 57 & 2 & 127.5 & 31.1 & 0.012 & 10.3 & 0.291 + run@xmath109 & 230 & 1.75 & @xmath110 & @xmath110 & 57 & 2 & 97.9 & 29.4 & 0.027 & 13.5 & 0.161 + run@xmath111 & 25 & 0 & @xmath110 & @xmath110 & 79 & 1 & 16.3 & @xmath110 & 0.034 & @xmath112 & 0.545 + run@xmath113 & 25 & 0 & @xmath110 & @xmath110 & 79 & 1 & 12.5 & @xmath110 & 0.1011 & @xmath112 & 0.485 + run@xmath114 & 25 & 0 & @xmath110 & @xmath110 & 79 & 1 & 9.9 & @xmath115 & 0.0005 & @xmath112 & 0.906 + run@xmath116 & 50 & 0 & @xmath110 & @xmath110 & 57 & 2 & 46.1 & 38.6 & 0.0136 & 22.8 & 0.740 + run@xmath117 & 50 & 0 & @xmath110 & @xmath110 & 57 & 3 & 27.1 & 21.2 & 0.0084 & 17.3 & 0.738 + run@xmath118 & 50 & 0 & @xmath110 & @xmath110 & 57 & 3 & 27.1 & 17.5 & 0.0031 & 45.0 & 0.705 + run@xmath119 & 50 & 0 & @xmath110 & @xmath110 & 57 & 3 & 42.7 & 26.3 & 0.0081 & 15.8 & 0.717 + run@xmath120 & 50 & 0 & @xmath110 & @xmath110 & 57 & 3 & 32.1 & 15.6 & 0.0013 & 59.9 & 0.734 + run@xmath121 & 50 & 0 & @xmath110 & @xmath110 & 57 & 3 & 28.0 & 11.4 & 0.0006 & 87.1 & 0.750 + run@xmath122 & 50 & 0 & @xmath110 & @xmath110 & 57 & 2 & 28.0 & 21.5 & 0.0014 & 43.3 & 0.767 + run@xmath123 & 50 & 0 & @xmath110 & @xmath110 & 57 & 1 & 40.5 & @xmath110 & 0.0194 & @xmath112 & 0.559 + run@xmath124 & 50 & 0 & @xmath110 & @xmath110 & 57 & 3 & 28.0 & 21.9 & 0.0148 & 16.4 & 0.755 + run@xmath125 & 100 & 0 & @xmath110 & @xmath110 & 41 & 2 & 61.3 & 23.4 & 0.0125 & 46.4 & 0.664 + run@xmath126 & 100 & 0 & @xmath110 & @xmath110 & 41 & 3 & 51.3 & 19.3 & 0.0020 & 37.2 & 0.628 + run@xmath127 & 100 & 0 & @xmath110 & @xmath110 & 41 & 4 & 50.3 & 18.5 & 0.0042 & 23.9 & 0.614 + run@xmath128 & 25 & 1.5 & @xmath110 & @xmath110 & 34 & 5 & 6.3 & 24.1 & 0.0114 & 6.2 & 0.401 + run@xmath129 & 25 & 1.5 & @xmath110 & @xmath110 & 34 & 3 & 10.2 & 36.6 & 0.0189 & 11.7 & 0.234 + run@xmath130 & 25 & 1.5 & @xmath110 & @xmath110 & 34 & 4 & 9.3 & 29.4 & 0.0136 & 6.9 & 0.380 + run@xmath90 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 3 & 17.3 & 32.1 & 0.0102 & 6.2 & 0.383 + run@xmath131 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 2 & 16.5 & 48.5 & 0.0417 & 11.3 & 0.156 + run@xmath132 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 2 & 23.3 & 48.7 & 0.0484 & 12.8 & 0.199 + run@xmath133 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 3 & 22.6 & 33.7 & 0.0033 & 5.1 & 0.513 + run@xmath134 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 4 & 19.9 & 24.6 & 0.0100 & 5.9 & 0.474 + run@xmath135 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 3 & 17.8 & 33.3 & 0.0059 & 8.4 & 0.304 + run@xmath136 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 4 & 11.7 & 30.4 & 0.0569 & 10.0 & 0.208 + run@xmath137 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 3 & 26.4 & 31.9 & 0.0395 & 7.4 & 0.341 + run@xmath138 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 4 & 15.9 & 34.0 & 0.0087 & 5.2 & 0.428 + run@xmath139 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 4 & 12.8 & 24.6 & 0.0094 & 8.6 & 0.286 + run@xmath140 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 3 & 23.3 & 33.9 & 0.0121 & 8.0 & 0.316 + run@xmath141 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 4 & 14.5 & 24.2 & 0.0076 & 6.4 & 0.384 + run@xmath142 & 100 & 1.5 & @xmath110 & @xmath110 & 18 & 5 & 28.3 & 16.8 & 0.0034 & 6.5 & 0.426 + run@xmath143 & 100 & 1.5 & @xmath110 & @xmath110 & 18 & 2 & 53.3 & 54.5 & 0.0067 & 13.6 & 0.588 + run@xmath144 & 100 & 1.5 & @xmath110 & @xmath110 & 18 & 4 & 37.3 & 20.5 & 0.0159 & 5.9 & 0.442 + run@xmath145 & 50 & 2.5 & @xmath110 & @xmath110 & 20 & 3 & 24.0 & 26.5 & 0.0093 & 13.8 & 0.132 + run@xmath146 & 50 & 2.5 & @xmath110 & @xmath110 & 20 & 3 & 19.4 & 30.9 & 0.0063 & 9.9 & 0.155 + run@xmath147 & 50 & 2.5 & @xmath110 & @xmath110 & 20 & 3 & 22.6 & 28.0 & 0.0125 & 16.1 & 0.123 + run@xmath148 & 50 & 2.5 & @xmath110 & @xmath110 & 20 & 4 & 19.8 & 19.4 & 0.0057 & 12.2 & 0.116 + run@xmath149 & 50 & 2.5 & @xmath110 & @xmath110 & 20 & 4 & 19.8 & 18.7 & 0.0040 & 14.8 & 0.102 + run@xmath150 & 50 & 2.5 & @xmath110 & @xmath110 & 20 & 5 & 16.6 & 17.3 & 0.0082 & 10.7 & 0.126 + run@xmath151 & 50 & 2.5 & @xmath110 & @xmath110 & 20 & 4 & 14.5 & 20.7 & 0.0023 & 13.8 & 0.110 + run@xmath152 & 50 & 2.5 & @xmath110 & @xmath110 & 20 & 4 & 14.5 & 21.2 & 0.0068 & 10.2 & 0.131 + run@xmath153 & 50 & 2.5 & @xmath110 & @xmath110 & 20 & 4 & 13.4 & 21.3 & 0.0039 & 11.8 & 0.133 + run@xmath154 & 100 & 2.5 & @xmath110 & @xmath110 & 14 & 4 & 40.5 & 21.0 & 0.0132 & 6.9 & 0.198 + run@xmath155 & 100 & 2.5 & @xmath110 & @xmath110 & 14 & 3 & 60.3 & 25.2 & 0.0099 & 9.3 & 0.165 + run@xmath156 & 100 & 2.5 & @xmath110 & @xmath110 & 14 & 4 & 49.7 & 18.6 & 0.0133 & 9.6 & 0.164 + + run@xmath157 & 50 & 1.5 & 2300 & 1.5 & 25 & 4 & 62.0 & 19.6 & 0.023 & 11.1 & 0.311 + run@xmath158 & 50 & 1.5 & 2300 & 1.5 & 25 & 4 & 60.7 & 20.6 & 0.011 & 10.9 & 0.267 + run@xmath159 & 50 & 1.5 & 2300 & 1.5 & 25 & 4 & 101.4 & 17.6 & 0.0047 & 9.6 & 0.354 + run@xmath160 & 50 & 1.5 & 2300 & 1.5 & 25 & 3 & 40.8 & 27.9 & 0.0230 & 6.8 & 0.315 + run@xmath161 & 50 & 1.5 & 2300 & 1.5 & 25 & 3 & 67.3 & 23.5 & 0.0091 & 12.8 & 0.230 + run@xmath162 & 50 & 1.5 & 2300 & 1.5 & 25 & 3 & 52.0 & 27.6 & 0.0146 & 6.8 & 0.301 + run@xmath163 & 50 & 1.5 & 2300 & 1.5 & 25 & 3 & 42.6 & 27.9 & 0.0145 & 8.1 & 0.259 + run@xmath164 & 50 & 1.5 & 2300 & 1.5 & 25 & 2 & 100.0 & 19.9 & 0.0059 & 34.3 & 0.410 + run@xmath165 & 40 & 1.5 & 767 & 1.0 & 30 & 4 & 22.1 & 23.9 & 0.0025 & 9.3 & 0.217 + run@xmath166 & 40 & 1.5 & 767 & 1.0 & 30 & 3 & 15.7 & 32.3 & 0.011 & 12.6 & 0.150 + run@xmath167 & 40 & 1.5 & 767 & 1.0 & 30 & 2 & 30.6 & 44.0 & 0.0025 & 11.3 & 0.196 + run@xmath168 & 40 & 1.5 & 767 & 1.0 & 30 & 4 & 28.9 & 24.0 & 0.0121 & 6.2 & 0.341 + run@xmath169 & 40 & 1.5 & 767 & 1.0 & 30 & 5 & 26.4 & 18.5 & 0.0891 & 7.4 & 0.313 + run@xmath170 & 40 & 1.5 & 767 & 1.0 & 30 & 5 & 13.9 & 20.2 & 0.0057 & 7.3 & 0.280 + run@xmath171 & 40 & 1.5 & 767 & 1.0 & 30 & 4 & 19.2 & 22.8 & 0.0097 & 6.7 & 0.317 + run@xmath172 & 40 & 1.5 & 767 & 1.0 & 30 & 3 & 18.7 & 35.5 & 0.0225 & 6.9 & 0.242 + run@xmath173 & 40 & 1.5 & 767 & 1.0 & 30 & 5 & 23.3 & 18.3 & 0.0063 & 5.6 & 0.350 + run@xmath174 & 40 & 1.5 & 767 & 1.0 & 30 & 3 & 32.9 & 34.1 & 0.0277 & 8.0 & 0.241 + run@xmath175 & 40 & 1.5 & 767 & 1.0 & 30 & 11 & 7.2 & 15.4 & 0.0007 & 6.1 & 0.442 + run@xmath176 & 40 & 1.5 & 767 & 1.0 & 30 & 1 & 43.2 & @xmath110 & 0.163 & @xmath112 & 0.106 + run@xmath177 & 50 & 1.0 & 767 & 1.5 & 30 & 4 & 28.2 & 20.9 & 0.0083 & 7.9 & 0.511 + run@xmath178 & 50 & 1.0 & 767 & 1.5 & 30 & 3 & 37.8 & 28.8 & 0.0011 & 10.6 & 0.344 + run@xmath179 & 50 & 1.0 & 767 & 1.5 & 30 & 3 & 51.1 & 27.9 & 0.0243 & 8.3 & 0.549 + run@xmath180 & 50 & 1.0 & 767 & 1.5 & 30 & 4 & 39.8 & 21.2 & 0.0042 & 8.6 & 0.511 + run@xmath181 & 50 & 1.0 & 767 & 1.5 & 30 & 3 & 55.4 & 25.9 & 0.0151 & 13.8 & 0.491 + run@xmath182 & 80 & 1.0 & 767 & 1.5 & 22 & 2 & 45.2 & 28.4 & 0.101 & 21.9 & 0.368 + run@xmath183 & 80 & 1.0 & 767 & 1.5 & 22 & 4 & 57.3 & 18.1 & 0.0078 & 9.1 & 0.580 + run@xmath184 & 80 & 1.0 & 767 & 1.5 & 22 & 3 & 49.5 & 25.7 & 0.0290 & 10.2 & 0.400 + run@xmath185 & 80 & 1.0 & 767 & 1.5 & 22 & 3 & 50.9 & 26.6 & 0.0352 & 10.6 & 0.339 + run@xmath186 & 80 & 1.0 & 767 & 1.5 & 22 & 2 & 57.3 & 31.2 & 0.0041 & 14.3 & 0.433 + run@xmath187 & 80 & 1.0 & 767 & 1.5 & 22 & 3 & 48.9 & 25.8 & 0.0086 & 7.4 & 0.500 + run@xmath188 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 3 & 11.8 & 33.3 & 0.0200 & 9.4 & 0.215 + run@xmath189 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 3 & 10.6 & 30.9 & 0.0305 & 10.4 & 0.194 + run@xmath190 & 50 & 1.5 & @xmath110 & @xmath110 & 25 & 3 & 14.5 & 30.8 & 0.0173 & 8.2 & 0.264 + lccccc|cccccc demo@xmath191 & 50 & 1.5 & @xmath110 & @xmath110 & 52 & 3 & 23.7 & 31.1 & 0.0254 & 6.5 & 0.379 + demo@xmath192 & 50 & 1.5 & @xmath110 & @xmath110 & 52 & 3 & 15.0 & 28.4 & 0.0214 & 13.1 & 0.291 + & 50 & 1.5 & @xmath110 & @xmath110 & 52 & 4 & 15.0 & 25.3 & 0.0199 & 9.7 & 0.356 + demo@xmath193 & 50 & 1.5 & @xmath110 & @xmath110 & 52 & 2 & 17.0 & 45.3 & 0.0467 & 13.2 & 0.169 + & 50 & 1.5 & @xmath110 & @xmath110 & 52 & 3 & 17.0 & 33.4 & 0.0471 & 7.6 & 0.272 + demo@xmath194 & 50 & 1.5 & @xmath110 & @xmath110 & 52 & 2 & 22.0 & 45.0 & 0.0495 & 13.9 & 0.174 + & 50 & 1.5 & @xmath110 & @xmath110 & 52 & 3 & 22.0 & 36.0 & 0.0498 & 5.0 & 0.422 + demo@xmath195 & 50 & 1.5 & @xmath110 & @xmath110 & 52 & 3 & 21.0 & 34.2 & 0.0218 & 8.4 & 0.291

materials with correlated electrons exhibit some of the most intriguing quantum states in condensed matter physics @xcite . since the number of electric charge carriers essentially determines such states , external electric or light field can be applied for controllable manipulations . stability of electric field or light induced states has been demonstrated in some novel systems , where switching occurs between neighboring equilibrium thermodynamic states @xcite . this powerful characteristic can be applied in electric devices such as transistors and memories , which are of great importance to not only the fundamental physics research but also the information processing technology @xcite . correlated materials with a rich set of quantum states delicately balanced on a similar energy scale will be promising platforms to realize such devices . ( a ) crystal structure of @xmath0-tas@xmath1 . ( b ) top view of ta - ta plane of @xmath0-tas@xmath1 . the unit cells of conventional @xmath0 structure and low temperature ccdw structure are denoted with green solid and dashed lines , respectively . the black hexagrams are the so - called `` david star '' pattern in ccdw phase . ( c ) schematic diagrams of `` david star '' ( left ) , ccdw ( middle ) , and nccdw ( right ) in @xmath0-tas@xmath1 . ] the transition metal dichalcogenide ( tmd ) @xmath0-tas@xmath1 is one of the promising candidate due to the multiple competing ground states in it @xcite . as shown in fig . [ fig_1_structure ] ( a ) , @xmath0-tas@xmath1 shows a cdi@xmath1-type layered crystal structure with ta atoms octahedrally coordinated by s atoms . a unit layer consists of one ta layer sandwiched between two s layers . at low temperatures , strong @xmath4-dependent electron - phonon coupling induced periodic lattice distortion makes a @xmath5 superlattice @xcite , in which ta atoms displace to make `` david star '' clusters ( fig . [ fig_1_structure ] ( b ) ) . the outer twelve atoms within each star move slightly towards the atom at the center , leading to the commensurate charge - density - wave ( ccdw ) ground state . in particular , in ccdw state , the correlation effect of @xmath6 electrons of ta atoms turns the system into mott insulating state @xcite . upon heating to 225 k , it undergoes a sequence of first order phase transition to a nearly commensurate ( nc ) cdw . the nccdw phase is composed of metallic incommensurate ( ic ) network and mott insulating ccdw domains . the ccdw domains shrink upon heating and finally disappear at 355 k , while the system transforms to iccdw state . the standard metallic @xmath0 structure appears above 535 k. moreover , when ccdw state is suppressed , superconductivity emerges in this system @xcite . one can expect that the controllable switching between those states will be helpful for figuring out the mechanism of cdw and superconductivity , and realizing the high performance memory and transistor in future technology . to meet this goal , many groups performed investigations on this novel material . yu et al . reported a gate - controlled li ion intercalation can suppress ccdw and introduce superconducivity @xcite . tsen et al . @xcite , hollander et al . @xcite , yoshida et al . @xcite , and mihailovic et al . @xcite reported the in - plane electric field induced transition from ccdw state to nccdw or some metastable hidden state . cho et al . applied perpendicular electric pulse on @xmath0-tas@xmath1 and found positive electric pulse can introduce ic network to suppress mott state at low temperature @xcite . moreover , zhang et al . @xcite , mihailovic et al . @xcite , and han et al . @xcite suggested that light can introduce a transition from ccdw state to nccdw or hidden state as well . in all those manipulations , the transitions are considered to be determined by the charge carrier doping . however , the mechanisms of such transitions are not fully clear yet . in this work , we simulated by first - principles calculations the charge carrier doping effect on cdw in @xmath0-tas@xmath1 . we found that ccdw is stable upon electron doping , while hole doping significantly suppresses ccdw instability , implying different mechanisms of recently reported electric and photoelectric manipulations of cdw in @xmath0-tas@xmath1 . we figured out such mechanism by analysis of carrier doping effects . furthermore , we show that superconductivity with @xmath2 about @xmath3 k can be introduced by hole doping in the system . the density functional theory ( dft ) calculations were carried out using quantum espresso package @xcite with ultrasoft pseudopotentials . the exchange - correlation interaction was treated with the generalized gradient approximation ( gga ) with pw91 parametrization @xcite . the energy cutoff for the plane - wave basis set was 35 ry . the marzari - vanderbilt fermi smearing method @xcite with a smearing parameter of @xmath7 ry was used for the calculations of the total energy and electron charge density . to simulate the monolayer , a vacuum layer more than 15 @xmath8 was introduced . for the bulk sample , brillouin zone sampling is performed on the monkhorst - pack ( mp ) mesh @xcite of @xmath9 , while a denser @xmath10 grid was used in the electron phonon coupling calculations . phonon dispersions were calculated using density functional perturbation theory ( dfpt ) @xcite with an @xmath11 mesh of @xmath12-points . for the monolayer sample , @xmath13-points grids of @xmath14 and @xmath15 and @xmath12-points grids of @xmath16 are used . since the frequency and stability of soft phonon modes were found to be very sensitive to the plane - wave cutoff for the charge density , a large cutoff of 1500 ry was used for the charge density in the phonon calculations @xcite . all the calculation parameters were well tested . in the investigation of carrier doping effects , the electron / hole doping was simulated by increasing / decreasing the total electron numbers of the system , together with a compensating uniform positive / negative background to maintain the charge neutrality . the crystal structure of each doped sample was optimized with respect to lattice parameters and atomic positions . for comparisons , we also calculated some doped samples by using the lattice parameters of pristine tas@xmath1 and relaxing the atomic positions , as described later . the low symmetry cdw structure is usually considered as the high symmetry phase with distortion introduced by some instability . therefore , we firstly investigated carrier doping effects in the bulk @xmath0-tas@xmath1 . the electronic structures and phonon properties for the samples with different doping level were calculated . as examples , figure [ fig_2_bulk ] shows the calculated band structures and fermi surfaces of the pristine bulk @xmath0-tas@xmath1 , the doped bulk @xmath0-tas@xmath1 with the doping level of @xmath17 electrons / f.u . , and the doped bulk @xmath0-tas@xmath1 with @xmath17 holes / f.u . as expected , electron doping increases the lattice parameters and hole doping shrinks the lattice ( figs . [ fig_2_bulk ] ( j ) and ( k ) ) . for the pristine tas@xmath1 , our results are of good agreement with the previous calculations @xcite . there is a gap of @xmath18 ev below fermi energy ( @xmath19 ) ( fig . [ fig_2_bulk ] ( b ) ) . in the @xmath20-@xmath21 direction , the two bands around the gap are nearly flat due to the quasi-2d nature of the layered structure . the band above the gap crosses fermi energy ( @xmath19 ) , forming a 2d electron pocket around @xmath22 point ( figs . [ fig_2_bulk ] ( b ) and ( e ) ) . such gap increases upon electron doping and decreases upon hole doping ( figs . [ fig_2_bulk ] ( a ) and ( c ) ) . electron doping increases @xmath19 . as shown in figs . [ fig_2_bulk ] ( a ) and ( d ) , for the doped bulk @xmath0-tas@xmath1 with @xmath17 electrons / f.u . the rise of @xmath19 expands and opens up the original electron pockets around @xmath22 , leaving hole pockets centered at @xmath23 point . besides the original band crossing @xmath19 , a band with higher energy starts to cross @xmath19 , forming a 2d cylinder - like electron pocket around the zone center . on the contrary , the hole doping enhances the dispersion in the @xmath20-@xmath21 direction and weakens the quasi-2d nature by shrinking the lattice and decreasing the interlayer distance . as shown in figs . [ fig_2_bulk ] ( c ) and ( f ) , for the doped bulk @xmath0-tas@xmath1 with @xmath17 holes / f.u . , the bands close to @xmath19 are not flat anymore . hole doping reduces @xmath19 . the decrease of @xmath19 shrinks the original electron pockets around m point . moreover , a lower energy band starts to cross @xmath19 , forming a 3d hole pocket around @xmath20 point . the phonon instability of the high symmetry structure is considered to be directly related to the cdw distortion : at high temperatures , the phonon of the high symmetry structure softens at cdw vector ( @xmath24 ) . above the transition temperature the phonon frequency near @xmath24 drops but does not go to zero . just below the transition temperature the phonon frequency near @xmath24 is imaginary , meaning there is a restructuring of lattice with a superlattice vector of @xmath24 @xcite . therefore , phonon dispersion without imaginary frequency implies that the structure is stable compared to cdw structure . the phonon calculation is proved to be an effective method to simulate the cdw instability : the calculated phonon dispersions show instability just locating at @xmath24 of some tmds @xcite . more specially , for the present @xmath0-tas@xmath1 , experimental report showed cdw can be suppressed by pressure @xcite , which is correctly simulated by liu s phonon calculation @xcite . in this work we also performed the phonon calculations on each sample . for the pristine @xmath0-tas@xmath1 , our calculation is of good agreement with liu s calculation ( fig . [ fig_2_bulk ] ( h ) ) . the phonon dispersion show instability very close to the ccdw vector @xmath25 . this instability persists at all values of @xmath26 , as shown in inset of fig . [ fig_2_bulk ] ( h ) . for the electron doped sample , the acoustic branches become more unstable . as shown in fig . [ fig_2_bulk ] ( g ) , the unstable modes in the doped bulk @xmath0-tas@xmath1 with with @xmath17 electrons / f.u . appear in @xmath23-@xmath22 and @xmath23-@xmath20 directions , indicating the area of instability are largely expanded . on the contrary , the hole doping significantly stabilizes lattice . as shown in fig . [ fig_2_bulk ] ( i ) , no unstable mode can be found in the phonon dispersion of the doped bulk @xmath0-tas@xmath1 with @xmath17 holes / f.u . since the lowest mode in the pristine tas@xmath1 locates near @xmath27 , we used such mode as an indicator of doping effect on ccdw in tas@xmath1 . the frequency variation of such mode under doping is shown in fig . [ fig_2_bulk ] ( l ) . one can note that upon electron doping the mode is always unstable in bulk @xmath0-tas@xmath1 , while hole doping significantly suppresses the instability . according to our calculation , the lattice becomes completely stable when the doping level is higher than @xmath28 electrons / f.u . since hole doping shrinks the lattice as the high pressure does , we should figure out the suppression of ccdw under hole doping is majorly attributed to whether the lattice compression or to the pure doping effect . we calculated a hypothetic doped @xmath0-tas@xmath1 with n=0.3 holes / f.u . , in which the lattice parameters are fixed to those of the undoped pristine @xmath0-tas@xmath1 . as shown in fig . [ fig_2_bulk ] ( l ) , ccdw can be suppressed by just doping holes into @xmath0-tas@xmath1 without changing its lattice volume . that demonstrates the suppression of ccdw in the present case is predominately by the hole doping effect . furthermore , we investigated the doping induced energy and the distortion variations in the ccdw state for the pristine tas@xmath1 , the doped tas@xmath1 with @xmath17 electrons / f.u . , and the doped tas@xmath1 with @xmath17 holes / f.u . we defined the ccdw formation energy @xmath29 as @xmath30 where @xmath31 and @xmath32 are the total energies of the relaxed ccdw structure and the @xmath0 structure . the distortion rations @xmath33 can be expressed as @xmath34 and @xmath35 where @xmath36 is the in - plane lattice parameter of the undistorted @xmath0 structure , @xmath37 and @xmath37 are radius of the inner and outer circles of `` david star '' , as shown in fig . 3 ( b ) . for @xmath0-tas@xmath1 with ccdw structure , the layers stacking order is not clear yet . therefore , we simply construct the @xmath38 supercell with 39 atoms to simulate the ccdw structure . we relaxed the atomic positions using the lattice parameters @xmath39 and @xmath40 , where @xmath36 and @xmath40 are from figs . [ fig_2_bulk ] ( j ) and ( k ) , respectively . in the relaxed ccdw structures , we found that for the pristine tas@xmath1 and the doped tas@xmath1 with @xmath17 electrons / f.u . [ fig_3_charge ] ( b ) and ( a ) ) , the distortion ratios are large , while the formation energies of those are negative . on the other hand , the distortion ratios of the doped tas@xmath1 with @xmath17 holes / f.u . are nearly zero ( fig . [ fig_3_charge ] ( c ) ) , and the energy difference between the relaxed supercell and the @xmath0 unit cell is very small ( @xmath41 mev ) , i.e. the ccdw structure relaxes to high symmetry @xmath0 structure . we also relaxed the atomic positions with @xmath36 and @xmath40 fixed to those in the pristine tas@xmath42 to prevent the volume variation . with the lattice parameters of the pristine tas@xmath42 , the doped tas@xmath42 with @xmath17 electrons / f.u can be seen as under pressure . however , the ccdw is still stable . on the other hand , without doping introduced volume shrinking , in the doped tas@xmath42 with @xmath17 holes / f.u . the ccdw is still not favorable in energy . such results of the ccdw supercell coincide well with our deductions based on the analysis of the phonon instabilities of the @xmath0 unit cell . if we simply consider that the roles of electron / hole doping on the electronic structures are to increase / decrease @xmath19 , we can estimate the population of the added / removed electrons when electron / hole doping and understand the calculated results more clearly . figure [ fig_3_charge ] ( d ) shows the density of states ( dos ) near @xmath19 of the pristine tas@xmath42 in the ccdw structure calculated directly by dft . based on the integrations of the dos upwards / downwards from @xmath19 , we can see that if one electron is doped into a `` david star '' ( the doping level is equal to @xmath43 electron / f.u . ) , it will be mainly added into the center of the `` david star '' . obviously , it will enhance the clustering of the charge density . therefore , ccdw is stable upon electron doping . on the contrary , doping one hole into a `` david star '' will notably decrease the charge density at the center and inner atoms of the `` david star '' , which will weaken the charge density clustering . therefore , ccdw is suppressed upon hole doping . [ cols="^,^,^,^,^,^,^",options="header " , ] the doping effects in the bulk tas@xmath1 investigated using the low symmetry ccdw structure . ( a ) , ( b ) , and ( c ) are the charge density in ta - ta plane of the doped tas@xmath1 with @xmath17 electrons / f.u . , pristine tas@xmath1 , and the doped tas@xmath1 with @xmath17 holes / f.u . , respectively . ( d ) dos of the pristine tas@xmath1 directly calculated in ccdw structure . the colored areas below and above @xmath19 denote the electron density which can be integrated to one electron . the related electron charge densities are plotted in the insets of ( d ) . ] the doping effects in monolayer @xmath0-tas@xmath1 . for the doped tas@xmath1 with @xmath17 electrons / f.u . , pristine tas@xmath1 , and the doped tas@xmath1 with @xmath17 holes / f.u . , ( a ) , ( b ) , and ( c ) are the band dispersions of those samples , ( d ) , ( e ) , and ( f ) are the fermi surfaces , and ( g ) , ( h ) , and ( i ) are the phonon dispersions , respectively . the red dashed lines show the phonon dispersion of undoped monolayer @xmath0-tas@xmath1 with tensile strain of 5% . ] schematic picture of suppression of ccdw by hole doping . ( a ) the situation of tas@xmath1 in the ccdw state when holes are doped locally . the solid orange `` david stars '' are the area under ccdw distortion . the hollow orange `` david stars '' are the area in which only one hole per a `` david star '' is doped , which are still under ccdw distortion . the dashed light orange `` david stars '' are the area with more doped holes , in which the ccdw distortion is fully suppressed . the black arrows denote the diffusion of the doped holes . ( b ) the schematic band structure of tas@xmath1 with ccdw distortion . ( c ) the schematic band structure of tas@xmath1 with ccdw distortion when one hole per `` david star '' is doped . ( d ) the schematic band structure of tas@xmath1 in @xmath0 structure . ] we also investigate the doping effect in monolayer @xmath0-tas@xmath1 . as examples , fig . [ fig_4_monolayer ] shows the calculated electronic structures and phonon properties of the monolayer pristine tas@xmath1 , the doped tas@xmath1 with @xmath17 electrons / f.u . , and the doped tas@xmath1 with @xmath17 holes / f.u in @xmath0 structure . for the undoped pristine monolayer @xmath0-tas@xmath1 , the band structure is similar to that of pristine bulk sample . one band crosses @xmath19 , forming a 2d electron pocket around @xmath22 point ( figs . [ fig_4_monolayer ] ( b ) and ( e ) ) . there is a gap of 0.8 ev below the band crossing @xmath19 . in the electrons and holes doped monolayer samples , the gap slightly changes , which is different from the case of bulk samples . for the doped monolayer @xmath0-tas@xmath1 with @xmath17 electrons / f.u . , the band structure is slightly different from that of the doped bulk @xmath0-tas@xmath1 with @xmath17 electrons / f.u . [ fig_4_monolayer ] ( a ) ) . although @xmath19 changes with electron doping , there is still only one band crossing @xmath19 . the 2d cylinder - like electron pocket in the doped bulk @xmath0-tas@xmath1 with @xmath17 electrons / f.u . does not appear in monolayer sample ( fig . [ fig_4_monolayer ] ( d ) ) . for the doped monolayer @xmath0-tas@xmath1 with @xmath17 holes / f.u . , there is only one band crossing @xmath19 ( fig . [ fig_4_monolayer ] ( c ) ) . similar to the case of bulk samples , the phonon calculation shows that the cdw instability in the monolayer @xmath0-tas@xmath1 can not be suppressed by electron doping ( fig . [ fig_4_monolayer ] ( g ) ) , while it can be suppressed under hole doping ( fig . [ fig_4_monolayer ] ( i ) ) . the result indicates the suppression of ccdw in @xmath0-tas@xmath1 is not due to the hole doping enhanced band dispersion along @xmath20-@xmath21 direction ( figs . [ fig_2_bulk ] ( c ) and ( f ) ) . the suppression should be attributed to the weakening of the electron - phonon coupling at @xmath24 upon hole doping . besides , one may note in the monolayer sample , a small instability near @xmath20 can be found in the phonon dispersion ( fig . [ fig_4_monolayer ] ( h ) ) . in a very recent calculation of phonon dispersion of monolayer @xmath0-tas@xmath1 by zhang et al . @xcite , there is a similar instability near @xmath20 . such instability is consistent with the instability against long - wavelength transversal waves @xcite . this instability is suggest to be fixed by defects , such as ripples or grain boundaries , which do not allow these waves by limiting the size @xcite . furthermore , we found tensile strain can suppress such instability near @xmath20 , which can be easily applied to monolayer materials @xcite . as an example , in fig . [ fig_4_monolayer ] ( h ) , the red dashed lines show the phonon dispersion of @xmath0-tas@xmath1 under a tensile strain of 5% , in which the instability near @xmath20 is suppressed . the observation indicates that the application of tensile strain is helpful for stabilizing the experimentally exfoliated monolayer or fewlayer @xmath0-tas@xmath42 . based on our calculations of phonon properties in @xmath0 structure , ccdw formation energies and distortion ratios in ccdw structure , we can conclude that when the doping level is above @xmath28 holes / f.u . ( 2.6 holes per a `` david stars '' ) , ccdw can be completely suppressed in @xmath0-tas@xmath1 for bulk and monolayer , i.e. all the `` david stars '' should be melted . experimentally , holes can directly be introduced by a positive electric field perpendicular to the sample @xcite , or by light @xcite . one can expect that in reality the doped holes will firstly distribute in a local area and gradually diffuse out , i.e. even the doping level might be much lower than 0.2 holes / f.u . , in the local area , the doping level could be high enough to melt the `` david stars '' in this area and destroy the long - range ccdw , as shown in fig . [ fig_5_hole_mechanism ] ( a ) . we describe the possible picture of such process here : in @xmath0 structure , the electronic structure near @xmath19 are formed predominantly from a single ta @xmath44 band , which splits into subbands by the formation of ccdw state . six of these subbands are fulfilled with 12 electrons per new ccdw unit cell , forming a manifold of occupied states . the 13th leftover electron is localized on the central ta atom of the `` david stars '' , forming a half - filled subband at @xmath19 . this half - filled subband further splits by the coulomb interaction into upper and lower hubbard bands ( uhb and lhb ) , as shown in fig . [ fig_5_hole_mechanism ] ( b ) . light or positive perpendicular electric filed can firstly excite one electron from lhb to uhb , and create one hole in lhb ( fig . [ fig_5_hole_mechanism ] ( c ) ) . in real space , the hole doped by light or positive perpendicular electric filed locates at the center of the `` david stars '' , leaving a polaron with an excess charge . the other 12 electrons in this polaron are still star - shaped around the central , thus screening the excess charge . when more holes are doped into the `` david stars '' , the ccdw distortion is suppressed locally , the structure transforms into @xmath0 in the doping area . the splitting subbands merge to a single band again ( fig . [ fig_5_hole_mechanism ] ( d ) ) . in this case , the `` david stars '' shaped clusters are annihilated and can not screen the holes any more . so the holes diffuse into neighboring `` david stars '' . therefore , upon hole doping in @xmath0-tas@xmath42 in ccdw phase , ccdw should firstly transform to a nccdw or metastable phase composed of ccdw domains and iccdw network . for example , cho et al . @xcite applied a very small positive perpendicular voltage pulse on @xmath0-tas@xmath1 single crystal sample within a typical scanning tunneling microscope ( stm ) set - up . since local hole concentration under the stm tip is largely enhanced , a pulse creates a textured cdw domain of a few tens of nanometers with an irregular domain wall network inside . they considered that such network is consistent with those in the thermally excited nccdw phase @xcite . besides , di / dv measured implies the weakening and broadening of the hubbard states together with the reduction of the mott gap inside the textured cdw domain induced by positive perpendicular electric filed @xcite . that is corresponding to our picture : holes diffuse from the domain wall into the neighboring area inside the domain . although the concentration of the diffused holes is not high enough to suppress ccdw distortion inside the domain , it can create the hole - electron pair in lhb and uhb to reduce the mott gap . moreover , the photoexcitation of electrons can be considered as directly doping holes into system . mihailovic s group found a laser pulse can introduce a transition from ccdw to a hidden state in @xmath0-tas@xmath1 thin flake @xcite . the raman spectrum of such hidden state is completely different from that in ccdw and nccdw phases @xcite . such hidden state is demonstrated to be a metastable phase , which is different from nccdw , but should be composed of ccdw domains and iccdw network as well @xcite . han et al . observed the electron diffraction spots varies from ccdw case to iccdw case when applying laser to tas@xmath1 in ccdw phase @xcite . very recently , zhang et al . showed laser can introduce a ccdw - nccdw transition in bulk @xmath0-tas@xmath1 single crystal @xcite . on the other hand , in experiments upon hole doping in @xmath0-tas@xmath1 in high temperature phase , the ccdw transition temperature is significantly lowered @xcite . both the two kinds of structural evidences can be well explained by our calculation . for the sake of illustration we draw , we drew a schematic diagram to describe the mechanism of the hole doping induced ccdw - nccdw / metastable phase transition , as shown in figs . [ fig_6_mechanism ] ( a ) , ( b ) and ( c ) . in the pristine @xmath0-tas@xmath1 , ccdw phase has the lowest energy ( fig . [ fig_6_mechanism ] ( a ) ) . the hole doping by perpendicular field or by light can stabilize lattice and largely increase the energy of ccdw phase . in this case , the system transforms to nccdw or other metastable phase ( figs . [ fig_6_mechanism ] ( b ) and ( c ) ) . on the other hand , according to our calculation , ccdw lattice distortion is stable upon electron doping . according to the report by cho et al . , negative perpendicular voltage pulses could not make change in ccdw state @xcite , which is consistent with our estimation . however , some recent experimental works suggest the opposite results . doped electron into @xmath0-tas@xmath1 by intercalation of li ions and found ccdw can be suppressed @xcite . the intercalated ions can not only carry electrons , but also strongly influence the lattice structures and induce disorder , which can suppress cdw in tmds @xcite . therefore , one can not simply attribute the suppression to the electron doping . to demonstrate the pure electron doping effect on cdw , more direct doping experiments by negative perpendicular electric field or liquid - gated method should perform . some works reported the suppression of ccdw by in - plane electric field @xcite . by measuring the temperature dependence of resistivity ( @xmath45 ) , yoshida et al . found the in - plane field can not affect the nccdw / iccdw transition in @xmath45 curve , but can introduce a metastable state with very low resistivity at low temperatures @xcite . if we consider that the @xmath45 in ccdw , nccdw , and iccdw reflects the related structural characteristics , one can infer that the in - plane field induced the metastable state with different structural characteristics , i.e. the in - plane field can suppress ccdw distortion as well . the suppressions of ccdw by the in - plane field are usually explained as electron injection effect @xcite . however , according to our calculation , ccdw could be stable upon electron doping . therefore , the effect of in - plane field should be more complex . we consider some potential mechanisms of the suppressions by in - plane field : besides the pure charge carrier injection , one can expect that the in - plane electric field can also force the electrons in the `` david star '' to be delocalized . moreover , the in - plane electric field might depinning the cdw , which is suggested in @xmath0-tas@xmath1 bulk single crystals @xcite . the suppression of ccdw by in - plane electric field might also be due to the thermal activation by local joule heating as current flows through the material . the real mechanism of the suppression of ccdw by in - plane field needs to be figured out by more experimental and theoretical works . here we only offer a general description of such process : although ccdw state might still have the lower energy , the in - plane electric field can drive the system to cross over the energy barrier from ccdw to nccdw or other metastable phase , as shown in the schematic diagram in fig . [ fig_6_mechanism ] ( d ) . eliashberg function ( left ) and the integrated electron - phonon coupling strength ( right ) for ( a ) the doped bulk @xmath0-tas@xmath1 with @xmath46 holes / f.u . and ( b ) the doped monolayer @xmath0-tas@xmath1 with @xmath47 holes / f.u . under a tensile strain of 2.5% , respectively . ] as described above , once long range ccdw is suppressed , the transition between mott insulating state to metallic state can be observed : when ccdw is suppressed , domain walls show up , which can be seen as conducting channels to induce metallic state @xcite . in the ccdw area near the domain walls , the hubbard states are weakening and broadening and the mott gap is reduced . @xcite . furthermore , superconductivity can emerge in percolated metallic ic network @xcite . while we are not able to directly model the textured ic phase , we can still estimate the superconductivity using @xmath0 structure qualitatively . approximately using the high symmetry doped bulk @xmath0-tas@xmath1 with @xmath46 holes / f.u . and the doped monolayer @xmath0-tas@xmath1 with @xmath47 holes / f.u . under a tensile strain of 2.5% , for which all the phonon instabilities are just suppressed , we estimated the potential carrier doping induced superconductivity by the electron - phonon coupling calculation . figure [ fig_7_a2f ] shows the calculated eliashberg spectral function @xmath48 where @xmath49 is the density of states at @xmath19 , @xmath50 is phonon frequency , @xmath51 is electronic energy , and @xmath52 is electron - phonon coupling matrix element . the total electron - phonon coupling strength is then @xmath53 the calculated @xmath54 for the two samples are 0.81 and 0.96 , respectively . we estimated @xmath2 based on the modified mcmillan formula @xcite : @xmath55 where the coulomb pseudopotential @xmath56 is set to a typical value of @xmath57 . the logarithmically averaged characteristic phonon frequency @xmath58 is defined as @xmath59 the calculated @xmath58 for the two samples are 136.7 and 98.4 k , respectively . using those parameters , we can estimate that the superconductivity with @xmath2 of @xmath3 k can emerge in the hole doped @xmath0-tas@xmath1 . to observe the superconductivity we predicted , the doping induced metallic network between ccdw domain in nccdw or other metastable phase must be percolated , which requires experimental devices with ability to dope more holes . recently , suda et al . suggested a high level hole doping technique using photoactive electric double layer @xcite , which might be applied to verify our prediction . in conclusion , based on the first - principles calculation , we simulated the carrier doping effect in the bulk and monolayer @xmath0-tas@xmath1 . we found that ccdw is stable upon electron doping , while hole doping can notably suppress the ccdw . according to our analysis , the different mechanisms of the reported electric and photoelectric manipulations of ccdw in @xmath0-tas@xmath1 are figured out : light or positive perpendicular electric field induced hole doping significantly increases the energy of ccdw , so that the system transforms to nccdw or similar metastable state . although the ccdw distortion is more stable upon the in - plane electric field induced electron injection , some accompanied effects can drive the system to cross over the energy barrier from ccdw to nccdw or similar metastable state . we also estimated that a potential superconductivity with @xmath2 of @xmath3 k can be introduced by hole doping in the system . by tuning the carrier density , controllable switching of different states such as ccdw / mott insulating state , metallic state , and even the superconducting state can be realized in @xmath0-tas@xmath1 , which makes the novel material have a very promising application in the future electronic devices . we thank prof . z. g. sheng and dr . y. z. zhang for helpful discussion . this work was supported by the national key research and development program under contract no . 2016yfa0300404 , the national nature science foundation of china ( grant no . 11674326 , 11304320 , 11274311 , 11404340 , 1408085ma11 , and u1232139 ) . 10 m. imada , a. fujimori , and y. tokura , rev . 70 , 1039 ( 1998 ) . e. dagotto , rev . 66 , 763 ( 1994 ) . p. a. lee , n. nagaosa , and x. -g . wen , rev . 78 , 17 ( 2006 ) . h. ohno , d. chiba , f. matsukura , t. omiya , e. abe , t. dietl , y. ohno , and k. ohtani , nature 408 , 944 ( 2000 ) . y. yamada , k. ueno , t. fukumura , h. t. yuan , h. shimotani , y. iwasa , l. gu , s. tsukimoto , y. ikuhara , and m. kawasaki , science 332 , 1065 ( 2011 ) . r. e. glover , iii and m. d. sherrill , phys . lett . 5 , 248 ( 1960 ) . n. takubo , y. ogimoto , m. nakamura , h. tamaru , m. izumi , and k. miyano , phys . 95 , 017404 ( 2005 ) . a. zakery , s. r. elliott , optical nonlinearities in chalcogenide glasses and their applications ( springer , new york , 2007 ) . m. nakano , k. shibuya , d. okuyama , t. hatano , s. ono , m. kawasaki , y. iwasa , and y. tokura , nature 487 , 459 ( 2012 ) . j. a. wilson , f. j. di salvo , and s. mahajan , adv . 24 , 117 ( 1975 ) . a. y. liu , phy . b 79 , 220515(r ) ( 2009 ) . y. ge and amy y. liu , phy . rev . 155133 ( 2010 ) . y. liu , d. f. shao , l. j. li , w. j. lu , x. d. zhu , p. tong , r. c. xiao , l. s. ling , c. y. xi , l. pi , h. f. tian , h. x. yang , j. q. li , w. h. song , x. b. zhu , y. p. sun , phys . b 94 , 045131 ( 2016 ) . p. darancet , a. j. millis , and c.a . marianetti , phys . b 90 , 045134 ( 2014 ) . r. ang , y. tanaka , e. ieki , k. nakayama , t. sato , l. j. li , w. j. lu , y. p. sun , and t. takahashi , phys . 109 , 176403 ( 2012 ) . r. ang , y. miyata , e. ieki , k. nakayama , t. sato , y. liu , w. j. lu , y. p. sun , and t. takahashi , phys . b 88 , 115145 ( 2013 ) . b. sipos , a. f. kusmartseva , a. akrap , h. berger , l. forr , and e. tuti , nature mater . 7 , 960 ( 2008 ) . y. liu , r. ang , w. j. lu , w. h. song , l. j. li , and y. p. sun , appl . 102 , 192602 ( 2013 ) . l. j. li , w. j. lu , x. d. zhu , l. s. ling , z. qu , and y. p. sun , europhys . 97 , 67005 ( 2012 ) . y. yu , f. yang , x. f. lu , y. j. yan , y. -h . cho , l. ma , x. niu , s. kim , y. -w . son , d. feng , s. li , s. -w . cheong , x. h. chen , and y. zhang , nat . 10 , 270 ( 2015 ) . a. w. tsen , r. hovden , d. wang , y. d. kim , j. okamoto , k. a. spoth , y. liu , w. j. lu , y. p. sun , j. hone , l. f. kourkoutis , p. kim , and a. n. pasupathy , pnas 8 , 15054(2015 ) . m. j. hollander , y. liu , w. j. lu , l. j. li , y. p. sun , j. a. robinson , and s. datta , nano lett . 15 , 1861 ( 2015 ) . masaro yoshida , ryuji suzuki , yijin zhang , masaki nakano , and yoshihiro iwasa , sci . adv . 1 , e1500606 ( 2015 ) . i. vaskivskyi , i. a. mihailovic , s. brazovskii , j. gospodaric , t. mertelj , d. svetin , p. sutar and d. mihailovic , arxiv:1409,3794 ( 2014 ) . i. vaskivskyi , j. gospodaric , s. brazovskii , d. svetin , p , sutar , e. goreshnik , i. a. mihailovic , t. mertelj , and d. mihailovic , sci . adv . 1 , e1500168 ( 2015 ) . d. cho , s. cheon , k. -s . kim , s. -h . lee , y. -h . cho , s. -w . cheong , and h. w. yeom , nature comm . 7 , 10453 ( 2016 ) . y. z. zhang , z. g. sheng , et al . , private communication . l. stojchevska , i. vaskivskyi , t. mertelj , p. kusar , d. svetin , s. brazovskii , and d. mihailovic , science 344 , 177 ( 2014 ) . t. -r . t. han , f.zhou , c. d. malliakas , p. m. duxbury , s. d. mahanti , m. g. kanatzidis , and c. -y . ruan , sci . adv . 1 , e1400173 ( 2015 ) . p. giannozzi , et al . , j. phys . : condens . matter 21 , 395502 ( 2009 ) . j. p. perdew , j. a. chevary , s. h. vosko , k. a. jackson , m. r. pederson , d. j. singh , and c. fiolhais , phys . rev . b 46 , 6671 ( 1992 ) . n. marzari , d. vanderbilt , a. de vita , and m. c. payne , phys . 82 , 3296 ( 1999 ) . h. j. monkhorst and j. d. pack , phys . b 13 , 5188 ( 1976 ) . s. baroni , s. de gironcoli , a. dal corso , and p. giannozi , rev . 73 , 515 ( 2001 ) . m. d. johannes , i. i. mazin , and c. a. howells , phys . rev . b 73 , 205102 ( 2006 ) . c. battaglia , h. cercellier , f. clerc , l. despont , m. g. garnier , c. koitzsch , p. aebi , h. berger , l. forr , and c. ambrosch - draxl , phys . b 72 , 195114 ( 2005 ) . r. bianco , m. calandra and f. mauri , phys . b 92 , 094107 ( 2015 ) . q. zhang , l. -y . gan , y. cheng , and u. schwingenschlcgl , phys . b 90 , 081103(r ) ( 2014 ) . s. cahangirov , m. topsakal , e. aktrk , h. ahin , and s. ciraci , phys . 102 , 236804 ( 2009 ) . h.ahin , s. cahangirov , m. topsakal , e. bekaroglu , e. akturk , r. t. senger , and s. ciraci , phys . b 80 , 155453 ( 2009 ) . a. j. mannix , x. -f . zhou , b. kiraly , j. d. wood , d. alducin , b. d. myers , x. liu , b. l. fisher , u. santiago , j. r. guest , m. jose yacaman , a. ponce , a. r. oganov , m. c. hersam , and n. p. guisinger , science 350 , 1513 ( 2015 ) . m. a. bissett , s. konabe , s. okada , m. tsuji , and h. ago , acs nano 7 , 10335 ( 2013 ) . s. -i . park , j. -h . ahn , x. feng , s. wang , y. huang , and j. a. rogers , adv . 18 , 2673 ( 2008 ) . e. morosan , h. w. zandbergen , b. s. dennis , j. w. g. bos , y. onose , t. klimczuk , a. p. ramirez , n. p. ong and r. j. cava , nature phys . 2 , 544 ( 2006 ) . k. e. wagner , e. morosan , y. s. hor , j. tao , y. zhu , t. sanders , t. m. mcqueen , h. w. zandbergen , a. j. williams , d. v. west , and r. j. cava , phys . b 78 , 104520 ( 2008 ) .

the in - medium properties of the vector mesons ( @xmath1 and @xmath0 ) in hot and dense matter have an important role to play in the low mass dilepton production resulting from heavy ion collision experiments . this has , hence , been a topic of great interest in the recent past , both experimentally @xcite and theoretically @xcite . the experimental observation of enhanced dilepton production @xcite in the low invariant mass regime could be due to a reduction in the vector meson masses in the medium . brown and rho suggested the hypothesis that the vector meson masses drop in the medium according to a simple ( br ) scaling law @xcite , given as @xmath2 , where @xmath3 is the pion decay constant and asterisk refers to in - medium quantities . in the framework of quantum hadrodynamics ( qhd ) as a description of the hot hadronic matter , it is seen that the dropping of the vector meson masses has its dominant contribution arising from the vacuum polarisation effects from the baryon sector @xcite , which is not observed in the mean field approximation . this means that the quantum effects do play an important role in the medium modification of vector meson properties . there have also been approaches based on qcd sum rules @xcite which confirm a scaling hypothesis @xcite as suggested by brown and rho , with a saturation scheme that leads to a delta function at the vector meson pole and a continuum for higher energies for the hadronic spectral function . it is , however , seen that such a simple saturation scheme for finite densities does not work and a more realistic description for the hadronic spectral function is called for @xcite . using an effective lagrangian model to calculate the hadronic spectral function , it is seen that such a universal scaling law is not suggested for in - medium vector meson masses @xcite . the medium modifications of the vector mesons have been thus a subject of several investigations . it has been emphasized in the literature @xcite that the properties of the hadrons are modified due to their interactions with the thermal bath , and such modifications are reflected in the dilepton and photon spectra emitted from a hot and dense matter @xcite . dileptons are interesting probes for the study of evolving matter arising from relativistic heavy ion collisions since they do not interact strongly and escape unthermalized from the hot and dense matter at all stages of the evolution . the temperature @xcite and density @xcite modifications of the dileptons from hot hadronic matter as well as from a quark gluon plasma ( qgp ) resulting from heavy ion collisions have been a subject of extensive investigations in the recent past . broadly , two types of modifications for the hadrons in the medium are expected : shift in the pole position for the hadron propagator , giving rise to a modification of the mass , and broadening of the spectral function . in the present work , we shall attempt to study the medium modification of vector meson masses and decay widths in the hot hadronic matter within the framework of quantum hadrodynamics ( qhd ) taking the vacuum polarisation effects into account , and their subsequent effect on the dilepton spectra . in the conventional hadronic models @xcite , the masses of the vector mesons stay constant or increase slightly , in the mean field approximation , i.e. , when the polarization from the dirac sea is neglected @xcite . with the inclusion of quantum corrections from the baryonic sector , however , in the walecka model one observes a drop in the vector meson masses in the medium @xcite . this medium modification of the vector meson masses plays an important role in the enhanced dilepton yield @xcite for masses below the @xmath1 resonance in the heavy ion collision experiments @xcite . it has been emphasized recently that the dirac sea contribution ( taken into account through summing over baryonic tadpole diagrams in the relativistic hartree approximation ( rha ) ) dominates over the fermi sea contribution and @xmath4 is caused by a large dressing of @xmath5 cloud in the medium @xcite . further quantum effects arising from the scalar mesons have been considered @xcite along with the rha for the baryon sector in the context of vector meson mass modifications in strange hadronic matter @xcite . the present method is a step in that direction of studying medium modifications of the vector meson properties including such quantum effects from the scalar mesons in a self consistent manner , along with those arising from the baryon sector . it was earlier demonstrated in a nonperturbative formalism that a realignment of the ground state with baryon - antibaryon condensates is equivalent to the relativistic hartree approximation @xcite . the ground state for the nuclear matter was extended to include sigma condensates to take into account the quantum correction effects from the scalar meson sector @xcite . such a formalism includes multiloop effects and is self consistent @xcite . the methodology was then generalized to consider hot nuclear matter @xcite as well as to the situation of hyperon - rich dense matter @xcite relevant for neutron stars . the effect of vacuum polarisations on the vector meson masses has also been recently studied @xcite . in the present work , we study the effect of such quantum corrections on the in - medium vector meson properties . this apart , for studying @xmath1 meson properties , we take into account the @xmath1-@xmath6-@xmath6 interactions , and their effects on the dilepton production in the low invariant mass regime . the pionic interactions are seen to modify the @xmath1 meson mass only marginally , as already stated in the literature @xcite . due to scalar meson contributions , there is broadening in the @xmath0 and @xmath1 peaks , as compared to rha . this is seen to be quite pronounced for the @xmath0 meson , which leads to smearing and ultimate disappearance of the @xmath0 peak at finite densities @xcite . we organize the paper as follows . we first briefly recapitulate the nonperturbative framework used for studying the quantum correction effects in hot nuclear matter in section ii . the medium modification of the @xmath0 and @xmath1 meson masses and decay widths are considered in section iii . we discuss the effect of these medium modifications on the low mass dilepton spectra in section iv . in section v , we summarize the results of the present work and give an outlook . we briefly recapitulate here the vacuum polarisation effects arising from the nucleon and scalar meson fields in hot nuclear matter in a nonperturbative variational framework @xcite . the method of thermofield dynamics ( tfd ) @xcite is used here to study the ground state " ( the state with minimum thermodynamic potential ) at finite temperature and density within the walecka model with a quartic scalar self interaction . the temperature and density dependent baryon and sigma masses are also calculated in a self - consistent manner in the formalism . the ansatz functions involved in such an approach are determined through functional minimisation of the thermodynamic potential . the lagrangian density in the walecka model is given as @xmath7 in the above , @xmath8 , @xmath9 , and @xmath10 are the fields for the nucleon , @xmath9 , and @xmath0 mesons with masses @xmath11 , @xmath12 , and @xmath13 respectively . since we are interested in symmetric nuclear matter , the isovector rho meson does not contribute . the quartic coupling term in @xmath9 is necessary for the sigma condensates to exist , through a vacuum realignment @xcite . our calculations thus include the quantum effects arising from the sigma meson in addition to the mean field contribution from the the quartic self interaction of the scalar meson . we retain the quantum nature of both the nucleon and the scalar meson fields , whereas the vector @xmath0 meson is treated as a classical field , using the mean field approximation for the @xmath0 meson , given as @xmath14 . the reason is that without higher order term for the @xmath0 meson , the quantum effects generated due to the @xmath0 meson through the present variational ansatz turn out to be zero . we then write down the expressions for the thermodynamic quantities including the quantum effects . the details regarding the formalism can be found in earlier references @xcite . the energy density , after carrying out the renormalisation procedures for the baryonic and scalar meson sectors @xcite , is obtained as @xmath15 with @xmath16 @xmath17 @xmath18 @xmath19 as the mean field result , contribution from the dirac sea , and contributions from the @xmath0 and @xmath9 mesons , respectively . we might note here that the quantum effects arising here from the scalar meson sector through @xmath9 meson condensates amounts to a sum over a class of multiloop diagrams and , does not correspond to the one meson loop approximation for scalar meson quantum effects considered earlier @xcite . in the above , @xmath20 and @xmath21 are the thermal distribution functions for the baryons and antibaryons , given in terms of the effective energy , @xmath22 , and the effective chemical potential , @xmath23 . @xmath24 is the effective nucleon mass and @xmath25 is the baryon number density after subtracting out the vacuum contribution . the spin - isospin degeneracy factor is @xmath26 for symmetric nuclear matter . in the expression for the energy density arising from the scalar meson sector , the field dependent effective sigma mass , @xmath27 , satisfies the gap equation in terms of the renormalised parameters as @xmath28 where @xmath29 where @xmath30 . the effective meson mass @xmath31 is determined through a self consistent solution of eq . ( [ mm2 ] ) . in eq . ( [ vph0 ] ) , @xmath32 and @xmath33 are the expressions as given by eqs . ( [ mm2 ] ) and ( [ if ] ) with @xmath34 . we might note here that the gap equation given by eq . ( [ mm2 ] ) is identical to that obtained through resumming the daisy and superdaisy graphs @xcite and hence includes higher order corrections from the scalar meson field in a self - consistent manner . extremisation of the thermodynamic potential , @xmath35 , with respect to the meson fields @xmath36 and @xmath37 yields the self consistency conditions for @xmath36 ( and hence for the effective nucleon mass @xmath38 ) and for the vector meson field @xmath37 . we now examine the medium modification to the masses and decay widths of the @xmath0 and @xmath1 mesons including the quantum correction effects in hot nuclear matter in the relativistic random phase approximation . the interaction vertices for these mesons with nucleons are given as @xmath39 where @xmath40 or @xmath41 , @xmath42 is the free nucleon mass , @xmath8 is the nucleon field and @xmath43 or @xmath44 , @xmath45 being the pauli matrices . @xmath46 and @xmath47 correspond to the couplings due to the vector and tensor interactions for the corresponding vector mesons to the nucleon fields . the vector meson self energy is expressed in terms of the nucleon propagator , @xmath48 modified by the quantum effects . this is given as @xmath49,\ ] ] where @xmath50 is the isospin degeneracy factor for nuclear matter , and @xmath51 represents the meson - nucleon vertex function . the vector meson self energy can be written as the sum of two parts @xmath52 where @xmath53 is the contribution arising from the vacuum polarisation effects , described by the coupling to the @xmath54 excitations and @xmath55 is the density dependent contribution to the vector self energy . for the @xmath0 meson , the tensor coupling is generally small as compared to the vector coupling to the nucleons @xcite , and is neglected in the present work . the feynman part of the self energy , @xmath56 is divergent and needs renormalization . we use dimensional regularization to separate the divergent parts and the renormalization condition @xmath57 . because of the tensor interaction , the vacuum self energy for the @xmath1 meson is not renormalizable . we employ a phenomenological subtraction procedure @xcite to extract the finite part using the renormalization condition @xmath58 , with ( @xmath59 ) . the effective mass of the vector meson is obtained by solving the equation , with @xmath60 , @xmath61 the decay width for the process @xmath62 is calculated from the imaginary part of the self energy using the cutkosky rule , and in the rest frame of the @xmath1 meson is given by @xmath63 \label { gmrho}\ ] ] where , @xmath64^{-1}$ ] is the bose - einstein distribution function . the first and the second terms in the equation ( [ gmrho ] ) represent the decay and the formation of the resonance , @xmath1 . the medium effects have been shown to play a very important role for the @xmath1 meson decay width . in the calculation for the @xmath1 decay width , the pion has been treated as free , and any modification of the pion propagator due to effects such as delta - nucleon hole excitation @xcite to yield a finite decay width for the pion , have not been taken into account . the coupling @xmath65 is fixed from the decay width of @xmath1 meson in vacuum ( @xmath66=151 mev ) decaying to two pions . for the nucleon - rho couplings , we use the vector and tensor couplings as obtained from the n - n forward dispersion relation @xcite . with the couplings as described above , we consider the temperature and density dependence of the @xmath0 and @xmath1 meson properties in the hot nuclear matter due to quantum correction effects . to calculate the decay width for the @xmath0 meson , we write down the effective lagrangian for the @xmath0 meson as @xcite @xmath67 in the above , the first term refers to the direct coupling of the vector meson @xmath0 to the photon , and hence to the dilepton pairs , as given by the vector dominance model . the decay width of the @xmath0 meson in vacuum is dominated by the channel @xmath68 . in the medium , the decay width for @xmath69 is given as @xmath70,\end{aligned}\ ] ] where @xmath71 , @xmath72 and @xmath73 s are 4-momenta and energies for the pions , and @xmath74 s are their thermal distributions . the matrix element @xmath75 has contributions from the channels @xmath76 ( described by the second term in ( [ lomg ] ) ) and the direct decay @xmath69 resulting from the contact interaction ( last term in ( [ lomg ] ) ) @xcite . for the @xmath77 coupling we take the value @xmath78 = 2 according to ref . @xcite , which is similar in value to the coupling taken in @xcite . we fix the point interaction coupling @xmath79 by fitting the partial decay width @xmath68 in vacuum ( 7.49 mev ) to be 0.24 . the contribution arising from the direct decay is upto around 15 % which is comparable to the results of @xcite . with the modifications of the vector meson masses in the hot and dense medium , a new channel becomes accessible , i.e. , the decay mode @xmath80 for @xmath81 , which is not possible in the vacuum ( since @xmath82 ) . the in - medium decay width for this process is found from the cutting rules at finite temperature to be @xmath83\ ] ] where @xmath84 arises from phase space considerations , and @xmath85 s are the thermal distribution functions for the pions and @xmath1 mesons . the first and second terms refer to the emission and absorption of pion and @xmath1 meson . the pion annihilation ( @xmath86 ) dominates dilepton production in the low invariant mass region . the dropping of the vector meson masses in the hot and dense matter also gives rise to substantial contribution to the dilepton yield from the direct decays of the vector mesons into dileptons ( @xmath87 , @xmath88 ) . the thermal production rate per unit four volume for lepton pairs is related to the imaginary part of the one particle irreducible photon self energy by @xcite @xmath89 where @xmath90 is the 4-momentum of the virtual photon , and @xmath91 is the fine structure constant . the invariant mass distribution for the lepton pairs from pion annihilation is given by @xmath92 where , @xmath93 , @xmath94 . in the above , @xmath95 is the cross section for the pion annihilation , and is given by @xmath96 where @xmath97 is the electromagnetic form factor of the pion . the photon can also couple to the hadrons through vector mesons , which is the basis of the vector dominance model @xcite . the photon can thus interact with the pions through a vector meson , which must be an isovector and is taken to be the @xmath1 meson . the effective lagrangian for the photon - pion - rho interaction can then be written as @xcite @xmath98 where @xmath99 is the field tensor for the @xmath1 field and @xmath100 . the coupling @xmath65 in the above can be determined from the decay @xmath62 , and the direct photon - rho coupling @xmath101 is determined from the process @xmath102 @xcite . these , including a finite decay width for the @xmath1 meson , then lead to the @xmath1-dominated pion form factor as @xmath103 in the rest frame of the @xmath1 meson , the invariant mass distribution of the lepton pair of invariant mass @xmath11 from pion annihilation is given as @xmath104 with the cross section @xmath105 for the pion annihilation , given in terms of the pion form factor at @xmath106 . the summation over @xmath107 arises from the bose - einstein distribution function for the @xmath1 meson . since we are mostly in the regime @xmath108 , the first term in the series , corresponding to the boltzmannian distribution function is dominant , with quantum corrections being of the order of upto 10 % . a vector meson @xmath109 in a thermal bath can decay into leptons and photons , described by the final state @xmath110 say , which escape the bath without thermalization . with @xmath111 as the four momentum of the nonhadronic state @xmath110 , a resonance peak should appear in the invariant mass ( @xmath11 ) plot for the number of lepton pair events , versus @xmath106 at @xmath112 , where @xmath113 is the mass of the resonance in the thermal bath . hence , the medium modifications of the vector mesons get reflected in the dilepton spectra through the shift and broadening of the peaks . the invariant mass distribution of lepton pairs from the decays of vector mesons is given by a generalization of the breit wigner formula as @xcite @xmath114 in the above , @xmath115 is the number of lepton pair events , say in the space time and four momentum element @xmath116 , @xmath117 being the spin of the resonance , @xmath118 is the formation width , @xmath119 is the total width and @xmath120 is the partial decay width for the off - shell @xmath109 ( i.e. of mass @xmath11 ) to go into the nonhadronic state @xmath110 . the total width @xmath121 is the rate at which the particles equilibriate and relax to chemical equilibrium . the invariant mass distribution of the lepton pairs from decays of vector mesons propagating with a momentum @xmath122 is obtained from equation ( [ gbw ] ) as @xmath123 q_0 \gamma _ { v \rightarrow e^+ e^-}^{\rm vac}\ ] ] in the rest frame of the vector meson , this reduces to @xmath124 in the above , @xmath125 is the partial width for the leptonic decay for the off - shell vector meson @xmath109 with mass @xmath11 , and is given as @xmath126 the coupling @xmath46 refers to the direct coupling of the vector meson @xmath109 to the photon as given in the equations ( [ lomg ] ) and ( [ lrho ] ) for the @xmath0 and @xmath1 mesons respectively . the coupling of @xmath0 to the photon taken to be one third of the coupling of @xmath1 to the photon gives reasonable agreement with the ratio of the observed partial widths @xmath127 @xcite . in the following we present numerical results on the influence of vacuum effects on hadronic masses , decay widths of vector mesons , and dilepton emission from hot hadronic matter . as described in the previous sections , the study is based entirely on a hadronic scenario and makes use of the original walecka model , extended by a nonlinear sigma self coupling . in this work we concentrate on one particular aspect , namely on the role of scalar vacuum effects . for each quantity studied , we compare the results of the relativistic hartree approximation ( rha ) , which was already treated in ref . @xcite , with those additional effects arising from a scalar vacuum condensate . the quartic self coupling @xmath128 is taken as a free parameter ; we have used @xmath129 and @xmath130 as representative values . the values of the incompressibility corresponding to these self couplings are 401 mev ( for @xmath128=1.8 ) and 329 mev ( for @xmath128=5 ) . these values are smaller than the mean field result of 545 mev as well as the relativistic hartree result ( @xmath128=0 ) of 450 mev . this is because an increase of @xmath128 leads to weaker potentials of the @xmath9 and @xmath0 mesons and to a softening of the equation of state . the parameters of the model are adjusted to fit nuclear matter , corresponding to a saturation density of @xmath131 = 0.193/@xmath132 . the fitted values for the scaled coupling constants are @xmath133 183.3 , 167.5 and 137.9 , and @xmath134 114.7 , 96.45 , and 63.7 for @xmath135 ( i.e. , rha ) , 1.8 , and 5 , respectively . when discussing dilepton production , we do not attempt to give a realistic description of specific experimental results . for this , the dynamic space - time evolution for the hot hadronic matter in the course of a nuclear collision would have to be addressed and additional effects contributing to the dilepton spectrum would have to be included . however , our study should help to elucidate , in which way in - medium vacuum effects can have an influence on experimentally observable quantities . we first discuss the effect of quantum corrections on the in - medium nucleonic properties for hot nuclear matter @xcite . in fig . 1a , we plot the temperature dependence of the nucleon mass for the baryon density @xmath136 and the nuclear matter density @xmath137 . the quantum corrections arising from the sigma field have the effect of softening the equation of state , and produce a higher value for the effective nucleon mass as compared to the rha . the effective nucleon mass was seen to increase with the vacuum polarisation effects arising from the baryons as compared to the mft calculations @xcite . though for @xmath136 , the effective nucleon mass decreases monotonically with temperature , for @xmath138 , the nucleon mass first increases and then decreases as a function of temperature , which has also been observed earlier to be the trend for higher densities @xcite . the variation in the nucleon mass is very slow up to a temperature @xmath139 150 mev , beyond which there is a fast decrease . though qualitative features are similar for the in - medium mass with inclusion of quantum effects from the scalar meson sector , it is observed that such effects lead to a higher value for the nucleon mass . we next study the temperature and density dependence of the vector mesons in hot nuclear matter . the @xmath0 meson mass as a function of the temperature for @xmath140=0 and @xmath131 , is plotted in fig . 1b . in the walecka model , the dirac sea has been shown to have a significant contribution over the fermi sea , leading to a substantial drop in the omega meson mass in nuclear matter @xcite . specifically , at saturation density and at zero temperature , the decrease in the omega meson mass from the vacuum value , is around 150 mev in rha , whereas with sigma quantum effects , the drop in the mass reduces to around 117 mev or 61 mev for the two values of @xmath128 studied . a similar reduction in the mass drop , as compared to the rha was also observed when scalar field quantum corrections were included within a one loop approximation @xcite , unlike the approximation adopted here , which is self - consistent and includes multi - loop effects . this indicates that the quantum corrections do play an important role in the medium modification of the vector meson masses . in the present work , it is seen that the quantum corrections from the scalar mesons , over those arising from the baryonic sector , have the effect of giving rise to a higher value for the @xmath0 meson mass . this again is related to the fact that the effective nucleon mass is higher when such quantum effects are taken into account . in figure [ figf2 m ] , we illustrate the medium modification for the @xmath1 meson mass with the vector and tensor couplings to the nucleons being fixed from the nn forward dispersion relation @xcite . the values for these couplings are given as @xmath141=0.55 and @xmath142=6.1 . we notice that the decrease in the @xmath1 meson mass with increase in temperature is much sharper than that of the @xmath0 meson . such a behaviour of the @xmath1 meson undergoing a much larger medium modification was also observed earlier @xcite within the relativistic hartree approximation for the nucleons . this indicates that the tensor coupling , which is negligible for the @xmath0 meson , plays a significant role for the @xmath1 meson . with inclusion of quantum corrections from the scalar meson , the qualitative comparison between the @xmath0 and @xmath1 vector mesons remains the same , though these effects are generally seen to lead to larger values for the vector meson masses . we have also calculated the contribution to the @xmath1 meson mass arising from pion loops through the @xmath143 coupling , but this was found to lead to a marginal increase of at most 1% , in agreement with @xcite . in figure [ figf3 m ] , the decay widths for the @xmath1 meson with and without the scalar quantum corrections , over the relativistic hartree approximation , are considered . such quantum effects lead to an appreciable increase in the decay widths , which , as we shall see , is observed as a broadening of the @xmath1-peak in the dilepton spectra . in figure [ figf4 m ] , the decay width for the @xmath0 meson is seen to have a positive contribution arising from the @xmath9-meson quantum effect as compared to rha . the decay width is calculated as arising from the processes @xmath69 as well as @xmath80 , a channel which becomes possible with the mass modifications of the vector mesons in the medium . it is seen that the channel @xmath144 opens for temperature greater than around 180 mev for @xmath136 , whereas it is observed even at zero temperature at nuclear matter saturation density . we might note that the collisional modification to the @xmath0 width , arising from the process @xmath145 @xcite , has not been taken into account in the present calculations . the dilepton spectra for the low invariant mass regime , at t=200 mev are shown in figures [ figf5 m ] and [ figf6 m ] calculated for densities @xmath136 and @xmath138 . the total dilepton spectra along with contributions arising from pion annihilation and direct decays of the @xmath1 and @xmath0 mesons , are displayed in figures 5a and 6a . the effect of sigma condensates shifts the peak position for the vector mesons to higher values as compared to rha , and also results in a broadening of the peaks in the dilepton spectra as can be seen from figures 5b and 6b . the @xmath1 and @xmath0 peaks are well separated due to the fact that the @xmath1 meson mass gets much larger medium modification as compared to the @xmath0 meson , which would not be the case in the absence of medium effects . the omega peak is seen to be smeared due to scalar meson quantum corrections . the broadening of the @xmath0 peak due to @xmath9-quantum effects is more clearly visible for @xmath138 than for zero density , in the dilepton spectra . the dominant contributions to the dilepton enhancement are from pion annihilation and @xmath1 meson decay , which are plotted in figures 5c and 6c in comparison with the rha results . the effect of @xmath9 condensates leads to a higher value for the peak position as well as broadening of the @xmath1-peak . the dilepton production rates due to the @xmath0 meson decay are displayed in figures 5d and 6d . it is observed that the broadening of the @xmath0 peak due to scalar quantum corrections is larger for @xmath131 than for zero density . this is a reflection of the fact that there is a larger modification of the omega width due to scalar quantum effects for higher densities . there have been various mechanisms under discussion which can lead to an increase in the @xmath0 width . it has been argued that the @xmath0 decay width can increase by an order of magnitude due to the reduction in the pion decay constant related with chiral symmetry restoration @xcite . the medium modification due to scalar meson quantum effects giving rise to higher values for the @xmath1 meson mass and decay width are reflected in the dilepton spectra . the medium modification for the vector meson masses depend more sensitively on the density than on temperature . this can give rise to an enhancement of the dilepton production rate when matter is highly compressed at the initial stage of a heavy ion collision . in the dls experiments for a - a collisions at 1 agev , there have been models that focus on density induced change for the @xmath1 meson mass to describe the dilepton pair production for invariant mass of dilepton pair , @xmath146gev @xcite . even though there is dilepton enhancement arising from the density modified @xmath1 mass , the results still underestimate the dls data by a factor of 2 - 3 @xcite . the mass modifications for the vector meson masses with density are plotted in figure [ figf7 ] , for temperature t= 100 mev . the trend is a sharp decrease of the masses with density up to @xmath147 , and then a rise for higher values of @xmath140 . the observed trend is due to the competing effects of vacuum polarisation which leads to a reduction in the mass , and the density dependent dressing for the vector self energy causing an increase in the meson mass . such a behaviour of the vector meson mass for the @xmath0 meson was also observed earlier @xcite . a higher value of the scalar self interaction coupling , @xmath128 , leads to a higher value for the vector meson masses . the change in the @xmath1 meson mass for higher densities is seen to be rather slow which means that there is a delicate cancellation between the feynman and the density dependent parts of the vector self energy . as depicted in figure [ figf7 ] , the @xmath0 meson mass seems to be more dominated by the density dependent contributions for higher densities . the decay widths are plotted in figure [ figf8 ] . the slow change in @xmath1 meson mass for higher densities is reflected in the in - medium @xmath1-decay width , while the @xmath0 decay width @xmath148 grows strongly with density and gets strongly enhanced by quantum effects from the @xmath9 meson . the dilepton spectra for @xmath149 mev and @xmath150 are plotted in figures [ figf91 ] and [ figf92 ] . these values should be typical for the compression zone of a nuclear collision at energies of a few gev per nucleon . for example , for inclusive ca+ca collisions at 1 agev studied by the dls collaboration @xcite , maximum temperatures and densities in this range were extracted from a transport model @xcite . in figure 9a , the total dilepton spectrum is plotted , for @xmath128=1.8 , displaying the individual contributions . figure 9b shows that the sigma quantum corrections lead to a broadening of the peaks in the dilepton spectra , which is seen to be more appreciable for the higher value of the scalar self coupling , @xmath128 . the individual contributions to the dilepton spectra , plotted in figure [ figf92 ] , illustrate the effect of @xmath9 quantum corrections . finally , the density dependence for the dilepton spectra is shown for temperature @xmath149 mev in figures [ figf101 ] and [ figf102 ] for couplings @xmath128 as 1.8 and 5 , respectively , in comparison with the rha results . the vector meson masses decreasing with density up to around twice the nuclear matter density and then slowly increasing or staying constant , are reflected in the dilepton spectra through the peak positions . one might notice that the broadening of the peaks due to @xmath9- quantum effects is larger for higher densities and is more pronounced for the @xmath0 peak . the curves for densities twice and four times nuclear matter density in figure [ figf102 ] are coincident up to a value of the invariant mass of about 0.6 gev . this is due to the fact that the rho meson properties whose effects are relevant for that mass region , remain almost constant for @xmath128=5 . beyond that however , they are different around the invariant mass region for the omega meson . due to substantial increase in the decay width of the @xmath0 meson with density the @xmath0 peak is almost smeared out even at nuclear matter density and the omega loses its quasi particle nature . the corrections from the scalar meson quantum effects are thus seen to be appreciable . to summarize , we have considered in the present paper , the modifications of the vector meson ( @xmath1 and @xmath0 ) properties due to quantum correction effects in hot and dense nuclear matter and their effect on dilepton spectra . the baryonic properties , as modified due to such effects , subsequently determine the vector meson masses and decay widths in the hot and dense hadronic matter . it has been recently emphasized that the dirac sea contribution dominates over the fermi sea , thus implying that the vacuum polarisation effects arising from the baryonic sector do play an important role in the vector meson properties in a medium . we consider the hadronic properties as modified arising due to the quantum effects from the baryon and scalar meson sectors and investigate their effect on the dilepton production rate . the effect of the scalar meson quantum corrections leads to an increase in the vector meson masses and decay widths . the effect on the omega meson decay width is particularly more pronounced at higher densities , than in the rha . the value for @xmath149 mev is modified from 67 mev ( for rha ) to 125 mev ( for @xmath128=1.8 ) and 251 mev ( for @xmath128=5 ) for @xmath151=2 , and for @xmath151=4 the rha value of 101 mev is modified to 175 mev and 370 mev , for @xmath128=1.8 and 5 , respectively . in the present work , we have studied the effect of quantum corrections arising from the scalar meson over those resulting from the baryon sector ( given through the summing over baryonic tadpole diagrams in the relativistic hartree approximation ) . the quantum corrections from the @xmath9-meson give rise to a higher value for the vector meson masses as compared to keeping only quantum effects from the baryons . there is also an appreciable increase in the decay widths in the presence of these quantum corrections , and this is reflected in the invariant mass plot for dilepton spectra through the broadening of the peaks . for the @xmath0 meson , an appreciable broadening arising due to sigma meson quantum effects leads to the disappearance of the peak from the spectra for high densities when @xmath0 ceases to be exist as a quasi particle . the density dependence of in - medium properties of the vector mesons and their subsequent effects on the dilepton spectra have been studied , which will be relevant for highly compressed matter at the early stage of a heavy ion collision . one of the authors ( am ) acknowledges financial support from alexander von humboldt foundation when the work was initiated . am would like to thank s. scherer , h. weber , d. dietrich for discussions , and would like to acknowledge the institut fr theoretische physik for warm hospitality and dfg ( sto 161/8 - 1 ) for financial support . 99 n. masera for the helios-3 collaboration , nucl . phys . * a 590 * , 93c ( 1995 ) . g. agakichiev et al ( ceres collaboration ) , phys . rev . lett . * 75 * , 1272 ( 1995 ) ; g. agakichiev et al ( ceres collaboration ) , phys . lett . * b 422 * , 405 ( 1998 ) ; g. agakichiev et al ( ceres collaboration ) , nucl . phys . * a 661 * , 23c ( 1999 ) . r. j. porter et al ( dls collaboration ) , phys . . lett . * 79 * , 1229 ( 1997 ) ; w. k. wilson et al ( dls collaboration ) , phys . c 57 , 1865 ( 1998 ) . d. p. morrison ( phenix collaboration ) , nucl . phys . * a 638 * , 565c ( 1998 ) . j. stroth ( hades collaboration ) , advances nuclear dynamics * 5 * , 311 ( 1999 ) . g. e. brown and m. rho , phys . lett . * 66 * , 2720 ( 1991 ) . t. hatsuda and su h. lee , phys . rev . * c 46 * , r34 ( 1992 ) ; t. hatsuda , s. h. lee and h. shiomi , phys . rev . * c 52 * , 3364 ( 1995 ) . x. jin and d. b. leinweber , phys . * c 52 * , 3344 ( 1995 ) ; t. d. cohen , r. d. furnstahl , d. k. griegel and x. jin , prog . * 35 * , 221 ( 1995 ) ; r. hofmann , th . gutsche , a. faessler , eur . j. * c 17 * , 651 ( 2000 ) ; s. mallik and k. mukherjee , phys . rev . * d 58 * , 096011 ( 1998 ) . s. mallik and a. nyffeler , phys . rev . * c 63 * , 065204 ( 2001 ) . f. klingl , n. kaiser , w. weise , nucl . phys . * a 624 * , 527 ( 1997 ) . c. ernst , s. a. bass , m. belkacem , h. stcker and w. greiner , phys * c 58 * , 447 ( 1998 ) . h. shiomi and t. hatsuda , phys . b 334 * , 281 ( 1994 ) . t. hatsuda , h. shiomi and h. kuwabara , prog . . phys . * 95 * , 1009 ( 1996 ) . jeans , j. piekarewicz and a. g. williams , phys . rev . * c 49 * , 1981 ( 1994 ) ; k. saito , k. tsushima , a. w. thomas , a. g. williams , phys . * b 433 * , 243 ( 1998 ) . jan - e alam , s. sarkar , p. roy , b. dutta - roy and b. sinha , phys . rev . * c 59 * , 905 ( 1999 ) . r. rapp and j. wambach , adv . phys . * 25 * , 1 ( 2000 ) . c. m. ko , v. koch and g. q. li , ann . * 47 * , 505 ( 1997 ) . jan - e alam , p. roy , s. sarkar and b. sinha , nucl - th/0106038 ; jan - e alam , s. sarkar , p. roy , t. hatsuda and b. sinha , ann . phys . * 286 * , 159 ( 2000 ) ; f. karsch , e. laermann , p. petreczky , s. stickan , i. wetzorke , hep - lat/0110208 . k. redlich , j. cleymans , v. v. goloviznin , in proceedings on nato advanced workshop on hot hadronic matter : theory and experiment , n.y . plenum press , 1995 , 562p ; a. dumitru , d. h. rischke , th . schnfeld , l. winckelmann , h. stcker and w. greiner , phys . lett . * 70 * , 2860 ( 1993 ) ; p. jaikumar , r. rapp and i. zahed , hep - ph/0112308 ; song gao , ru - keng su , xue- qian li , comm . * 28 * , 207 ( 1997 ) ; d. dutta , k. kumar , a. k. mohanty , r. k. choudhury , phys . rev . * c 60 * , 014905 , 1999 . b. d. serot and j. d. walecka , adv . nucl . phys . * 16 * , 1 ( 1986 ) . s. a. chin , ann . phys . * 108 * , 301 ( 1977 ) ; m. asakawa , c. m. ko , p. levai and x. j. qiu , phys . rev . * c 46 * , r1159 ( 1992 ) . s. pal , song gao , h. stcker and w. greiner , phys * b 465 * , 282 ( 1999 ) . a. mishra , p. k. panda , s. schramm , j. reinhardt and w. greiner , phys . * c 56 * , 1380 ( 1997 ) . a. mishra and h. mishra , j. phys . * g 23 * , 143 ( 1997 ) . s.y . pi and m. samiullah , phys . rev . * d 36 * , 3121 ( 1987 ) ; g. a. camelia and s.y . pi , phys . rev . * d 47 * , 2356 ( 1993 ) . a. mishra , p. k. panda and w. greiner , jour . g 27 * , 1561 ( 2001 ) . a. mishra , p. k. panda and w. greiner , jour . * g 28 * , 67 ( 2002 ) . a. mishra , j. c. parikh and w. greiner , jour . * g 28 * , 151 ( 2002 ) . c. gale and j. i. kapusta , nucl . * b 357 * , 65 ( 1991 ) . r. a. schneider , t. renk and w. weise , nucl . * a 698 * , 428 ( 2002 ) ; t. renk , r. a. schneider and w. weise , phys . rev . * c 66 * , 014902 ( 2002 ) . h. umezawa , h. matsumoto and m. tachiki , _ thermofield dynamics and condensed states _ ( north - holland , amsterdam , 1982 ) . m. asakawa , c. m. ko , p. levai and x. j. qiu , phys . rev . * c 46 * , r1159 ( 1992 ) ; m. herrmann , b. l. friman and w. nrenberg , nucl . phys . * a 560 * , 411 ( 1993 ) ; g. chanfray and p. shuck , nucl . phys . * a 545 * , 271c ( 1992 ) . w. grein , nucl . * b 131 * , 255 ( 1977 ) ; w. grein and p. kroll , nucl . phys . * a 338 * , 332 ( 1980 ) . j. j. sakurai , _ currents and mesons _ ( the university od chicago press , chicago , 1969 ) . m. gell - mann , d. sharp , and w. d. wagner , phys . rev . lett . * 8 * , 261 ( 1962 ) . b. a. li , phys . rev . * d 52 * , 5165 ( 1995 ) . f. klingl , n. kaiser and w. weise , z. phys . * a 356 * , 193 ( 1996 ) . . kaymakcalan , s. rajeev and i. schechter , phys . rev . * d 30 * , 594 ( 1984 ) . m. le bellac , _ thermal field theory _ ( cambridge university press , new york , 1996 ) . h. b. connell , b. c. pearce , a. w. thomas and a. g. williams , prog . . phys . * 39 * , 201 ( 1997 ) . h. a. weldon , ann . ( n. y. ) * 228 * , 43 ( 1993 ) . g. q. li , c. m. ko , g. e. brown , nucl . phys . * a 606 * , 568 ( 1996 ) . s. sarkar , jan - e alam , p. roy , a.k . dutt - mazumder , b. dutta - roy and b. sinha , nucl . phys . * a 634 * , 206 ( 1998 ) . r. d. pisarski , hep - ph/9503330 . g. chanfrey and p. schuck , nucl . phys . * a 555 * , 329 ( 1993 ) ; g. q. li and c. m. ko , nucl . phys . * a 582 * , 731 ( 1995 ) . e. l. bratkovskaya , w. cassing , r. rapp and j. wambach , nucl . phys . * a 634 * , 168 ( 1998 ) . j. zhang and c. gale , phys . rev . * c 50 * , 1617 ( 1994 ) .

seeing a quantum effect , and even making practical use of it , does not necessarily imply recognising it as such . magnetic compasses have been around for millenia , but quantum nature of ferromagnetism could not be understood till the discovery of the electron spin and the pauli principle . more than forty years elapsed between stewart s and kirchhoff s experiments with thermal radiation and planck s dissertation . already in the 20th century , it took three years for einstein to come up with a quantum explanation of lenard s experiments . clearly there must exist certain _ mathematical patterns _ in quantum equations of motion allowing one to look at the quantum without recognising it . an immediate reservation is in place here . this paper concerns , and is limited to , phase - space formulation of quantum electrodynamics based on the time - normal operator ordering @xcite . this is only one example in the multitude of analogies existing between the quantum and classical mechanics ( see , e.g. , @xcite ) . our work should be kept distinct from attempts to modify quantum electrodynamics by making arbitrary physically motivated assumptions about properties of matter ( cf . @xcite ) , as well as from attempts to imitate quantum behaviour in classical stochastic electrodynamics ( cf . @xcite ) . another well established tradition is applying methods of quantum field theory to classical statistical mechanics @xcite . worth mentioning are also attempts to implement classical mechanics in the hilbert space , see @xcite and references therein . the result of this paper in a nutshell is that _ quantum electrodynamics and classical stochastic electrodynamics share their macroscopic dynamical structure . _ more specifically , all relations for generalised phase - space quasi - distributions @xcite we obtain in this paper lack planck s constant and , _ ipso facto _ , coincide with the corresponding relations of the classical stochastic electrodynamics for probability distributions . this lack of planck s constant is exactly the aforementioned mathematical pattern . one way of formulating our results is that truly quantum dynamics is limited to microscopic conditions of indistinguishability and/or equations of motion of matter . for purposes of this paper , _ macroscopic _ refers to electromagnetic interactions under conditions of distinguishability . etymologies aside , we associate `` macroscopic '' and `` microscopic '' with `` distinguishable '' and `` indistinguishable '' , and not with `` large '' and `` small '' . `` distinguishability '' has its standard meaning ( formalised as commutativity of relevant dynamical variables in the interaction picture ) . on the commonly used term _ mesoscopic _ see endnote @xcite ; we treat macroscopic and mesoscopic as synonyms . the logic of the paper revolves around two concepts : `` obscured macroscopic view '' and `` doing quantum electrodynamics while thinking classically '' . the former refers to a closed albeit , strictly speaking , phenomenological quantum dynamical framework confined to the electromagnetic field and current operators . the latter expresses the critical property of such framework : on rewriting it in the so - called _ response picture _ @xcite , planck s constant drops out . this is another manifestation of the said mathematical pattern . any relation within `` obscured macroscopic view '' may therefore be obtained by formal quantisation of its classical counterpart . we stress that `` obscured macroscopic view '' is not an approximation , nor is it a modification of conventional quantum mechanics . it is a _ voluntary limit _ we impose on information extracted from a microscopic theory . it does not impose any restrictions on the theory itself , except that it be consistent with conventional quantum electrodynamics . implications of these results , with a multitude of specifications and reservations , are discussed at length in the paper . here we allow ourselves only a brief comment so as to clarify our motivation . particulars aside , quantum measurement is interaction of classical apparata with a quantum system . nice as it may sound , this definition has two obvious problems . classical apparata in the strict meaning of the term do not exist in nature , and it is not at all clear which laws quantum or classical should govern such interaction . both problems disappear if we limit `` measurement '' to electromagnetic interaction of an observer e.g . , a human being with the rest of the world . `` obscured macroscopic view '' is a formal expression of limitations of such `` electromagnetic observer '' . the difference between the classical and the quantum for such observer is _ only _ in whether currents he / she sees may or may not be _ phenomenologically _ interpteted as stochastic c - numbers . if such interpretation happens to be possible for some macroscopic quantum system , the latter becomes phenomenologically indistinguishable from a classical system . we term such systems classically behaving . `` classical apparata '' are classically behaving quantum systems , i.e. , _ inherently quantum _ systems which _ appear classical _ to the observer . their interaction with other quantum systems is governed by laws of quantum electrodynamics , which in this case are structurally identical with classical electrodynamics . if all devices involved behave classically , the whole situation reverts to classical electrodynamics . in particular , the classical world we perceive in everyday life is a collection of classically behaving quantum systems . as we demonstrate in this paper , these leading considerations are fully justified by the formal structure of quantum electrodynamics . the goal of this paper is twofold . firstly , we extend the earlier analyses @xcite to distinguishable matter subsystems . we show that the _ formal _ response characterisation of a solitary electromagnetic device @xcite becomes its _ physical _ characterisation as a radiation scatterer were this device interacting electromagneticaly with other macroscopic devices . this proves overall physical consistency of the response viewpoint . secondly , we establish the connection between our approach and the conventional phase - space techniques , making all results , the old as well as the new , intuitive . to this end we reformulate everything in terms of _ conditional p - functionals _ @xcite ; these quantities generalise both _ conditional probability distributions _ of classical stochastics and _ quasiprobability distributions _ of the conventional phase - space techniques , to arbitrary nonlinear non - markovian bosonic quantum systems . our analyses imply generalisation of the conventional phase - space techniques , firstly , beyond the resonance , or rotating wave , approximation ( rwa ) , and , secondly , to quantum response problems . coherent states of the harmonic oscillator , which traditionally serve as an entry point to phase - space approaches , were introduced by schrdinger as early as 1926 @xcite . that quantum dynamics of free bosonic systems maps to classical dynamics irrespective of the quantum state was firstly noticed by feynman in his review on path integrals @xcite . this understanding was instrumental in developing quantum theory of photodetection by glauber @xcite . glauber s theory was initially formulated for free electromagnetic fields , then extended to interacting fields by kelley and kleiner @xcite . however , de haan @xcite and later bykov and tatarskii @xcite pointed out that kelley - kleiner s results are limited to the rwa . taking them outside the rwa leads to causality violations ( we associate causality with retardation , cf . also endnote @xcite ) . lifting this restriction by amending the glauber - kelley - kleiner definition is one of the key results of . the critical generalisation is inclusion of response . it establishes a link to a host of powerful ideas , notably , to the real - time quantum field theory , schwinger - perel - keldysh s closed - time - loop formalism @xcite , and , last but not least , to relativistic quantum field theory @xcite . ( this link was a subject of @xcite . ) from a physical perspective , it allows one to make propagation of quantum signals in space - time the guiding principle of the whole investigation . in particular , propagation of the electromagnetic field underlies the aforementioned amendment of the concept of time - normal ordering . it also reveals the inherent link between operator noncommutativity and response @xcite . the pivotal formal point of these analyses is that the description of a quantum system in terms of time - normal averages of heisenberg operators conditional on external sources turns out to be equivalent to the closed - time - loop framework . this kind of description was termed in @xcite _ response formulation _ of a quantum system ( also _ response picture _ , _ response viewpoint _ , or _ response characterisation _ ) . the said equivalence was proven in @xcite for interacting bosons and in @xcite for interacting fermions ( see also endnote @xcite ) . the closed - time - loop formulation is the most general formal framework known in quantum field theory . _ ipso facto _ , our approach may be seen as an ultimate generalisation of the phase - space techniques in quantum electrodynamics ( cf . endnote @xcite ) . its ability to tackle relativistic problems with renormalisations was demonstrated in @xcite . the way the paper is structured reflects our wish to make the results intuitive . `` obscured macroscopic view '' as a formal viewpoint is mostly relegated to appendices . in the main body of the paper , we proceed by `` doing quantum electrodynamics while thinking classically '' . we start from a brief overview of the results in . in , we develop the `` classical yardstick '' : a collection of general formulae , describing properties of devices and their interactions in classical stochastic electrodynamics . quantum mechanics starts in , which summarises formal definitions . conditional p - functionals are introduced in . the conjecture that classical formulae of are in fact _ exact _ relations for p - functionals is verified in and [ ch:91rj ] . in , we review the formal approach , concentrating on the interplay of the formal viewpoint and approximations . the results are discussed in . in appendix [ ch : reg ] we briefly touch upon the role of causality and regularisations in the classical electromagnetic self - action problem . appendix [ ch : qr ] summarises the results `` imported '' from earlier papers . in appendix [ ch : lm ] , we outline redefinitions allowing one to exclude passive linear devices ( mirrors etc . ) from explicit consideration . appendices [ ch:2dc ] and [ ch : phdet ] provide formal support to `` pictorial derivations '' in . in this paper we concern ourselves with the _ formal structure _ of quantum electrodynamics . rather than attempting to calculate something measurable , we look at restrictions measurable quantities must obey , assuming the only means of accessing the world is the electromagnetic interaction . moreover , we are only interested in relations which are independent of any particulars of nonelectromagnetic nature . this greatly limits the kind of question we are able to ask . consider for instance electromagnetic interaction of two devices depicted schematically in . for the time being , ignore hats and think about the devices as classical electromagnetic scatterers . in any real problem , there must exist microscopic dynamical models allowing one to calculate properties of the devices . these models necessarily rely on information about properties of matter , which is by itself nonelectromagnetic ( e.g. , the dirac equation in spinor qed , or various models in solid state theory ) . within `` obscured macroscopic view '' , we are not inquiring about such details . we assume that properties of individual devices are known ( a ) , and ask how they should be formally combined to determine properties of the pair ( b ) . this way , our investigation is to a large extent a highly formalised electronic - engineering viewpoint . when planning an experiment , an engineer is only interested in properties of a photodiod ( say ) such as frequency range , efficiency , dark count , dead time , saturation limit , and so on . from our perspective , all these are _ response properties _ of the photodiod . we mention features such as dark count and saturation limit to emphasise that proper formalisation of the engineering viewpoint can not be too simple . it should include , e.g. , such nastiness as _ nonlinear non - markovian self - noise_. this is the minimal level of theoretical sophistication accommodating for shot noise , dark count , dead time , and saturation . at the same time , the problem we face is largely pedagogical . we need a formal framework which would not impose any restrictions on models of devices . this implies a relatively high level of abstraction . physically , all we can say about the arrangements in a , b is that , * in `` obscured macroscopic view '' , each device is characterised by a c - number random current ; * statistics of this current is dependent ( conditional ) on the incoming field radiated by some external sources and affecting the device ; * when characterising a particular device , the incoming field should be regarded given ; * for each device in the pair , the incoming field is the external field plus radiation of the other device ; * the current characterising the pair is a sum of currents in the devices ; * the field emitted by the pair is a sum of fields emitted by the devices . in , we take these trivialities in classical electrodynamics to a high level of formal abstraction . in the following sections , we demonstrate that quantum - electrodynamical relations in the `` obscured macroscopic view '' express nothing but the same set of trivialities . it is this fact_and not the trivialities themselves_that constitutes the result of this paper . in this section we construct a `` classical yardstick '' for future quantum analyses . we develop a formal response characterisation of a device in classical stochastic electrodynamics , including the formal solution to the self - action problem ( cf . a ; ignore hats ) , then derive formulae reducing properties of an interacting pair to its components ( cf . we employ an intuitive characterisation of radiation scatterers in terms of probability distributions conditional on external sources , making all our derivations next to trivial . from the point of view of an `` electromagnetic observer '' , the world is a maze of electromagnetic currents . these currents are mostly stochastic ( we do not need quantum mechanics to arrive at this conclusion , experimental evidence is aplenty ) . thinking of a device as a macroscopic radiation scatterer ( a ) , we characterise it by a probability distribution over a random current conditional on the incoming ( external ) field . for simplicity , we suppress all arguments of the fields and currents except time ; for a generalisation beyond single mode see below . in a general nonlinear non - markovian case , is a nontrivial _ functional _ of two variables . its full name is _ conditional functional probability distribution . _ in classical stochastic electrodynamics , @xmath0}\geq 0 , % } { % \label{eq:35at}\end{aligned}\]]but we never use this in our reasoning otherwise `` doing quantum electrodynamics while thinking classically '' would certainly fail . viewed literally , all formulae in this paper are written for a one - mode case . to adapt them to more complicated situations it suffices to `` expand '' the time variable . a multi - mode case is recovered replacing , @xmath1where @xmath2 is the mode index . ( here and hereafter , omitted integration limits indicate the maximal possible area of integration . ) a nonrelativistic 3d formulation emerges on replacing , @xmath3where is now the 3d index . this covers all cases of interest in macroscopic optics . relativistic reformulation takes changing a few signs and factors , see @xcite and appendix [ ch : gz ] . averages of the random current may be written as functional ( path ) integrals , @xmath4\ , p\protect { [ \mathcal{j}| a_{\mathrm{e } } ] } \mathcal{j}(t ) , \\ \protect \langle \mathcal{j}(t)\mathcal{j}(t ' ) \rangle & = \int \mathcal{d}[\mathcal{j}]\ , p \protect { [ \mathcal{j}| a_{\mathrm{e } } ] } \mathcal{j}(t ) \mathcal{j}(t ' ) , \end{aligned } \label{eq:27ak}\end{aligned}\]]etc . averages ( [ eq:27ak ] ) are by definition conditional on the source . we stress that we have not invented anything new , but simply explained what the innocuous symbol of classical averaging actually means . we think about a path integral as a multidimensional integral in discretised time , @xmath5 = \prod_t \bigg [ \int d\mathcal{j}(t ) \bigg ] , % } { % \label{eq:28al}\end{aligned}\]]cf . the remark on omitted integration limits after . to generalise ( [ eq:28al ] ) beyond a single mode , use formulae of . such definition makes all manipulations straightforward . a mathematically sound limit of continuous time is outside the scope of this paper . the minimal requirement for any physical model is _ nonrelativistic causality_. namely , the external field at time @xmath6 may only influence the current for times ( cf . endnote @xcite ) . varying for times should have no effect on . formally , this is expressed by the zero variational ( functional ) derivative , @xmath7for the averages , @xmath8and similarly for higher - order averages . it should not be missed that neither nor ( [ eq:54nf ] ) assume setting to zero after differentiation . the latter is characteristic of the definition of generalised susceptibilities , @xmath9from ( [ eq:54nf ] ) we obtain , @xmath10however , afford a stronger result , @xmath11unlike ( [ eq:89nh ] ) , this condition , as well as the underlying conditions ( [ eq:86nd ] ) and ( [ eq:54nf ] ) , remain physically meaningful also in a vicinity of a phase transition . the scattered field is the field radiated by the random current , @xmath12where is the transfer function of the linear medium ( or vacuum ) in which the device is immersed . response of the linear medium should be causal , @xmath13we do not assume the linear medium to be stationary ( nor homogeneous subject to generalisations as per ) . we however do assume it to be passive : its radiation ( e.g. , thermal ) must be negligible . for simplicity , we assume that is not affected by the presence of devices . this assumption is utterly impractical . in a laser theory ( say ) , it means that the mirrors and the active medium are introduced on an equal basis . it is much more common to reassign linear susceptibilies of devices to the `` free '' field . all passive linear devices ( mirrors , beam - splitters and _ tutti quanti _ ) are then eliminated from explicit consideration , being accounted for in . for purposes of this paper , practicality is the least of concerns , so we stick with simplicity . we return to this assumption in ; a formal way of lifting it is outlined in appendix [ ch : lm ] . whether a particular physical entity is regarded `` linear medium '' or `` device '' depends on the degree of insight . for instance , in nonrelativistic theories , vacuum is `` linear medium '' . in relativistic qed , vacuum is rather a `` nonlinear device '' , of which the _ low frequency limit of linear response _ is separated from `` vacuum corrections '' . this low frequency limit is that what one calls `` the free electromagnetic field in vacuum '' . the whole procedure is highly nontrivial because of divergences and renormalisations . for details , see textbooks @xcite and papers @xcite . mixed averages of the field and current reduce to current averages , @xmath14and similarly for higher - order averages . formally , these averages imply existence of a joint conditional probability functional over the current and field , @xmath15 \mathcal{d } [ \mathcal{a}]\ , p \protect { [ \mathcal{j},\mathcal{a}| a_{\mathrm{e } } ] } \mathcal{j}(t ) \mathcal{a}(t ' ) , % } { % \label{eq:29am}\end{aligned}\]]etc . the radiation law ( [ eq:95rn ] ) reduces to , @xmath16}= \delta [ \mathcal{a}-g_{\text{r } } \mathcal{j } ] p\protect { [ \mathcal{j}| a_{\mathrm{e } } ] } , % } { % \label{eq:30an}\end{aligned}\]]where we use a condensed notation , @xmath17the @xmath18-functional is defined as a multidimensional @xmath18-function , @xmath19 = \prod_t \delta ( f(t ) ) . % } { % \label{eq:31ap}\end{aligned}\]]this matches , making all algebraic manipulations straightforward . distribution provides a macroscopic characterisation of a device in terms of statistics of a random current , conditional on radiation of _ other devices_. in a microscopic model of a device , evolution of matter depends on the local ( microscopic ) field rather than on the external field . the local field includes self - radiation of matter , i.e. , @xmath20where we used condensed notation ( [ eq:77nv ] ) . the microscopic model allows one to calculate statistics of the current conditional on the local field . formally , it is described by the probability distribution , which we assume to be known ( here is a given quantity ; is disregarded ) . the self - action problem is then solved by the `` stohastic dressing formula '' , @xmath21}= p^{\mathrm{i}}\protect { [ \mathcal{j}| a_{\mathrm{e}}+g_{\text{r } } \mathcal{j}]}. % } { % \label{eq:38aw}\end{aligned}\]]the apparent simplicity of this relation is extremely deceptive . its consistency is subject to causality and suppression of instantaneous self - action . for a brief discussion consider now a pair of devices , labeled @xmath22 and @xmath23 ( cf . b ; again ignore hats ) . each is characterised by a pair of distributions connected by , macroscopically by , and microscopically by . extends to individual devices , @xmath24}= p_{a , b}^{\mathrm{i}}\protect { [ \mathcal{j}| a_{\mathrm{e}}+g_{\text{r } } \mathcal{j}]}. % } { % \label{eq:77lh}\end{aligned}\]]as a compound device , the pair is characterised by and . now , what does it mean that the devices are distinguishable ? since indistinguishability in the true meaning of the term does not occur in classical mechanics , the only assumption we have to make is that their models do not contain correlated noise sources . in other words , the currents @xmath25 and @xmath26 should be statisticaly independent if considered conditional on the local field . in formal terms , the corresponding joint probability distribution should factorise , @xmath27 = p^{\mathrm{i}}_{a } [ \mathcal{j}_a | a_{\mathrm{loc } } ] p^{\mathrm{i}}_{b } [ \mathcal{j}_b | a_{\mathrm{loc } } ] . \end{aligned } \label{eq:4kr}\end{aligned}\]]this realised , solution to the interaction problem in terms of the microscopic distributions is trivial . the total random current characterising the composite device is a sum of the currents in devices , @xmath28integrating ( [ eq:4kr ] ) over the redundant information we find the probability distribution over the full current , @xmath29 & = \int \mathcal{d}[\mathcal{j}_a ] \mathcal{d}[\mathcal{j}_b]\delta [ \mathcal{j}-\mathcal{j}_a-\mathcal{j}_b ] \nonumber\\&\quad \times p^{\mathrm{i}}_a[\mathcal{j}_a | a_{\mathrm{loc}}]p^{\mathrm{i}}_b[\mathcal{j}_b | a_{\mathrm{loc } } ] . % } { % \label{eq:5ks}\end{aligned}\]]we refrain from integration over or to maintain symmetry of formulae . the expression for the macroscopic functional by and also follows with ease . applying the stochastic dressing formula ( [ eq:38aw ] ) to ( [ eq:5ks ] ) we obtain , @xmath30 & = \int \mathcal{d}[\mathcal{j}_a ] \mathcal{d}[\mathcal{j}_b]\delta [ \mathcal{j}-\mathcal{j}_a-\mathcal{j}_b ] \nonumber\\&\quad \times p^{\mathrm{i}}_a[\mathcal{j}_a | a_{\mathrm{e}}+g_{\text{r } } \mathcal{j}]p^{\mathrm{i}}_b[\mathcal{j}_b | a_{\mathrm{e}}+g_{\text{r } } \mathcal{j } ] . % } { % \label{eq:6kt}\end{aligned}\]]we then `` dress '' the components of the device by making use of , @xmath31 & = p^{\mathrm{i}}_a[\mathcal{j}_a | a_{\mathrm{e}}+g_{\text{r } } ( \mathcal{j}_a+\mathcal{j}_b ) ] \\ & = p_a[\mathcal{j}_a | a_{\mathrm{e}}+g_{\text{r } } \mathcal{j}_b ] , \\ p^{\mathrm{i}}_b[\mathcal{j}_b | a_{\mathrm{e}}+g_{\text{r } } \mathcal{j } ] & = p^{\mathrm{i}}_b[\mathcal{j}_b | a_{\mathrm{e}}+g_{\text{r } } ( \mathcal{j}_a+\mathcal{j}_b ) ] \\ & = p_b[\mathcal{j}_b | a_{\mathrm{e}}+g_{\text{r } } \mathcal{j}_a ] . \end{aligned } \label{eq:7ku}\end{aligned}\]]finally , @xmath32 & = \int \mathcal{d}[\mathcal{j}_a ] \mathcal{d}[\mathcal{j}_b]\delta [ \mathcal{j}-\mathcal{j}_a-\mathcal{j}_b ] \\&\quad \times p_a[\mathcal{j}_a | a_{\mathrm{e}}+g_{\text{r } } \mathcal{j}_b]p_b[\mathcal{j}_b | a_{\mathrm{e}}+g_{\text{r } } \mathcal{j}_a ] . \end{aligned } \label{eq:8kv}\end{aligned}\]]apart from , this formula expresses the simple fact that external field affecting either device is plus radiation of the other device ( cf . the list of `` trivialities '' in ) . in this section we reiterate formal definitions , allowing us to embed our analyses in a conventional quantum framework . we start with a brief overview of the general formal viewpoint and terminology . _ distinguishable subsystems _ in quantum mechanics `` inhabit '' orthogonal subspaces of the hilbert space . as a mathematical requirement , this definition applies in the interaction picture . all models in this paper comprise the _ electromagnetic - field _ subsystem and one or more _ matter _ subsystems . for brevity , we call matter subsystems _ devices_. the term _ bare device _ applies to characterisation of a matter subsystem in the interaction picture . a _ ( solitary ) dressed device _ refers to a model where one of the matter subsystems is made to interact with the field subsystem . _ interacting devices _ emerge by coupling more than one device to the field . direct nonelectromagnetic interaction of devices is not allowed . dressed solitary and interacting devices are characterised by the corresponding heisenberg field and current operators ( which are interaction - dependent , unlike the interaction - picture operators ) . of main interest to us is how properties of interacting devices are related to those of the solitary ones . in formal terms , we look at relations between heisenberg operators defined according to different interaction hamiltonians . this point ought to be well noted : _ we compare models _ , rather than work with a single model . majority of the formal concepts we rely upon are standard , or at least well known . to the standard ones belong quantisation of the electromagnetic field , the electromagnetic hamiltonian as a sum of matter , field and interaction hamiltonians , the interaction _ vs _ heisenberg picture , free and heisenberg operators , and the initial ( heisenberg ) state of the system specified in remote past . the latter gives rise to quantum averaging . we do not use the schrdinger picture . implicit in our techniques are also the evolution operator and the @xmath33-matrix ( they underlie the results `` imported '' from @xcite ; cf . appendix [ ch : qr ] ) . all these concepts are amply covered in textbooks @xcite . among the well - known concepts are the time - normal ordering of operators , introduced by kelley and kleiner and glauber in the context of photodetection theory @xcite , and placing c - number sources into hamiltonians , which we borrow from kubo @xcite . it should not be missed that , unlike kubo , we do not limit our analyses to the linear response . `` proprietary '' formal concepts we use are the _ response transformation _ of quantum mechanics @xcite , and the redefinition of the conventional time - normal operator ordering this transformation prompts @xcite . the latter takes care of causality violations @xcite plagueing the conventional definition ; this assures strict causality in our approach @xcite . key points of the `` proprietary '' concepts will be reiterated where required . for details see the quoted papers . throughout the paper we assume the standard electromagnetic hamiltonian , @xmath34hereinafter m , f and i stand for _ matter _ , _ field _ and _ interaction_. quantum averaging is defined with respect to the factorised heisenberg @xmath35-matrix , @xmath36where and describe the initial states of field and matter . the same symbol ( angle brackets ) is used for the classical and quantum averages ; what we have in mind should be clear in the context . for simplicity we write the electromagnetic interaction in a single - mode form , @xmath37 , \label{eq:36mm}\end{aligned}\]]where and are the current and the electromagnetic field operators in the interaction picture , and is the aforementioned kubo - style source . this hamiltonian may be understood literally , or as a shorthand notation according to . the latter implies summation over all field labels except time . e.g. , in the multimode case , @xmath38 = \sum_{k } \hat j_k(t ) [ \hat a_k(t)+a_{\mathrm{e}k}(t ) ] , % } { % \label{eq:61px}\end{aligned}\]]in the nonrelativistic 3d case , @xmath39 \nonumber\\&\quad = \int d^3{\mbox{\rm\boldmath$r$}}\sum_{k=1}^3 \hat j_k({\mbox{\rm\boldmath$r$}},t ) [ \hat a_k({\mbox{\rm\boldmath$r$}},t)+a_{\mathrm{e}k}({\mbox{\rm\boldmath$r$}},t ) ] , % } { % \label{eq:18zz}\end{aligned}\]]etc . irrespective of all other formal details , and commute for arbitrary @xmath6 and @xmath40 , and all their mixed quantum averages factorise ( cf . the remarks on the `` overall viewpoint '' in above ) . that this does not extend to the heisenberg operators denoted and is due solely to the interaction . the heisenberg operators and their quantum averages are by construction dependent ( conditional ) on the external source . it should not be overlooked that interaction ( [ eq:36 mm ] ) is nonresonant : and are hermitian operators , not broken into their frequency - positive and frequency - negative parts . we adhere to certain notational conventions . italics are consistently used for interaction - picture operators ; this concerns hamiltonians ( [ eq:37mn ] ) and ( [ eq:36 mm ] ) , as well as all other hamiltonians in the paper . calligraphic letters are equally consistently used for heisenberg operators . apart from the minimal assumptions of hermiticity of all operators and positivity of -matrices , the matter hamiltonian @xmath41 , the current operator @xmath42 and the state of the device @xmath43 may be arbitrary . consistency of the response formulation @xcite hinges on two properties of : that it is a free hermitian bosonic operator , of which the two - point commutator is a c - number , and that its frequency - positive part annihilates vacuum . we assume that the field is in a vacuum state , @xmath44physically , assumptions ( [ eq:40bj ] ) and ( [ eq:42bl ] ) matter little , because the system is allowed infinite time to develop . _ in no way do we assume that a closed system of microscopic equations of motion may be written solely in terms of the mesoscopic variables and . _ some underlying dynamical variables must exist , bosonic or fermionic , by which all matter variables are expressed , and for which closed equations of motion may be written . however , apart from the assumption that it may in principle be consistenly formulated , microscopic dynamics plays no role in our analyses . macroscopic dynamics is observed , but in general does not allow one to construct a closed theory . microscopic dynamics specifies the system completely , but may only be inferred from what is observed ( cf . however ) . it is worthy of stressing that _ the possibility of separation of the mesoscopic from microscopic dynamics is not an assumption_. it is a structural property of conventional quantum field theory ( specifically , of the feynman - dyson approach ) @xcite , cf . @xcite and appendix [ ch : qr ] in this paper . all parallels between classical and quantum electrodynamics are ultimately rooted on the _ wave quantisation relation _ @xcite . it introduces the concept of running electromagnetic wave in quantum electrodynamics : @xmath45 = -i\hbar [ g_{\text{r } } ( t , t')-g_{\text{r } } ( t',t ) ] . % } { % \label{eq:40mr}\end{aligned}\]]the linear response function ( a.k.a . transfer function or retarded propagator ) of the electromagnetic field is found replacing the quantum current in ( [ eq:36 mm ] ) by a c - number source and considering the linear response of the free field : @xmath46 . % } { % \label{eq:41ms}\end{aligned}\]]the first formula here is the definition of and the last one is kubo s relation for it @xcite ; we dropped quantum averaging of the commutator which is a c - number anyway . the wave quantisation formula follows trivially by inverting kubo s one . it is not associated with any approximation , but its implications are much more profound under conditions of distinguishability . note that the causality condition ( [ eq:92nl ] ) is inherent to kubo s response . in the main body of the paper , we assume that is not affected by the presence of devices . this assumption was discussed in . apart from simplicity , the reason we stick to it is that it can not be lifted in a hamilton formulation . the corresponding formal redefinitions are summarised in appendix [ ch : lm ] . ( caveat : this appendix relies on definitions yet to be given in . ) a pair of interacting distinguishable devices are introduced postulating , @xmath47 , and commute for arbitrary @xmath6 , @xmath40 and @xmath48 , and all their mixed averages factorise ( cf . the remarks on hilbert - space factorisation in ) . we shall also need models of solitary devices , @xmath49for device @xmath22 , and @xmath50for device @xmath23 . majority of the above definitions are formal . they serve to embed our analyses in a conventional dynamical quantum approach , and to perform its response transformation ( cf . ) ; with that their role ends . assumptions `` making it '' into the discussion are as follows : * the field is fully characterised by its response properties , reflected by . * matter is fully characterised by quantum averages of products of the bare current operator , @xmath51 * for interacting devices , averages ( [ eq:31ta ] ) factorise into products of the corresponding averages for devices , e.g. , @xmath52 [ \hat j_a(t ' ) + \hat j_b(t ' ) ] \rangle \nonumber\\ & = \protect \langle \hat j_a(t)\hat j_a(t ' ) \rangle + \protect \langle \hat j_a(t ) \rangle \protect \langle \hat j_b(t ' ) \rangle \nonumber\\&\quad + \ { a\leftrightarrow b \ } . % } { % \label{eq:84vj}\end{aligned}\ ] ] that these assumptions suffice is a formal expression of consistency of `` obscured macroscopic view '' in quantum electrodynamics . the time - normal ordering was introduced by kelley and kleiner and glauber in the context of quantum photodetection theory @xcite ( for a pedagogical discussion see , e.g. , mandel and wolf @xcite ) . there are well - known causality violations associated with taking the glauber - kelley - kleiner definition beyond the rotating wave approximation ( rwa ) @xcite , so that the idea that some amendments are in order should come naturally to the reader . these amendments were suggested by l.p . and s.s . in @xcite , see also @xcite . as an example , consider the time - normally ordered product of two current operators , . the notation for the time - normal ordering is borrowed from mandel and wolf @xcite . we assume that the reader is familiar with the concept of separation of the frequency - positive and frequency - negative parts of functions . the latter is conveniently expressed as an integral transformation , ( with being an arbitrary function ) @xmath53for more on this operation see , e.g. , appendix a in @xcite . the beyond - the - rwa expression for follows in two steps . firstly , expand operators into the frequency - positive and frequency - negative parts , @xmath54and apply the kelley - kleiner definition , @xmath55 in this formula , stand for the forward and reverse time orderings ( often denoted by @xmath56 and @xmath57 ) . secondly , change the order of time orderings and the operations , @xmath58time - normal products of many bosonic operators follow similarly . note that @xmath59 in ( [ eq:19sn ] ) , we moved the operations from under the averaging also in the second and third terms , where this does not matter . this serves to emphasise that the rhs of ( [ eq:19sn ] ) is expressed by `` entire '' operators , not broken in their frequency - positive and frequency - negative parts ( cf . sec . 3.2 in @xcite ) . the reader familiar with the schwinger - perel - keldysh closed - time - loop approach of quantum field theory @xcite should have recoginised operator structures typical for this framework . indeed , formally , and the like follow if using the wave quantisation formula to replace the matrix propagator in the closed - time - loop approach by the retarded propagator @xcite . this association with quantum field theory assures a solid formal ground for our analyses . it is straightforward to show that the time - normally ordered product of hermitian operators is hermitian ; quantum averages of time - normally ordered products of hermitian operators ( time - normal averages , for short ) are real . causality is a subtler property . the following `` no - peep - into - the - future theorem '' was proven in @xcite : _ a time - normally ordered product depends on operators it comprises only for times not larger than its largest time argument_. for a relativistic extension of this theorem see @xcite ; cf . also endnote @xcite . in addition to , we introduce three more cases of condensed notation : @xmath60where @xmath61 and @xmath62 are c - number or q - number functions . the `` products '' @xmath63 and @xmath64 denote scalars , while @xmath65 and @xmath66functions . the effect of condensed notation is twofold : it diminishes the bulk of formulae and reduces them visually to a scalar case . the latter much simplifies all calculations ( we borrowed this trick from vasilev @xcite ) . in this section we introduce the formal concepts , allowing one , so to speak , to implement in quantum electrodynamics . we also use this opportunity to acquaint the reader with the key results of papers @xcite . the quantum counterpart of the classical averages ( [ eq:27ak ] ) are time - normal averages of the heisenberg current operator , @xmath67etc . the classical averages were dependent ( conditional ) on an external field by definition . the quantum average exhibits this dependence by construction , due to the external field in the interaction ( [ eq:36 mm ] ) . introduces _ response formulation _ @xcite ( also response picture , response viewpoint , or response characterisation ) of an electromagnetic device . specifications beyond the rwa aside , one borrows the concepts of the photodetection theory to characterise the _ output _ of a device , while using kubo - style sources to define the _ input _ ( cf . also endnote @xcite ) . as remarked in above , the beyond - the - rwa time - normal ordering emerges in fact within the closed - time - loop formalism of quantum field theory @xcite . moreover , the response formulation was shown to be formally equivalent to this formalism @xcite . for details see the quoted papers and appendix [ ch : qr ] . an immediate word of caution is in place here . in our approach external sources are used as functional arguments . this implies that they are _ arbitrary_. informally speaking , we allow c - number sources to be placed inside atomic nuclei . this extent of control is certainly impossible with real light sources . _ arbitrary external sources in quantum mechanics are unphysical quantities . _ this controversy is resolved by the so - called _ consistency condition _ @xcite . for an electromagnetic system , it is expressed by and ( [ eq:38fv ] ) in appendix [ ch : qr ] . it stipulates that , * response properties of a system are formally expressed in terms of heisenberg operators defined with zero sources ( cf . kubo s ) . * these properties are `` encoded '' in commutators of these operators ( again , cf . ) . these facts hold with given sources . with arbitrary sources , the consistency condition also shows that , * the information lost due to fixing the time - normal order of operators is exactly that re - entering through the dependence of the ordered operator products on the sources . all these facts warrant the formal consistency of the response formulation , hence the term . since arbitrary sources are unphysical , response formulation should be regarded an _ intuitive interpretation _ of a conventional quantum approach not relying on sources ( specifically , of the aforementioned closed - time - loop formalism ) . this interpretation enables `` doing quantum electrodynamics while thinking classically '' , making it extremely useful ; nonetheless it is only formal . the interplay of the formal and the physical in response formulation will be further discussed in . being of fundamental importance for introduction of the response formulation , the consistency condition is of little if any relevance to analyses within it . for all practical purposes only its existence matters . the critical property making the analogy ( [ eq:53ne ] ) meaningful is _ causality_. the no - peep - into - the - future theorem mentioned in assures validity of the quantum counterparts of , @xmath68etc . apart from the said theorem , they rely on the quantum counterpart of , @xmath69conditions ( [ eq:55nh ] ) may be verified without making any assumptions about the quantum matter @xcite . in other words , they constitute in essence a kinematical property of quantum mechanics . relativistic extention of ( [ eq:55nh ] ) takes a minimal assumption of commutativity of operators at space - like intervals . this is one of wightman s axioms of quantum field theory @xcite . for details see @xcite and endnote @xcite . generalised to quantum mechanics , define the _ functional conditional quasiprobability distribition _ , or _ conditional p - functional _ , , @xmath70\ , p\protect { [ \mathcal{j}| a_{\mathrm{e } } ] } \mathcal{j}(t ) , \\ \protect \langle { \mathcal t}{\mbox{\rm\boldmath$:$}}\protect{\hat{\mathcal j}}(t)\protect{\hat{\mathcal j}}(t'){\mbox{\rm\boldmath$:$}}\rangle & = \int \mathcal{d}[\mathcal{j}]\ , p \protect { [ \mathcal{j}| a_{\mathrm{e } } ] } \mathcal{j}(t ) \mathcal{j}(t ' ) , \end{aligned } \label{eq:33ar}\end{aligned}\]]etc . these quantities generalise both _ conditional probability distributions _ of classical stochastics and _ quasiprobability distributions _ of the conventional phase - space techniques to arbitrary nonlinear non - markovian quantum systems . we do not notationally distinguish the quasidistributions from probability distributions , assuming that , in classical mechanics , they are positive , while in quantum mechanics may become nonpositive . the term p - functional was introduced by w. vogel @xcite for a functional generalisation of the conventional p - distribution @xcite . our definition generalises vogel s , firstly , beyond rwa , and , secondly , to response . in classical mechanics , the conditional probability distribution is a primary quantity , while averages are defined in its terms . in quantum mechanics , all dynamical relations are derived for averages , and the corresponding p - functionals are to be defined in terms of the latter . we follow the association between probability distributions and characteristic functionals . for example , for the classical we have , @xmath71\ , p\protect { [ \mathcal{j}| a_{\mathrm{e } } ] } \exp(i\zeta \mathcal{j } ) . % } { % \label{eq:43tp}\end{aligned}\]]we use condensed notation ( [ eq:3vs ] ) . the quantum counterpart of ( [ eq:43tp ] ) reads , @xmath72 } & = \protect \langle { \mathcal t}{\mbox{\rm\boldmath$:$}}\exp ( i\zeta \protect{\hat{\mathcal j } } ) { \mbox{\rm\boldmath$:$}}\rangle \label{eq:20sp}\\ & = \int \mathcal{d}[\mathcal{j}]\ , p\protect { [ \mathcal{j}| a_{\mathrm{e } } ] } \exp ( i\zeta \mathcal{j } ) . \label{eq:41az}\end{aligned}\]]apart from defining the conditional p - functional , we have also introduced notation for the corresponding characteristic one . we also need a quantum definition of the joint quasiprobability distribution , @xmath73 } & = \protect \langle { \mathcal t}{\mbox{\rm\boldmath$:$}}\exp ( i\eta \protect{\hat{\mathcal a}}+i\zeta \protect{\hat{\mathcal j } } ) { \mbox{\rm\boldmath$:$}}\rangle \label{eq:44tq}\\ & = \int \mathcal{d}[\mathcal{a}]\mathcal{d}[\mathcal{j } ] \ , p[\mathcal{a},\mathcal{j}|a_{\mathrm{e}}]\exp ( i\eta \mathcal{a}+i\zeta \mathcal{j } ) . % } { % \label{eq:45tr}\end{aligned}\]]cf . the classical . now , how should we define ? to start with , here is a given quantity ; should be disregarded . fixing means that self - radiation is somehow prevented from affecting the device . this is achieved by suppressing interaction of the device with the quantised field , i.e. , replacing interaction ( [ eq:36 mm ] ) by , @xmath74this way , @xmath75 & = \protect \langle { \mathcal t}{\mbox{\rm\boldmath$:$}}\exp ( i\zeta \hat j_{\mathrm{loc } } ) { \mbox{\rm\boldmath$:$}}\rangle , \label{eq:21sq}\\ & = \int \mathcal{d}[\mathcal{j}]\ , p^{\mathrm{i}}\protect { [ \mathcal{j}| a_{\mathrm{loc } } ] } \exp ( i\zeta \mathcal{j } ) . \label{eq:24st}\end{aligned}\]]where is the heisenberg current operator with respect to ( [ eq:47bf ] ) . an immediate reservation is in place here . definitions ( [ eq:20sp])([eq:24st ] ) are a rare occasion when the consistency condition ( see ) must be at least acknowledged . as explained there , ([eq:24st ] ) constitute an intuitive interpretation of all functionals involved , while their rigorous definitions can not rely on external sources . such definitions may be found in appendices [ ch : upc ] and [ ch:25su ] . for the time being , awareness of their existence suffices . this reservation is especially important for . it is in fact a primary quantity , with everything else expressed by it . that it can be _ directly _ expressed by averages of is critical for consistency of our approach , cf . and comments thereon . as an example of working with functionals we prove . substituting it into ( [ eq:45tr ] ) results in the following relation between and , @xmath73}= \phi_{\mathrm{m}}\protect { [ \eta g_{\text{r } } + \zeta | a_{\mathrm{e } } ] } . % } { % \label{eq:47tt}\end{aligned}\]]we use notation ( [ eq:78nw ] ) . is a special case of the general formula ( [ eq:53ze ] ) in appendix [ ch : upd ] , verified in @xcite . has thus been proven as a quantum formula for p - functionals . differentiating ( [ eq:47tt ] ) and using the chain rule for functional derivatives we obtain quantum counterparts of , @xmath76etc . a detailed parallelism between classical and quantum response characterisations of a solitary device is evident . it is worthy of a remark that , while and ( [ eq:47tt ] ) are equivalent , physical transparency of the former compares favourably to obscure formality of the latter . consider now the self - action problem for a solitary device . in quantum electrodynamics , it is expressed by the _ dressing formula _ @xcite , relating averages of the heisenberg operator to those of the `` free '' current , @xmath77 % % \\ = \exp \bigg ( -i\frac{\delta } { \delta a_{\mathrm{e}}}g_{\text{r } } \frac{\delta } { \delta \zeta } \bigg ) % \\ \times \phi_{\mathrm{m}}^{\mathrm{i } } \big [ \zeta \big| a_{\mathrm{e } } \big ] . \label{eq:54zf}\end{aligned}\]]we use condensed notation ( [ eq:76nu ] ) . substituting ( [ eq:24st ] ) into the dressing formula we find , @xmath78 } \nonumber\\&\quad = \exp \big ( -i\frac{\delta } { \delta a_{\mathrm{e}}}g_{\text{r } } \frac{\delta } { \delta \zeta } \big ) \int \mathcal{d}[\mathcal{j } ] \,p^{\mathrm{i}}\protect { [ \mathcal{j}| a_{\mathrm{e } } ] } \exp ( i\zeta \mathcal{j } ) \nonumber\\&\quad = \int \mathcal{d}[\mathcal{j } ] \exp \big ( \frac{\delta } { \delta a_{\mathrm{e}}}g_{\text{r } } \mathcal{j}\big ) p^{\mathrm{i}}\protect { [ \mathcal{j}| a_{\mathrm{e } } ] } \exp ( i\zeta \mathcal{j } ) \nonumber\\&\quad = \int \mathcal{d}[\mathcal{j}]\ , p^{\mathrm{i}}\protect { [ \mathcal{j}| a_{\mathrm{e}}+g_{\text{r } } \mathcal{j}]}\exp ( i\zeta \mathcal{j } ) . % } { % \label{eq:43bb}\end{aligned}\]]the last transformation here is an application of a functional shift operator , @xmath79}= p^{\mathrm{i}}\protect { [ \mathcal{j}| a_{\mathrm{e}}+g_{\text{r } } \mathcal{j}]}. % } { % \label{eq:44bc}\end{aligned}\]]the dressing formula ( [ eq:54zf ] ) is thus equivalent to the _ quasistochastic dressing formula _ , which is the stochastic dressing formula ( [ eq:38aw ] ) regarded as a relation for p - functionals . two remarks are in order here . firstly , the dressing formula expresses the nontrivial part of perturbative calculations . in full - space problems , this is associated with suppression of infinities , which in turn are due to instantaneous self - actions . further discussion of this issue may be found in appendix [ ch : reg ] . secondly , the very possibility to introduce quasidistribution as a characterisation of a bare device is yet another manifestation of the concept of obscured macroscopic view in the formal structure of quantum electrodynamics . microscopic dynamics is screened from the `` electromagnetic observer '' from the start , by mere assumption that interaction with the quantised electromagnetic field is facilitated by the current . this happens _ before _ the field and interaction are actually introduced . the critical step to physics is from a solitary device to an interacting pair ( b ) . allow one to confine the discussion to the quantum current . the `` dressed '' and `` bare '' p - functionals for devices @xmath22 and @xmath23 and the corresponding characteristic functionals are introduced extending definitions ( [ eq:20sp ] ) , ( [ eq:41az ] ) and ( [ eq:21sq ] ) , ( [ eq:24st ] ) to the models of devices ( [ eq:47my ] ) and ( [ eq:48mz ] ) . in practice , this boils down to assigning labels @xmath22 and @xmath23 to everything , e.g. , @xmath80 } & = \protect \langle { \mathcal t}{\mbox{\rm\boldmath$:$}}\exp ( i\zeta \protect{\hat{\mathcal j } } ) { \mbox{\rm\boldmath$:$}}\rangle _ a % \eqlabel{20sp } % % \nonumber % % \z \nonumber\\ & = \int \mathcal{d}[\mathcal{j}]\ , p_a[\mathcal{j}|a_{\mathrm{e}}]\exp ( i\zeta \mathcal{j } ) . % } { % \label{eq:27sw}\end{aligned}\]]the symbol denotes averages defined in the model of solitary device @xmath22 , cf . . similar definitions apply to other `` labeled '' quantities . as the first step to verification of and ( [ eq:8kv ] ) as quantum relations for p - functionals we rewrite them in terms of the corresponding characteristic functionals . substituting the former into ( [ eq:24st ] ) we get , @xmath81 } & = \int \mathcal{d}[\mathcal{j}_a ] \mathcal{d}[\mathcal{j}_b ] \exp[i\zeta ( \mathcal{j}_a+\mathcal{j}_b ) ] \nonumber\\&\quad \times p^{\mathrm{i}}_a[\mathcal{j}_a | a_{\mathrm{loc}}]p^{\mathrm{i}}_b[\mathcal{j}_b | a_{\mathrm{loc } } ] . % } { % \label{eq:9kw}\end{aligned}\]]the integral factorises , and we arrive at , @xmath82 = \phi_{\mathrm{m}a } ^{\mathrm{i } } [ \zeta | a_{\mathrm{loc } } ] \phi_{\mathrm{m}b } ^{\mathrm{i } } [ \zeta | a_{\mathrm{loc } } ] . \end{aligned } \label{eq:63bf}\end{aligned}\ ] ] takes more work . substituting it into ( [ eq:41az ] ) yields , @xmath83 & = \int \mathcal{d}[\mathcal{j}_a ] \mathcal{d}[\mathcal{j}_b ] \exp[i\zeta ( \mathcal{j}_a+\mathcal{j}_b ) ] \\&\quad \times p_{a } [ \mathcal{j}_a \big| a_{\mathrm{e } } + g_{\text{r } } \mathcal{j}_b ] p_{b } [ \mathcal{j}_b \big| a_{\mathrm{e } } + g_{\text{r } } \mathcal{j}_a ] . \end{aligned } \label{eq:10kx}\end{aligned}\]]we then take the route as in , but in reverse . we start from writing , @xmath84 % % \np = \exp \bigg ( \frac{\delta } { \delta a_{\mathrm{e } } } g_{\text{r } } \mathcal{j}_b \bigg ) p_{a } [ \mathcal{j}_a \big| a_{\mathrm{e } } ] , % } { % \label{eq:11ky}\end{aligned}\]]then pull the exponential differential operator out of the integral . similar manipulations are applied to . as a result we find , @xmath85 & = \exp \bigg ( -i\frac{\delta } { \delta a_{\mathrm{e } } ' } g_{\text{r } } \frac{\delta } { \delta \zeta } -i\frac{\delta } { \delta a_{\mathrm{e } } } g_{\text{r } } \frac{\delta } { \delta \zeta ' } \bigg ) \nonumber\\&\quad \times \int \mathcal{d}[\mathcal{j}_a ] \mathcal{d}[\mathcal{j}_b ] \exp ( i\zeta \mathcal{j}_a + i\zeta ' \mathcal{j}_b ) \nonumber\\&\qquad \times p_{a } [ \mathcal{j}_a \big| a_{\mathrm{e } } ] p_{b } [ \mathcal{j}_b \big| a_{\mathrm{e } } ' ] |_{\zeta ' = \zeta , a_{\mathrm{e}}'=a_{\mathrm{e } } } . % } { % \label{eq:14lb}\end{aligned}\]]introducing pairs of variables @xmath86 and @xmath87 allows all differentiations to hit the right targets . the integral in ( [ eq:14 lb ] ) is factorised . recalling ( [ eq:27sw ] ) we arrive at the relation sought , @xmath85 & = \exp \bigg ( -i\frac{\delta } { \delta a_{\mathrm{e } } ' } g_{\text{r } } \frac{\delta } { \delta \zeta } -i\frac{\delta } { \delta a_{\mathrm{e } } } g_{\text{r } } \frac{\delta } { \delta \zeta ' } \bigg ) \nonumber\\&\quad \times \phi_{\mathrm{m}a } [ \zeta \big | a_{\mathrm{e } } ] \phi_{\mathrm{m}b } [ \zeta ' \big | a_{\mathrm{e } } ' ] \settoheight{\auxlv}{$ |$}% \raisebox{-0.3\auxlv}{$ |_{\zeta ' = \zeta , a_{\mathrm{e}}'=a_{\mathrm{e}}}$}. % } { % \label{eq:66bk}\end{aligned}\ ] ] follows from factorisation of quantum averages for two distinguishable noninteracting subsystems , cf . . to prove it , use the definition of by , and the factorisation assumption ( [ eq:84vj ] ) . we regard verified . in turn , this proves as a formula for p - functionals . substituting ( [ eq:63bf ] ) into the dressing formula ( [ eq:54zf ] ) we obtain , @xmath88 & = \exp \bigg ( -i\frac{\delta } { \delta a_{\mathrm{e } } } g_{\text{r } } \frac{\delta } { \delta \zeta } \bigg ) \nonumber\\&\quad \times \phi_{\mathrm{m}a } ^{\mathrm{i } } [ \zeta \big| a_{\mathrm{e } } ] \phi_{\mathrm{m}b } ^{\mathrm{i } } [ \zeta \big| a_{\mathrm{e } } ] . % } { % \label{eq:95fm}\end{aligned}\]]we now apply the relation @xcite , @xmath89 \mathcal{f}_3 [ f ] = \mathcal{f}_1 \bigg ( \frac{\delta } { \delta f}+ \frac{\delta } { \delta f ' } \bigg ) \mathcal{f}_2 [ f ] \mathcal{f}_3 [ f ' ] |_{f'=f } , % } { % \label{eq:96fn}\end{aligned}\]]where @xmath90 $ ] , @xmath91 $ ] , @xmath91 $ ] are arbitrary functionals and @xmath92 are auxiliary functional variables . equation ( [ eq:96fn ] ) is a compact way of formulating general rules of product differentiation . it may be verified expanding @xmath90 $ ] , @xmath91 $ ] , @xmath91 $ ] in functional taylor series . using ( [ eq:96fn ] ) we rewrite ( [ eq:95fm ] ) as , @xmath88 & = \exp \bigg [ -i \bigg ( \frac{\delta } { \delta a_{\mathrm{e } } } + \frac{\delta } { \delta a_{\mathrm{e } } ' } \bigg ) g_{\text{r } } \bigg ( \frac{\delta } { \delta \zeta } + \frac{\delta } { \delta \zeta ' } \bigg ) \bigg ] \nonumber\\&\quad \times \phi_{\mathrm{m}a } ^{\mathrm{i } } [ \zeta \big| a_{\mathrm{e } } ] \phi_{\mathrm{m}b } ^{\mathrm{i } } [ \zeta ' \big| a_{\mathrm{e } } ' ] \settoheight{\auxlv}{$\big|$}% \raisebox{-0.3\auxlv}{$\big|_{\zeta ' = \zeta , a_{\mathrm{e}}'= a_{\mathrm{e}}}$}. % } { % \label{eq:97fp}\end{aligned}\]]expanding the bilinear form in the exponent we have , @xmath93 \nonumber\\&\quad = \exp \bigg ( -i\frac{\delta } { \delta a_{\mathrm{e } } ' } g_{\text{r } } \frac{\delta } { \delta \zeta } \bigg ) \exp \bigg ( -i\frac{\delta } { \delta a_{\mathrm{e } } } g_{\text{r } } \frac{\delta } { \delta \zeta ' } \bigg ) \nonumber\\&\qquad \times \exp \bigg ( -i\frac{\delta } { \delta a_{\mathrm{e } } } g_{\text{r } } \frac{\delta } { \delta \zeta } \bigg ) \exp \bigg ( -i\frac{\delta } { \delta a_{\mathrm{e } } ' } g_{\text{r } } \frac{\delta } { \delta \zeta ' } \bigg ) . % } { % \label{eq:98fq}\end{aligned}\]]the last two factors here `` dress '' the devices , @xmath94 & = \phi_{\mathrm{m}a } [ \zeta \big | a_{\mathrm{e } } ] , \\ \exp \bigg ( -i\frac{\delta } { \delta a_{\mathrm{e } } ' } g_{\text{r } } \frac{\delta } { \delta \zeta ' } \bigg ) \phi_{\mathrm{m}b}^{\mathrm{i } } [ \zeta ' \big | a_{\mathrm{e } } ' ] & = \phi_{\mathrm{m}b } [ \zeta ' \big | a_{\mathrm{e } } ' ] , \end{aligned } \label{eq:65bj}\end{aligned}\]]cf . , and we arrive at . in turn , this proves as a formula for p - functionals . extension to many devices is achieved by applying and ( [ eq:8kv ] ) recursively . so , for a set - up of @xmath95 devices , labeled by their number in place of @xmath22 and @xmath23 , we obtain , @xmath29 & = \int \prod_{k=1}^n \big \ { \mathcal{d}[\mathcal{j}_k ] \,p^{\mathrm{i}}_k[\mathcal{j}_k | a_{\mathrm{loc } } ] \big \ } \nonumber\\&\quad \times \delta[\mathcal{j}_1+\cdots+\mathcal{j}_n -\mathcal{j } ] , % } { % \label{eq:51tx}\end{aligned}\]]and @xmath30 & = \int \prod_{k=1}^n \big \ { \mathcal{d}[\mathcal{j}_k ] \,p_k[\mathcal{j}_k | a_{\mathrm{e}}+g_{\text{r } } ( \mathcal{j}-\mathcal{j}_k ) ] \big \ } \nonumber\\&\quad \times \delta [ \mathcal{j}_1+\cdots+\mathcal{j}_n -\mathcal{j } ] , % } { % \label{eq:52ty}\end{aligned}\]]for the corresponding characteristic functionals we have , @xmath96 = \prod_{k=1}^n \phi_{\mathrm{m}k } ^{\mathrm{i } } [ \zeta | a_{\mathrm{loc } } ] , % } { % \label{eq:54ua}\end{aligned}\]]and @xmath85 & = \bigg \ { \exp \bigg ( -i\sum_{l\neq n}\frac{\delta } { \delta a_{\mathrm{e}l } } g_{\text{r } } \frac{\delta } { \delta \zeta_n } \bigg ) \nonumber\\&\quad \times \prod_{k=1}^n \phi_{\mathrm{m}k } [ \zeta_k | a_{\mathrm{e}k } ] \bigg \ } \big| _ { \ { \zeta , a_{\mathrm{e } } \ } } , % } { % \label{eq:55ub}\end{aligned}\]]where stands for the substitution , @xmath97these formulae generalise and ( [ eq:66bk ] ) . they may be verified either applying them recursively , or adapting the method of to @xmath95 devices . recursive nature of ([eq:48bh ] ) reflects design of real macroscopic ( e.g. , electronic ) devices . the latter are made of a large number of components , which , in our terms , are themselves electromagnetic devices ( cf . the `` engineering viewpoint '' in ) . true quantum mechanics only starts when this kind of subdivisibility stops . * * * ([eq:48bh ] ) give a formal expression to the statement made in the intro , that , _ under macroscopic conditions , quantum electrodynamics and classical stochastic electrodynamics have identical dynamical structure . _ one may also say that _ flow of quantum information in a macroscopic setup is governed by laws of classical stochastic electrodynamics_. truly quantum dynamics is limited to microscopic conditions of indistinguishability and/or equations of motion of matter . as was explained in , the way external c - number sources are employed in the response formulation is unphysical . this raises two questions : whether one can introduce response formulation without relying on sources , and whether one can reintroduce sources as a consistent macroscopic approximation . both questions are answered in affirmative in this section . we also take this opportunity to reiterate the key technical points of our approach , with the emphasis on their interplay with the approximations . the actual calculations underlying our resuts utilize the functional aprroach to the schwinger - perel - keldysh closed - time - loop techniques of quantum field theory @xcite ( for a recent review see @xcite ) . this is the minimal framework that allows one to include such concepts as _ arbitrary macroscopic electromagnetic device _ in a dynamical approach . the aforementioned key technical points of our approach are , * _ response transformation _ of the closed - time - loop techniques @xcite , prompted by the wave quantisation formula ( [ eq:40mr ] ) . * _ wick s theorem _ in its functional ( hori s ) form @xcite . both `` key technical points '' are exact and formally independent , but even a cursory inspection reveals that they are deeply connected . indeed , consider the problem with the hamiltonian ( cf . @xcite ) , @xmath98it is easily solved , @xmath99cf . . the no - backaction nature of a c - number source automatically makes the electromagnetic interaction directed . the question whether directedness may also be introduced in nonlinear problems is answered in the affirmative by the wave quantisation formula ( [ eq:40mr ] ) . in turn , applying response transformation to wick s theorem reveals its profound association with response properties of free quantised fields @xcite . is the simplest example of strict causality in the response of a quantum system to sources @xcite . `` strict causality '' means that it holds on all scales , macroscopic as well as microscopic . it is the``strict causality '' that makes the response formulation intuitive . paradoxical as it may sound , it is also the `` strict causality '' that makes it unphysical , because it implies one s ability to vary sorces arbitrarily on microscopic scales . application of the responce formulation to a solitary electromagnetic device @xcite and its extention to electromagnetic interactions of distinguishable devices ( of this paper ) preserve both its advantages ( causality and intuitiveness ) and its main drawback ( unphysical nature of c - number souces ) . ) . the `` electromagnetic arrows '' are directional as in . ] can one rederive the results in a manner preserving the advantages but without the drawbacks ? such alternative derivation is the subject of this section . in the main body of the paper , we resort to a `` pictorial argument '' . all calculations are relegated to the appendices . the starting point of the alternative derivation is the two - device model ( [ eq:46mx ] ) without sources , @xmath100one also has to change the order in which the `` key technical points '' ( cf . ) are applied to the two - device problem . firstly , construct a perturbative solution to this problem in the closed - time - loop framework with wick s theorem in its conventional form , secondly , impose the two - device approximation , and only in the very end perform the response transformation . the first step of this schedule may be borrowed from preprint @xcite , by dropping sources in the closed perturbative formula ( 56 ) . the other two steps are outlined in appendix [ ch:2dc ] . dressed devices as scattering entities emerge in such model due to an interplay of distinguishability with the wave quantisation formula ( [ eq:40mr ] ) . indeed , consider the formal structure of electromagnetic interaction _ before _ the response transformation ( fig . [ fig : birdseye10]a ) . this picture is a schematic representation of in the appendix . assumed distinguishability of the devices leads to separation of the electromagnetic self - action problems for the devices from the problem of their electromagnetic interaction . `` separation '' means that the self - action problems for the devices may be formulated within models ( [ eq:47my ] ) and ( [ eq:48mz ] ) without any reference to the two - device model . neither fig . [ fig : birdseye10]a nor the underlying `` know '' anything about directedness of the electromagnetic interaction . graphically , this is emphasized by making dashed arrows symbolising the electromagnetic interaction bidirectional . the concept of `` dressed macroscopic device '' relies only _ on the assumption of distinguishability and on the formal structure of quantum electrodynamics_. it is not limited to any kind of response characterisation of a quantum system . the structure of electromagnetic interactions _ after _ the response transformation is illustrated in fig . [ fig : birdseye10]b . this picture is a schematic representation of . in response picture , free electromagnetic field is characterised by the retarded propagator . consequently all `` electromagnetic arrows '' in fig . [ fig : birdseye10]b are directional as in . the way fig . [ fig : birdseye10]b is drawn emphasises that the wave quantisation formula gives rise to characterisation of the individual devices as _ dressed response entities_. this happens _ irrespective of c - number sources_which , we remind , are absent in the approach illustrated in . field entity , operating independently of the detector . formally , this entity is characterised by functional ( [ eq:70qh ] ) in appendix [ ch : phdet ] . ( b ) the four - device arrangement , to which response characterisation of the two - device model ( cf . b ) is an approximation . neglected macroscopic actions and back - actions are shown as dotted arrows . ] the central formal result of this paper is that the dressed response entities in b are identical with those emerging in formal models of the solitary devices with sources ( a ) . as formal concepts , these models were discussed in . the formal expression of this identity is the consistency condition . reintroducing sources in this approach starts from recognising the drastically different roles of single - directedness in the self - action and in the interaction problems . in the case of self - action , both ends of @xmath101 are attached , so to speak , to the same electron . there is no way to control microscopic actions and back - actions associated with emitting and reabsorbing the field by matter at a truly microscopic level . whereas in case of interaction , the signals travelling from device @xmath22 to device @xmath23 may be _ experimentally distinguished _ from those travelling from device @xmath23 to device @xmath22 . consider , e.g. , the generalised photodetection problem ( a ) . calculations underlying this figure are relegated to appendix [ ch : phdet ] ( cf . also endnote @xcite ) . one starts from two scattering entities depicted in b. the physical assumption turning them into _ source _ and _ detector _ is that the macroscopic electromagnetic backaction ( shown in a as a dotted arrow ) can be neglected . this does not interfere with self - action problems for the source and detector , nor with the accuracy with which their properties may be accounted for . neglecting macroscopic backaction gives rise to the concept of detected field . the corresponding formal entity is enclosed in a in the dashed oval . formally , the detected field is the radiated field in the model of the solitary device @xmath22 . it is described by the time - normal field averages , @xmath102the additional subscript 0 reminds that all c - number sources should be put to zero . in general , averages ( [ eq:13sf ] ) can not be mimicked by any classical phenomenology ( quantum states of detected radiation ) . the semiclassical photodetection theory is recoved by assuming that such phenomenology ( the _ classical doppelganger _ of the source , see below ) does exist . in this phenomenology one introduces a random classical field , such that , @xmath103etc . this defines sources of radiation in a classical state . for coherent sources , averages ( [ eq:13sf ] ) factorise , @xmath104etc . as far as the detection process is concerned , a coherent source is fully accounted for by the interaction with the external field in the hamiltonian . similar arguments apply to other arrangements with external sources . in particular , response characterisation of a two - device model ( b ) is physically an approximation to the four - device arrangement ( b ) , where some of the macroscopic actions and backactions ( shown as dotted arrows ) have been neglected . external sources introduced in this manner are `` coarse - grained '' quantities limited to macroscopic scales and conditions of distinguishability . they suffice for a macroscopic characterisation of response properties of devices . upgrading this characterisation to a response formulation requires external sources placed inside atoms and electrons . this _ physically unjustified _ leap of logic recovers response formulation as a formal viewpoint , while also making it unphysical . in particular , theoretical response characterisation of a device is more detailed than anything one can measure in a real experiment . recovering the former from the latter is always a matter of inference . two remarks are in place here . firstly , we called the photodetection arrangement in a _ generalised _ because we did not require classicality of the output current of the detector . such classicality is an additional assumption , distinguishing real photodetectors in the true meaning of the term ( cf . below ) . on the other hand , that this classicality is not needed allows one to easily generalise a to _ cascaded systems _ @xcite . secondly , in practice , control of macroscopic electromagnetic actions and back - actions is very much a matter of engineering . examples are control of stray currents and fields in electronics and prevention of reflection of light toward laser sources in optics ( to name just two ) . we take the liberty of starting this section with a somewhat lengthy quotation from w. lamb s renowned paper on operational approach to quantum mechanics @xcite : + `` a discussion of the interpretation of quantum mechanics on any level beyond [ regarding it a set of rules for calculating physical properties of matter ] almost inevitably becomes rather vague . the major difficulty involves the concept of measurement . i have taught graduate courses in quantum mechanics for over 20 years , and for almost all of them have dealt with measurement in the following manner . on beginning of lectures i told the students , ` you must first learn the rules of calculation in quantum mechanics , and then i will tell you about the theory of measurement and discuss the meaning of the subject . ' my attitude towards such problems has no doubt been influenced by contact with some research in experimental physics in which atomic states are `` manipulated '' by microwave and radiofrequency fields . in discussion of measurement of some dynamical variable of a physical system i want to know what apparatus is necessary for the task and how to use it , at least in principle . '' this paper follows lamb s remarks , albeit more in spirit than to the letter . to begin with , it does not seem to be possible to introduce quantum measurement as a primary concept . the reason is the indistinguishability of microscopic objects ( elementary particles ) . it is difficult to imagine how measurement could be discussed if one can not tell the measured system from the measurement apparatus and from the observer . _ measurement is limited to macroscopic interactions under conditions of distinguishability . _ quantum measurement as a formal concept may therefore only emerge as an _ approximation _ within a fully developed theory of microscopic interactions``rules of calculation '' first , `` measurement '' second . furthermore , quantum measurement is much more dependent on fundamental interactions available in nature , and on limitations of the human observer , than it is commonly acknowleged . the only kind of interaction available to human beings without use of auxiliary quantum systems as measurement apparata is the electromagnetic one . it is a long - range interaction , which acts at distances where all short - range interactions may be neglected . the latter remain hidden inside macroscopic devices , being responsible for their observable properties . it is the chierarchy of characteristic interaction lengths in nature that is ultimately responsible for the very existence of the macroscopic world , as perceived by a human observer . the critical distinction to be made in a quantum measurement theory is between _ the observed _ and _ the inferred_. no particular macroscopic experiment gives full access to the underlying quantum dynamics ( nor is this access necessary think of quarks and gluons in quantum optics ) . one should therefore distinguish the directly _ measured _ mesoscopic quantities , and the microscopic ones , the properties of which may only be _ inferred _ from the experimental data ( we remind that we treat `` mesoscopic '' and `` macroscopic '' as synonyms , cf . endnote @xcite ) . we call measurable quantities _ mesoscopic coordinates _ of the quantum system in question . real macroscopic experiments give access to properties of mesoscopic coordinates , thus allowing one only a fairly obscured and limited view of microscopic quantum dynamics . it is therefore not surprising that certain properties of mesoscopic coordinates ( notably , those associated with propagation of quantum information in macroscopic experiments ) turn out to be classical no matter what . given the actual set of fundamental interactions in nature , mesoscopic coordinates of real devices may only be electromagnetic . however , the idea of separation of the macroscopic and microscopic levels of insight is not confined to quantum mechanics , nor to electrodynamics , nor is it anything new . it underlies thermodynamics , whether classical or quantum . it may also be found in any case where some kind of `` coarse - grained '' description is introduced ( e.g. , in hydrodynamics ) . sometimes our view of a particular quantum system is obscured to the extent that its quantum properties become fully hidden , and what we see looks perfectly classical . this was exactly the situation before first experimental indications of the underlying quantum world properties of the black - body radiation and of the photoelectric effect emerged in the experiment . we then say that the _ seemingly _ classical entity we are observing is the _ classical doppelganger _ of the quantum system in question . by definition , such classicality is limited to properties of mesoscopic coordinates : an electron can not look classical while an electron current can . a _ classically behaving quantum system _ is a system that has a classical doppelganger ; in other words , it is a system that _ appears _ classical in some experiment . a system that does not behave classically is said to exhibit quantum behaviour ( _ quantum - behaved systems _ ) . an immediate word of caution is in place here . as mentioned in the introduction , we associate `` macroscopic '' and `` microscopic '' with `` distinguishable '' and `` indistinguishable '' , and not with `` large '' and `` small '' . for example , electrons in an epr pair remain indistinguishable until environmental influences , such as collisions with backgroud gas , destroy their entanglement . the epr pair must be regarded a microscopic object , irrespective of how large it becomes over its lifetime ( there is more to this than mere terminology , see below ) . on the other hand , in a hydrogen atom , the electron and proton are distinguishable unless pair creation needs to be taken into account . this allows one to go as far in the theory as the leading ( bethe ) contribution to the lamb shift @xcite ( cf . also endnote @xcite ) . this example also shows that observable quantum effects have two origins : properties of individual quantum particles and their indistinguishability . in this paper , we do not include collisions of distinguishable matter subsystems . `` collision '' is any situation where conditions of distinguishability in a setup vary with time . collisions of elementary particles are obviously covered by this definition . whether one regards the epr pair `` microscopic '' or `` macroscopic '' depends on whether collisions are explicitly accounted for . for purposes of this paper , collisions belong to microscopic dynamics . integration of the latter into our approach remains subject to further work ( there does not seem to be any formal obstacle to that , cf . @xcite ) . one more reservation is in order here . that only the electromagnetic current is observed in real experiments is a limitation of humans , and not a property of quantum mechanics . a creature with the sixth sense of strong interaction would see the world very differently , and require a very different measurement theory . in , we defined a classically behaving quantum system as a system that has a classical doppelganger . in quantum electrodynamics , this implies nonnegativity of the relevant quasiprobability distribution , allowing one to formally interpret it as a probability distribution . so , device @xmath22 in is classically behaving , if the p - functional characterising it is nonnegative , @xmath105 \geq 0 . % } { % \label{eq:57ud}\end{aligned}\]]this does not change the fact that this device is a quantum system , nor does it in any way affect validity of the general quantum formula for p - functionals ( [ eq:52ty ] ) . classically behaving devices interact with the world according to laws of quantum , not classical , electrodynamics . that under macroscopic conditions these two in essence coincide is a _ property _ of the conventional quantum electrodynamics . continuing this logic , `` classical world '' emerges when all devices within reach of the observer are classically behaving . the whole world looked classical to the human observer before first experimental indications of the underlying quantum world properties of the black - body radiation and of the photoelectric effect emerged in the experiment . the above definition of a classically behaving device calls for an extensive comment . this definition obviously depends on what we term `` quantum '' . formal techniques naturally puts the boundary at what we can / cannot _ describe _ in terms of classical stochastic electrodynamics ; this is all what is about . according to such definition , a thermal source of radiation behaves classically ( unlike , e.g. , a kdp crystal ) . this is in agreement with the general distinction between the observed and the inferred . indeed , quantum nature of the thermal radiation is recognised only so far as one attempts to _ explain _ it in some sort of classical dynamical model , and realises futility of all such attempts due to the equidistribution theorem . an important reservation is that the same real system may behave classically in some experiments and exhibit quantum behaviour in others . so , in quantum optics , the output current of a photodetector is commonly treated as a classical stochastic process . a deeper insight reveals that macroscopic currents are antibunched @xcite . the classical doppelganger of a classically behaving device is a classical system introduced regarding ( say ) a conditional probability distribution . this definition is strictly phenomenological ; a dynamical classical model of a classically behaving device is not bound to exist . an example is again the black - body radiation . in certain cases such model may be constructed ; this mostly depends on the degree of insight . so , an ideal mirror may be introduced by postulating boundary conditions for maxwell equations , but an attempt to fully explain properties of real mirrors in classical terms is doomed . as a rule , taking the limit of a classically behaving device does not recover its classical doppelganger . indeed , such limit turns a quantum hamilton evolution into a classical hamilton one . the latter is not stochastic . thus all dynamical noises ( such as shot noise ) will be lost . the noises associated with quantum averaging ( e.g. , thermal noise ) may survive such limit but get distortet ( cf . the thermal p - function of the harmonic oscillator _ versus _ thermal distribution of the classical oscillator @xcite ) . worse still , such limit may not exist , as , for instance , for the black - body radiation . while interpreting planck s curve as a probability distribution makes perfect phenomenological sense . actually , it is not even guarantied that a particular classically behaving device is subject to a semiclassical model . this again depends on the degree of insight . as a nontrivial example we point to the linear response theory of the dirac vacuum in @xcite . in the first nonvanishing approximation , vacuum polarisation may be calculated , with the necessary regularisation and renormalisation , solely as a property of free fermions . corrections of higher order would involve quantised electromagnetic field as part of the dynamical model . ) , `` intermediate '' ( @xmath106 ) , and `` strict '' ( @xmath107 ) . ] equations ( [ eq:51tx])([eq:55ub ] ) allow one to mix classically and nonclassically behaving devices at will . however , not every such mix is a `` measurement '' . the latter implies existence of a `` classical observer '' , who is only capable of interacting with `` classical measurement apparata '' ( cf . ) . in physical terms , these are classically behaving devices , but this is not enough . for a set - up to become a measurement , it should exhibit the _ quantum - classical boundary _ and contain _ macroscopic detectors . _ indeed , consider the flow of information in . the observer interacts with a computer , which controls the laser source , and processes infomation coming from the photodetector . the latter detects radiation of the quantum system pumped by the laser . devices encircled by solid lines are classically behaving ; similarly , electromagnetic radiation in a classical state is depicted schematically by solid lines with arrows . the quantum system is encircled by dashed line ; its radiation , which may be in a quantum state , is shown as dashed arrow . possible positions of the quantum - classical boundary in this set - up are shown by thick dotted lines . the `` minimal '' boundary separates the observer from the rest of the world , the `` intermediate '' one separates the observer and computer from the laser , detector and quantum system . the `` strict '' quantum - classical boundary separates the system and the detector from the rest the world which in this case includes the observer . what is shared by all three is that the observer is always on one side of the boundary , while the system _ and the detector_on the other . for a classically behaving device to be a detector , its output radiation must be in a classical state irrespective of the state of the detected radiation . this is only possible if the detector adds noise of its own , such as shot noise of real photodetectors . the same requirement may be put as that the _ compound device _ comprising the quantum system and the detector be a classically behaving one . treating the quantum system and the detector as an entity turns the measurement setup into a `` classical world '' . positioning of possible quantum - classical boundaries in depends on assumptions about the devices . if the laser radiation were in a quantum state , the strict quantum - classical boundary would shift to the `` intermediate '' one . in the future where the controlling device may be a quantum computer , the only possible choice of the quantum - classical boundary may become the `` minimal '' one . however , even in this future the observer will interact with what he perceives as a classically behaving device . in conclusion , quantum theory of the electromagnetic interaction under macroscopic conditions of distinguishability of devices and of controlled actions and back - actions between them is constructed . this theory is shown to be subject to `` doing quantum electrodynamics while thinking classically '' , allowing one to substitute essentally classical considerations for quantum ones without any loss in generality . implications of these results for the quantum measurement theory are discussed . as we have already mentioned , eq . ( [ eq:38aw ] ) is consistent only due to causality properties of the p - functionals . here is a simple example . assume that all quantities in ( [ eq:38aw ] ) do not depend on time . the external field shifts the gaussian distribution of the current , @xmath108 , \end{aligned}}}% \label{eq:3zz}\end{aligned}\]]where @xmath109 and @xmath110 are real constants . in place of ( [ eq:37av ] ) we postulate a scalar formula , @xmath111where @xmath112 is one more real constant . for the `` dressed current '' we find , @xmath113 . \end{aligned}}}% \label{eq:5ab}\end{aligned}\]]this function is not normalised , @xmath114and can not be a probability distribution for anything . to see how causality breaks this vicious circle of same - time interactions consider another simple example . we generalise ( [ eq:3zz ] ) to two currents , @xmath115 . \hspace{0.4\columnwidth}\hspace{-0.4\twocolumnwidth } \label{eq:7ad}\end{gathered}\]]the primed current preceds the unprimed one in time ; therefore it may affect the latter but not _ vice versa_. same - time interactions are not allowed either . the simplest case of such interaction is , @xmath116for the dressed currents we then have , @xmath117 . \hspace{0.4\columnwidth}\hspace{-0.4\twocolumnwidth } \label{eq:8ae}\end{gathered}\]]unlike ( [ eq:5ab ] ) , this function is normalised , @xmath118it is therefore a genuine two - dimensional conditional probability distribution for a correlated pair of currents . the order of integrations in ( [ eq:9af ] ) is chosen so as to make the result obvious . indeed , ( [ eq:8ae ] ) has the structure , @xmath119where @xmath120 , \\ p ' \big ( j ' \big | a_{\mathrm{e } } ' \big ) & = \frac{1}{\sqrt{2\pi}\ , j_0 } % \\ \times \exp \bigg [ -\frac { \big ( j ' - \chi a_{\mathrm{e } } ' \big)^2 } { 2 j_0 ^ 2 } \bigg ] . \end{aligned}}}% \label{eq:13al}\end{aligned}\]]the later current is conditional on the earlier one and the external field . the earlier current is conditional only on the external field . similar structures should emerge for any time sequence of currents irrespective of any detals of the interaction . the only requirement is that each current depends only on those preceding it in time , and not on itself . in real problems with continuous time , critical for cancellation of same - time interactions are _ regularisations_. so , in , _ causal regularisation _ was applied to the retarded green function of the emerging equation for phase - space amplitudes . this made noise sources present in the said equation independent of the amplitudes at the same time , resulting in the ito calculus being chosen . the effect of causal regularisation is thus twofold : to introduce an infinitesimal delay into the phase - space equation , which is in essence time discretisation , and to prevent same - time interactions . a conceptual connection between the above simple examples and the causal regularisation is obvious . an example of causal regularisation applied to a relativistic problem is the linear response theory of the dirac vacuum in @xcite , see also @xcite . further discussion of this issue is outside the scope of this paper . here we reiterate the relevant results of @xcite . for purposes of this appendix , we assume to be , @xmath121while interaction ( [ eq:36 mm ] ) suffices for all physical purposes , interaction ( [ eq:25ah ] ) leads to a more symmetric formulation . in paper @xcite a covariant formulation was used . adapting its results to interaction ( [ eq:25ah ] ) requires a certain amount of care . for simplicity , we assume ( anyway @xmath122 must disappear in a nonrelativistic formulation ) . equally trivial replacements apply to the arguments of functions . the obvious part of adapting @xcite thus is , @xmath123to identify more subtle replacements we compare formulae for the s - matrix . dropping the sources , the s - matrix associated with interaction ( [ eq:25ah ] ) is , @xmath124the corresponding formula in @xcite reads , ( with ) @xmath125where condensed notation ( [ eq:3vs ] ) is redefined according to the covariant standard , @xmath126 . % } { % \label{eq:56br}\end{aligned}\]]changed sign of the exponent leads to redefinition of the retarded propagator , @xmath127this implies correspondences ( [ eq:57bs ] ) . for more details on sign and factor conventions in nonrelativistic and relatistic approaches see appendix c.1 in @xcite . the simplest way of adapting the formulae of is thus to change the metric signature from to . this leads to replacements , @xmath128for all occurences of these kinds of condensed notation . combined with ( [ eq:57bs ] ) , this allows one to adapt any formula in @xcite to interaction ( [ eq:25ah ] ) . we work in the _ closed - time - loop _ formalism of the quantum field theory @xcite . the system is described by the double - time - ordered averages of the heisenberg field and current operators , ( are arbitrary integers ) @xmath129where are the forward and reverse time orderings ( often denoted @xmath56 and @xmath130 ) . by definition , bosonic operators commute under all kinds of ordering . we note in passing that definition ( [ eq:10bu ] ) may cause mathematical problems , see the concluding remark in appendix a1 in @xcite . averages of quantities ( [ eq:10bu ] ) are conveniently accessed through their characteristic functional , @xmath131 \\&\quad = \langle t_- \exp ( -i{\eta}_- { \hat{\mathcal e } } -i{\zeta}_- { \hat{\mathcal d } } ) t_+ \exp ( i{\eta}_+ { \hat{\mathcal e } } + i{\zeta}_+ { \hat{\mathcal d } } ) \rangle \label{eq:51zc}\end{aligned}\]]where @xmath132 and @xmath133 are auxiliary complex c - number functions , and c.v . ( short for _ causal variables _ ) refers to the set of _ response substitutions _ @xcite , @xmath134the symbols @xmath135 denote separation of the frequency - positive and negative parts as per eq . ( [ eq:07pa ] ) . the averaging in ( [ eq:51zc ] ) is over the initial ( heisenberg ) state of the system ( [ eq:40bj ] ) . we remind that the initial state of the field is vacuum , cf . . in terms of @xcite , this means that . the corresponding argument of is redundant and has been omitted . note also that the choice of signs in ( [ eq:51zc ] ) matches ( [ eq:55bq ] ) . the critical property of functional ( [ eq:51zc ] ) is that it depends only on sums of the external sources and the corresponding auxiliary variables , and @xcite , @xmath136 % % \\ = \phi \big [ \eta , \zeta \big | 0,0 \big | j_{\mathrm{e}}+j_{\mathrm{e } } , a_{\mathrm{e}}+a_{\mathrm{e } } \big ] . \label{eq:29fl}\end{aligned}\]]alternatively , @xmath136 % % \\ = \phi \big [ \eta , \zeta \big | j_{\mathrm{e}}+j_{\mathrm{e } } , a_{\mathrm{e}}+a_{\mathrm{e } } \big | 0,0 \big ] . \label{eq:38fv}\end{aligned}\]]these relations are a generalisation of _ consistency conditions _ derived in . equations ( [ eq:29fl ] ) and ( [ eq:38fv ] ) express the same functional but much differ in their interpretation . equation ( [ eq:29fl ] ) introduces the formal _ response viewpoint _ , the one we mostly adhere to in this paper . equation ( [ eq:38fv ] ) shows that , mathematically , external c - number sources in quantum equations of motion are redundant : information contained in the heisenberg operators conditional on the sources is already present in the operators defined without the sources . this allows one to reintroduce response viewpoint without relying on sources . for a discussion see . in view of eq . ( [ eq:29fl ] ) we may set the redundant auxiliary variables to zero , @xmath137this does not lead to any loss of generality ( assuming the souces may be arbitrary , cf . ) . complete quantum formulae may be recovered replacing , @xmath138under ( [ eq:24fe ] ) , the formal description of the system is given by the reduced characteristic functional , @xmath139 % \\ = \phi \big [ \eta , \zeta \big | 0,0 \big | j_{\mathrm{e } } , a_{\mathrm{e } } \big ] . \end{aligned}}}% \label{eq:77jn}\end{aligned}\]]we _ postulate _ this functional to be the characteristic one for the the time - normal averagea of the quantised current and field , @xmath140 = \protect \langle { \mathcal t}{\mbox{\rm\boldmath$:$}}\exp ( i\eta \protect{\hat{\mathcal a}}+ i \zeta \protect{\hat{\mathcal j } } ) { \mbox{\rm\boldmath$:$}}\rangle . % } { % \label{eq:26aj}\end{aligned}\]]the heisenberg operators depend on the sources by construction . postulating ( [ eq:26aj ] ) recovers the beyond - the - rwa definition of the time - normal ordering @xcite ; cf . . full electromagnetic properties of the quantum device may be expressed by the properties of the heisenberg ( `` dressed '' ) current operator . the latter are contained in the functional , @xmath141 = \phi \big [ 0 , \zeta \big | 0 , a_{\mathrm{e } } \big ] . \label{eq:52zd}\end{aligned}\]]namely @xcite , @xmath142 % % \\ = \exp \big ( i\eta g_{\text{r } } j_{\mathrm{e } } \big ) % \\ \times \phi_{\mathrm{m } } \big [ \zeta+\eta g_{\text{r } } \big | a_{\mathrm{ext } } \big ] , \label{eq:53ze}\end{aligned}\]]where is the natural combinations of the sources , @xmath143with , is nothing but a compact form of . nontrivial part of perturbative calculations is expressed by the `` dressing formula '' ( [ eq:54zf ] ) . the no - sources definitions of the functionals and read , @xmath144 & = \protect \langle t_-\exp ( -i{\zeta}_- \protect{\hat{\mathcal j } } ) t_+\exp ( i{\zeta}_+ \protect{\hat{\mathcal j } } ) \rangle |_{a_{\mathrm{e}}=0 } , \label{eq:22sr}\\ \phi_{\mathrm{m}}^{\mathrm{i } } [ \zeta | a_{\mathrm{e } } ] & = \protect \langle t_-\exp ( -i{\zeta}_- \hat j ) t_+\exp ( i{\zeta}_+ \hat j ) \rangle , \label{eq:62zq}\end{aligned}\]]cf . the `` immediate reservation '' at the end of . consequently , rigorous formal definitions of and are given by , ( [ eq:22sr ] ) and ( [ eq:24st ] ) , ( [ eq:62zq ] ) , respectively . to account for the linear susceptibility of matter in properties of the field , it suffices to replace defining functional by , @xmath145 & = \protect \langle t_-\exp ( -i{\zeta}_- \hat j ) t_+\exp ( i{\zeta}_+ \hat j ) \rangle , % } { % \label{eq:75uy}\end{aligned}\]]where @xmath146 \rangle . % } { % \label{eq:76uz}\end{aligned}\ ] ] implies substitution ( [ eq:23ss ] ) . then , for passive linear media , @xmath147 = 1 . % } { % \label{eq:77va}\end{aligned}\]]susceptibility may be accounted for by including it in the dyson equation for the retarded propagator , cf . appendix d.1.2 in @xcite , or by any other method one would care to employ . for interacting devices , redefinition ( [ eq:75uy ] ) extends to individual devices , @xmath148 \\&\quad = \protect \langle t_-\exp ( -i{\zeta}_- \hat j_a ) t_+\exp ( i{\zeta}_+ \hat j_a ) \rangle , \\ & \pi _ { \mathrm{r}a}(t , t ' ) = \frac{i}{\hbar } \theta(t - t ' ) \protect \langle [ \hat j_a(t),\hat j_a(t ' ) ] \rangle , \end{aligned } \label{eq:80vd}\end{aligned}\]]and similarly for device @xmath23 . the all - important persists , with susceptibility of matter being a sum of those of devices , @xmath149as expected . if a particular device ( say , @xmath23 ) is a linear passive one , it is fully accounted for in and may be excluded from the explicit consideration , @xmath150 = 1 . % } { % \label{eq:81ve}\end{aligned}\ ] ] redefinition ( [ eq:75uy ] ) is not compulsory , nor is ( [ eq:80vd ] ) , nor does anything preclude one from applying ( [ eq:80vd ] ) selectively , only to certain devices . the latter may be a matter of physical clarity . for example , for a coherent linear quantum amplifier , separates its amplification properties ( expressed by the imaginary part of ) from its noise properties ( accounted for in ) . in quantum mechanics , these two properties are connected @xcite . formally separating them is not an error , but may obscure the physical picture . for simplicity we confine the discussion to current . as a starting point we employ eq . ( 56 ) of preprint @xcite . with dropped sources , reduced to current and adapted to the present notation it reads , @xmath151 % \nonumber\\ & & = \exp\bigg ( -\frac{i\hbar}{2}\frac{\delta } { \delta a_+ } g _ { \mathrm{f } } \frac{\delta } { \delta a_+ } + \frac{i\hbar}{2}\frac{\delta } { \delta a_- } g_{\mathrm{f}}^ * \frac{\delta } { \delta a_- } \nonumber \\ & \quad -i\hbar \frac{\delta } { \delta a_- } g^{(+ ) } \frac{\delta } { \delta a_+ } \bigg ) \nonumber\\ & \qquad \times \phi_{\mathrm{m}}^{\mathrm{i } } \bigg [ \frac { a_+ } { \hbar } + \zeta_+ , \frac { a_- } { \hbar } + \zeta_- \bigg]\big| _ { a_{\pm}=0 } , \label{eq:86au } % % \nonumber % % \z \end{aligned}\]]where and are defined by and ( [ eq:62zq ] ) , are auxilliary functions , and the kernels and are the so - called contractions , @xmath152we use an obvious extension of notation ( [ eq:76nu ] ) . originates in wick s theorem . for details see @xcite . adapting manipulations ( [ eq:95fm])([eq:65bj ] ) to results in the closed - time - loop prototype of , @xmath151 & = \exp\bigg ( -i\hbar\frac{\delta } { \delta a_+ } g _ { \mathrm{f } } \frac{\delta } { \delta a_+ ' } + i\hbar\frac{\delta } { \delta a_- } g_{\mathrm{f}}^ * \frac{\delta } { \delta a_- ' } % \nonumber \\ & \quad -i\hbar \frac{\delta } { \delta a_- } g^{(+ ) } \frac{\delta } { \delta a_+ ' } -i\hbar \frac{\delta } { \delta a_- ' } g^{(+ ) } \frac{\delta } { \delta a_+ } \bigg ) \nonumber\\ & \quad \times \phi_{\mathrm{m}a } \bigg [ \frac { a_+ } { \hbar } + \zeta_+ , \frac { a_- } { \hbar } + \zeta_- \bigg ] \phi_{\mathrm{m}b } \bigg [ \frac { a_+ ' } { \hbar } + \zeta_+ , \frac { a_- ' } { \hbar } + \zeta_- \bigg ] \big| _ { a_{\pm}=a_{\pm}'=0 } , \label{eq:73uw}\end{aligned}\ ] ] where we also used symmetry of . its structure is illustrated symbolically in a. to recover ( [ eq:66bk ] ) , ( [ eq:23ss ] ) should be supplemented by two more substitutions @xcite , @xmath153this turns ( [ eq:73uw ] ) into , @xmath154 & = \exp \bigg ( -i\frac{\delta } { \delta a_{\mathrm{e } } '' } g_{\text{r } } \frac{\delta } { \delta \zeta ' } -i\frac{\delta } { \delta a_{\mathrm{e } } ' } g_{\text{r } } \frac{\delta } { \delta \zeta '' } \bigg ) % % \np \times \phi_{\mathrm{m}a } [ \zeta + \zeta ' a_{\mathrm{e } } + a_{\mathrm{e } } ' ] % % \np \times \phi_{\mathrm{m}b } [ \zeta+ \zeta '' a_{\mathrm{e } } + a_{\mathrm{e } } '' ] |_{\zeta ' = \zeta''= a_{\mathrm{e}}'=a_{\mathrm{e}}''=0 } , % } { % \label{eq:74ux}\end{aligned}\]]which is equivalent to ( [ eq:66bk ] ) . the generalised photodetection problem ( a ) is the only case in the paper when the single - mode approximation must be lifted . it suffices to assume that the electromagnetic field comprises two modes , and , where i , o refer to the _ input _ and _ output _ of the photodetector . all other quantities are similarly doubled , , etc . ; applies with . we shall also refer to the input and output modes as the optical and photocurrent ones . at this stage , all such terms are conditional . adapted to two modes , becomes , @xmath155 & = \int \mathcal{d}[\mathcal{j}_{a\mathrm{i } } ] \mathcal{d}[\mathcal{j}_{a\mathrm{o } } ] \mathcal{d}[\mathcal{j}_{a\mathrm{i } } ] \mathcal{d}[\mathcal{j}_{b\mathrm{o } } ] \delta [ \mathcal{j}_{\mathrm{i}}-\mathcal{j}_{a\mathrm{i}}-\mathcal{j}_{b\mathrm{i } } ] \delta [ \mathcal{j}_{\mathrm{o}}-\mathcal{j}_{a\mathrm{o}}-\mathcal{j}_{b\mathrm{o } } ] \nonumber\\&\quad \times p_a[\mathcal{j}_{a\mathrm{i}},\mathcal{j}_{a\mathrm{o } } | a_{\mathrm{ei}}+g_{\mathrm{ri}}\mathcal{j}_{b\mathrm{i}},a_{\mathrm{eo}}+g_{\mathrm{ro}}\mathcal{j}_{b\mathrm{o } } ] p_b[\mathcal{j}_{b\mathrm{i}},\mathcal{j}_{b\mathrm{o } } | a_{\mathrm{ei}}+g_{\mathrm{ri}}\mathcal{j}_{a\mathrm{i}},a_{\mathrm{eo}}+g_{\mathrm{ro}}\mathcal{j}_{a\mathrm{o } } ] , % } { % \label{eq:60pw}\end{aligned}\ ] ] where , @xmath156 , \\ g_{\mathrm{r}{\mathrm{o}}}(t - t ' ) & = \frac{i}{\hbar } \theta(t - t ' ) \big [ \hat a_{{\mathrm{o}}}(t),\hat a_{{\mathrm{o}}}(t ' ) \big ] . \end{aligned } \label{eq:54at}\end{aligned}\]]the only assumption made when obtaining ( [ eq:60pw ] ) is that and commute equation ( [ eq:60pw ] ) is not yet a photodetection formula . it is symmetric with respect to the modes , as well as to the devices ; electromagnetic interaction in it is bi - directional . it also contains a lot of irrelevant information , in particular , full quantum response properties of the input and output fields . it `` knows '' how an attempt to measure the input field would affect the detection , and how simultaneous measurements of the input field and output current would be correlated ( say ) . subject to valid quantum models of the devices , it is also `` aware '' of all limitations imposed by quantum mechanics on such simultaneous measurements . to suppress the irrelevant information we limit our attention to the quantity , @xmath157 = \int \mathcal{d}[\mathcal{j}_{\mathrm{i}}]\,p [ \mathcal{j}_{\mathrm{i}},\mathcal{j}_{\mathrm{o } } | 0 , 0 ] . % } { % \label{eq:62py}\end{aligned}\]]that is , we do not attempt to measure the `` optical '' current , nor to probe the system by radiation in either mode . in fact , it would be easy to preserve as an argument , thus covering photodetection with local oscillators , but we are not currently interested in such generalisation . the critical assumption turning device @xmath22 ( say ) into a source is that it is not sensitive to the incoming field in either mode , @xmath158 \approx p_a[\mathcal{j}_{a\mathrm{i}},\mathcal{j}_{a\mathrm{o } } | 0,0 ] . % } { % \label{eq:63pz}\end{aligned}\]]we shall also assume that its contribution to the photocurrent may be neglected , @xmath159 p_{\mathrm{s}}[\mathcal{j}_{a\mathrm{i}}]\,\delta [ \mathcal{j}_{a\mathrm{o } } ] , \label{eq:64qa}\end{gathered}\]]where @xmath160 = \int \mathcal{d}[\mathcal{j}_{a\mathrm{o}}]\,p_a[\mathcal{j}_{a\mathrm{i}},\mathcal{j}_{a\mathrm{o } } | 0,0 ] . % } { % \label{eq:66qc}\end{aligned}\]]combining ( [ eq:60pw ] ) , ( [ eq:62py ] ) , ( [ eq:63pz ] ) , and ( [ eq:64qa ] ) we obtain , @xmath161 = \int \mathcal{d}[\mathcal{j}_{a\mathrm{i}}]\,p_{\mathrm{s}}[\mathcal{j}_{a\mathrm{i}}]\,p_{\mathrm{d}}[\mathcal{j}_{\mathrm{o}}|g_{\mathrm{ri}}\mathcal{j}_{a\mathrm{i } } ] , % } { % \label{eq:67qd}\end{aligned}\]]where @xmath162 = \int \mathcal{d}[\mathcal{j}_{a\mathrm{i}}]\,p_b[\mathcal{j}_{a\mathrm{i}},\mathcal{j}_{\mathrm{o } } | a_{\mathrm{i}},0 ] . % } { % \label{eq:68qe}\end{aligned}\]]classical collocations of are unmistakeable . it could be guessed from the very beginning . the actual result of this appendix are and ( [ eq:68qe ] ) . they define properties of the source and detector in strictly quantum terms , including _ quantisation of the photocurrent_. to establish a closer connection with , we introduce the characteristic functional of averages ( [ eq:13sf ] ) . the operator occuring there should be identified with , defined within the model of solitary device @xmath22 . this way , @xmath163\,p_{\mathrm{s}}[\mathcal{j}_{a\mathrm{i}}]\,\exp ( i\eta g_{\mathrm{ri}}\mathcal{j}_{a\mathrm{i } } ) . % } { % \label{eq:70qh}\end{aligned}\]]the additional subscript 0 stands for the condition , @xmath164functional ( [ eq:70qh ] ) is a formal expression of the source@xmath165field entity , encircled in the dashed oval in a. with it , eq . ( [ eq:67qd ] ) may be rewritten as , @xmath157 = \phi _ { \mathrm{s } } \big ( -i\frac{\delta } { \delta a_{\mathrm{ei } } } \big ) p_{\mathrm{d}}[\mathcal{j}_{\mathrm{o}}|a_{\mathrm{ei}}]|_{a_{\mathrm{ei}}=0 } . % } { % \label{eq:69qf}\end{aligned}\]]that averages ( [ eq:13sf ] ) indeed fully characterise the source is evident . the term `` mesoscopic '' refers to quantum manifestations under macroscopic conditions . common as it actually is , this term is confusing . it implies that there exists something `` else '' that distinguishes macroscopic and mesoscopic conditions . in fact , there is nothing `` else '' apart from the hard - working experimental physicist , rescuing fragile quantum effects from being destroyed by environmental and measurement noises . we regard these terms as synonyms , with a tendency to apply `` macroscopic '' to experiment and `` mesoscopic''to theory ( as in `` macroscopic experiments are described by mesoscopic phenomenologies '' ) . the actual logic in @xcite was as follows . _ response transformation _ of the closed - time - loop formalism for bosons was firstly constructed in @xcite for the harmonic oscillator , then postulated for interacting bosons in @xcite following analogies with @xcite . this resulted , in particular , in the amended definition of the time - normal ordering . its equivalence to the conventional definition under rwa was then subject to verification . in turn , extention of the results to fermions in @xcite was based on analogies with bosons . we talk about _ an _ ultimate generalisation because there exist alternative phase - space mappings , based on orderings other than time - normal @xcite . for a discussion of comparative merits of different mappings see the quoted paper . the distinction between the mesoscopic and microscopic coordinates is very much blurred for linear bosonic systems . this results , we believe , in a lot of confusion , because such systems are commonly used as a testing ground for fundamental concepts . for pedagogical reasons , we oversimplify the discussion of photodetection in section [ ch:61np ] , brushing under the carpet a number of physically important points . we therefore strongly recommend the reader to give proper attention to appendix [ ch : phdet ] . we do not claim that description of the electromagnetic field binding the electron to proton in terms of running waves is _ useful _ , only that it is _ formally possible_. all practical approaches we are aware of depart from such viewpoint .

one of the spectacular predictions of quantum electrodynamics is the emergence of forces that arise purely from quantum fluctuations of the electromagnetic vacuum @xcite . also known as london - van der waals @xcite or casimir forces @xcite in different regimes , these forces are often dominant at short distances and can give rise to undesirable effects in nanoscale systems , such as nanomechanical stiction @xcite . these forces have also attracted increasing attention in the fields of atomic physics and quantum optics . in particular , significant efforts have been made in recent years to interface cold atoms with the evanescent fields of dielectric micro- and nano - photonic systems @xcite . these systems are expected to facilitate strong , tunable interactions between individual atoms and photons for applications such as quantum information processing @xcite and the investigation of quantum many - body physics @xcite . in practice , efficient atomic coupling to the evanescent fields of the nanophotonic systems requires that atoms be trapped within sub - wavelength distances of these structures . at these scales , quantum vacuum forces can overwhelm the forces associated with conventional optical dipole traps , typically resulting in a loss of trap stability at distances @xmath0 nm from dielectric surfaces . given the ability to engineer the properties of nanophotonic structures , an interesting question arises as to whether such systems could be used to significantly modify vacuum forces , perhaps changing their sign from being attractive to repulsive , or even creating local potential minima . the strength of nanoscale vacuum forces should lead to unprecedented energy and length scales for atomic traps , which would find use beyond nanophotonic interfaces , such as in quantum simulation protocols based upon ultracold atoms @xcite and control of inter - atomic interactions @xcite . here we propose a novel mechanism in which engineered vacuum forces can enable the formation of a nanoscale atomic trap . while a recent no - go theorem @xcite forbids a vacuum trap for atoms in their electronic ground states , tailored nanophotonic systems can yield strong repulsive potentials for atomic excited states . we show that a weak external optical field can give rise to an overall trapping potential for a dressed state . remarkably , absent fundamental limits on the losses of the surrounding dielectric structure , the fraction of excited - state population in the dressed state can become infinitesimal , which greatly enhances the trapping lifetime and stability . we identify and carefully analyze the actual limiting mechanisms . while we present calculations on a simple model where analytical results can be obtained , we also discuss the generality of our protocol to realistic systems such as photonic crystal structures . the no - go theorem for non - magnetic media states that a dielectric object in vacuum can not be stably trapped with vacuum forces for any surrounding configuration of dielectric objects , provided that the system is in thermal equilibrium @xcite . in analogy with earnshaw s theorem , which prohibits trapping of charged objects with static electric potentials , at best one can create a saddle - point potential ( see , _ e.g. _ , ref . @xcite involving metal particles and ref . @xcite for a hybrid vacuum / optical trap for atoms ) . for dielectric objects , potential loopholes involve embedding the system in a high - index fluid @xcite , or using blackbody radiation pressure associated with strong temperature gradients @xcite . applied to atoms , this theorem essentially forbids stable vacuum trapping near a dielectric structure when the atom is in its electronic ground state . no such constraint exists for atoms in their excited states @xcite , although the robustness of any possible trap would be limited by the excited state lifetime . here , we show that an atom only weakly dressed by its excited state can be trapped by properly tailoring the dispersion of the underlying structure . we begin by briefly reviewing how quantum vacuum fluctuations give rise to forces on the ground and excited states . within an effective two - level approximation of an isotropic atom with ground and excited states @xmath1 ( see appendix for description of how isotropy is incorporated into the model ) , the interaction between the electromagnetic field and an atom at position @xmath2 is given by the dipole hamiltonian @xmath3 . following a complete decomposition of the electric field into its normal modes @xmath4 , one can write @xmath5 , where @xmath6 . the energy non - conserving terms ( @xmath7 and @xmath8 ) enable an atom in its ground state @xmath9 to couple virtually to the excited state and create a photon , @xmath10 , which is subsequently re - absorbed . the corresponding frequency shift for the ground state within second - order perturbation theory is @xmath11 , where @xmath12 is the unperturbed atomic transition frequency . this shift results in a mechanical `` vacuum '' potential when translational symmetry is broken due to dielectric surfaces . using a quantization technique for electrodynamics in the presence of dispersive dielectric media , the shift can be expressed in terms of the scattered component of the dyadic electromagnetic green s function evaluated at imaginary frequencies @xmath13 @xcite ( see appendix ) , _ g()=_0^dug_(,,iu ) , where @xmath14 is the free - space spontaneous emission rate of the atom . a similar process can occur for an atom in its excited state @xmath15 , in which the atom virtually emits and re - absorbs an off - resonant photon ( @xmath16 ) . however , the excited state is unique in that it can also emit a resonant photon . the total excited state shift is given by @xcite _ e()=--g_(,,_0).[eq : omegae ] the first term on the right arises from off - resonant processes , while the second term can be interpreted as the interaction between the atom and its own resonantly emitted photon . naturally , the dielectric environment can modify the spontaneous emission rate as well , ( ) = _ 0+g_(,,_0).[eq : gamma ] note that the modified emission and resonant shifts are complementary in that they emerge from different quadratures of the green s function . the versatility of micro- and nanophotonic systems has already been exploited to greatly modify emission rates of atoms and molecules in a variety of contexts @xcite . we propose that these same systems can be used to tailor excited - state potentials in order to facilitate vacuum trapping . of course , the excited - state shift can be affected by coupling to even higher electronic levels . however , since our goal is to engineer giant resonant shifts of @xmath17 , the off - resonant couplings to higher - lying states yield only minor corrections and consequently are ignored here . in what follows , we present a simple model system that enables one - dimensional trapping in a plane parallel to a semi - infinite dielectric slab , as illustrated in fig . [ fig : schematic ] . the use of a homogeneous dielectric ( as opposed to , _ e.g. _ , a photonic crystal ) greatly simplifies the calculation and enables one to understand the relevant trap properties analytically , although we later argue why the qualitative features should still hold in more complex geometries . in the dielectric slab model , dispersion engineering will be realized via the frequency - dependent electric permittivity @xmath18 . while this particular system achieves a one - dimensional trap along @xmath19 , our electrodynamic calculations are performed in three dimensions . enables large repulsion in the excited state @xmath20 , in contrast to the attractive vacuum potential experienced by an atom in its ground state . an overall trapping potential normal to the surface ( along @xmath19 ) can be generated by driving the atom with an external laser of frequency @xmath21 , which is greater than the natural resonance frequency @xmath12 of the atom . the laser comes into local resonance with the atom at @xmath22 . an atom far from @xmath22 is unaffected by the largely detuned field and feels an attractive force @xmath23 toward the dielectric interface . b ) the atom comes closer to local resonance with the laser as @xmath24 . this creates an atomic dressed state , whose excited state component yields a net repulsive force on the atom . c ) a slowly moving atom experiences the adiabatic dressed state potential @xmath25 shown in red , which characterizes the position - dependent mixing between ground and excited states . the minimum of the dressed potential is located at @xmath26.[fig : schematic],width=491 ] a short - distance expansion of the green s function , valid for sub - wavelength scales , shows that the excited - state emission rate @xmath27 and resonant contribution @xmath28 to the shift are given by & & ( ) , [ eq : gammageneral ] + & & -(),[eq : omegaegeneral ] where the permittivity @xmath29 is evaluated at the atomic resonance frequency , @xmath19 is the distance from the surface , and @xmath30 is the resonant free - space wavevector . a large repulsive potential is generated when @xmath31 , which is analogous to the interaction between a classical dipole at position @xmath19 and its large induced image dipole in the dielectric . the modified spontaneous emission rate originates from the absorption or quenching of the atomic emission due to material losses . we choose a drude model , @xmath32 , as a simple dielectric function . note that this function satisfies kramers - kronig relations ( causality ) , as a choice of a non - causal function could result in an apparent violation of the no - go theorem @xcite as well . in the limit of vanishing material loss parameter @xmath33 , the system passes through @xmath34 at the plasmon resonance frequency @xmath35 . the system is conveniently parameterized by the quality factor @xmath36 and a dimensionless detuning @xmath37 . then , for large @xmath38 and @xmath39 , @xmath40 and @xmath41 . the dispersion and dissipation scale like @xmath42 and @xmath43 , respectively . thus , for high @xmath38 , it is possible to choose detunings where the atom still sees significant repulsive forces , but where spontaneous emission into the material is not significantly enhanced ( in addition to material - induced emission of eq . ( [ eq : gammageneral ] ) , there still is emission into free - space at a rate @xmath44 ) . this scaling behavior is in close analogy with conventional optical trapping of atoms , where simultaneously using large trapping intensity and detuning maintains reasonable trap depths but suppresses unwanted photon scattering . the ground - state shift for atomic frequencies @xmath45 and in the near - field is given by @xmath46 . we now describe how a trap can form for an atom subject to these potentials , and in the presence of a weak driving laser of frequency @xmath21 and rabi frequency @xmath47 ( see fig . [ fig : schematic ] ) . for conceptual simplicity , we choose the rabi frequency to be spatially uniform , such that all atomic forces are attributable to the vacuum potentials alone . qualitatively , the strong position dependence of the ground and excited state frequencies causes the laser to come into resonance with the atom at a single , tunable point @xmath48 . under certain conditions ( described in detail later ) , an atom starting at position @xmath49 will be far detuned , such that it is essentially in the ground state and is attracted by the pure ground - state potential . however , moving closer to @xmath22 brings the atom closer to resonance . the dressing of the atom with a small fraction of excited state population causes a repulsive barrier to form near @xmath50 , yielding a metastable trap . significantly , our approach does not attempt to directly counteract the attractive ground - state potential with large external optical potentials , in sharp contrast with other trapping schemes near dielectric surfaces @xcite . the order of our calculation of the trap properties is as follows . we first find the adiabatic potential experienced by a slowly - moving atom . we derive relevant properties around the location of the trap minimum @xmath51 ( fig . [ fig : schematic ] ) . in particular , we show that absent material constraints ( arbitrarily high @xmath38 ) , the dressed state can in principle have infinitesimal excited state population and scattering . we then quantize the motion to find the motional eigenstates , binding energy , and position uncertainty @xmath52 . finally , we analyze the mechanisms that limit the trap lifetime . first treating the motion classically , the force experienced by the atom is given by @xmath53 , while the internal atomic dynamics satisfies the usual bloch equations , _ e.g. _ , = ( i_a(z)-(z)/2)_ge+i(_ee-_gg)[eq : bloch ] . here @xmath54 is the detuning between the laser and the local atomic resonance frequency . when the atom moves slowly on the time scales of the internal dynamics , the atomic coherence can be adiabatically eliminated , @xmath55 , such that the atomic populations are functions of position alone , @xmath56 . we are primarily interested in the regime where the atom is weakly driven , @xmath57 . a more careful calculation reveals that the green s function @xmath58 for the excited state resonant shift and emission rate in eqs . ( [ eq : omegae ] ) and ( [ eq : gamma ] ) must in fact be evaluated at the laser frequency in this regime , which reflects that the atom primarily acts as a rayleigh scatterer ( see appendix ) . consequently , @xmath59 in eqs . ( [ eq : gammageneral ] ) and ( [ eq : omegaegeneral ] ) is replaced with @xmath60 . the adiabatic potential seen by a slowly - moving atom is @xmath61 . three independent frequencies characterize the system ( @xmath62 , and @xmath63 ) , and two relative frequencies must be specified . one is determined by selecting the trap barrier position , determined by the relation @xmath64 . a second is the dimensionless detuning between the plasmon resonance and laser , @xmath65 , which we treat as a free parameter that determines the relative strengths of the excited state shift and dissipation . in the near - field , the excited state repulsion exceeds the ground state attraction by a factor @xmath66 . the population at @xmath22 must then exceed the inverse of this quantity , @xmath67 , in order to provide a classical barrier in the adiabatic potential , which translates to a minimum rabi frequency @xmath68 . since @xmath69 , this scheme potentially enables atom trapping near surfaces with greatly reduced intensities as compared to conventional trapping . in the latter case , an optical dipole potential requires a minimum driving field of @xmath70 to overcome the attractive ground state potential . the excited - state population at the trap minimum is fixed by the ratio of ground to excited state forces , @xmath71 , and is surprisingly independent of @xmath47 ( provided that @xmath72 ) . an increasing rabi frequency has the effect of increasing the local detuning @xmath73 to preserve the same population . this manifests itself as a `` self - selection '' of the trapping position and an increasing distance from the barrier , ( ^2 - 1)^1/2 , where we have defined the dimensionless rabi frequency @xmath74 . the photon scattering rate at the trap position , @xmath75 , leads to lifetime limitations of our trapping scheme . fixing other parameters , the detuning @xmath76 from the plasmon resonance can be optimized to yield the minimum scattering rate , @xmath77 . thus , absent fundamental limitations to the magnitude of @xmath38 , the photon scattering rate can in principle be suppressed to an arbitrary degree . this implies that an `` infinitesimal '' violation of the assumptions underlying the no - go theorem can in fact allow for trapping . in a conventional optical dipole trap , the potential @xmath78 seen by an atom depends on the local field intensity and is proportional to a spatially - constant atomic polarizability evaluated at the laser frequency . in contrast , here we exploit a novel mechanism arising from strong spatial variations in the polarizability itself , induced through the vacuum level shifts . this creates a back - action effect analogous to that producing a dynamical `` optical spring '' in opto - mechanical systems @xcite , and enables extremely steep trap barriers around @xmath22 ( as in fig . [ fig : schematic]c ) . in particular , the repulsive force is given by @xmath79 , where @xmath80 is engineered to be large . however , the excited state population @xmath81 itself varies strongly with position through the spatially - dependent detuning ( for simplicity , here we omit the varying linewidth @xmath27 ) . this provides an overall repulsive force that scales like @xmath82 , f~-()^2 . and an infinite barrier at @xmath48 . the maximum depth of the potential is denoted by @xmath83 . the quantum wave - function amplitudes @xmath84 ( in arbitrary units ) for the two lowest eigenstates are illustrated in red , while the binding energy @xmath85 and position uncertainty @xmath52 of the ground state are depicted as well . b ) properties of the ideal trapping potential versus barrier position @xmath22 , for the case of a cesium atom . the blue solid and dashed curves give the potential depth at the barrier position , @xmath86 , and the ground - state binding energy @xmath85 , respectively . the green curve depicts the ground - state uncertainty @xmath52 . c ) for the non - ideal trapping potential , the depth of the classical potential @xmath87 is numerically calculated . here we plot it normalized by the maximum possible value @xmath83 , versus detuning parameter @xmath76 and normalized rabi freuqency @xmath88 . d ) the trapping potential , ground - state binding energy @xmath85 , and wave - function amplitude @xmath84 for the parameter sets a , b , as illustrated in c ) . the color coding is the same as in a).[fig : properties],width=566 ] given the possible steepness of this barrier , an `` ideal '' limit for the overall potential is shown in fig . [ fig : properties]a . the attractive part consists of the pure ground - state potential and arises from quantum fluctuations , while an infinite `` hard wall '' is created at @xmath48 due to the combination of strong resonant excited - state shifts ( originating from the interaction of the atom with a `` classical '' image dipole ) and opto - mechanical back - action . this idealized model enables one to understand the best possible scaling of the trap properties . in particular , the trap depth @xmath87 ( the energy required for a classical particle to become unbound ) can not exceed the value of the ground - state potential at @xmath22 , @xmath89 ( _ e.g. _ , @xmath90 mk when @xmath91 nm for a cesium atom ) . this trapping potential can be easily quantized numerically . in fig . [ fig : properties]b , we plot the quantum binding energy @xmath85 ( where @xmath92 ) and position uncertainty @xmath52 for the motional ground - state wave function . for the numerical results in figs . [ fig : properties ] and [ fig : lifetime ] , we use atomic properties corresponding to cesium ( @xmath93 nm , @xmath94 mhz , recoil frequency @xmath95 ) . the strength of the vacuum potentials is reflected in the strong confinement of the wave - functions ( _ e.g. _ , @xmath96 nm for distances @xmath97 nm ) , which are an order of magnitude smaller than in conventional optical traps . to contrast our results for cesium with those of a less massive atom , we have considered lithium as well . results for li similar to those for cs in fig . [ fig : properties]a , b show that at @xmath91 nm , the potential depth is smaller by roughly a factor of two , while the confinement @xmath52 sees roughly a five - fold increase . generally , within the wkb approximation and to lowest order in @xmath52 , the ground - state localization scales like @xmath98 , while the binding energy @xmath99 can be nearly the entire classical potential depth . this ideal model describes qualitatively well the actual potentials over a large parameter regime . in fig . [ fig : properties]c , we plot the ratio of the numerically obtained trap depth to the theoretical maximum , @xmath100 , as functions of @xmath76 and dimensionless driving amplitude @xmath101 . here , we have chosen a quality factor of @xmath102 and a barrier position of @xmath91 nm . although we have provided near - field approximations for the energy shifts above , the plots in figs . [ fig : properties]c , d and fig . [ fig : lifetime ] are calculated using the full green s functions . a trap depth comparable to @xmath86 is achievable over a large range of values . for low values of @xmath101 , the trap depth is weakened due to escape over the potential barrier , while for large values , the trap position becomes shallow as @xmath51 is pulled away from @xmath22 ( as discussed previously ) . two representative trap potentials , along with the ground - state wave functions and confinement @xmath52 are illustrated in fig . [ fig : properties]d . we have identified four sources of motional heating recoil heating , non - adiabatic motion of the atom within the trap , fluctuations of the atomic coherence , and tunneling over the finite - height potential barrier that limit the trap lifetime in absence of external cooling mechanisms . tunneling is exponentially suppressed with barrier height and thus is easily suppressed with increasing rabi frequency . this calculation can be found in the appendix , while we discuss the more important mechanisms here . as in conventional optical trapping , the random momentum recoil imparted on the atom by scattered photons yields an increase in motional energy at the rate @xmath103 . two qualitative differences emerge relative to free space , however . first , the photon scattering rate @xmath104 is modified in the presence of dielectric surfaces . second , the effective momentum @xmath105 imparted by a photon is enhanced compared to the free - space momentum . this momentum scales at close distances like @xmath106 ( see appendix ) , due to emission into high - wavevector photons in the dielectric . starting from the motional ground state , this heating causes the atom to become unbound over a time scale @xmath107 . it should be noted that the heating rate @xmath108 does not depend on the atom confinement ( _ i.e. _ , whether we operate in the lamb - dicke limit ) , as the suppression of motional jumps in tight traps is compensated by the increase in energy gained per jump @xcite . our derivation of the atomic potential assumes that the atomic coherence adjusts rapidly to the local trap properties , @xmath109 , which results in a purely conservative potential . as we now show , non - adiabatic corrections result in a momentum - dependent force , @xmath110 ( anti - damping ) . we can solve eq . ( [ eq : bloch ] ) for the atomic motion perturbatively by writing @xmath111 , where @xmath112 is the adiabatic solution . it depends only on the instantaneous position and is given by @xmath113 in the weak saturation limit . here , @xmath114 is the complex detuning . substituting this solution into eq . ( [ eq : bloch ] ) yields a velocity - dependent correction to the coherence , @xmath115 . this in turn yields a new contribution to the force , @xmath116 , where the anti - damping rate is given by @xmath117 $ ] evaluated at the trap position @xmath51 . intuitively , the atom is anti - damped ( @xmath118 ) because the laser frequency is blue - detuned relative to the atomic frequency at @xmath51 , giving rise to preferential stokes scattering over anti - stokes . aside from the term proportional to @xmath119 , the heating rate is exactly analogous to that occurring in an opto - mechanical system in the blue - detuned regime @xcite . we define a characteristic time @xmath120 needed for the energy increase produced by anti - damping to equal the ground - state binding energy @xmath85 . thus far , we have focused on the motion of the atomic dressed state , in which a trap emerges due to weak mixing with the excited state potential . following a spontaneous emission event , however , the atom returns to the ground state and sees the pure ground - state potential for a transient time @xmath121 until the atomic coherence is restored , during which the atomic wave - packet accelerates toward the surface and gains energy . while a simulation of the full dynamics of the atom ( including the spatial and internal degrees of freedom ) is quite challenging , here we estimate the heating rate from these transient processes . in particular , we first find the spatial propagator under the pure ground - state potential , @xmath122 ( see appendix ) . then , assuming that a spontaneous emission event occurs at @xmath123 with the atom initially in the motional ground - state @xmath124 , the new wave - function that emerges once the coherence equilibrates is given by @xmath125 . the probability of atom loss @xmath126 per spontaneous emission event is estimated from the overlap between the initial and final states , @xmath127 , with a corresponding trap lifetime of @xmath128 . in fig . [ fig : lifetime ] , we plot the overall trap lifetime , which is obtained by adding the individual lifetimes in parallel , @xmath129 . the overall trap lifetime exhibits a complicated dependence on rabi frequency @xmath47 and detuning @xmath76 due to the different scalings of the constituent heating mechanisms . as an example , however , for external parameters @xmath130 and @xmath131 for atomic cs , one achieves a lifetime of @xmath132 ms , ground - state confinement @xmath133 nm , trap depth @xmath134 mk , and trap distance @xmath135 nm . versus dimensionless rabi frequency @xmath74 and detuning @xmath76 . here we have assumed a plamson resonance quality factor of @xmath102 and atomic properties corresponding to cesium . [ fig : lifetime],width=302 ] we have described a protocol for an atomic trap based upon engineered vacuum forces , and have analyzed in detail a model case of one - dimensional trapping near a drude material , where analytical results are possible to obtain . the drude response approximates well a number of metals , such as silver and gold , in the optical domain @xcite . in practice , however , these metals are limited by their low quality factors ( @xmath136 ) , and the thermal fluctuations of such conducting materials can give rise to strong noise - induced trap heating and decoherence @xcite . on the other hand , photonic crystal structures enable engineering of resonances through geometry , and quality factors approaching @xmath137 have been observed @xcite . while the field profiles of such structures are more complex , we anticipate that our trapping protocol could be quite generally applied , and the resulting trap behavior would qualitatively remain the same . in particular , the scaling of the engineered excited - state shift ( @xmath138 ) and dissipation ( @xmath139 ) are generic to coupling between an atom and resonator , and only the spatial dependence is determined by the field profiles . photonic crystals should thus offer great flexibility in tailoring the properties of vacuum traps ( such as dimensionality ) , which we plan to investigate in detail in future work . we note that related efforts have already been made to design realistic hybrid atom traps in photonic crystal structures , where a combination of vacuum forces and optical forces are required to stabilize the trap in all dimensions @xcite . in addition , while we have focused on trapping of atoms weakly dressed by an external laser , it should also be possible to achieve laser - free traps by using atoms pumped into metastable electronic levels and engineering resonant shifts . the ability to create atomic traps with parameter sets ( such as depth , confinement , and proximity to surfaces ) that are not possible in conventional traps ( whether optical , electrical , or magnetic ) should lead to a number of intriguing applications . for example , it has been proposed that the trapping of atoms near dielectric surfaces with nano - scale features ( such as a sub - wavelength lattice ) would facilitate large interaction strengths and long - range interactions for ultracold atoms @xcite . however , the major limitation of optical trapping is the divergent intensity needed to overcome the ground - state vacuum forces of these structures . the use of resonantly - enhanced excited state repulsion to create a trap could significantly reduce the power requirements and enable smaller lattice constants . furthermore , traps with localization on the nanometer scale can induce inter - atomic forces that are of comparable strength to molecular van der waals forces , which enables novel opportunities to control ultracold atomic collisions @xcite and realize exotic interactions such as @xmath140-wave scattering of fermions @xcite . atoms trapped near surfaces could also act as exquisite nanoscale probes of surface physics , including the precision measurement of vacuum forces themselves . the authors thank n. stern and o. painter for helpful discussions . dec acknowledges support from fundaci privada cellex barcelona . ks was funded by the nsf physics frontier center at the jqi . jmt acknowledges funding from the nsf physics frontier center at the jqi and the us army research office muri award w911nf0910406 . hjk acknowledges funding from the iqim , an nsf physics frontier center with support of the moore foundation , by the afosr qumpass muri , by the dod nsseff program , and by nsf phy-1205729 . in this section we briefly describe the quantization scheme of the electromagnetic field in the presence of arbitrary linear dielectric media , which follows that of ref . the free field is characterized by a set of bosonic modes with frequencies @xmath141 and annihilation operators @xmath142 ( @xmath143 ) , with a corresponding hamiltonian h_f = _ jd _ j^(,)_j ( , ) . the bosonic operators satisfy canonical commutation relations @xmath144=\delta_{jk}\delta(\bfr-\bfr')\delta(\omega-\omega')$ ] and are physically associated with noise polarization sources in the material . the electric field operator at frequency @xmath141 is driven by the sources at the same frequency and is given in the coulomb gauge by @xcite e_i(,)= i_0 ^2 d g_ij(,,)_j(,).[eq : e ] here and in the following , it is assumed that all repeated vector or tensor indices are summed over , _ e.g. _ , @xmath145 . @xmath146 is the frequency - dependent permittivity at position @xmath2 , and @xmath147 is the dyadic green s function which is the ( gauge - invariant ) solution to g(,,)=(-)i.[eq : greeneq ] the total electric field is @xmath148 similarly , the magnetic field is @xmath149 . it can be verified that these expressions preserve the same field commutation relations as in free - space field quantization . in the following , we consider a two - level atom at position @xmath150 and with ground and excited states @xmath151,@xmath17 interacting with the electromagnetic field within the electric dipole approximation . the interaction hamiltonian is given by h _ & = & -*d*(_a ) + & = & -_i(_ege_i(_a)+h.c.),[eq : haf ] where @xmath152 is the atomic dipole matrix element . our system of interest consists of an atom in a region of vacuum with nearby dielectric objects . in general , for @xmath153 in the vacuum region , one can express the green s function as a sum @xmath154 , where the free component is the solution in uniform vacuum ( @xmath155 ) and the scattered component physically describes waves scattered from the dielectric objects . in a planar geometry such as that considered in the main text , the green s function can be evaluated exactly using a plane wave expansion . specifically , suppose that we have an interface with vacuum in the region @xmath156 and a dielectric with permittivity @xmath18 in the region @xmath157 . the free and scattered components of the dyadic green s function in the region @xmath158 are then given by g _ & = & -d _ e^i_(-),[eq : gfree ] + g _ & = & d_e^i_(- ) e^i(z+z ) .[eq : gsc ] here @xmath159 and @xmath160 are the wavevector and position along the direction parallel to the interface , @xmath161 , @xmath162 , and @xmath163 and @xmath164 are the fresnel reflection coefficients for @xmath165 and @xmath140 polarized waves , respectively . the interaction hamiltonian @xmath166 couples the ground state of the bare system @xmath9 ( consisting of the atomic ground state and vacuum ) to states @xmath167 . within second - order perturbation theory , this interaction induces a shift of the bare ground state energy by an amount _ g(_a)=-_jdd , where @xmath12 is the bare resonance frequency of the atom . substituting eqs . ( [ eq : e ] ) and ( [ eq : haf ] ) into the expression above , and using analyticity of the integrand to rotate the frequency integral onto the positive imaginary axis @xmath13 , one finds @xcite _ g(_a ) = _ 0^ du _ i g_,ij(_a,_a , iu ) _ j.[eq : omegag ] here we have only included the scattered component of the green s function , as the free component gives a contribution that is independent of position . while we have considered a two - level atom for simplicity , a more realistic model of an isotropic atom consists of multiple excited states and equal polarizabilities in each direction ( _ i.e. _ , the ground state can emit virtual photons of any polarization with equal strength ) . since the contribution to the ground - state shift from each excited state transition is additive , eq . ( [ eq : omegag ] ) can be modified to account for this by setting @xmath168 for all @xmath169 and associating @xmath170 with the free - space spontaneous emission rate of any excited state . this yields eq . ( 1 ) in the main text . in this section , because we want to focus exclusively on the effect of vacuum interactions with the excited state , it is sufficient to include only energy - conserving terms in the atom - field interaction hamiltonian @xmath166 ( _ i.e. _ , allowing a transition from the excited to ground state accompanied by creation of a photon from vacuum ) . we also want to consider the specific case where the atom is weakly driven by a classical field of frequency @xmath21 , which motivates working in a rotating frame where the free atomic evolution is given by @xmath171 and @xmath172 is the detuning between the laser frequency and atomic resonance frequency @xmath12 . as will be seen , the inclusion of the laser frequency in the free hamiltonian causes the effective excited state shift and decay rates to depend on the green s function evaluated at @xmath21 once the vacuum modes are eliminated . physically , this describes the effect of the weakly driven atom ( which primarily acts as a rayleigh scatterer ) interacting with its own rayleigh - scattered field , which can be altered due to the presence of nearby dielectric surfaces . the standard technique to derive the excited - state shift and decay rate follows that of appendix [ sec : ground ] for the ground state and uses time - independent second - order perturbation theory @xcite . in this approach , however , one can not find the jump operator associated with photon emission , which is needed in order to calculate the effect of emission on atomic motion ( _ e.g. _ , recoil heating ) . to rectify this , we employ an alternative approach based on deriving an atomic master equation within the born - markov approximation @xcite . in particular , the equation of motion for the reduced atomic density matrix @xmath173 is given by _ a = _ f_0^,[eq : rhodot ] where @xmath174 is the interaction hamiltonian in the interaction picture ( with respect to the free hamiltonian @xmath175 ) and @xmath176 is the electromagnetic vacuum state . any exponential of time appearing in the integral can be evaluated using the relation @xmath177 . using the field expansion of eq . ( [ eq : e ] ) in @xmath174 and following some manipulation , eq . ( [ eq : rhodot ] ) can be written in the form _ a=-(i/)[h_,_a]+[_a ] . the effective hamiltonian describes coherent interactions between the atom and vacuum field and takes the form h_=_e(_a)_ee , which can be interpreted as a position - dependent energy shift of the excited state . a realistic model of an atom allows for the possibility for an excited state to emit photons of different polarizations and transition into different states in a ground - state manifold . the different clebsch - gordan coefficients of these transitions , along with the symmetry breaking introduced by the nearby dielectric structure , imply that different excited states can experience different shifts absent some special engineering . to arrive at a simpler two - level atomic model that captures the essential physics of our trapping mechanism , we assume that the excited state is equally coupled to all polarizations ( but still with total emission rate @xmath14 ) , such that one should average over all dipole orientations . in this case , the excited - state shift is given by _ e(_a)=-_0^dug_(_a,_a , iu)-g_(_a,_a,_l),[eq : omegaeappendix ] as stated in the main text . the liouvillian term @xmath178 $ ] can be written in a lindblad form [ _ a]=_jd _ j_a _ j^-(_j^_j_a+_a_j^_j ) . the jump operators take the form _ i g^_ij(_a,,_l)_ge.[eq : ojump ] the excited - state spontaneous emission rate is in turn given through the relation @xmath179 , or ( _ a ) = _ 0+g_(_a,_a,_l).[eq : gammaappendix ] eq . ( [ eq : gammaappendix ] ) again averages over dipole orientations and reproduces the expression for the emission rate provided in the main text . to derive this , we have used the identity d ( ( , ) ) g_ij(,,)g_jk^(,,)=g_ik(, , ) . the expressions for the excited - state emission rate @xmath180 and level shift @xmath181 agree with previous derivations @xcite , if the laser frequency @xmath21 is replaced by the atomic resonance frequency @xmath12 . the appearance of the laser frequency here is associated with the restriction to the weak - driving limit , which was not considered in earlier derivations . the main new result obtained here , however , is an explicit expression for the jump operator associated with photon emission , which we will use to calculate photon recoil heating in appendix [ sec : recoil ] . here we derive in more detail the equations relating to heating of the atomic motion around the minimum of the vacuum force trap . recoil heating can be calculated by treating the atomic position @xmath150 in eq . ( [ eq : ojump ] ) as an operator . for notational simplicity , we shall consider recoil heating only along one direction , say @xmath19 , although our results can be easily generalized . in the main text , @xmath19 corresponds to the direction in which the atom is trapped ( normal to the dielectric surface ) . the increase in momentum uncertainty due to photon scattering is given by = ( p_z^2 _ a ) . here we will focus solely on the liouvillian term in the density matrix evolution ( _ i.e. _ , momentum incurred from spontaneous emission ) . it can be shown that the liouvillian term produces no net force , _ i.e. _ , @xmath182 . using the fact that @xmath183=-i\hbar\frac{\partial f}{\partial z_a}$ ] for any function @xmath184 , one can derive the following general expression for the recoil heating rate in terms of green s functions , = _ 0_l^2 _ i _ j ( _ 1 ^ 2 + 2_1_2-_2 ^ 2).[eq : recoil ] here , the derivatives @xmath185 and @xmath186 act on the first and second spatial arguments of the green s functions , respectively , and we have suppressed the atomic spatial variables in the directions that are not of interest . we further assume that the internal and spatial degrees of freedom can be de - correlated , @xmath187 . formally , we can write the above equation in the convenient form = ( k_)^2 ( _ a ) p_e , where @xmath188 is the excited - state population . this form is intuitive as it describes the momentum increase arising from a random walk process , where the atom experiences random momentum kicks of size @xmath189 at a rate @xmath190 ( _ i.e. _ , the photon scattering rate ) . @xmath191 is a derived quantity that thus characterizes the effective momentum associated with scattered photons in the vicinity of a dielectric surface . as a simple example , we can consider an atom in free space polarized parallel to the surface ( say along @xmath192 ) . evaluating eq . ( [ eq : recoil ] ) with the free - space green s function @xmath193 from eq . ( [ eq : gfree ] ) yields @xmath194 ( and @xmath195 ) , recovering the known result @xcite . this recoil momentum is smaller than the full momentum @xmath196 of the scattered photon because the atom emits in a dipole pattern , but only the projection of the photon momentum along @xmath19 contributes to motional heating in that direction . as a more complicated example , we can consider the recoil heating near a surface in the trapping scheme considered in the main text . for the drude model and at small distances between the atom and surface , we find that the scattered component of the green s function dominates the recoil heating , leading to the asymptotic expansion of @xmath197 given in the main text . in all of our numerical calculations , the full green s function is used rather than asymptotic results . as described in the main text , following a spontaneous emission event , the spatial wave - function of the atom is temporarily untrapped and evolves under the pure ground - state potential for a characteristic time @xmath198 before the internal dynamics equilibrates . here we discuss how to calculate an approximate propagator @xmath122 for the wave - function evolving under the ground - state potential . the problem significantly simplifies if the ground - state potential @xmath199 is linearized around the atom trapping position @xmath200 , h+u_g(z_t)+mg(z - z_t ) , where @xmath201 . this linearization is justified in our regime of interest where the atom is tightly trapped and the wave - function undergoes little evolution in the time @xmath198 . for simplicity , here we will ignore the constant energy and position offsets and consider the simplified hamiltonian @xmath202 , which corresponds to a mass accelerating under a constant gravitational force . this observation enables one to solve for the propagator by utilizing the equivalence principle , that the system of interest is equivalent to that of a free particle viewed in an accelerating frame . one finds that the propagator is given by u(z_f , z_i , t,0)=u_(z_f , z_i , t)(--),[eq : prop ] where @xmath203 is the well - known propagator for a free particle in an inertial frame @xcite . in our simulations , the ground state wave - function for an arbitrary trapping potential is obtained numerically on a lattice , and so the propagator must be spatially discretized as well . in particular , while the exponential term in eq . ( [ eq : prop ] ) is well - behaved on a lattice , the free propagation @xmath204 is not at short times . we instead replace @xmath204 with the propagator @xmath205 for the free discrete schrodinger equation , = i(_n+1(t)-2_n(t)+_n-1(t ) ) , where @xmath206 , @xmath207 is the lattice site index , and @xmath208 is the lattice constant . the propagator for this equation ( defined as the matrix satisfying @xmath209 ) is given by d_mn(t)=i^m - nj_m - n(2t)e^-2it , where @xmath210 is the @xmath207-th order bessel function . the escape rate of an atom from the metastable potential formed by the vacuum trapping scheme due to quantum tunneling is calculated in the wkb approximation in two steps . first , we find an approximation to the real part of the bound state energy @xmath211 . this is implemented by replacing the actual metastable dressed - state potential @xmath25 with a stable one , @xmath212 , according to the formula ( also see fig . [ fig : wkb ] ) u_s(z ) & = & u_d(z ) ( z > z _ ) , + & = & u_d(z _ ) ( zz _ ) . here , @xmath213 is the position corresponding to the maximum of the metastable barrier . the ground - state energy of @xmath212 is solved numerically . once the approximate ground - state binding energy is obtained , the tunneling probability @xmath214 for a particle hitting the metastable barrier is calculated using the wkb approximation , p_t=4(-2_z_1^z_2 dz ) . here @xmath215 are the solutions to @xmath216 and are thus the classical turning points of a particle with energy @xmath211 . the rate that the trapped atom collides with the barrier is approximated by @xmath217 , where the position and momentum uncertainties are obtained from the numerical ground - state solution . thus we estimate the inverse of the tunneling - limited lifetime to be @xmath218 . ( in arbitrary units ) seen by the atom . the maximum of the metastable barrier is located at @xmath213 . red curve : stable potential @xmath212 used to approximately calculate the binding energy @xmath211 ( blue line ) of the atom in the trap . the classical turning points @xmath215 of a particle with this energy , which are used to evaluate the tunneling rate in the wkb approximation , are labeled as well.[fig : wkb],width=264 ]

very unstable nuclei with unusual ratios of neutron number ( @xmath2 ) to proton number ( @xmath5 ) are currently the subject of intense theoretical and experimental work . a particularly important issue is the effect on nuclear properties of a large neutron excess and the accompanying low - lying continuum of excited states . in ground states these conditions sometimes lead to halos , thick skins , and changes in magic numbers . less is known about excited states , which we study here by calculating transition strengths and densities . most of the work on excited states is in light neutron - rich nuclei ; a low - energy strength - function peak has been studied for years . recent work @xcite has focused on enhanced @xmath4 strength in nuclei like @xmath6li or in neutron - rich oxygen isotopes @xcite . but in heavier nuclei , calculations are more spotty . several groups have calculated the isovector ( iv ) @xmath4 strength functions , or photoabsorption cross sections , in the ca isotopes @xcite . references @xcite predict a small peak at around @xmath7 mev in @xmath8ca . the first paper also calculates the strength function in @xmath9ca , finding no low - energy peak , but ref . @xcite , a calculation based on density - functional theory , finds a very broad bump at @xmath1010 mev . it is not clear , therefore , at what value of @xmath2 the low - energy strength is first discernible . the method we use here , the quasiparticle random phase approximation ( qrpa ) , has been applied mainly to selected @xmath4 excitations in particular isotopes of o , ca , and ni ; one example is ref.@xcite , which asserts that the low - energy excitations reflect strong particle - particle ( dineutron " ) correlations . references @xcite show extensive and systematic qrpa calculations of photoabsorption cross sections , but focus primarily on radiative neutron - capture cross sections . the relativistic qrpa @xcite has been applied more with an eye to nuclear structure issues ( for a recent review , see ref . @xcite ) , asserting , for example , that low - lying isoscalar ( is ) @xmath4 states exhibit toroidal flow . references @xcite go beyond rpa , altering details of predicted low - lying strength . other channels have been discussed less than the iv @xmath4 . the authors of ref . @xcite calculate @xmath3 strength functions in ca isotopes up to the neutron drip line , finding an enhancement at low energy in @xmath8ca . reference @xcite examines transition strengths in the @xmath3 , @xmath4 , and @xmath1 channels near the neutron drip line , finding , for example , a sharp @xmath1 peak near @xmath11 in @xmath12ni . the above represents a considerable amount of work on excitations in exotic nuclei . its range and results can be summarized , roughly , as follows : 1 ) most of the calculations are for spherical nuclei , and the iv @xmath4 channel has been studied the most . 2 ) a low - energy peak grows as @xmath2 increases . 3 ) the low - energy excitations do not seem collective in the @xmath4 channel ( e.g. ref . @xcite ) , i.e. they do not contain more than a few particle - hole or two - quasiparticle configurations . 4 ) the corresponding transition densities have long neutron tails extending beyond the bulk nuclear radius . 5 ) the neutron skin thickness is correlated with the amount of low - energy strength @xcite . 6 ) calculations reproduce measurements of the amount of low - energy strength fairly well if they include many - particle many - hole correlations @xcite . most of this work has focused on nuclei right at the drip line and , as already noted , has emphasized the iv @xmath4 channel . it is still not known whether theory makes similar predictions for other neutron - rich nuclei , or to what extent the properties of the low - energy peaks vary from one multipole to another . the purpose of this paper , therefore , is to calculate strength functions in as many nuclei as possible , in all multipoles up to @xmath1 , and investigate how the properties of the strength functions vary as @xmath2 and the multipolarity change . we use the qrpa with skyrme density functionals ( which we refer to as interactions from here on ) and volume - type delta pairing interactions . currently , the qrpa is the most sophisticated method that can be used systematically for resonances of all energies in heavy nuclei without assuming that part of the nucleus forms an inert core . here we will display results of qrpa calculations for ca , ni , and sn isotopes from the proton drip line to the neutron drip line , describing the corresponding strength functions and transition densities in detail . we will also examine whether results depend on which skyrme interaction we use , so that we can identify important measurements / nuclei that will help distinguish among them . section [ sec : calculation ] below describes our method . section [ sec : strength function ] shows strength functions in the three isotopic chains and looks for @xmath13-dependent trends in their behavior . section [ sec : transition density ] discusses transition densities to states in low - lying peaks . section [ sec : conclusion ] is a conclusion . our approach to the qrpa is described in detail in ref . @xcite , and we discuss it here only briefly . the starting point is a coordinate - space hartree - fock - bogoliubov ( hfb ) calculation , carried out with the code used in refs . the code outputs quasiparticle wave functions , with spherical symmetry assumed . it takes the continuum into account by placing the nucleus in a 20-fm box . we use as large a single - particle space as possible and solve the hfb equations accurately , so as to ease the removal of spurious states from the qrpa and satisfy sum rules . we diagonalize the hfb density matrix to obtain a set of canonical states as many as there are quasiparticle states that make up our qrpa basis . the qrpa - hamiltonian matrix then contains off - diagonal elements from the quasiparticle one - body hamiltonian as well as the matrix elements of the residual two - body interaction . the two - body matrix elements are functional derivatives of the energy with respect to the density and pairing tensor , after diagonalizing the qrpa hamiltonian , we use the eigenvectors to obtain strength functions @xcite and transition densities ( see appendix ) . we can not handle an infinite number of canonical quasiparticle states , and so truncate our basis in the following way : for @xmath14 the maximum single - particle angular momentum @xmath15 is 15/2 , and for @xmath16 @xmath15 is 21/2 . if the ground state has a finite pairing gap in the hfb calculation we truncate the canonical quasiparticle basis in the qrpa by omitting states with very small occupation probability @xmath17 . in the @xmath18 and @xmath4 channel we keep states with @xmath19 , and in the @xmath20 channel we keep those with @xmath21 . if the mean field has no pairing we use the hartree - fock energy to truncate ; in the @xmath18 and @xmath4 channels we keep single - particle states with energy @xmath22 mev , and in the @xmath20 channel we keep those with @xmath23 mev . in the hfb calculations we always cut off the quasiparticle spectrum at 200 mev . the largest qrpa - hamiltonian matrix that we encounter here ( in the @xmath4 channel for sn near the neutron drip line ) is about 14,000 , and the minimum is about 2,000 ( in the @xmath3 channel for ca near the proton drip line ) . and @xmath24 mev were used for the @xmath1 channel , and then the maximum dimension of the hamiltonian matrix was about 20,000 . ] the separation of spurious excitations from physical states is as complete as in ref . @xcite in most of cases . we use 3 skyrme interactions : skm@xmath25 , sly4 , and skp . our pairing interaction has zero range with no density dependence ( this is `` volume pairing '' ) . for reasons we do not understand , we can not obtain solutions of the hfb equation with skp for ca and ni isotopes . the other interactions give solutions that are unstable against qrpa quadrupole vibrations for isotopes with : @xmath26 in ni ( skm@xmath25 ) ; @xmath27 ( skm@xmath25 ) and @xmath28 , @xmath29 ( sly4 ) in sn , and so we do not discuss these nuclei . strength functions are important because they capture information about all excited states and can often be measured . we define the strength functions through the transition operators @xmath30 : @xmath31 @xmath32 : @xmath33 @xmath34 : @xmath35 where @xmath36 is total - nucleon number , and @xmath37 is a single - particle index . the upper limit for @xmath5 ( @xmath2 ) on a summation sign means that the sum is over protons ( neutrons ) . in the next few sections we show the strength functions , with spurious states removed , produced by skm@xmath25 , and discuss those produced by the other skyrme interactions when appropriate . we then display energy - weighted sums ( up to several different energies ) produced by all our interactions , in all isotopes . figures [ fig : str_ca_0+][fig : str_sn_0 + ] show the is and iv strength functions in the @xmath18 channels . in ca ( fig.[fig : str_ca_0 + ] ) , most of the functions are smooth except for small low - energy spikes in several nuclei and the large low - energy peaks in @xmath39ca . the is giant resonances peak at @xmath40 mev for @xmath41ca and gradually move to 15 mev by @xmath42ca . as @xmath2 approaches the neutron drip line , a peak near @xmath11 grows , becoming quite large at the drip line . the iv giant resonance is broad in the light isotopes and develops a low - energy component as @xmath2 increases , apparently increasing the summed strength . the larger peak varies only slightly in energy over the same range of @xmath2 . near the drip line the low - energy iv strength develops a peak that exactly mirrors the is peak , indicating that the strength in both channels is produced solely by neutrons this distinct low - energy bump seems to appear first at 10 mev in @xmath43ca . this nucleus is not as short - lived as the isotopes near the drip line ; perhaps the appearance of the low - energy mode is experimentally testable . the same nucleus marks the beginning of the low energy peak when we use sly4 . the @xmath2 dependence of the strength function of ni ( fig . [ fig : str_ni_0 + ] ) is similar in many ways to that of ca ; the energy of the is giant - resonance energy , for example , falls in a similar way past @xmath44 . a large low - energy peak again develops near the neutron - drip line , though not as dramatically as in ca , and with the complication that the strength shrinks after @xmath45ni . this peak first is noticeable in both the is and iv channel at 6 mev in @xmath46ni , here and in calculations with sly4 . this example and that in ca suggest more generally that the nucleus with 2 neutrons outside a closed shell will mark the appearance of the low - energy peak . in sn ( fig . [ fig : str_sn_0 + ] ) , the peak near zero energy around the drip line is much smaller than in ca and ni . the bump first appears in @xmath47sn at @xmath48 mev ( violating by two nucleons the conjecture made a few lines up ) in both the is and iv channels . this threshold nucleus is the same with other skyrme interactions . although we can make these kinds of comparisons in words , we would like to display the results of both ( or all three where possible ) skyrme interactions in a more quantitative and graphic way . to limit the number of our figures , we do so by showing predictions for the sum of energy weighted strength up to 4 successively higher energies . we define the `` partial '' sum , up to an energy @xmath49 as @xmath50 where @xmath51 is a strength function , eq . ( [ eq : part_esum ] ) overestimates the energy - weighted sum , because the tail in the negative energy region is not included . this problem occurs near the neutron drip line , and the energy - weighted sum is then overestimated by 10 % at most . we did not calculate @xmath52 from negative @xmath53 , because then @xmath54 is sometimes negative . ] ( see eq . ( 1 ) of ref . figure [ fig : part_esum_is_ca0 + ] shows @xmath52 with @xmath55 , 20 , 30 , and 100 mev for the is @xmath3 channel in ca . all skyrme interactions show similar @xmath2-dependence and @xmath49-dependence ; the difference between the @xmath56 and 100 mev curves is comparable with or smaller than the differences among the other curves ( separated by 10 mev ) , reflecting the eventual vanishing of the strength as @xmath49 increases . the @xmath40 mev curve increases more rapidly with @xmath2 than the others in the region @xmath57 , indicating that the strength around @xmath40 mev is shifted down at large @xmath2 . the differences among the interactions are most apparent in the location of the neutron drip line , a static property . the differences in strength functions , at least in this channel , are for the most part minor ( though large enough to make significant differences in , e.g , the compression modulus ) . figure [ fig : part_esum_iv_ca0 + ] shows the partial sums in the iv channel . here the two interactions do yield somewhat different results . though the @xmath2-dependence is similar , skm@xmath25 produces more integrated strength than sly4 . to see how the total energy - weighted strength can differ from one interaction to another , we examine the iv energy - weighted sum rule ( ewsr ) @xmath58 @xmath59 @xmath60 @xmath61 @xmath62 @xmath63 here @xmath64 is the energy of the excited state @xmath65 , @xmath66 is the ground state , @xmath67 is the nucleon mass , and @xmath68 @xmath69 is the mean square proton ( neutron ) ground - state radius . the sum over states @xmath70 does not include spurious states . the second term @xmath71 in eq . ( [ eq : ewsr_iv ] ) , with @xmath72 and @xmath73 of eq . ( [ eq:9 ] ) the proton and neutron ground - state densities , is a correction arising from the momentum - dependent terms of the skyrme interaction . the skyrme parameters @xmath74 are defined in refs . @xcite ( skm@xmath25 ) , @xcite ( sly4 ) , and @xcite ( skp ) . another term in the sum rule depending on the proton - neutron mixed density is always zero in our work . in tab . [ tab : ewsr_iv_ca0 + ] we show @xmath75 , @xmath76 , @xmath77 , and ewsr ( the full energy - weighted sum ) for @xmath8ca . the interactions reproduce ground - state observables about equally well , so the difference in the ewsr , as the table shows , comes from the term involving @xmath78 . the degree by which the ewsr exceeds @xmath79 is in line with expectations ( see , e.g. , chap . 10 of @xcite and refs . @xcite ) . the properties of ni are similar to those of ca . @xmath80 in the is channel of ni increases more rapidly than the other curves ( fig . [ fig : part_esum_is_ni0 + ] ) . skm@xmath25 again has the larger iv sum - rule value . [ fig : part_esum_iv_ni0 + ] ) . @xmath80 in the is channel of sn is much higher than @xmath81 ( fig . [ fig : part_esum_is_sn0 + ] ) , reflecting the lower energy of the giant resonance . figure [ fig : part_esum_iv_sn0 + ] shows that the iv sum rules follow the same order as in the other isotopes , with skp slightly larger than sly4 . .[tab : ewsr_iv_ca0 + ] components of the ewsr of iv @xmath3 channel of @xmath8ca . [ cols="^,^,^,^,^ " , ] is ( solid ) and iv ( dashed ) 0@xmath82 strength functions for even ca isotopes ( skm@xmath25).,scaledwidth=100.0% ] the same as fig . [ fig : str_ca_0 + ] but for ni isotopes . , scaledwidth=100.0% ] the same as fig . [ fig : str_ca_0 + ] but for sn isotopes . , scaledwidth=100.0% ] @xmath2-dependence of the partial energy - weighted sums in the is @xmath3 channel of ca . the upper limits of the sums are 10 mev ( plus symbol ) , 20 mev ( x ) , 30 mev ( asterisk ) , and 100 mev ( open square ) . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_is_ca0 + ] but for the iv channel . the open squares now correspond to an upper limit of 150 mev . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_is_ca0 + ] but for ni . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_iv_ca0 + ] but for ni . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_is_ca0 + ] but for sn . , scaledwidth=100.0% ] the same as fig . [ fig : part_esum_iv_ca0 + ] but for sn . , scaledwidth=100.0% ] having analyzed the @xmath3 strength functions in detail , we will , for the most part , let the higher - multipole figures speak for themselves . just a few remarks about the @xmath4 channel : the is strength functions in ca appear in fig . [ fig : str_ca_1- ] . here , unlike in the @xmath3 channel , there is a clear low - energy peak in all nuclei , increasing in sizeca are so sharp because we smooth the strength functions with a smaller width at bound states than in the continuum ( see ref . @xcite ) . ] after @xmath9ca . ( the is strength functions of @xmath84ca are multiplied by 0.1 in the figure . ) in @xmath42ca 45 % of the energy - weighted strength is below 10 mev . the iv strength functions also have increasing low - energy strength at large @xmath2 . the small low - energy bump in @xmath43ca is produced by sly4 as well as skm@xmath85 . we see similar @xmath2-dependence in ni ( see fig . [ fig : str_ni_1- ] ) , though after @xmath45ni the low - energy is peak shrinks . the threshold for rapid growth with both skyrme interactions is the closed - shell nucleus @xmath86ni . in sn , once again , the low - energy is peak grows with @xmath2 , though more smoothly than in the other isotopes ( see fig . [ fig : str_sn_1- ] ) . in particular , there is apparently no sudden increase at the closed - shell - plus - two nucleus @xmath87sn . recently , ref . @xcite reported a clear iv bump around 10 mev in that nucleus . the bumps in our calculated spectra do not seem to really set in until higher @xmath2 . the @xmath4 @xmath52 tell a slightly different story ( figs . [ fig : part_esum_is_ca1-][fig : part_esum_iv_sn1- ] ) . the partial sums in the is channel of sn ( fig . [ fig : part_esum_is_sn1- ] ) show a noticeable kink at @xmath88 , due clearly to an increase in low - energy strength . as remarked above , the increase is not obvious in the strength - function figures . the choice of interaction makes very little difference in the @xmath52 except in the blank regions ( indicating quadrupole deformation ) and near the neutron drip line . according to the analytical argument in the previous subsection , if the iv ewsr depends on the interaction in one multipole , then it should in all multipoles . as figs . [ fig : part_esum_iv_ca1- ] , [ fig : part_esum_iv_ni1- ] , and [ fig : part_esum_iv_sn1- ] show , this is indeed the case . we have not commented on the question of whether protons are involved in the transitions to low - lying states in this channel because the is and iv operators do not have the same radial dependence . we will address the issue in later when we display transition densities . is ( solid , scale on left ) and iv ( dashed , scale on right ) 1@xmath89 strength functions for even ca isotopes ( skm@xmath25 ) . the is strength functions in @xmath84ca are reduced by a factor of 10.,scaledwidth=100.0% ] is ( solid , scale on left ) and iv ( dashed , scale on right ) 1@xmath89 strength functions for even ni isotopes ( skm@xmath25 ) . , scaledwidth=100.0% ] the same as fig . [ fig : str_ni_1- ] but for sn isotopes . , scaledwidth=100.0% ] the same as fig . [ fig : part_esum_is_ca0 + ] but for the @xmath4 channel . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_iv_ca0 + ] but for the @xmath4 channel . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_is_ni0 + ] but for the @xmath4 channel . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_iv_ni0 + ] but for the @xmath4 channel . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_is_sn0 + ] but for the @xmath4 channel . , scaledwidth=100.0% ] the same as fig . [ fig : part_esum_iv_sn0 + ] but for @xmath4 channel . , scaledwidth=100.0% ] the strength functions in the @xmath1 channels are qualitatively different from those in the lower - multipole channels . in ca , for example ( fig . [ fig : str_ca_2 + ] ) , all isotopes strength functions of @xmath8ca without low - energy peaks ; we do not know the reasons for the difference between that calculation and ours . ] except @xmath91 have considerable low - energy strength , in many instances because of sharp quadrupole vibrations . the low - energy strength grows with @xmath2 , but unlike in the @xmath3 channel , both neutrons and protons are involved in the transitions to these states , even near the drip line , because is and iv distributions do not coincide . this is so for all skyrme interactions . in fact the is strength at low energies is much larger than the iv strength in all isotopes except @xmath42ca . even there , the is strength to the low - energy peak with skm@xmath85 is 256 @xmath92@xmath93 and the iv strength only 136 @xmath92@xmath93 . the @xmath1 strength functions in ni ( fig . [ fig : str_ni_2 + ] ) are similar , though with two differences : all the nickel isotopes have low - energy peaks , and for @xmath94ni the iv peak is at a different energy than the is peak . the lowest - lying peak seems to be nearly pure is , like the familiar surface - quadrupole vibrations in stable nuclei . in the next section we will discuss the extent to which the states near the neutron - drip line display the same vibrational features we observe near stability . sly4 puts the drip line at @xmath95ni , much closer to stability , and in isotopes close to @xmath95ni puts the is and iv peaks at about the same location . in the lighter ( @xmath96 ) ni isotopes , the lowest peak always has a little bit of iv strength mixed with the is strength , as with skm@xmath25 . in @xmath46ni ( skm@xmath25 ) , the iv strength in the peak at 400 kev is 5 % of the is strength . things are quite similar in sn ( fig . [ fig : str_sn_2 + ] ) ; for @xmath97sn the lowest - energy peak has both is and the iv components but in @xmath98sn it has no iv component . the other interactions give the same results near the neutron drip line , indicating again that proton excitations are important in these modes everywhere . the predictions for @xmath52 in ca ( fig . [ fig : part_esum_is_ca2 + ] ) show that sly4 evinces slightly more of a closed shell ( a small kink in the 10-mev curve ) at @xmath91 than does skm@xmath85 , while showing less of a shell effect at @xmath99 ( no small kink in the 20-mev curve ) . as was the case in the @xmath3 and @xmath4 channels , the iv @xmath52 ( fig . [ fig : part_esum_iv_ca2 + ] ) depends on the interaction . similar differences in shell - strength among the skyrme interactions are apparent in the ni and sn @xmath52 curves for the is channel ( figs . [ fig : part_esum_is_ni2 + ] and [ fig : part_esum_is_sn2 + ] ) , while the difference is not obvious in the iv channels ( figs . [ fig : part_esum_iv_ni2 + ] and [ fig : part_esum_iv_sn2 + ] ) . the is @xmath52 for sn ( fig . [ fig : part_esum_is_sn2 + ] ) indicate that skp has more low - energy strength @xmath100 than the other parameter sets . the iv curves in sn show interaction differences similar to those in the @xmath4 channel . is ( solid ) and iv ( dashed ) 2@xmath82 strength functions in even ca isotopes ( skm@xmath25 ) . , scaledwidth=100.0% ] the same as fig . [ fig : str_ca_2 + ] but for ni isotopes . , scaledwidth=100.0% ] the same as fig . [ fig : str_ca_2 + ] but for sn isotopes . , scaledwidth=100.0% ] the same as fig . [ fig : part_esum_is_ca0 + ] but for the @xmath1 channel . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_iv_ca0 + ] but for the @xmath1 channel . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_is_ni0 + ] but for the @xmath1 channel . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_iv_ni0 + ] but for the @xmath1 channel . , scaledwidth=80.0% ] the same as fig . [ fig : part_esum_is_sn0 + ] but for the @xmath1 channel . , scaledwidth=100.0% ] the same as fig . [ fig : part_esum_iv_sn0 + ] but for @xmath1 channel . , scaledwidth=100.0% ] giant resonances lie in the continuum and their @xmath2-dependence is therefore smooth . the low - energy peaks , on the other hand , can vary dramatically with @xmath2 , especially near the drip line . in this section , to learn more about their collectivity and @xmath2 dependence , we examine transition densities to representative low - lying states . the transition density @xmath101 is defined in the appendix . here we focus on the radial transition density @xmath102 the transition amplitude depends on this quantity through the relation @xmath103 where @xmath104 is the radial part of the multipole operator @xmath105 . the radial transition density therefore specifies the radius at which the strength is large or small . the transition density , as eq . ( [ eq : tr_den ] ) shows , is proportional to the product of particle and hole wave functions and so is localized as long as all hole states are localized , i.e. as long as the ground state is bound . in this section , as before , we will be using skm@xmath25 unless we state otherwise . figure [ fig : tr_den_ca0+n16 ] shows the transition densities to 4 states below the giant resonances in @xmath106ca with large strengths ( the energies label the results of the qrpa calculation before smoothing ) . proton excitations are important , particularly because of the factor @xmath107 in the @xmath3 transition operator . the rms radii of the protons and neutrons in the ground state are 3.40 and 3.19 fm . the important part for the strength comes from outside the rms radius . this is not true of the giant resonances . figure [ fig : tr_den_gr_ca0+n16 ] shows transition densities in the peaks of the giant resonances of @xmath106ca . the upper two panels represent the is giant resonance , and as is expected the protons and neutrons have similar transition densities . in the lower panels , corresponding to the iv giant resonance , the two densities are out of phase . the peaks of the proton transition densities are closer to the center of the nucleus than those of the lower - energy states , particularly in the is channel . we saw that a small bump appeared after magic numbers in neutron - rich nuclei . in ca , the relevant isotope is @xmath43ca . [ fig : tr_den_ca0+n30 ] ) shows that the excitation is created entirely by neutrons . but the state is not at all collective . the largest two - quasiparticle component in the qrpa wave function is @xmath108(@xmath109 mev , @xmath110(@xmath111 mev , @xmath112 , where @xmath113 is the quasiparticle energy in the hfb quasiparticle basis and @xmath114 is the diagonal ground - state occupation probability ( norm of the lower component of the quasiparticle wave function ) . the radial quantum numbers are assigned in order of @xmath113 ( @xmath115 for those having @xmath116 ) . @xmath117(@xmath118 ) is the forward ( backward ) amplitude of the qrpa solution . the squared amplitude @xmath119 of the @xmath120 component is @xmath121 . the component corresponds to the excitation of a neutron from a @xmath122 orbit near the fermi surface to another @xmath122 orbit about 9 mev above . the first orbit is strongly affected by pairing correlations . the two - quasiparticle energy is 10.055 mev , and the energy of the excited state itself is 10.005 mev ; the excitation is clearly almost unaffected by the residual interaction . yet this nearly pure two - quasiparticle state , because the neutron transition density has such a long tail , carries 5 % of the total is transition strength . furthermore , protons contribute essentially nothing to the transition ; the maximum of the @xmath119 for protons is @xmath123 . this remarkable situation dominance by a single two - neutron - quasiparticle configuration with very large strength , occurs in the @xmath3 channel repeatedly . take , for example , the drip - line nucleus @xmath42ca . two low - lying states , at @xmath124 and 2.995 mev , together account for 25.8 % of the total is strength and 10.2 % of the ewsr . transition densities , shown in fig . [ fig : tr_den_ca0+n56 ] , are again dominated by neutrons , with outer peaks at 911 fm . the rms ground - state radii are 3.738 fm for protons and 4.615 fm for neutrons , so the highest ( inner ) peaks of @xmath125 are close to the neutron rms radius . the main component of the excited state with @xmath124 mev , with a squared amplitude of 0.807 , is @xmath126 mev , @xmath127 mev , @xmath128 ) . the main component of the state at @xmath129 mev , with a squared amplitude of 0.930 , is @xmath130 mev , @xmath131 mev , @xmath132 . like before , these are nearly pure two - neutron - quasiparticle states with long tails in their product . turning now to ni near the closed @xmath133 shell , we see that @xmath46ni has a small bump around 5 mev , while @xmath86ni does not ( fig . [ fig : str_ni_0 + ] ) . the transition density in fig . [ fig : tr_den_ni0+n52 ] again shows neutron dominance . this excited state has 2.8 % of the total is strength , and its main component is @xmath134 mev , @xmath135 mev , @xmath136 , with @xmath137 . at the drip - line nucleus @xmath138ni , the main component of the excited state at @xmath139 mev , which is one of the states forming a low - energy bump , is @xmath140 mev , @xmath141 ; it has @xmath142 . in sn the nucleus just past the closed neutron shell is @xmath87sn . here , as we noted earlier , a new bump is not visible ( fig . [ fig : str_sn_0 + ] ) , until @xmath47sn . we show the transition density to the corresponding state in fig . [ fig : tr_den_sn0+n86 ] . the excitation s main component is @xmath143 mev , @xmath144 mev , @xmath145 , with @xmath146 . @xmath147 mev is the smallest neutron quasiparticle energy of the hfb calculation . figure [ fig : tr_den_sn0+n126 ] shows the transition densities to three states in the low - energy bump in @xmath148sn . the behavior is similar to that in the drip - line ca and ni nuclei . we turn finally to the dependence of the transition densities on the skyrme interaction . figure [ fig : tr_den_ca0+n16_sly4_skoprime ] shows , unsurprisingly by now , that skm@xmath25 and sly4 give similar transition densities to states below the giant resonances in @xmath106ca . we show transition densities for some giant - resonance states in fig . [ fig : tr_den_ca0+n16_gr_sly4_skoprime ] . the is / iv character of those states is clear . next we compare predictions for @xmath43ca . figure [ fig : tr_den_ca0+n30_sly4_skoprime ] shows the sly4 transition densities to the states in the small bump in the low - energy continuum . the bump emerges there with both interactions . @xmath149ca is the neutron - drip - line nucleus obtained with sly4 ; figure [ fig : str_ca0+n46_sly4 ] displays the strength functions in that nucleus . apparently they are similar not to those of the skm@xmath85 drip - line nucleus @xmath42ca but rather to those of the skm@xmath85 @xmath149ca ( see fig . [ fig : str_ca_0 + ] ) . clearly the quantum numbers of the quasiparticle orbits involved in the transition affect the transition density much more than the chemical potential . figures [ fig : tr_den_ca0+n46_sly4 ] ( sly4 ) and [ fig : tr_den_ca0+n46 ] ( skm@xmath25 ) show the transition densities to the excited states in the shoulder at about 9 mev in @xmath149ca . the positions of the outer neutron peak are nearly the same in both , but the neutron transition densities for @xmath150 fm have some differences . the proton transition density is small except in the right panel of fig . [ fig : tr_den_ca0+n46 ] . the tails of the transition densities are quite different from those in fig . [ fig : tr_den_ca0+n56 ] , for the nucleus with 10 more neutrons . the two - quasiparticle components with the largest amplitude involve either neutron @xmath151 s or @xmath122 s , depending on the state and interaction , and the largest amplitude itself ranges from 0.5 to 0.99 . proton and neutron transition densities to the four states in @xmath106ca in the peaks of the strength function ( @xmath152 mev ) in fig . [ fig : str_ca_0 + ] . ] the same as fig . [ fig : tr_den_ca0+n16 ] but for the giant resonances . the states in the upper panels correspond to the peaks of the is giant resonance , and in the lower to the iv giant resonance . ] transition densities to the excited state in the small low - energy bump in the @xmath3 strength function of @xmath43ca . ] transition densities to the two excited states in the low - energy peak in the @xmath3 strength functions of @xmath42ca . ] transition densities to the excited state in the small low - energy bump in the @xmath3 strength function of @xmath46ni . ] transition densities to the excited state in a small bump in the @xmath3 strength function of @xmath47sn . ] transition densities to the excited state in a small bump in the @xmath3 strength function of @xmath148sn . ] transition densities to the @xmath3 excited states below the giant resonances in @xmath106ca with relatively large strength ( sly4 ) . ] transition densities to the @xmath3 excited states of @xmath106ca in the giant resonances ( sly4 ) . ] transition densities to the @xmath3 excited states of @xmath43ca in a small bump at low - energy ( sly4 ) . ] 0@xmath82 strength functions of @xmath149ca ( sly4 ) . ] transition densities to the excited states in the low - energy shoulder ( is ) or peak ( iv ) around @xmath7 mev of the strength function of fig . [ fig : str_ca0+n46_sly4 ] ( sly4 ) . ] transition densities to the excited states in the low - energy shoulders around @xmath7 mev and 11 mev in the @xmath3 strength function of @xmath149ca ( skm@xmath25 ) . ] as we have just seen , the large neutron tails in the transition density are responsible for enhanced low - lying strength near the drip line . the robust enhancement of low - lying is @xmath4 strength is due in large part to the factor @xmath153 , which emphasizes large radii , in the transition operator . one stark difference between these transition densities and those of the @xmath3 channel is the role played by protons , which is non - negligible here . figure [ fig : tr_den_ca1-n30 ] shows the transition densities of two states in the low - lying is peak in @xmath43ca . protons and neutrons contribute coherently . the component of the state at @xmath154 mev with the largest amplitude is @xmath155 mev , @xmath156 mev , @xmath157 . the squared amplitude @xmath119 is only 0.36 . two other configurations , @xmath158 and @xmath159 have @xmath160 and 0.12 , not much smaller . and since the two - quasiparticle energy of the largest component is 9.099 mev , the excited state gains more correlation energy than those in the @xmath3 channel . the proton contributions remain even at the neutron drip line . figure [ fig : tr_den_ca1-n56 ] shows the transition densities to 3 excited states in the large low - energy is peak in @xmath42ca ( fig . [ fig : str_ca_1- ] ) . these states have is strengths 9227 @xmath92@xmath161 mev ) , 3219 @xmath92@xmath162 mev ) , and 4856 @xmath92@xmath163 mev ) , representing 42 % , 15 % , and 22 % of the total strength . while protons play no role for @xmath164 fm , they are active around @xmath165 fm . the main difference between these states and those of @xmath43ca is that here the neutrons both extend much further out and contribute more inside the nucleus . it is the tail that is mainly responsible for the increased strength . interestingly , the states in @xmath42ca are mainly of two - quasiparticle character even though there is a coherent proton contribution . the component in the state at @xmath166 mev with the largest amplitude ( @xmath167 ) is @xmath168 mev , @xmath169 mev , @xmath170 . in the state at @xmath171 mev the component with the largest amplitude ( @xmath172 ) is @xmath173 mev , @xmath174 , and at @xmath175 mev it is @xmath176 mev , @xmath177 , with @xmath178 . the proton two - quasiparticle configuration with the largest amplitude has @xmath179 . in ni , a low - energy peak rises suddenly at @xmath46ni ( fig . [ fig : str_ni_1- ] ) . the transition densities of the states in the peak , shown in fig . [ fig : tr_den_ni1-n52 ] , are similar to those of the neutron - rich ca isotopes . the density in @xmath138ni , at the drip line , is similar though larger . in sn there is no sudden increase in the low - energy peak around @xmath88 ( see fig . [ fig : str_sn_1- ] ) . fig . [ fig : tr_den_sn1-n80_n84 ] shows the transition densities to low - energy states in three isotopes around @xmath88 . here the proton contributions stretch to larger radii than in ca or ni , even though they are dwarfed by the neutron tails there . the other skyrme interactions give very similar results , both in the is and iv channels and for all three isotope chains . transition densities to the excited states in the low - energy peak of the is @xmath4 strength function of @xmath43ca . ] transition densities to the excited states in the low - energy peak of the is @xmath4 strength function of @xmath42ca . ] transition densities to the excited states in the low - energy peak of the is @xmath4 strength function of @xmath46ni . ] transition densities to the excited states in the low - energy peak of the is @xmath4 strength function of @xmath180sn ( upper panels ) , @xmath181sn ( middle ) , and @xmath87sn ( lower ) . ] figures [ fig : str_ca_2+][fig : str_sn_2 + ] showed that at the neutron drip line the is and iv strength functions in ni and sn have distinct low - energy peaks ( though the is strength is always larger , even at the low energy iv peak ) while in ca the peaks coincided . this is true , except in ni with sly4ni has a small iv peak at the is peak . ] , no matter which skyrme interaction we use . with skm@xmath85 , the is strength of the large low - energy peak at @xmath182 mev in @xmath138ni is 92 % of the total strength ( 8.9 % of the ewsr ) , and that of the peak at @xmath183 mev in @xmath148sn is 23 % of the total strength ( 2.2 % of the ewsr ) . in fig . [ fig : tr_den_sn2+n126_ni2+n70_ca2+n56 ] we compare the transition densities of the states in these low - energy peaks with those in @xmath42ca . as fig . [ fig : str_ca_2 + ] suggests , though the protons play a role , the neutrons are proportionately more important in ca and do not seem to correlate with the protons . in @xmath138ni and @xmath148sn , by contrast , the neutrons and protons are completely in phase ( as they must be if the iv strength is negligible ) . the other skyrme interactions bear out these conclusions , though in the sly4 calculation of the drip - line ca isotope the ratio of proton transition density to neutron transition density is larger than that with skm@xmath25 ( shown in the left panel of fig . [ fig : tr_den_sn2+n126_ni2+n70_ca2+n56 ] ) . in all nuclei near the neutron drip line , the states in low - energy peaks are mostly above neutron - emission threshold , yet the neutron tails cut off at much smaller radii than to those in the @xmath3 and @xmath4 channels . in addition , the transition densities in those other channels generally have a node at about @xmath165 fm , while the @xmath1 transition densities show no such feature . we put off for just a moment the obvious question of whether these low - lying is peaks are familiar quadrupole vibrations and turn briefly to the iv channel . figure [ fig : tr_den_2_sn2+n126_ni2+n70 ] shows the transition densities to states in the low - energy iv strength - function peak in @xmath148sn and @xmath138ni . the states in @xmath148sn display remarkably is character , despite being at the peak of the iv distribution . the fact that the iv distribution peaks there is a little misleading because the is strength is still considerably larger , even though it is larger still at lower energies . we get some iv strength because the ratio of the transition density of protons to that of neutrons is not exactly equal to @xmath184 ( see eq . ( [ eq : transition_operator_2 + ] ) ) . returning to the is strength , we note , in accord with fig.[fig : tr_den_sn2+n126_ni2+n70_ca2+n56 ] , that the transitions appear more collective in the ni and sn drip - line isotopes than in that of ca . the component with the largest amplitude ( @xmath185 ) in the state at @xmath183 mev of @xmath42ca is @xmath186 mev , @xmath187 mev , @xmath188 , and that with the second largest ( @xmath189 ) is @xmath190 . this is still a predominantly single two - quasiparticle excitation . by contrast , the components of the state with at @xmath182 mev in @xmath138ni with the largest amplitudes are @xmath191 while still not very collective , this state is a little more so than its counterpart in ca . ( @xmath138ni may be in the spherical - deformed transition region . ) in @xmath148sn , the component of the state at @xmath192 mev with the largest amplitude is ( fig . [ fig : tr_den_sn2+n126_ni2+n70_ca2+n56 ] ) @xmath193 mev , @xmath194 mev , @xmath195 , ( @xmath196 ) and that with the next largest is @xmath197 mev , @xmath198 mev , @xmath199 , ( @xmath200 ) . this excitation is mainly a single neutron particle and neutron hole . still , the proton contribution is much larger than that in other channels at the neutron drip line . this excited state has 23.7 % of the total is strength and 8.6 % of the ewsr . we see both in this channel and others that the large transition strength near the neutron drip line does not always indicate the coherent contributions of many two - quasiparticle excitations ( i.e. , collectivity ) . yet the transition densities make these strong low - lying states , particularly in ni and sn , look very much like the surface vibrations that characterize nuclei closer to stability . to get a better handle on the degree to which the states we examine here can be called vibrations , we show in fig . [ fig : n - e - be2-largest_sn_ni_ca2 + ] the energy , transition probability @xmath201 , and largest @xmath119 for neutrons in the lowest - energy @xmath1 states of ca , ni , and sn . the ca panel shows , first of all , that we do not reproduce experimental data very well except in a few nuclei , and second , that the states are dominated by a single two - quasiparticle excitation for @xmath202 . for @xmath203 something peculiar happens ; when the energy increases , so does the @xmath201 , and vice versa , with the largest @xmath119 , a measure of collectivity , varying irregularly . in nuclei with collective vibrations , transition strength is usually correlated inversely with energy @xcite ; we see the opposite here . near the neutron drip line of ni , the classic signs of increasing quadrupole collectivity decreasing energy , rising @xmath201 , and decreasing maximum @xmath119 are all present . the sn isotopes look much more familiar , exhibiting the usual trends , except at the magic numbers . and the states close to the drip line with @xmath204 are as classically collective as ( or more so than ) those in more stable isotopes with @xmath205 . finally we show the interaction dependence of the collectivity measures for the lowest @xmath1 states in sn ( the easiest chain to understand ) in fig . [ fig : n - e - be2-largest_sly4_skp_skoprime ] . in the mid - shell region , the energies are low , the @xmath201 s are high , and the largest amplitudes low , as expected from systematics in stable nuclei in regions of large dynamical deformation . the values of these quantities depend significantly on the interaction , however . ca and ni show similar behavior but with more irregularities . transition densities to the excited states in the low - energy peak of the is @xmath1 strength functions in the neutron - drip - line nuclei ( wth skm@xmath25 ) @xmath42ca , @xmath138ni , and @xmath148sn . both proton and neutron curves are reduced by a factor of 5 in @xmath138ni to fit on the figure .. ] transition densities to the excited states in the low - energy peak of the iv @xmath1 strength functions ( skm@xmath25 ) in @xmath148sn and @xmath138ni . ] @xmath2-dependence of the energy ( upper ) , @xmath201 ( middle ) , and largest @xmath119 ( lower ) of the lowest @xmath1 states in ca , ni , and sn , with skm@xmath25 . experimental data are also shown where available . , scaledwidth=100.0% ] the same as fig . [ fig : n - e - be2-largest_sn_ni_ca2 + ] but for sn only , sly4 and skp . ] the feature that leaps out of our calculations is the enhancement of low - energy strength as the neutron drip line is approached . it is present in all multipoles ( and is most pronounced in the is @xmath4 channel , where the transition operator contains a factor @xmath153 ) , with all skyrme interactions , and in every isotopic chain , beginning just after a closed neutron shell . interestingly , the recent measurement of an iv @xmath4 strength function at gsi shows a low - energy peak before a closed shell , in @xmath180sn . this may be due to correlations not included in the qrpa . the transition densities to the low - lying states reveal some interesting features . in the @xmath3 and @xmath4 channels , the low - energy peaks are usually created essentially entirely by a single two - neutron - quasiparticle configuration . in the @xmath3 channel , where the total number of two - quasiparticle configurations is restricted , even small admixtures of two - proton - quasiparticle states are absent ; the proton transition density is nearly zero . protons play a noticeably larger role in the transition density of the @xmath4 channel despite the lack of any two - proton - quasiparticle configurations with sizable amplitude . in both these channels , the transition densities have a peak around the rms radius and very long tail outside the radius , which allows single quasiparticle configuration to have such large strength . the transition densities to @xmath1 states are different ; they have no real nodes and a proton component that is of the same order as the neutron component . the transition densities near the neutron - drip line resemble those of surface vibrations , and in many cases the corresponding states clearly are vibrational . but the usual systematics of collective vibrations is not always present ; energies and @xmath492 strengths are sometimes correlated in the wrong way . some of the most significant differences in the predictions of the three skyrme interactions we used are in static properties : the locations of the neutron drip line and quadrupole - deformed regions . but some dynamical properties also depend significantly on the interaction , most notably ( and easiest to measure ) the energies of and transition - strengths to the lowest @xmath1 states near regions of deformation and the total energy - weighted iv strength . this paper , in a way , is a catalog of predicted dynamic properties of nuclei near the drip line . for the most part we have not offered simple explanations for the anomalies that appear . that , we hope , will be the subject of a future paper . this work was supported in part by the u.s . department of energy under grant de - fg02 - 97er41019 . we used computers belonging to the national center for computational sciences at oak ridge national laboratory and information technology services at university of north carolina at chapel hill . parts of this research were done when one of us ( j.t . ) was at the university of north carolina at chapel hill and at riken . the transition density is defined by @xmath206 where @xmath66 and @xmath65 denote the ground and excited states of the system , and @xmath207 in the right - hand side is an operator acting on the nucleon @xmath37 . the transition density is evaluated in the qrpa ( we omit @xmath208 ) as @xmath209 where @xmath210 and @xmath211 are the forward and backward amplitudes of the qrpa solution @xmath70 ( see appendix a of ref . @xcite ) , and @xmath212 stands for a single - particle canonical - basis wave function . the @xmath213 factors associated with the basis are represented by @xmath214 and @xmath215 . assuming spherical symmetry for the system and using @xmath216 with @xmath217 a set of quantum numbers , @xmath218 a radial wave function , and @xmath219 a spin wave function , one can derive @xmath220 where @xmath221 , and @xmath222 denotes the angular momentum of the state @xmath70 . @xmath223 and @xmath224 are the forward and backward amplitudes in the spherical representation .

the last seven years have been a very exciting time for string theory . a new understanding of nonperturbative features of string theory , such as d - branes , has led to exciting new developments relating string theory to physically interesting systems such as black holes and supersymmetric gauge field theories , as well as to a new understanding of the relationship between yang - mills theories and quantum theories of gravity . despite remarkable progress in these directions , however , a consistent nonperturbative background - independent formulation of string theory is still lacking . this situation makes it impossible at this point , even in principle , to directly address cosmological questions using string theory . string field theory is a nonperturbative approach to string theory which holds some promise towards providing a background - independent definition of the theory . these lecture notes give an introduction to string field theory and review some recent work which incorporates d - branes into the framework of string field theory . this work shows that string field theory is a sufficiently robust framework that distinct string backgrounds can arise as disconnected solutions of the theory , at least for open strings . it remains to be seen whether this success can be replicated in the closed string sector . in this section we review briefly the situation in string theory as a whole , and summarize the goals of this set of lectures . in section 2 we review some basic aspects of d - branes . in section 3 , we describe a particular d - brane configuration which exhibits a tachyonic instability . this tachyon can be seen in the low - energy super yang - mills description of the d - brane geometry . this field theory tachyon provides a simple model which embodies much of the physics of the more complicated string field theory tachyon discussed in the later lectures . in section 4 we give an introduction to witten s cubic bosonic open string field theory and summarize the conjectures made by sen in 1999 , which suggested that the tachyonic instability of the open bosonic string can be interpreted in terms of an unstable space - filling d - brane , and that this system can be analytically described through open string field theory . section 5 gives a more detailed analytic description of witten s cubic string field theory . in section 6 we summarize evidence from string field theory for sen s conjectures . section 7 contains a brief review of some more recent developments . section 8 contains concluding remarks and lists some open problems . much new work has been done in this area since these lectures were presented at valdivia in january 2002 . except for a few references to more recent developments in footnotes and in the last two sections , these lecture notes primarily cover work done before january 2002 . previous articles reviewing related work include those of ohmori @xcite , de smet @xcite , and arefeva _ et al . an expanded set of lecture notes , based on lectures given by the author and barton zwiebach at tasi 01 , will appear in @xcite ; the larger set of notes will include further details on a number of topics . to understand the significance of developments over the last seven years , it is useful to recall the situation of string theory as it was in early 1995 . at that time it was clearly understood that there were 5 distinct ways in which a supersymmetric closed string could be quantized to give a microscopic definition of a theory of quantum gravity in ten dimensions . each of these approaches to quantizing the string gives a set of rules for calculating scattering amplitudes between on - shell string states which describe gravitational quanta as well as an infinite family of massive particles in a ten - dimensional spacetime . these five string theories are known as the type iia , iib , i , heterotic @xmath0 , and heterotic @xmath1 superstring theories . while these string theories give perturbative descriptions of quantum gravity , in 1995 there was little understanding of nonperturbative aspects of these theories . in the years between 1995 and 2000 , several new ideas dramatically transformed our understanding of string theory . we now briefly summarize these ideas and mention some aspects of these developments relevant to the main topic of these lectures . * dualities : * the five different perturbative formulations of superstring theory are all related to one another through duality symmetries @xcite , whereby the degrees of freedom in one theory can be described through a duality transformation in terms of the degrees of freedom of another theory . some of these duality symmetries are nonperturbative , in the sense that the string coupling @xmath2 in one theory is related to the inverse string coupling @xmath3 in a dual theory . the web of dualities relating the different theories gives a picture in which , rather than describing five distinct possibilities for a fundamental theory , each of the perturbative superstring theories appears to be a particular perturbative limit of some other , as yet unknown , underlying theoretical structure . * m - theory : * in addition to the five perturbative string theories , the web of dualities also seems to include a limit which describes a quantum theory of gravity in eleven dimensions . this new theory has been dubbed `` m - theory '' . although no covariant definition for m - theory has been given , this theory can be related to type iia and heterotic @xmath1 string theories through compactification on a circle @xmath4 and the space @xmath5 respectively @xcite . for example , in relating to the type iia theory , the compactification radius @xmath6 of m - theory becomes the product @xmath7 of the string coupling and string length in the 10d iia theory . thus , m - theory in flat space , which arises in the limit @xmath8 , can be thought of as the strong coupling limit of type iia string theory . it is also suspected that m - theory may be describable as a quantum theory of membranes in 11 dimensions @xcite , although a covariant formulation of such a theory is still lacking . * branes : * in addition to strings , all five superstring theories , as well as m - theory , contain extended objects of higher dimensionality known as `` branes '' . m - theory has m2-branes and m5-branes , which have two and five dimensions of spatial extent ( whereas a string has one ) . the different superstring theories each have different complements of d - branes as well as the fundamental string and neveu - schwarz 5-brane ; in particular , the iia / iib superstring theories contain d - branes of all even / odd dimensions . the branes of one theory can be related to the branes of another through the duality transformations mentioned above . through an appropriate sequence of dualities , any brane can be mapped to any other brane , including the string itself . this suggests that none of these objects are really any more fundamental than any others ; this idea is known as `` brane democracy '' . * m(atrix ) theory and ads / cft : * one of the most remarkable results of the developments just mentioned is the realization that in certain space - time backgrounds , m - theory and string theory can be completely described through simple supersymmetric quantum mechanics and field theory models related to the low - energy description of systems of branes . the m(atrix ) model of m - theory is a simple supersymmetric matrix quantum mechanics which is believed to capture all of the physics of m - theory in asymptotically flat spacetime ( in light - cone coordinates ) . a closely related set of higher - dimensional supersymmetric yang - mills theories are related to string theory in backgrounds described by the product of anti - de sitter space and a sphere through the ads / cft correspondence . it is believed that these models of m - theory and string theory give true nonperturbative descriptions of quantum gravity in space - time backgrounds which have the asymptotic geometry relevant to each model . for reviews of m(atrix ) theory and ads / cft , see @xcite . the set of ideas just summarized have greatly increased our understanding of nonperturbative aspects of string theory . in particular , through m(atrix ) theory and the ads / cft correspondences we now have nonperturbative definitions of m - theory and string theory in certain asymptotic space - time backgrounds which could , in principle , be used to calculate any local result in quantum gravity . while these new insights are very powerful , however , we are still lacking a truly background - independent formulation of string theory . the goal of these lectures is to describe progress towards a nonperturbative background - independent formulation of string theory . such a formulation is needed to address fundamental questions such as : what is string theory / m - theory ? how is the vacuum of string theory selected ? i.e. _ , why can the observable low - energy universe be accurately described by the standard model of particle physics in four space - time dimensions with an apparently small but nonzero positive cosmological constant ? ) , and other questions of a cosmological nature . obviously , aspiring to address these questions is an ambitious undertaking , but we believe that attaining a better understanding of string field theory is a useful step in this direction . more concretely , in these lectures we will describe recent progress on open string field theory . it may be useful here to recall some basic aspects of open and closed strings and the relationship between them . closed strings , which are topologically equivalent to a circle @xmath4 , give rise upon quantization to a massless set of spacetime fields associated with the graviton @xmath9 , the dilaton @xmath10 , and the antisymmetric two - form @xmath11 , as well as an infinite family of massive fields . for the supersymmetric closed string , further massless fields associated with the graviton supermultiplet appear these are the ramond - ramond @xmath12-form fields @xmath13 and the gravitini @xmath14 . thus , the quantum theory of closed strings is naturally associated with a theory of gravity in space - time . on the other hand , open strings , which are topologically equivalent to an interval @xmath15 $ ] , give rise under quantization to a massless gauge field @xmath16 in space - time . the supersymmetric open string also has a massless gaugino field @xmath17 . it is now understood that generally open strings should be thought of as ending on a dirichlet @xmath12-brane ( d@xmath12-brane ) , and that the massless open string fields describe the fluctuations of the d - brane and the gauge field living on the world - volume of the d - brane . it may seem , therefore , that open and closed strings are quite distinct , and describe disjoint aspects of the physics in a fixed background space - time containing some family of d - branes . at tree level , the closed strings indeed describe gravitational physics in the bulk space - time , while the open strings describe the d - brane dynamics . at the quantum level , however , the physics of open and closed strings are deeply connected . indeed , historically open strings were discovered first through the form of their scattering amplitudes @xcite . looking at one - loop processes for open strings led to the first discovery of closed strings , which appeared as _ poles _ in nonplanar one - loop open string diagrams @xcite . the fact that open string diagrams naturally contain closed string intermediate states indicates that in some sense all closed string interactions are implicitly defined through the complete set of open string diagrams . this connection underlies many of the important recent developments in string theory . in particular , the m(atrix ) theory and ads / cft correspondences between gauge theories and quantum gravity are essentially limits in which closed string physics in a fixed space - time background is captured by a simple limiting yang - mills description of an open string theory on a family of branes ( d0-branes for m(atrix ) theory , d3-branes for the cft describing ads@xmath18 , etc . ) the fact that , in certain fixed space - time backgrounds , quantum gravity theories can be encoded in terms of open string degrees of freedom through the m(atrix ) and ads / cft correspondences leads to the question of how a global change of the space - time background would appear in the quantum field theory describing the appropriate limit of the open string model in question . if such a change of background could be described in the context of m(atrix ) theory or ads / cft , it would indicate that these models could be generalized to a background - independent framework . unfortunately , however , such a change in the background involves adding nonrenormalizable interactions to the field theories in question . at this point in time we do not have the technology to understand generically how a sensible quantum field theory can be described when an infinite number of nonrenormalizable interaction terms are added to the lagrangian . one example of a special case where this can be done is the addition of a constant background @xmath19 field in space - time . in the associated yang - mills theory , such as that on a system of @xmath20 d3-branes in the case of the simplest ads / cft correspondence , this change in the background field corresponds to replacing products of open string fields with a noncommutative star - product . the resulting theory is a noncommutative yang - mills theory . such noncommutative theories are the only well - understood example of a situation where adding an infinite number of apparently nonrenormalizable terms to a field theory action leads to a sensible modification of quantum field theory ( for a review of noncommutative field theory and its connection to string theory , see @xcite ) . string field theory is a nonperturbative formulation in target space of an interacting string theory , in which the infinite family of fields associated with string excitations are described by a space - time field theory action . for open strings , this field theory is a natural extension of the low - energy yang - mills action describing a system of d - branes , where the entire hierarchy of massive string fields is included in addition to the massless gauge field on the d - brane . integrating out all the massive fields from the string field theory action gives rise to a nonabelian born - infeld action for the d - branes , including an infinite set of higher - order terms arising from string theory corrections to the simple yang - mills action . like the case of noncommutative field theory discussed above , the new terms appearing in this action are apparently nonrenormalizable , but the combination of terms must work together to form a sensible theory . in the 1980 s , a great deal of work was done on formulating string field theory for open and closed , bosonic and supersymmetric string theories . most of these string field theories are quite complicated . for the open bosonic string , however , witten @xcite constructed an extremely elegant string field theory based on the chern - simons action . this cubic bosonic open string field theory ( osft ) is the primary focus of the work described in these lectures . although witten s osft can be described in a simple abstract language , practical computations with this theory rapidly become extremely complicated . despite a substantial amount of work on this theory , little insight was gained in the 1980 s regarding how this theory could be used to go beyond standard perturbative string methods . work on this subject stalled out in the late 80 s , and little further attention was paid to osft until several years ago . one simple feature of the 26-dimensional bosonic string has been problematic since the early days of string theory : both the open and closed bosonic strings have tachyons in their spectra , indicating that the usual perturbative vacua used for these theories are unstable . in 1999 , ashoke sen had a remarkable insight into the nature of the open bosonic string tachyon @xcite . he observed that the open bosonic string should be thought of as ending on a space - filling d25-brane . he pointed out that this d - brane is unstable in the bosonic theory , as it does not carry any conserved charge , and he suggested that the open bosonic string tachyon should be interpreted as the instability mode of the d25-brane . this led him to conjecture that witten s open string field theory could be used to precisely determine a new vacuum for the open string , namely one in which the d25-brane is annihilated through condensation of the tachyonic unstable mode . sen made several precise conjectures regarding the details of the string field theory description of this new open string vacuum . as we describe in these lectures , there is now overwhelming evidence that sen s picture is correct , demonstrating that string field theory accurately describes the nonperturbative physics of d - branes . this new nonperturbative application of string field theory has sparked a new wave of work on witten s cubic open string field theory , revealing many remarkable new structures . in particular , string field theory now provides a concrete framework in which disconnected string backgrounds can emerge from the equations of motion of a single underlying theory . although so far this can only be shown explicitly in the open string context , this work paves the way for a deeper understanding of background - independence in quantum theories of gravity . in this section we briefly review some basic features of d - branes . the concepts developed here will be useful in describing tachyonic d - brane configurations in the following section . for more detailed reviews of d - branes , see @xcite . d - branes can be understood in two ways : _ a _ ) as extended extremal black brane solutions of supergravity carrying conserved charges , and _ b _ ) as hypersurfaces on which strings have dirichlet boundary conditions . _ a _ ) the ten - dimensional type iia and iib supergravity theories each have a set of @xmath21-form fields @xmath22 in the supergraviton multiplet , with @xmath12 even / odd for type iia / iib supergravity . these are the ramond - ramond fields in the massless superstring spectrum . for each of these @xmath21-form fields , there is a solution of the supergravity field equations which has @xmath21-dimensional lorentz invariance , and which has the form of an extremal black hole solution in the orthogonal @xmath23 space directions plus time ( for a review see @xcite ) . these `` black @xmath12-brane '' solutions carry charge under the r - r fields @xmath24 , and are bps states in the supergravity theory , preserving half the supersymmetries of the theory . _ b _ ) in type iia and iib string theory , it is possible to consider open strings with dirichlet boundary conditions on some number @xmath23 of the spatial coordinates @xmath25 . the locus of points defined by such dirichlet boundary conditions defines a @xmath21-dimensional hypersurface @xmath26 in the ten - dimensional spacetime . when @xmath12 is even / odd in type iia / iib string theory , the spectrum of the resulting quantum open string theory contains a massless set of fields @xmath27 and @xmath28 . these fields can be associated with a gauge field living on the hypersurface @xmath26 , and a set of degrees of freedom describing the transverse fluctuations of this hypersurface in spacetime . thus , the quantum fluctuations of the open string describe a fluctuating @xmath21-dimensional hypersurface in spacetime a dirichlet - brane , or `` d - brane '' . the remarkable insight of polchinski in 1995 @xcite was the observation that dirichlet - branes carry ramond - ramond charges , and therefore should be described in the low - energy supergravity limit of string theory by precisely the black @xmath12-branes discussed in _ a _ ) . this connection between the string and supergravity descriptions of these nonperturbative objects paved the way to a dramatic series of new developments in string theory , including connections between string theory and supersymmetric gauge theories , string constructions of black holes , and new approaches to string phenomenology . in this subsection we briefly review the low - energy super yang - mills description of the dynamics of one or more d - branes . as discussed in the previous subsection , the massless open string modes on a d@xmath12-brane in type iia or iib superstring theory describe a @xmath29-component gauge field @xmath30 , @xmath23 transverse scalar fields @xmath31 , and a set of massless fermionic gaugino fields . the scalar fields @xmath31 describe small fluctuations of the d - brane around a flat hypersurface . if the d - brane geometry is sufficiently far from flat , it is useful to describe the d - brane configuration by a general embedding @xmath32 , where @xmath33 are @xmath34 coordinates on the d@xmath12-brane world - volume @xmath35 , and @xmath36 are ten functions giving a map from @xmath35 into the space - time manifold @xmath37 . just as the einstein equations governing the geometry of spacetime arise from the condition that the one - loop contribution to the closed string beta function vanish , a set of equations of motion for a general d@xmath12-brane geometry and associated world - volume gauge field can be derived from a calculation of the one - loop open string beta function @xcite . these equations of motion arise from the classical born - infeld action @xmath38 where @xmath39 , @xmath19 and @xmath10 are the pullbacks of the 10d metric , antisymmetric tensor and dilaton to the d - brane world - volume , while @xmath40 is the field strength of the world - volume @xmath41 gauge field @xmath42 . @xmath43 represents a set of chern - simons terms which will be discussed in the following subsection . this action can be verified by a perturbative string calculation @xcite , which also gives a precise expression for the brane tension @xmath44 where @xmath45 is the string coupling , equal to the exponential of the dilaton expectation value . a particular limit of the born - infeld action ( [ eq : dbi ] ) is useful for describing many low - energy aspects of d - brane dynamics . take the background space - time @xmath46 to be flat , and all other supergravity fields ( @xmath47 ) to vanish . we then assume that the d - brane is approximately flat , and is close to the hypersurface @xmath48 , so that we may make the static gauge choice @xmath49 . we furthermore assume that @xmath50 and @xmath51 are small and of the same order . in this limit , the action ( [ eq : dbi ] ) can be expanded as @xmath52 where @xmath53 is the @xmath12-brane world - volume and the coupling @xmath54 is given by @xmath55 including fermionic terms , the second term in ( [ eq : action - expansion ] ) is simply the dimensional reduction to @xmath21 dimensions of the 10d @xmath56 super yang - mills action @xmath57 where for @xmath58 , @xmath59 is the world - volume u(1 ) field strength , and for @xmath60 , @xmath61 ( setting @xmath62 ) . when multiple d@xmath12-branes are present , the d - brane action is modified in a fairly simple fashion @xcite . consider a system of @xmath20 coincident d - branes . for every pair of branes @xmath63 there is a set of massless fields @xmath64 associated with strings stretching from the @xmath65th brane to the @xmath66th brane ; the indices @xmath67 are known as chan - paton indices . treating the fields ( [ eq : nonabelian - fields ] ) as matrices , the analogue for multiple branes of the born - infeld action ( [ eq : dbi ] ) takes the form @xmath68 this action is known as the nonabelian born - infeld action ( nbi ) . in order to give a rigorous definition to the nonabelian born - infeld action , it is necessary to resolve ordering ambiguities in the expression ( [ eq : nbi ] ) . since the spacetime coordinates @xmath31 associated with the d - brane positions in space - time become themselves matrix - valued , even evaluating the pullbacks @xmath69 involves resolving ordering issues . much work has been done recently to resolve these ordering ambiguities ( see @xcite for some recent papers in this direction which contain further references to the literature ) , but there is still no consistent definition of the nonabelian born - infeld theory ( [ eq : nbi ] ) which is valid to all orders . the nonabelian born - infeld action ( [ eq : nbi ] ) becomes much simpler in the low - energy limit when the background space - time is flat . in the same limit discussed above for the single d - brane , where we find a low - energy limit giving the u(1 ) super yang - mills theory in @xmath34 dimensions , the inclusion of multiple d - branes simply leads in the low - energy limit to the nonabelian u(n ) super yang - mills action in @xmath70 dimensions . this action is the dimensional reduction of the 10d u(n ) super yang - mills action ( analogous to ( [ eq : sym1 ] ) , but with an overall trace ) to @xmath34 dimensions . in this reduction , as before , for @xmath58 , @xmath59 is the world - volume u(1 ) field strength , and for @xmath60 , @xmath61 , where now @xmath71 and @xmath59 are @xmath72 matrices . we furthermore have , for @xmath73 , @xmath74 $ ] in the dimensional reduction . the low - energy description of a system of @xmath20 coincident flat d - branes is thus given by @xmath75 super yang - mills theory in the appropriate dimension . this connection between d - brane actions in string theory and super yang - mills theory has led to many new developments , including new insights into supersymmetric field theories , the m(atrix ) theory and ads / cft correspondences , and brane world scenarios . in this subsection we describe a remarkable feature of d - brane systems , namely a mechanism by which one or more d - branes of a fixed dimension can be used to construct additional d - branes of higher or lower dimension . in our discussion of the d - brane action ( [ eq : dbi ] ) above , we mentioned a group of terms @xmath43 which we did not describe explicitly . for a single d@xmath12-brane , these chern - simons terms can be combined into a single expression of the form @xmath76 where @xmath77 represents a formal sum over all the ramond - ramond fields @xmath78 of various dimensions . in this integral , for each term @xmath78 , the nonvanishing contribution to ( [ eq : chern - simons ] ) is given by expanding the exponential of @xmath79 to order @xmath80 , where the dimension of the resulting form saturates the dimension of the brane . for example , on a d@xmath12-brane , there is a coupling of the form @xmath81 this coupling implies that the u(1 ) field strength on the d@xmath12-brane couples to the r - r field associated with @xmath82-branes . thus , we can associate magnetic fields on a d@xmath12-brane with dissolved @xmath82-branes living on the d@xmath12-brane . this result generalizes to a system of multiple d@xmath12-branes by simply performing a trace on the rhs of ( [ eq : chern - simons ] ) for example , on @xmath20 compact d@xmath12-branes , the charge @xmath83 which is the first chern class of the u(n ) bundle described by the gauge field on the @xmath20 branes , is quantized and measures the number of units of d@xmath82-brane charge living on the d@xmath12-branes , which are encoded in the field strength @xmath59 . similarly , @xmath84 encodes d@xmath85-brane charge on the d@xmath12-branes . just as lower - dimensional branes can be described in terms of the degrees of freedom associated with a system of @xmath20 d@xmath12-branes through the field strength @xmath59 , higher - dimensional branes can be described by a system of @xmath20 d@xmath12-branes in terms of the commutators of the matrix - valued scalar fields @xmath31 . just as @xmath86 measures @xmath82-brane charge , the matrix @xmath87 \label{eq : up - charge}\ ] ] measures @xmath88-brane charge @xcite . the charge ( [ eq : up - charge ] ) should be interpreted as a form of local charge density . the fact that the trace of ( [ eq : up - charge ] ) vanishes for finite sized matrices corresponds to the fact that the net d@xmath12-brane charge of a finite - size brane configuration in flat spacetime vanishes . a simple example of the mechanism by which a system of multiple d@xmath12-branes form a higher - dimensional brane is given by the matrix sphere . if we take a system of d0-branes with scalar matrices @xmath31 given by @xmath89 where @xmath90 are the generators of su(2 ) in the @xmath20-dimensional representation , then we have a configuration corresponding to the `` matrix sphere '' . this is a d2-brane of spherical geometry living on the locus of points satisfying @xmath91 . the `` local '' d2-brane charge of this brane is given by ( [ eq : up - charge ] ) . the d2-brane configuration given by ( [ eq : matrix - sphere ] ) is rotationally invariant ( up to a gauge transformation ) . the restriction of the brane to the desired locus of points can be seen from the relation @xmath92 . we conclude our discussion of d - branes with a brief description of t - duality . t - duality is a perturbative symmetry which relates the type iia and type iib string theories . this duality symmetry was in fact crucial in the original discovery of d - branes @xcite . a more detailed discussion of t - duality can be found in the textbook by polchinski @xcite . using t - duality , we construct an explicit example of a brane within a brane encoded in super yang - mills theory , illustrating the ideas of the previous subsection . this example will be used in the following section to construct an analogous configuration with a tachyon . consider type iia string theory on a spacetime of the form @xmath93 where @xmath94 is a generic 9-manifold of lorentz signature , and @xmath4 is a circle of radius @xmath95 . t - duality is the statement that this theory is precisely equivalent , at the perturbative level , to type iib string theory on the spacetime @xmath96 , where @xmath97 is a circle of radius @xmath98 . t - duality is most easily understood in terms of closed strings , where it amounts to an exchange of winding and momentum modes of the string . the string winding modes on @xmath4 have energy @xmath99 , where @xmath100 is the winding number . the t - dual momentum modes on @xmath97 have @xmath101 ; it is straightforward to check that the spectrum of closed string states is unchanged under t - duality . t - duality can also be understood in terms of open strings . under t - duality , an open string with neumann boundary conditions on @xmath4 is mapped to an open string with dirichlet boundary conditions on @xmath97 , and vice versa . thus , a dirichlet @xmath12-brane which is wrapped around the circle @xmath4 is mapped under t - duality to a dirichlet @xmath102-brane of one lower dimension which is localized to a point on the circle @xmath97 . at the level of the low - energy theory on the d - brane , the @xmath29-dimensional yang - mills theory on the @xmath12-brane is replaced under t - duality with the @xmath12-dimensional yang - mills theory on the dual @xmath102-brane . mathematically , the covariant derivative operator in the direction @xmath4 is replaced under t - duality with an adjoint scalar field @xmath31 . formally , this adjoint scalar field is an infinite size matrix , containing information about the open strings wrapped an arbitrary number of times around the compact direction @xmath97 . we can summarize the relevant mappings under t - duality in the following table [ cols=">,^ , > " , ] these numerical calculations indicate that level truncation of string field theory leads to a good systematic approximation scheme for computing the nonperturbative tachyon vacuum . up to level 12 , at level ( 14 , 42 ) the energy becomes @xmath103 , and that the energy continues to decrease , reaching @xmath104 at level ( 18 , 54 ) . in @xcite , it was shown that this calculation could be theoretically extrapolated to higher levels using the result found in @xcite that perturbative amplitudes converge in level truncation with errors described by a power series in @xmath105 . this extrapolation suggests that the energy turns around again near @xmath106 , and then increases again , asymptotically approaching @xmath107 as @xmath108 . further analysis supporting this conclusion was given in @xcite , where the effective tachyon potential was extrapolated to higher order using results calculated up to level 18 . ] it is interesting to consider the tachyon condensation problem from the point of view of the effective tachyon potential . if instead of trying to solve the quadratic equations for all @xmath20 of the fields appearing in ( [ eq : scalar - potential ] ) , we instead fix the tachyon field @xmath109 and solve the quadratic equations for the remaining @xmath110 fields , we can determine a effective potential @xmath111 for the tachyon field . this was done numerically up to level @xmath112 in @xcite@xmath113 . at each level , the tachyon effective potential smoothly interpolates between the perturbative vacuum and the nonperturbative vacuum near @xmath114 . for example , the tachyon effective potential at level ( 2 , 6 ) is graphed in figure [ f : potential ] . in all level truncations other than ( 0 , 0 ) and ( 2 , 4 ) , the tachyon effective potential has two branch point singularities at which the continuous solution for the other fields breaks down ; for the level ( 2 , 6 ) truncation , these branch points occur at @xmath115 and @xmath116 ; the lower branch point is denoted by a solid circle in figure [ f : potential ] . as a result of these branch points , the tachyon effective potential is only valid for a finite range of @xmath109 , ranging between approximately @xmath117 and @xmath118 . in @xcite it was demonstrated numerically that the branch points in the tachyon effective potential arise because the trajectory in field space associated with this potential encounters the boundary of the region of feynman - siegel gauge validity . as mentioned earlier , feynman - siegel gauge is only valid in a finite - size region around the perturbative vacuum . it seems almost to be a fortunate accident that the nonperturbative vacuum lies within the region of validity of this gauge choice . it is also worth mentioning here that in the `` background - independent '' formulation of sft , the tachyon potential can be computed exactly @xcite . in this formulation , there is no branch point in the effective potential , which is unbounded below for negative values of the tachyon . on the other hand , the nontrivial vacuum in the background - independent approach arises only as the tachyon field goes to infinity , so it is harder to study the physics of the stable vacuum from this point of view . another interesting perspective on the tachyon effective potential is found by performing a perturbative computation of the coefficients in this effective potential in the level - truncated theory . this gives a power series expansion of the effective tachyon potential @xmath119 in @xcite , the coefficients up to @xmath120 were computed in the level truncations up to ( 10 , 20 ) . because of the branch point singularity near @xmath121 , this series has a radius of convergence much smaller than the value of @xmath109 at the nonperturbative vacuum . thus , the energy at the stable vacuum lies outside the naive range of perturbation theory . thus , the stable vacuum is in some sense accessible from purely perturbative calculations . ] . we have seen that numerical results from level - truncated string field theory strongly suggest the existence of a classically stable vacuum solution @xmath122 to the string field theory equation of motion ( [ eq : sft - eom ] ) . the state @xmath122 , while still unknown analytically , has been determined numerically to a high degree of precision . this state seems like a very well - behaved string field configuration . while there is no positive - definite inner product on the string field fock space , the state @xmath122 certainly has finite norm under the natural inner product @xmath123 , and is even better behaved under the product @xmath124 . thus , it is natural to assume that @xmath122 defines a classically stable vacuum for the theory , around which we can expand the action to find a new `` vacuum string field theory '' . expanding @xmath125 we get the action @xmath126 where @xmath127 this string field theory around the stable vacuum has precisely the same form as witten s original cubic string field theory , only with a different brst operator @xmath128 , which so far is only determined numerically . note that this formulation of vacuum string field theory is distinct from the vsft model of rastelli , sen , and zwiebach ( rsz ) @xcite . these authors make an ansatz that the brst operator takes a pure ghost form , along the lines of @xmath129 , and they conjecture that the theory with such a brst operator is equivalent to the vsft model given by the brst operator ( [ eq : vsft - q ] ) . we discuss the rsz model again briefly in the next section . sen s third conjecture states that there should be no open string excitations of the theory around @xmath130 . this implies that there should be no solutions of the linearized equation @xmath131 in the vsft ( [ eq : vsft - action ] ) other than pure gauge states of the form @xmath132 . in this subsection we discuss evidence for this conjecture . it may seem surprising to imagine that _ all _ the perturbative open string degrees of freedom will vanish at a particular point in field space , since these are all the degrees of freedom available in the theory . this is not a familiar phenomenon from quantum field theory . to understand how the open strings can decouple , it may be helpful to begin by considering the simple example of the ( 0 , 0 ) level - truncated theory . in this theory , the quadratic terms in the action become @xmath133 \phi ( p ) \ , . \ ] ] taking @xmath134 , we find that the quadratic term is a transcendental expression which does not vanish for any real value of @xmath135 . thus , this theory has no poles , and the tachyon has decoupled from the theory . of course , this is not the full story , as there are still finite complex poles . it does , however suggest a mechanism by which the nonlocal parts of the action ( encoded in the exponential of @xmath135 ) can remove physical poles . to get the full story , it is necessary to continue the analysis to higher level . at level 2 , there are 7 scalar fields , the tachyon and the 6 fields associated with the fock space states @xmath136 note that in this analysis we can not fix feynman - siegel gauge , as we only believe that this gauge is valid for the zero - modes of the scalar fields in the vacuum @xmath122 . an attempt at analyzing the spectrum of the theory in feynman - siegel gauge using level truncation was made in @xcite , with no sensible results . diagonalizing the quadratic term in the action on the full set of 7 fields of level @xmath137 , we find that poles develop at @xmath138 and @xmath139 ( in string units , where the tachyon has @xmath140 ) @xcite . these poles correspond to states satisfying @xmath141 . the question now is , are these states physical ? if they are exact states , of the form @xmath132 , then they are simply gauge degrees of freedom . if not , however , then they are states in the cohomology of @xmath128 and should be associated with physical degrees of freedom . unfortunately , we can not precisely determine whether the poles we find in level truncation are due to exact states , as the level - truncation procedure breaks the condition @xmath142 . thus , we can only measure _ approximately _ whether a state is exact . a detailed analysis of this question was carried out in @xcite . in this paper , all terms in the sft action of the form @xmath143 were determined , where @xmath144 is a scalar zero - mode , and @xmath145 are nonzero - momentum scalars . in addition , all gauge transformations involving at least one zero - momentum field were computed up to level ( 6 , 12 ) . at each level up to @xmath146 , the ghost number 1 states in the kernel @xmath147 were computed . the extent to which each of these states lies in the exact subspace was measured using the formula @xmath148 where @xmath149 are an orthonormal basis for @xmath150 , the image of @xmath128 acting on the space of ghost number 0 states in the appropriate level truncation . ( note that this measure involves a choice of inner product on the fock space ; several natural inner products were tried , giving roughly equivalent results ) . the result of this analysis was that up to the mass scale of the level truncation , @xmath151 , all the states in the kernel of @xmath152 were @xmath153 within the exact subspace , for @xmath154 . this result seems to give very strong evidence for sen s third conjecture that there are no perturbative open string excitations around the stable classical vacuum @xmath122 . this analysis was only carried out for even level scalar fields ; it would be nice to check that a similar result holds for odd - level fields and for tensor fields of arbitrary rank . another more abstract argument that there are no open string states in the stable vacuum was given by ellwood , feng , he and moeller @xcite . these authors argued that in the stable vacuum , the identity state @xmath155 in the sft star algebra , which satisfies @xmath156 for a very general class of string fields @xmath157 , seems to be an exact state , @xmath158 if indeed the identity is exact , then it follows immediately that the cohomology of @xmath128 is empty , since @xmath159 then implies that @xmath160 so to prove that the cohomology of @xmath128 is trivial , it suffices to show that @xmath161 . while there are some subtleties involved with the identity string field , ellwood _ et al . _ found a very elegant expression for this field , @xmath162 ( recall that @xmath163 . ) they then looked numerically for a state @xmath164 satisfying ( [ eq : i - exact ] ) . for example , truncating at level @xmath165 , @xmath166 while the only candidate for @xmath167 is @xmath168 for some constant @xmath169 . the authors of @xcite showed that the state ( [ eq : identity-2 ] ) is best approximated as exact when @xmath170 ; for this value , their measure of exactness becomes @xmath171 which the authors interpreted as a @xmath172 deviation from exactness . generalizing this analysis to higher levels , they found at levels 5 , 7 , and 9 , a deviation from exactness of @xmath173 and @xmath174 respectively . at level 9 , for example , the identity field has 118 components , and there are only 43 gauge parameters , so this is a highly nontrivial check on the exactness of the identity . like the results of @xcite , these results strongly support the conclusion that the cohomology of the theory is trivial in the stable vacuum . in this case , the result applies to fields of all spins and all ghost numbers . given that the witten string field theory seems to have a classical solution with no perturbative open string excitations , in accordance with sen s conjectures , it is quite interesting to ask what the physics of the vacuum string field theory ( [ eq : vsft - action ] ) should describe . one natural assumption might be that this theory should include closed string states in its quantum spectrum . unfortunately , addressing this question requires performing calculations in the quantum theory around the stable vacuum . such calculations are quite difficult ( although progress in this direction has been made in the @xmath12-adic version of the theory @xcite ) . even in the perturbative vacuum , it is difficult to systematically study closed strings in the quantum string from theory . we discuss the question again briefly in the final section . in this section we review briefly some further developments which we do not have time to explore in great detail in these lectures . in subsection 7.1 we discuss the pure ghost brst operator ansatz of rsz ( rastelli , sen , and zwiebach ) for vacuum string field theory . in subsection 7.2 we discuss `` sliver '' states and related states ; these states are projectors in the sft star algebra , and are closely related to d - branes in the rsz vsft model . these topics will be discussed in further detail in @xcite in @xcite , rastelli , sen , and zwiebach made an intriguing proposal regarding the form of witten s string field theory around the stable tachyon vacuum . since the exact form of the brst operator @xmath128 given by ( [ eq : vsft - q ] ) is not known analytically , and is difficult to work with numerically , these authors suggested that it might be possible to `` guess '' an appropriate form for this operator ( after suitable field redefinition ) , using the properties expected of the brst operator in any vacuum . they suggested a simple class of brst operators @xmath175 which satisfy the properties ( a - c ) described in section 4.2 ( actually , they impose the slightly weaker but still sufficient condition @xmath176 instead of condition ( b ) ) . in particular , they propose that after a field redefinition , the brst operator of the string field theory in the classically stable vacuum should be an operator @xmath175 expressable purely in terms of ghost operators . for example , the simplest operator in the class they suggest is @xmath177 , which clearly satisfies @xmath178 , and which also satisfies condition ( c ) and the weaker form of condition ( b ) mentioned above . the rsz model of vacuum string field theory has a number of attractive features . * this model satisfies all the axioms of string field theory , and has a brst operator with vanishing cohomology . * in the rsz model , the equation of motion factorizes into the usual equation of motion @xmath179 for the ghost part of the field , and a projection equation @xmath180 for the matter part of the field , where the full string field is given by @xmath181 thus , finding a solution of the equation of motion reduces to the problem of solving the equation of motion in the ghost sector and identifying projection operators in the string field star algebra . it was also recently shown @xcite that by taking the brst operator @xmath175 to be given by a ghost insertion localized at the string midpoint , the ghost equation also has essentially the form of the projection equation . thus , this seems to be a very natural choice for the brst operator of the rsz model . * a number of projection operators have been identified in the string field star algebra . these projection operators have many of the properties desired of d - branes . we will briefly review some aspects of these projection operators in the next subsection . * given the projection operators just mentioned , the ratio of tensions between d - branes of different dimensionality can be computed and has the correct value @xcite . despite the successes of the rsz model , there are some difficult technical aspects of this picture . first , it seems very difficult to actually prove that this model is related to the vsft around the stable vacuum in the witten model , not least because we lack any analytic control over the witten theory . second , the rsz model seems to have a somewhat singular structure in several respects . formally , the action on any well - behaved fock space state satisfying the equation of motion will vanish @xcite . further , the natural solutions of the projection equation corresponding to the matter sector of the equation of motion have rather singular properties @xcite . some of these singular properties are related to the fact that some of the physics in the rsz model seems to have been `` pushed '' into the midpoint of the string . in the witten model , the condition that , for example , @xmath182 involves a fairly subtle anomaly cancellation between the matter and ghost sectors at the midpoint . in the rsz model , the matter and ghost sectors are essentially decoupled , so that the theory seems to have separate singularities in each sector , which cancel when the sectors are combined . these are all indications of a theory with problematic singularities . while the witten theory seems to be free of singularities of this type , it remains to be seen whether resolving the singularities of the rsz model or finding an analytic approach to the witten theory will be a more difficult problem to solve . from the point of view of the rsz model of vsft just discussed , projection operators in the matter sector of the star algebra play a crucial role in constructing solutions of the equations of motion . such projection operators may also be useful in understanding solutions in the original witten theory . quite a bit of work has been done on constructing and analyzing projectors in the star algebra since the rsz model was originally proposed . without going into the technical details , we now briefly review some of the important features of matter projectors . the first matter projector which was explicitly constructed is the `` sliver '' state . this state was identified in conformal field theory in @xcite , and then constructed explicitly using matter oscillators in @xcite . the sliver state takes the form of a squeezed state @xmath183| 0 \rangle \ , . \label{eq : sliver - squeezed}\ ] ] by requiring that such a state satisfy the projection equation @xmath184 , and by making some further assumptions about the nature of the state , an explicit formula for the matrix @xmath185 was found in @xcite . projectors like the sliver have many properties which are reminiscent of d - branes . this relationship between projection operators and d - branes is familiar from noncommutative field theory , where projectors also play the role of d - brane solitons @xcite ( for a review of noncommutative field theory , see @xcite ) . in the rsz model , by tensoring an arbitrary matter projector with a fixed ghost state satisfying the ghost equation of motion ( [ eq : eom - ghosts ] ) , states corresponding to an arbitrary configuration of d - branes can be constructed . particular projectors like the sliver can be constructed which are localized in any number of space - time dimensions , corresponding to the codimension of a d - brane . under gauge transformations , a rank 1 projector can be rotated into an orthogonal rank 1 projector , so that configurations containing multiple branes can be constructed as higher rank projectors formed from the sum of orthogonal rank one projectors @xcite . this gives a very suggestive picture of how arbitrary d - brane configurations can be constructed in string field theory . while this picture is quite compelling , however , there are a number of technical obstacles which make this still a somewhat incomplete story . as mentioned above , in the rsz model , many singularities appear due to the separation of the matter and ghost sectors . in the context of the matter projectors , these singularities manifest as singular properties of the projectors . for example , the sliver state described above has a matrix @xmath185 which has eigenvalues of @xmath186 for any dimension of d - brane @xcite . such eigenvalues cause the state to be nonnormalizable elements of the matter fock space . in the dirichlet directions , this lack of normalizability occurs because the state is essentially localized to a point and is analogous to a delta function . in the neumann directions , the singularity manifests as a `` breaking '' of the strings composing the d - brane , so that the functional describing the projector state is a product of a function of the string configurations on the left and right halves of the string , with no connection mediated through the midpoint . these geometric singularities seem to be generic features of matter projectors , not just of the sliver state @xcite . these singular geometric features are one of the things which makes direct calculation in the rsz model somewhat complicated , as all these singularities must be sensibly regulated . these singularities do not seem to appear in the witten theory , where the brst operator and numerically calculated solutions seem to behave smoothly at the string midpoint . on the other hand , it may be that further study of the matter projection operators and their cousins in the ghost sector which satisfy ( [ eq : eom - ghosts ] ) will lead to analytic progress on the witten theory . the work described in these lectures has brought our understanding of string field theory to a new level . we now have fairly conclusive evidence that open string field theory can successfully describe distinct vacua with very different geometrical properties , which are not related to one another through a marginal deformation . the resulting picture , in which a complicated set of degrees of freedom defined primarily through an algebraic structure , can produce different geometrical backgrounds at different solutions of the equations of motion , represents an important step beyond perturbative string theory . such an approach , where different backgrounds with different low - energy degrees of freedom arise from a single underlying formalism , is clearly necessary to discuss questions of a cosmological nature in string theory . it is clearly essential , however , to generalize from the work described here in which the theory describes distinct _ open _ string backgrounds , to a formalism where different _ closed _ string backgrounds appear as solutions to an equation of motion for a single set of degrees of freedom . clearly , it is an important goal to have a formulation of string / m - theory in which all the currently understood vacua can arise in terms of a single well - defined set of degrees of freedom . it is not yet clear , however , how far it is possible go towards this goal using the current formulations of string field theory . it may be that the correct lesson to take from the work described here is simply that there _ are _ nonperturbative formulations in which distinct vacua can be brought together as solutions of a single classical theory , and that one should search for some deeper fundamental algebraic formulation where geometry , and even the dimension of space - time emerge from the fundamental degrees of freedom in the same way that d - brane geometry emerges from the degrees of freedom of witten s open string field theory . a more conservative scenario , however , might be that we could perhaps use the current framework of string field theory , or some limited refinement thereof , to achieve this goal of providing a universal nonperturbative definition of string theory and m - theory . following this latter scenario , we propose here a series of questions aimed at continuing the recent developments in open string field theory as far as possible towards this ultimate goal . it is not certain that this research program can be carried to its conclusion , but it will be very interesting to see how far open string field theory can go in reproducing important nonperturbative aspects of string theory . some open problems : 1 . the first important unsolved problem in this area is to find an analytic description of the tachyonic vacuum . despite several years of work on this problem , great success with numerical approximations , and some insight from the rsz vacuum string field theory model , we still have no good analytic understanding of the difference between the d - brane vacuum and the empty vacuum in witten s open cubic string field theory . it seems almost unbelievable that there is not some elegant analytic solution to this problem . finding such an analytic solution would almost certainly greatly enhance our understanding of this theory , and would probably lead to other significant advances . 2 . another interesting and important unsolved problem is to find , either analytically or numerically , a solution of the witten theory describing _ two _ d25-branes . if open string field theory is truly a background - independent theory , at least in the open string sense , it should be just as feasible to go from a vacuum with one d - brane to a vacuum with two d - branes as it is to go from a vacuum with one d - brane to the empty vacuum ( or from the vacuum with two d - branes to the vacuum with one d - brane , which is essentially the same problem as going from one to none ) . despite some work on this problem @xcite , there is as yet no evidence that a double d - brane solution exists for the witten theory on a single d - brane . several approaches which have been tried ( and will be described in more detail in @xcite ) include : _ i _ ) following a positive mass field upward , looking for a stable point ; this method seems to fail because of gauge - fixing problems the effective potential often develops a singularity before reaching the energy @xmath187 , _ ii _ ) following the intuition of the rsz model and constructing a gauge transform of the original d - brane solution which is @xmath188orthogonal to the original d - brane vacuum . it can be shown formally that such a state , when added to the original d - brane vacuum gives a new solution with the correct energy for a double d - brane ; unfortunately , however , we have been unable to identify such a state numerically in level truncation . + there are several other problems closely related to the double d - brane problem . one related problem is the problem of studying a d0-brane lump solution from the tachyon field on a d1-brane wrapped on a small circle . when the circle is sufficiently small , the mass of the d0-brane is larger than that of the wrapped d1-brane . in this case , it seems much more difficult to construct the d0-brane lump solution than it is when the d0-brane has mass smaller than the d1-brane @xcite . another possibly related problem is the problem of translating a single d - brane of less than maximal dimension in a transverse direction . it was shown by sen and zwiebach @xcite ( in a t - dual picture ) that after moving a d - brane a finite distance of order of the string length in a transverse direction , the level - truncated string field theory equations develop a singularity . thus , in level truncation it does not seem possible to move a d - brane a macroscopic distance in a transverse direction . in this case , a toy model @xcite suggests that the problem is that the infinitesimal marginal parameter for the brane translation ceases to parameterize the marginal trajectory in field space after a finite distance , just as the coordinate @xmath189 ceases to parameterize the circle @xmath190 near @xmath191 . this is similar in spirit to the breakdown of feynman - siegel gauge along the tachyon potential discussed in section 6.1 . + to show that open string field theory is sufficiently general to address arbitrary questions involving different vacua , it is clearly necessary to show that the formalism is powerful enough to describe multiple brane vacua , the d0-brane lump on an arbitrary radius circle , and translated brane vacua . it is currently unclear whether the obstacles to finding these vacua are conceptual or technical . it may be that the level - truncation approach is not well - suited to finding these vacua . if this is true , however , we may need a clearer mathematical formalism for describing the theory . there is currently some ambiguity in the definition of the theory , in terms of precisely which states are allowed in the string field . level - truncation in some sense gives a regularization of , and a concrete definition to , the theory . without level truncation , we would need some more definitive mathematical tools for analyzing various features of the theory , such as the other vacua mentioned here . another open question involves the role that closed strings play in open string field theory . as has been known since the earliest days of the subject , closed strings appear as poles in perturbative open string scattering amplitudes . this was shown explicitly for witten s sft in @xcite , where it was shown that closed string poles arise in the one - loop 2-point function . if witten s theory is well - defined as a quantum theory , it would follow from unitarity that the closed string states should also arise in some natural sense as asymptotic states of the quantum open string field theory . it is currently rather unclear , however , whether , and if so how , this might be realized . there are subtleties in the quantum formulation of the theory which have never completely been resolved @xcite . both older sft literature @xcite and recent work @xcite have suggested ways in which closed strings might be incorporated into the open string field theory formalism , but a definitive resolution of this question is still not available . if it is indeed possible to encode closed string degrees of freedom in some way in the quantum open string field theory , it suggests that one could use the witten formalism in principle to not only compute general closed string scattering amplitudes , but perhaps even to address questions of closed string vacua . this is clearly an optimistic scenario , but one can imagine that the open string theory might really contain all of closed string physics as well as open string physics . this scenario is perhaps not so farfetched , as it really represents simply a lifting to the level of string field theory of the ads / cft story , where the massive as well as the massless modes are included . furthermore , the fact that , as discussed in section 5.5 , the open string diagrams precisely cover the moduli space of riemann surfaces with an arbitrary number of handles ( and at least one boundary ) , suggests that by shrinking the boundaries to closed strings , one might neatly describe all perturbative closed string amplitudes in the open string language . on the other hand , it seems quite possible that the closed string sector of the theory is encoded in a singular fashion ( like the encoding of the d - brane in the rsz vsft model ) , so that extracting the closed string physics from the open string field theory may involve such complicated manipulations that one is better off directly working with a closed string formalism . it would certainly be nice to have a clearer picture of how far one can go in this direction purely from the open string point of view . another obvious , but crucial , question is how this whole story can be generalized to superstrings . the naive witten cubic superstring field theory has technical problems arising from contact terms between picture - changing operators @xcite . it has been suggested that these problems can be resolved directly in the cubic theory @xcite . berkovits has also suggested a new non - polynomial string field theory framework which seems to deal successfully with the contact term problem , at least in the ns - ns sector @xcite . some preliminary work indicates that numerical calculations on the tachyon condensation problem for the open superstring can be carried out in the berkovits model with analogous results to those described here for the bosonic open string , although the results to date for the superstring are much more limited @xcite . it would be nice to have a more complete picture for the superstring , and some sense of how issues like the closed string question would work in the supersymmetric framework . perhaps the most important lesson we have learned from the body of work discussed in these lectures is that open string field theory is a consistent framework in which geometrically distinct open string backgrounds can arise as classical solutions of a single theory . a fundamental outstanding problem in string theory is to find a framework in which different closed string backgrounds arise in a similar fashion from some fixed set of degrees of freedom within a single well - defined theory . in principle , we would hope that all the different closed string backgrounds would arise as solutions of the equations of motion for the fundamental underlying degrees of freedom of string field theory , either by incorporating closed strings into the open string field theory framework as described above , or by working directly in some formulation of closed string field theory . it is quite challenging to imagine a single set of degrees of freedom which would encode , in different phases , all the possible string backgrounds we are familiar with . a particularly pressing case is that of m - theory . in principle , a nonperturbative background - independent formulation of type ii string theory should allow one to take the string coupling to infinity in such a way that the fundamental degrees of freedom of the theory are still actually at some finite point in their configuration space in the limit . this would lead to the vacuum associated with m - theory in flat space - time . it would be quite remarkable if this can be achieved in the framework of string field theory . given the nontrivial relationship between string fields and low - energy effective degrees of freedom , however , such a result can not be ruled out . if this picture could be successfully implemented , it would give a very satisfying understanding of how the complicated network of dualities of string and m - theory could be represented in terms of a single underlying set of degrees of freedom . i would like to thank cecs and the organizers of the school on quantum gravity for their support and hospitality , and for an extremely enjoyable summer school experience . thanks to erasmo coletti , ian ellwood , david gross , nicolas moeller , joe minahan , greg moore , leonardo rastelli , martin schnabl , ashoke sen , jessie shelton , ilya sigalov , and barton zwiebach , for many discussions and collaborations which provided the material for these lectures . thanks also to tasi 01 , where some of this material was presented prior to this school . particular thanks to barton zwiebach for suggestions and contributions to these lecture notes , which have substantial overlap with a more extensive set of lecture notes based on lectures by zwiebach and myself at tasi 01 , which will appear presently @xcite . this work was supported by the doe through contract # de - fc02 - 94er40818 . w. taylor and m. van raamsdonk , `` multiple d0-branes in weakly curved backgrounds , '' _ phys . lett . _ * b558 * ( 1999 ) , 63 , hep - th/9904095 ; `` multiple dp - branes in weak background fields , '' _ nucl . phys . _ * b573 * , 703 ( 2000 ) , hep - th/9910052 . n. berkovits , `` super - poincare invariant superstring field theory , '' _ nucl . phys . _ * b450 * ( 1995 ) , 90 , hep - th/9503099 ; `` review of open superstring field theory , '' hep - th/0105230 ; `` the ramond sector of open superstring field theory , '' jhep * 0111 * , 047 ( 2001 ) , hep - th/0109100 . s. b. giddings , e. j. martinec and e. witten , `` modular invariance in string field theory , '' phys . b * 176 * , 362 ( 1986 ) . b. zwiebach , `` a proof that witten s open string theory gives a single cover of moduli space , '' commun . phys . * 142 * , 193 ( 1991 ) . d. z. freedman , s. b. giddings , j. a. shapiro and c. b. thorn , `` the nonplanar one loop amplitude in witten s string field theory , '' nucl . b * 298 * , 253 ( 1988 ) . j. a. harvey and p. kraus , `` d - branes as unstable lumps in bosonic open string field theory , '' jhep * 0004 * , 012 ( 2000 ) , hep - th/0002117 ; a. a. gerasimov and s. l. shatashvili , `` on exact tachyon potential in open string field theory , '' jhep * 0010 * , 034 ( 2000 ) , hep - th/0009103 ; d. kutasov , m. marino and g. moore , `` some exact results on tachyon condensation in string field theory , '' jhep * 0010 * , 045 ( 2000 ) , hep - th/0009148 ; d. ghoshal and a. sen , normalisation of the background independent open string field jhep * 0011 * , 021 ( 2000 ) , hep - th/0009191 ; d. kutasov , m. marino and g. moore , d. kutasov , m. marino and g. w. moore , `` remarks on tachyon condensation in superstring field theory , '' , hep - th/0010108 ; s. moriyama and s. nakamura , descent relation of tachyon condensation from boundary string phys . b * 506 * , 161 ( 2001 ) , hep - th/0011002 ; a. a. gerasimov and s. l. shatashvili , `` stringy higgs mechanism and the fate of open strings , '' jhep * 0101 * , 019 ( 2001 ) , hep - th/0011009 ; i. y. arefeva , a. s. koshelev , d. m. belov and p. b. medvedev , `` tachyon condensation in cubic superstring field theory , '' nucl . b * 638 * , 3 ( 2002 ) , hep - th/0011117 ; p. kraus and f. larsen , `` boundary string field theory of the d d - bar system , '' phys . d * 63 * , 106004 ( 2001 ) , hep - th/0012198 ; m. alishahiha , h. ita and y. oz , `` on superconnections and the tachyon effective action , '' phys . lett . b * 503 * , 181 ( 2001 ) , hep - th/0012222 ; g. chalmers , `` open string decoupling and tachyon condensation , '' jhep * 0106 * , 012 ( 2001 ) , hep - th/0103056 ; m. marino , `` on the bv formulation of boundary superstring field theory , '' jhep * 0106 * , 059 ( 2001 ) , hep - th/0103089 ; v. niarchos and n. prezas , `` boundary superstring field theory , '' nucl . phys . b * 619 * , 51 ( 2001 ) , hep - th/0103102 ; k. s. viswanathan and y. yang , `` tachyon condensation and background independent superstring field theory , '' phys . d * 64 * , 106007 ( 2001 ) , hep - th/0104099 ; m. alishahiha , `` one - loop correction of the tachyon action in boundary superstring field theory , '' phys . b * 510 * , 285 ( 2001 ) , hep - th/0104164 . m. schnabl , `` wedge states in string field theory , '' jhep * 0301 * , 004 ( 2003 ) , hep - th/0201095 ; `` anomalous reparametrizations and butterfly states in string field theory , '' nucl . b * 649 * , 101 ( 2003 ) , ep - th/0202139 . j. kluson , `` marginal deformations in the open bosonic string field theory for n d0-branes , '' hep - th/0203089 ; `` exact solutions in open bosonic string field theory and marginal deformation in cft , '' hep - th/0209255 . j. a. shapiro and c. b. thorn , `` brst invariant transitions between open and closed strings , '' _ phys . _ * d36 * 432 ( 1987 ) ; `` closed string - open string transitions in witten s string field theory , '' phys . b * 194 * , 43 ( 1987 ) . p. de smet and j. raeymaekers , level four approximation to the tachyon potential in superstring jhep * 0005 * , 051 ( 2000 ) , hep - th/0003220 ; a. iqbal and a. naqvi , `` tachyon condensation on a non - bps d - brane , '' , hep - th/0004015 ;

rna is an important biopolymer critical to all living systems @xcite and may be the crucial entity in prebiotic evolution @xcite . like for dna , there are four types of nucleotides ( or bases ) a , c , g , and u which , when polymerized can form double helical structures consisting of stacks of stable watson - crick pairs ( a with u or g with c ) . however unlike a long polymer of dna , which is often accompanied by a complementary strand and forms otherwise featureless double helical structures , a polymer of rna usually `` operates '' in the single - strand mode . it bends onto itself and forms elaborate spatial structures in order for bases located on different parts of the backbone to pair with each other , similar conceptually to how the sequence of an amino acid encodes the structure of a protein . understanding the encoding of structure from the primary sequence has been an outstanding problem of theoretical biophysics . most theoretical work in the past decade have been focused on the problem of protein folding , which is very difficult analytically and numerically @xcite . here , we study the problem of rna folding , specifically the formation of rna _ secondary structures_. for rna , the restriction to secondary structures is meaningful due to a separation of energy scales . it is this restriction that makes the rna folding problem amenable to detailed analytical and numerical studies @xcite . there exist efficient algorithms to compute the exact partition function of rna secondary structures @xcite . together with the availability of carefully measured free energy parameters @xcite describing the formation of various microscopic structures ( e.g. , stacks , loops , hairpins , etc . ) , the probable secondary structures formed by any given rna molecule of up to a few thousand bases can be obtained readily . on the experimental side , rna molecules of @xmath1 @xmath2 bases in length are available . furthermore , the restriction to secondary structures can be physically enforced in a salt solution with monovalent ions , e.g. , @xmath3 , so that controlled experiments are in principle possible @xcite . in this work , we are not concerned with the structure formed by a specific sequence . instead , we will study the statistics of secondary structures formed by the ensemble of _ long _ _ random _ rna sequences ( of at least a few thousand bases in length in practice ) . such knowledge may be of value in detecting important structural components in messenger rnas which may otherwise be regarded as random from the structural perspective , in understanding how functional rnas arise from random rna sequences @xcite , or in characterizing the response of a long single - stranded dna molecule to external pulling forces @xcite . more significantly from the theoretical point of view , the rna secondary structure problem presents a rare tractable model of a random heteropolymer where concrete progress can be made regarding the thermodynamic properties @xcite . nevertheless , there are many gaps in our understanding . this paper is a detailed report of our on - going effort in this regard . it provides a self - contained introduction of the random rna problem to statistical physicists as a novel problem of disordered systems , and depicts several approaches we have tried to characterize this system . the manuscript is organized as follows : in sec . [ sec_secondarystructure ] , we provide a detailed introduction to the phenomenology of rna secondary structure formation . we review the key simplifications which form the basis of efficient computing as well as exact solutions in some cases . we also review the properties of the molten phase " , which is the simplest possible phase of the system assuming sequence disorder is not relevant . in sec . [ sec_randomnesshight ] , we consider the effect of sequence disorder at high temperatures . we show numerical evidence that the random rna sequence is in the molten phase at sufficiently high temperatures , and support this conclusion by solving the two - replica system which can be regarded as a perturbative study on the stability of the molten phase . in sec . [ sec_randomnesslowt ] , we provide a scaling argument , and show why the molten phase should break down at low enough temperatures . this is followed by a detailed numerical study of the low temperature regime . we apply the droplet picture and characterize the statistics of large - scale , low - energy excitations of the secondary structures from the ground state structure . our results support the existence of a very weak ( i.e. , marginal ) glass phase characterized by logarithmic excitation energies . finally , we describe the intermediate temperature regime where the system makes the transition from the glass phase to the molten phase . the solution of the two - replica problem is relegated to the appendices . we present two approaches : in appendix [ app_intuitive ] , we provide a mapping of the two - replica problem to the denaturation of an effective single rna in @xmath0-dimensional embedding space ; this approach highlights the connection of the rna problem to the self - consistent hartree theory and should be most natural to field theorists . in appendices [ app_solution ] and [ app_derivez1 ] , we present the exact solution . it is hoped that the two - replica solution may be helpful in providing the intuition needed to tackle the full @xmath4-replica problem . the secondary structure of an rna describes the configuration of base pairings formed by the polymer . if the pairing of the @xmath5 and @xmath6 bases in a polymer of @xmath7 total bases is denoted by @xmath8 with @xmath9 , then each secondary structure @xmath10 is defined by a list of such pairings , with each position appearing at most once in the list , and with the pairs subject to a certain restriction to be described shortly below . each such structure can be represented by a diagram as shown in fig . [ fig_diagrams ] , where the solid line symbolizes the backbone of the molecule and the dashed lines stand for base pairings . the structure shown can be divided into _ stems _ of consecutive base pairs and _ loops _ which connect or terminate these stems . in naturally occurring rna molecules , the stems typically comprise on the order of five base pairs . they locally form the same double helical structure as dna molecules . however , while the latter typically occur in complementary pairs and bind to each other , rna molecules are mostly single - stranded and hence must fold back onto themselves in order to gain some base pairings . by a secondary structure , one often considers only the restricted set of base pairings where any two base pairs @xmath8 and @xmath11 in a given secondary structure are either independent , i.e. , @xmath12 , or nested , i.e. , @xmath13 . this excludes the so - called pseudo - knots ( as exemplified by fig . [ fig_pseudoknot ] ) and makes analytical and numerical studies much more tractable . for an rna molecule , the exclusion of pseudo - knots is a reasonable approximation because the long pseudo - knots are kinetically difficult to form , and even the short ones occur infrequently in natural rna structures @xcite . the latter is due to their relatively low binding energies for short sequences and the strong electrostatic repulsion of the backbone because the polymer backbone is highly charged and pseudo - knotted configurations increase the density of the molecule , their formation can be relatively disfavored in low salt solution . similarly , the tertiary structures which involve additional interactions of paired bases are strongly dependent on electrostatic screening and can be turned off " experimentally by using monovalent salt solution @xcite . indeed , the pseudo - knots are often deemed part of the tertiary structure of an rna molecule . throughout this study , we will exclude pseudo - knots in our definition of secondary structures . without the pseudo - knots , a secondary structure can alternatively be represented by a diagram of non - crossing arches or by a mountain " diagram as shown in fig . [ fig_strreps ] . : in ( a ) the solid line symbolizes the stretched out backbone of the molecule while the dashed arches stand for the base pairs formed . due to the no pseudo - knot constraint two arches never cross . ( b ) shows an equivalent representation as a `` mountain diagram '' . it is a line derived from the arch diagram by going along the backbone from left to right and going one step up for every beginning arch , horizontally for each unbound base , and one step down for each ending arch . such a mountain always stays above the baseline and comes back to the baseline at base @xmath7 . ] in order to calculate boltzmann factors within an ensemble of secondary structures , we need to assign an energy @xmath14 $ ] to each structure @xmath10 . each secondary structure can be decomposed into elementary pieces such as the stems of base pairs and the connecting loop regions as shown in fig . [ fig_diagrams ] . a common approach is to assume that the contributions from these structural elements to the total energy are independent of each other and additive . within a stem of base pairs , the largest energy contribution is the _ stacking energy _ between two adjacent base pairs ( g - c , a - u , or g - u ) , and the total energy of the stem is the sum of stacking energies over all adjacent base pairs . since each secondary structure is defined as a _ single _ state in our ensemble , it is necessary to integrate out all other microscopic degrees of freedom of the bases within a given secondary structure and use an effective energy parameter for each base stacking . the most convenient one to use is the gibbs free energy of stacking @xcite , which contains an enthalpic term due to base stacking , and an entropic term due to the loss of single - stranded degrees of freedom ( as well as the additional conformational change of the backbone and even the surrounding water molecules ) due to base pairing . the magnitudes of these stacking free energies actually depend on the identity of all four bases forming the two base pairs bracketing the stack and are dependent on temperature themselves . while their typical values are on the order of @xmath15 at room temperature , the enthalpic and entropic contributions are each on the order of @xmath16 . thus , upon moderately increasing the temperature from room temperature to about @xmath17 , the stacking free energies become repulsive and the rna molecule denatures . the stacking free energies account for most but not all of the entropic terms for a given secondary structure . there is an additional ( logarithmic ) `` loop energy '' term associated with the entropy loss of each _ closed loop _ of single - stranded rna formed by the secondary structure , as well as the energy necessary to bend the single strand . all of these energy parameters have been measured in great detail @xcite . when incorporated into an efficient dynamic programming algorithm ( to be described below ) , they can rather successfully _ predict _ the secondary structures of many rna molecules of up to several hundred bases in length @xcite . in this paper , we investigate the statistical properties of long , random rna sequences far below the denaturation temperature . we are interested in generic issues such as the existence of a glass phase and various scaling properties . guided by experiences with other disordered systems @xcite , we believe these generic properties of the system should not depend on the specific choice of the model details . since the full model used in refs @xcite makes analytical and numerical studies unnecessarily clumsy , we will examine a number of simplified models , while preserving the most essential feature of the system , namely , the pattern of matches and mismatches between different positions of the sequence . as in the realistic model described above , we choose our reference energy to be the unbound state , so that each unbound base in a secondary structure is assigned the energy @xmath18 . we will neglect the logarithmic loop energy , which is important very close to the denaturation transition @xcite where the average binding energy is close to zero , but not far below the denaturation temperature where most bases are paired . moreover , we will radically simplify the energy rules for base pairing : we neglect the stacking energies and instead associate an interaction energy @xmath19 with every pairing @xmath8 . thus , @xmath20=\sum_{(i , j)\in s}\varepsilon_{i , j}\ ] ] is the total energy of the structure @xmath10 . within this model , it remains to be decided how to choose the energy parameters @xmath19 s . one possibility is to choose each of the bases @xmath21 randomly from the ` alphabet ' set @xmath22 and then assign @xmath23 with @xmath24 being the match or mismatch energy respectively . here , the value of @xmath25 is actually not essential as long as it is repulsive , since the two bases always have the energetically preferred option to not bind at all . thus the energetics of the system is set by @xmath26 . in our numerical study to be reported in secs . [ sec_randomnesshight ] and [ sec_randomnesslowt ] , we will primarily use this model ( as long as it is larger than @xmath27 as explained below ) . indeed the alphabet size and the choice of the matching rule can be used as tuning parameters to change the strength of sequence disorder . but in our study , we choose to minimize the number of parameters and tune the effective strength of disorder by changing temperature . ] with @xmath28 . we will refer to this as the `` sequence disorder '' model . for analytical calculations , it is preferable to treat all the @xmath19 s as _ independent _ identically distributed random variables , i.e. , to assume @xmath29=\prod_{1\le i < j\le n}\rho(\varepsilon_{i , j})\ ] ] for the joint distribution function @xmath30 $ ] of all the @xmath19 s . this choice neglects the correlations between @xmath19 and @xmath31 which are generated through the shared base @xmath32 ; it is an additional approximation on the model ( [ eq_energychoice ] ) . however , we do not anticipate universal quantities to depend on such subtle correlation of the @xmath19 s . this will be tested numerically by comparing the behavior of the model ( [ eq_energychoice ] ) with that of the model defined by eq . ( [ eq_uncorrelated ] ) together with @xmath33 this distribution is chosen to mimic the random sequence model ( [ eq_energychoice ] ) with a 4-letter alphabet , but it does not contain any correlation between the different @xmath19 s . we will refer to this model as defined by eqs . ( [ eq_uncorrelated ] ) and ( [ eq_uncorrelateddisorder ] ) as the `` energy disorder '' model . in the actual analytical calculations , we will go even one step further and take the @xmath19 to be _ gaussian _ random variables specified by @xmath34 where @xmath35 is the average binding energy and @xmath36 is the variance . in this model ( referred to below as gaussian disorder " model , ) the parameter @xmath36 provides us with a convenient measure of the disorder strength . again , universal quantities should not depend on the choice of the distribution functions . we will test this directly by performing numerical studies for these gaussian random energies , with @xmath37 chosen to match the first two moments of the distribution eq . ( [ eq_uncorrelateddisorder ] ) . in contrast to prior numerical studies @xcite , we do not exclude base pairing between neighboring bases @xmath38 , i.e. , we do not set a minimal allowed length for the hairpins . since the results are qualitatively identical to the results of the simpler model presented here , we do not show this data . ] . setting a constraint on the minimal hairpin length would make the analytical study much more cumbersome . however , in the study by pagnani _ et al . _ @xcite , it has been argued that the system will not be frustrated ( and hence will not form a glass ) without this additional constraint . we believe this is an artifact of the 2-letter alphabet used by pagnani _ _ in order to generate the binding energy @xmath19 s via a rule similar to eq . ( [ eq_energychoice ] ) : it is simple to see that for any 2-letter sequence in which the like letters repel and unlike letters attract , one can always find the minimal total binding energy by pairing up neighboring bases of opposite types and removing them from the sequence if no additional constraints such as the minimal hairpin length are enforced . as we will discuss in detail in sec . [ sec_proofglass ] this is not a problem if the alphabet size is larger than 2 . thus , in our study , we use the sequence disorder model with a 4-letter alphabet , or the energy disorder model , without enforcing the minimal hairpin length constraint . while the minimal hairpin length ( of 3 bases ) is known for real rna folding , it should not change the universal properties of long rna sequences . once the energy of each secondary structure is defined , we can study the partition function @xmath39}\ ] ] of the molecule where @xmath40 denotes the set of _ all _ allowed secondary structures of a polymer of @xmath7 bases , and @xmath41 . to calculate this partition function , it is useful to study the _ restricted _ partition function @xmath42 of the substrand from position @xmath43 to position @xmath44 of the rna molecule . given the model ( [ eq_energy ] ) , the restricted partition functions can be split up according to the possible pairings of position @xmath44 . this leads to the recursive equation @xcite @xmath45 with @xmath46 being the total partition function of the molecule . in terms of the arch diagrams introduced in fig . [ fig_strreps](a ) this can be represented as @xmath47{receq.eps}}\ ] ] where the wavy lines stand for the restricted partition functions . this is easily recognized as a hartree equation . since the restricted partition functions on the right hand side of this equation all correspond to shorter pieces of the rna molecule than the left hand side , this equation allows one to calculate the exact partition function of an rna molecule of length @xmath7 with _ arbitrary _ interactions @xmath19 in @xmath48 time . this is accomplished by starting with the partition functions for single bases and recursively applying eq . ( [ eq_partfuncrec ] ) , and is known as a dynamic programming algorithm @xcite . this algorithm allows one to compute numerically the partition function involving all secondary structures , for arbitrary rna molecules of up to @xmath49 bases . it also forms the basis of analytical approaches to the problem as we will see shortly . apart from the partition function itself , we will use additional observables in order to characterize the behavior of rna secondary structures . one such quantity of interest is the binding probability @xmath50 , i.e. , the probability that positions @xmath43 and @xmath44 are paired given the @xmath19 s , @xmath51 where @xmath52 is given by the recursion equation ( [ eq_partfuncrec ] ) and @xmath53 is the partition function of the sequence @xmath54 . the latter can be calculated as the quantity @xmath55 when applying the recursion eq . ( [ eq_partfuncrec ] ) to the duplicated sequence @xmath56 . thus , all @xmath57 such constraint partition functions can be calculated with the same recursion in @xmath48 time . the logarithms @xmath58 of these binding probabilities have a natural interpretation : they can be read as the `` pinching free energies '' , i.e. , as the free energy cost of a pinch between positions @xmath43 and @xmath44 and the unperturbed state . we will make extensive use of this concept of pinched structures in our discussion of the low temperature behavior of rna secondary structures in sec . [ sec_randomnesslowt ] . in our numerical investigations , we will choose as a representative of all the pinching energies for different positions by @xmath59 which is the free energy cost of the largest possible pinch that splits the molecule of length @xmath7 into two pieces of length @xmath60 each . another quantity which describes a secondary structures is its `` size profile '' . as an intrinsic measure of the size of a given secondary structure @xmath10 , we use the `` ladder distance '' @xmath61 between the base at position @xmath62 and the base at position @xmath43 , which is the the number of pairings ( or ladders ) one has to cross to go from a pair involving base @xmath62 to the base @xmath43 ; see fig . [ fig_hdef ] . it can be defined for each secondary structure @xmath10 as the total number of pairings @xmath63 that bracket @xmath43 , i.e. , @xmath64 this quantity can be very easily visualized as the `` height '' at position @xmath43 of the mountain representation of the secondary structure @xmath10 as shown in fig . [ fig_strreps](b ) . a quantity characterizing the full ensemble of secondary structures is the _ thermal average _ @xmath65 of this size profile over all secondary structures with their respective boltzmann factors ; it can be straightforwardly calculated from the probabilities @xmath66 as @xmath67 since we expect all positions in the sequence to behave in a similar way , in our numerics we will summarize the properties of the size profile by the ladder distance from the first to the middle base , i.e. , we will study @xmath68 as a quantity representing the overall `` size '' of an ensemble of secondary structures . of a secondary rna structure : the size profile measures the extension of the structure if drawn as a planar diagram . as an intrinsic definition of @xmath69 which captures this notion of the size of a secondary structure at position @xmath43 , we use the `` ladder distance '' of base @xmath43 from the end of the molecule , i.e. , the number of base pairs which have to be crossed when connecting position @xmath43 to position @xmath62 along the folded structure as indicated by the dashed line . ] if sequence disorder does not play an important role , we may describe the rna molecule by replacing all the binding energies @xmath19 by some effective value @xmath70 . as we will see later , this will be an adequate description of our random rna models at high enough temperatures ( but before denaturation . ) for the real rnas , this provides a coarse grained description of repetitive , self - complementary sequences , e.g. , cagcag ... cag , which are involved in a number of diseases @xcite . we will refer to rna which is well described by this model without sequence disorder as being in the `` molten '' phase . it serves as a starting point for modeling non - specific self - binding of rna molecules , and its properties will form the basis of our study of the random rna at low temperatures . since in the absence of sequence disorder , the energy of a structure @xmath10 depends only on the number of paired bases @xmath71 of this structure , we can write the partition function in the molten phase as @xmath72\ ] ] the partition functions of the sub - strands @xmath42 become translationally invariant and can be written as @xmath73 where @xmath74 is only a function of the length @xmath7 . the recursion equation ( [ eq_partfuncrec ] ) then takes the form @xmath75 where @xmath76 upon introducing the @xmath77-transform @xmath78 the convolution can be eliminated and the recursion equation turns into a quadratic equation @xmath79 with the solution @xmath80 performing the inverse @xmath77-transformation in the saddle point approximation yields the expression @xcite @xmath81 in the limit of large @xmath7 , with the exponent @xmath82 and the non - universal quantities @xmath83 and @xmath84^{1/2}$ ] . this result characterizes the state of the rna where a large number of different secondary structures of equal energy coexist in the thermodynamic ensemble , and the partition function is completely dominated by the configurational entropy of these secondary structures . while the result is derived specifically for the special case @xmath85 , we will argue below that it is applicable also to random @xmath19 s at sufficiently high temperatures , in the sense that for long rna molecules , the partition function is dominated by an exponentially large number of secondary structures all having _ comparable _ energies ( within @xmath86 ) that are smoothly related to each other in configuration space . the latter is what we meant by the `` molten phase '' . the exponent @xmath82 is an example of a scaling exponent characteristic of the molten phase . this and other exponents can be derived in a geometric way by the `` mountain '' representation of secondary structures as illustrated in fig . [ fig_strreps](b ) . each such mountain corresponds to exactly one secondary structure . in the molten phase , the weight of a secondary structure @xmath10 is simply given by @xmath87 . this can be represented in the mountain picture by assigning a weight of @xmath88 to every upward and downward step and a weight of @xmath62 to every horizontal step . since the only constraints on these mountains are ( i ) staying above the baseline , and ( ii ) returning to the baseline at the end , the partition function of an rna of length @xmath7 is then simply that of a random walk of @xmath7 steps , constrained to start from and return to the origin , in the presence of a _ hard wall _ at the origin , with the above weights ( @xmath89 or @xmath62 ) assigned to each allowed step . this partition function is well - known to have the characteristic @xmath90 behavior which we formally derived in the last section @xcite . in this framework , it also becomes obvious why imposing a minimal hairpin length does not change the universal behavior of rna at least in this molten phase : if the minimal allowed size of a hairpin is @xmath91 , this enforces a potentially strong _ penalty _ for the formation of a hairpin , since with every hairpin @xmath91 bases are denied the possibility of gaining energy by base pairing . this tends to make branchings less favorable and thus leads to longer stems . however , this additional constraint translates in the mountain representation into the rule that an upwards step may not be followed by a downwards step within the next @xmath91 steps . this is clearly a _ local _ modification of the random walk . thus , it does not change universal quantities although the above mentioned suppression of branchings will require much longer sequences in order to observe the asymptotic universal behavior . for real rna parameters , the crossover length is very long due to this effect . for example , it is several hundred nucleotides for the cag repeat , and even longer for some other repeats . another characteristic exponent describes the scaling of the ladder size @xmath92 with the sequence length @xmath7 . as already mentioned in its definition ( [ eq_hdef ] ) , @xmath92 is equivalent to the average `` height '' of the mid - point of the sequence in the mountain picture . in the molten phase , the random walk analogy immediately yields the result @xmath93 where @xmath94 denotes ensemble average in the molten phase . as should be clear from the coarse - grained view depicted in fig . [ fig_hdef ] , the ensemble of rna secondary structures in the molten phase can be mapped directly to the ensemble of _ branched polymers_. these branched polymers are _ rooted _ at the bases @xmath95 and @xmath96 of the rna . in this context , @xmath82 is known as the configuration exponent of the rooted branched polymer @xcite . additionally from the result ( [ eq_hscaling ] ) , we see that the ladder length of the branched polymer scales , a well - known result for the branched polymer in the absence of self - avoidance @xcite . ] as @xmath97 . because of the very visual analogy of the secondary structures to branched polymer , we refer to the configurational entropy of the secondary structures as the `` branching entropy '' . finally , the binding probabilities @xmath50 defined in eq . ( [ eq_returnprobability ] ) only depend on the distance @xmath98 in the molten phase , i.e. , @xmath99 . the behavior of this function can be derived explicitly by inserting the result eq . ( [ eq_moltenz ] ) for the partition function into eq . ( [ eq_returnprobability ] ) . alternatively , one just needs to recognize that @xmath100 corresponds in the random walk analogy to the first - return probability of a random walk after @xmath101-steps . in either case , one finds the result @xmath102 i.e. , the return probability decays with the separation @xmath101 of the two bases as a power law with the configuration exponent @xmath82 . for the pinching free energy @xmath103 , we simply set @xmath104 and obtain @xmath105 for large @xmath7 , i.e. , it scales _ logarithmically _ in the molten phase . this logarithmic dependence merely reflects the loss in branching entropy due to the pinching constraint and is a manifestation of the configuration exponent @xmath82 . there are in principle three different scenaria for the behavior of long random rna sequences . ( i ) disorder is irrelevant at any finite temperature , so that the molten phase description presented in sec . [ sec_molten ] applies to long rnas at all temperatures . ( ii ) disorder is relevant at all temperatures , and the molten phase description would be completely inadequate . ( iii ) there is a finite temperature @xmath106 above which the molten description of random rna is correct , while below @xmath106 a qualitatively different description is needed . in accordance with the statistical physics literature , we will refer to the non - molten phase as the glass phase , and @xmath106 as the glass transition . the purpose of the study is to determine which of these three scenaria is actually realized , and to characterize the glass phase if either ( ii ) or ( iii ) is realized . in this section , we study the high temperature behavior and demonstrate that the molten phase is stable with respect to weak sequence disorder . this ensures that the molten description of rna given in sec . [ sec_molten ] is at least valid at high enough temperatures , thereby ruling out scenario ( ii ) . we will address the question of whether there is a glass phase at low but finite temperatures in sec . [ sec_randomnesslowt ] . before we engage in detailed calculations , we want to convince ourselves with the help of some numerics that weak disorder does not destroy the molten phase . to this end , we study the observables introduced in sec . [ sec_observables ] . we generate a large number of disorder configurations , i.e. , interaction energies @xmath19 using the 3 models introduced in sec . [ sec_intenergies ] : sequence disorder , energy disorder , and gaussian disorder as described by eq . ( [ eq_energychoice ] ) , eqs . ( [ eq_uncorrelated ] ) and ( [ eq_uncorrelateddisorder ] ) , and eqs . ( [ eq_uncorrelated ] ) and ( [ eq_gaussiandisorder ] ) respectively , with @xmath107 . then , we calculate the observables @xmath92 and @xmath103 for each disorder configuration at the relatively large temperature of @xmath108 and average the obtained values over many disorder configurations . in order to keep the numerical effort manageable , we average over @xmath109 random sequences for @xmath110 , over @xmath111 sequences for @xmath112 , and over @xmath113 sequences for @xmath114 and @xmath115 . [ fig_moltennumerics ] shows the results ; disorder averaged quantities are denoted by an overline throughout the text . we see that the data for @xmath116 follows a power law with a fitted exponent @xmath117 , with the exponent value decreasing for larger @xmath7 s . this result is consistent with the prediction eq . ( [ eq_hscaling ] ) for the molten phase . also , the pinching free energy follows the predicted logarithmic behavior eq . ( [ eq_moltenpinch ] ) without any noticeable difference between the three choices of disorder . taken together , these results indicate that the three models of disorder belong to the same universality class , i.e. , the molten phase description of the uniformly attracting rna , at high temperatures . of the sequence at @xmath108 . each plot shows data for three different choices of disorder according to eqs . ( [ eq_energychoice ] ) , ( [ eq_uncorrelateddisorder ] ) , and ( [ eq_gaussiandisorder ] ) . ( a ) shows the scaling of the average size @xmath116 and the dashed line is the best fit @xmath117 to a power law . ( b ) shows the free energy of the largest pinch as defined in ( [ eq_largestpinchfreeenergy ] ) . the dashed line is up to an additive constant the logarithmic behavior @xmath118 predicted in eq . ( [ eq_moltenpinch ] ) . the statistical fluctuations are smaller than the size of the symbols in both plots . all plots suggest that the behavior of rna secondary structures at high temperatures is well described by the molten phase picture and independent of the disorder . ] now , we will establish the stability of the molten phase against weak disorder by an analytical argument . we will use gaussian disorder characterized by eqs . ( [ eq_uncorrelated ] ) and ( [ eq_gaussiandisorder ] ) . as we have shown above , the different microscopic models of binding energies all yield the same scaling behaviors . with the uncorrelated gaussian energies , it is possible to perform the ensemble average of the partition function @xmath119 of @xmath4 rna molecules sharing the same disorder . the disorder - averaged free energy can then in principle be obtained via the `` replica - trick '' @xmath120 , by solving the @xmath4-replica problem @xcite . the @xmath4-replica partition function can be written down formally as @xmath121}\\ & = & \sum_{\{s_1\}}\ldots\sum_{\{s_n\ } } \prod_{k=1}^n\exp\left[-\beta\avereps|s_k|\right ] \exp\left[\frac{1}{2}\beta^2\sum_{k=1}^n\sum_{l=1}^n\sum_{(i , j)\in s_k } \sum_{(r , s)\in s_l } \overline{(\varepsilon_{i , j}-\avereps)(\varepsilon_{r , s}-\avereps ) } \right]\\ & = & \sum_{\{s_1\}}\ldots\sum_{\{s_n\}}\prod_{k=1}^n \exp\left[-\beta\avereps|s_k|\right ] \exp\left[\frac{1}{2}\beta^2d\sum_{k=1}^n\sum_{l=1}^n|s_k\cap s_l|\right]\\ & = & \sum_{\{s_1\}}\ldots\sum_{\{s_n\}}\prod_{k=1}^n q^{|s_k| } \prod_{1\le k < l\le n}\widetilde{q}^{|s_k\cap s_l| } \ ] ] where @xmath122 are the two relevant `` boltzmann factors '' . this effective partition function has a simple physical interpretation : it describes @xmath4 rna molecules subject to a _ homogeneous _ attraction with effective interaction energy @xmath123 between any two bases of the same molecule . as before , this effective attraction is characterized by the factor @xmath124 . in addition , there is an inter - replica attraction characterized by the factor @xmath125 for each bond _ shared _ between any pair of replicas . the inter - replica attraction is induced by the same sequence disorder shared by all replicas . for example , if the base composition in one piece of the strand matches particularly well with another piece , then there is a tendency to pair these pieces together in all replicas . thus , the inter - replica attraction can potentially force the different replicas to `` lock '' together , i.e. , to behave in an correlated way . indeed , the distribution of inter - replica correlations , usually measured in terms of `` overlaps '' , is a common device used to detect the existence of a glass phase in disordered systems @xcite . the full @xmath4-replica problem is difficult to solve analytically . we will examine this problem in the regime of small @xmath36 , aiming to resolve the relevancy of disorder in a perturbative sense . since the lowest order term of the fully random problem in a perturbation expansion in @xmath36 corresponds to the two - replica ( @xmath126 ) problem , we will focus on the latter in order to study the small-@xmath36 behavior of the full problem . the solution of the two - replica problem will also illustrate explicitly the type of interaction one is dealing with , thereby providing some intuition needed to tackle the full problem . it turns out that the two - replica problem can be solved exactly . here , we outline the saline features of the solution . details of the calculation and analysis are provided in the appendices . we will find that the two - replica system has a phase transition between the molten phase in which the two replicas are uncorrelated and a nontrivial phase in which the two replicas are completely locked together in the thermodynamic limit . the transition occurs at a finite temperature @xmath127 which approaches zero as @xmath128 . thus , the effect of weak disorder is _ irrelevant _ at finite temperatures . let us denote the two - replica partition function @xmath129 for two strands each of length @xmath7 by @xmath130 , where we keep the dependence on @xmath124 implicit . then , @xmath131 the key observation which allows us to solve the two - replica problem is that for each given pair of secondary structures , the bonds shared by two replicas ( hereafter referred to as `` common bonds '' ) form a valid secondary structure by themselves ( see fig . [ fig_combonds ] . ) thus , we can rearrange the summation over the pairs of secondary structures in the following way : we first sum over all possible secondary structures of the common bonds . for a _ given configuration _ of the common bonds , we then sum over the remaining possibilities of intra - replica base pairings for each replica , with the constraint that no new common bonds are created . note that the common bonds partition the diagram into a number of `` bubbles '' , shown as the shaded regions in fig . [ fig_combonds ] . due to the exclusion of pseudo - knots from the valid secondary structures , only bases belonging to the same bubble can be paired with each other . thus , the two - replica partition function can be written as @xmath132 where the factor @xmath133 is the weight of each common bond , and @xmath134 is the sum of all possible intra - replica pairings of of the @xmath5 bubble of @xmath135 bases in @xmath10 , with the restriction that there are no common bonds . it should be clear that @xmath136 neither depends on the number of stems branching out from the bubble @xmath43 nor the positions of these stems relative to the bases within the bubble . it depends on @xmath10 only through the number of bases @xmath135 in the bubble and is given by a single function @xmath137 independent of @xmath43 . this function can be written explicitly as @xmath138 with eqs . ( [ eq_gintermsofloops ] ) and ( [ eq_defq ] ) , the two - replica problem is reduced to an effective _ single _ homogeneous rna problem , with an effective boltzmann weight @xmath139 for each pairing , and an effective weight @xmath137 for each single stranded loop . as described in appendix [ app_intuitive ] , this problem becomes formally analogous to that of an rna in the vicinity of the denaturation transition , with @xmath137 being the weight of a single polymer loop fluctuating in @xmath0-dimensional embedding space . the competition between pairing energy and the bubble entropy leads to a _ phase transition _ for the two - replica problem , analogous to the denaturation transition for a single rna . the details of this transition are given in appendix [ app_solution ] , where the partition function ( [ eq_gintermsofloops ] ) is solved exactly . the exact solution exploits the relation @xmath140 which follows from the definitions ( [ eq_defg ] ) and ( [ eq_defq ] ) , and turns eq . ( [ eq_gintermsofloops ] ) into a recursive equation for @xmath141 . the solution is of the form @xmath142 for large @xmath7 , with two different forms for @xmath143 and @xmath144 depending on whether @xmath125 is above or below the critical value @xmath145 here @xmath74 is the molten phase partition function , whose large @xmath7 asymptotics is given by eq . ( [ eq_moltenz ] ) and whose values for small @xmath7 can be calculated explicitly from the recursion eq . ( [ eq_g0recursion ] ) . thus , the actual value of @xmath146 can be found for any given @xmath124 . for @xmath147 , we have @xmath148 and @xmath149 where @xmath150 according to eqs . ( [ eq_gatsingdef ] ) and ( [ eq_moltenzresult ] ) . in this regime , the two - replica partition function @xmath141 is essentially a product of two single - replica partition functions @xmath151 . compared to eq . ( [ eq_moltenz ] ) , we can identify @xmath143 as @xmath152 , and @xmath144 as a modified version of @xmath153 . since there is no coupling of the two replicas beyond a trivial shift in the free energy per length , @xmath154 , we conclude that the disorder coupling is irrelevant . hence the two - replica system is in the molten phase in this regime . for @xmath155 , we have @xmath156 and @xmath144 is given as the implicit solution of an equation involving only single - replica partition functions as shown in eqs . ( [ eq_zkdef ] ) and ( [ eq_zetaresult ] ) . here , the partition function of the two - replica system is found to have the _ same _ form as that of the single - replica system in ( [ eq_moltenz ] ) . this result implies that the two replicas are _ locked _ together via the disorder coupling , and the molten phase is no longer applicable in this regime . of course , as already explained above , only the weak - disorder limit ( i.e. , @xmath157 ) of the two - replica problem is of relevance to the full random rna problem . in this limit , @xmath158 while @xmath146 is found by evaluating eq . ( [ qc ] ) with @xmath159 . it can be easily verified that @xmath160 as long as @xmath124 is finite . thus in the weak disorder limit , we have @xmath147 , indicating that the molten phase is an appropriate description for the random rna . unfortunately , the two - replica calculation can not be used in itself to deduce whether the molten phase description breaks down at sufficiently strong disorder or low temperature . based on this analysis , we can not conclude whether the type of phase transition obtained for the two - replica problem is present in the full problem . having established the validity of the molten phase description of random rna molecules at weak disorder or high temperatures , we now turn our focus onto the low temperature regime . first , we will give an analytical argument for the existence of a glass phase at low temperatures . then , we will present extensive numerical studies confirming this result and characterizing this glass phase . we will start by showing that the molten phase can not persist for all temperatures down to @xmath161 . to this end , we will _ assume _ that long random rna is in the molten phase for _ all _ temperatures , i.e. , that the partition function for any substrand of large length @xmath162 is given by @xmath163\ ] ] with some effective temperature - dependent prefactor @xmath164 and free energy per length @xmath165 . then , we will show that this assumption leads to a contradiction below some temperature @xmath166 . this contradiction implies that the molten phase description breaks down at some finite @xmath167 . to be specific , we will consider the sequence disorder model ( [ eq_energychoice ] ) in this analysis . the quantity we will focus on is again the free energy @xmath168 of the largest possible pinch . under the assumption that the random sequences are described by the molten phase , it is given by @xmath169 for large @xmath7 and all @xmath170 _ independently _ of the values of the effective prefactor @xmath164 and the free energy per length @xmath165 ( see eq . ( [ eq_moltenpinch ] ) . ) on the other hand , we can study this pinching free energy for each given sequence of bases drawn from the ensemble of random sequences . for each such sequence , we can look for a continuous segment of @xmath171 watson - crick pairs @xmath32@xmath172@xmath173@xmath174 where the bases @xmath175 are within the first half of the molecule and the bases @xmath176 are in the second half ( see fig . [ fig_goodmatch](a ) . ) for random sequences , the probability of finding such exceptional segments decreases exponentially with the length @xmath101 , with the largest @xmath101 in a sequence of length @xmath7 being typically @xmath177 for exact complementary matches , the proportionality constant is known to be @xmath178 @xcite . and @xmath179 and the other of which is between positions @xmath180 and @xmath7 . such a piece of length @xmath181 can be found for almost all sequences . ( b ) shows how restricting configurations to those in which the good match forms watson - crick base pairs splits the molecule into two loops which can still form base pairs within the loops independently from each other . ] now , we calculate the pinching free energy @xmath182 by evaluating the two terms separately . the partition function for the unpinched sequence contains _ at least _ all the configurations in which the two complementary segments @xmath183 and @xmath176 are completely paired ( see fig . [ fig_goodmatch](b ) ) . thus , @xmath184 where @xmath185 is the free energy of the ensemble of structures in which the two complementary segments are paired . the latter is the sum of the free energy of the paired segments and those of the two remaining substrands @xmath186 of length @xmath187 and @xmath188 ( wrapping around the end of the molecule ) of length @xmath189 , i.e. , @xmath190.\ ] ] the free energy @xmath191 in the presence of the pinch is , by the assumption of the molten phase , the interaction energy of the pinched base pair @xmath192@xmath193 plus the molten free energy of the substrand @xmath194 and the molten free energy of the substrand @xmath195 , i.e. , according to eq . ( [ eq_moltenzeff ] ) @xmath196 up to terms independent of @xmath7 . combining this with eqs . ( [ eq_deltafasdifference ] ) , ( [ eq_unpinchedandpaired ] ) , and ( [ eq_fpaired ] ) , we get @xmath197 + \ell [ u_{\mathrm{m}}\!+\!2f_0(t)].\ ] ] using the result ( [ eq_lambda ] ) and that @xmath198 and @xmath199 are typically proportional to @xmath7 , we finally obtain @xmath200 \lambda^{-1 } \ln n\ ] ] for large @xmath7 . this is only consistent with eq . ( [ eq_assumedmoltenpinch ] ) if @xmath201.\ ] ] now , @xmath165 is a free energy and is hence a monotonically decreasing function of the temperature . thus the validity of the inequality ( [ eq_assumedmoltenpinch ] ) depends on the behavior of its right hand side at low temperatures . as @xmath202 , the inequality can only hold if @xmath203 . since the average total energy at @xmath204 is @xmath205 times the average total number of matched pairs of a random sequence , then @xmath206 is simply the fraction of matches and @xmath207 is the fraction of bases not matched ( for @xmath208 . ) clearly , @xmath207 can not be negative , and the inequality ( [ eq_assumedmoltenpinch ] ) must fail at some finite temperature unless @xmath209 . we can make a simple combinatorial argument to show that in most cases the fraction @xmath207 of unbound bases must strictly be positive . to illustrate this , let us generalize the `` alphabet size '' of the sequence disorder model of sec . [ sec_intenergies ] from @xmath210 to an arbitrary even integer @xmath211 . we will still adopt the energy rule ( [ eq_energychoice ] ) where each of the @xmath212 bases can form a `` watson - crick '' pair exclusively with one other base . let us estimate the number of possible sequences for which the fraction of unmatched bases @xmath207 is zero in the limit of long sequence length @xmath7 at @xmath204 . since at @xmath204 , only watson - crick ( w - c ) pairs can be formed , we only need to count the number of sequences for which the fraction of w - c paired bases is @xmath62 . this means that except for a sub - extensive number of bases , all have to be w - c paired to each other . from the mountain picture ( fig . [ fig_strreps ] ) , it is clear that the number of possible secondary structures for such sequences must scale like @xmath213 , since the fraction of horizontal steps is non - extensive so that at each step , there are only the possibilities for the mountain to go up or down . for each of the @xmath179 pairings in one of these @xmath213 structures , there are @xmath212 ways of choosing the bases to satisfy the pairing . so for each structure , there are @xmath214 ways of choosing the sequence that would guarantee the structure . since there are a total of @xmath215 sequences , it is clear that the fraction of sequences with all ( but a sub - extensive number of ) w - c pairs becomes negligible if @xmath216 thus , for @xmath217 , we must have @xmath218 . for @xmath219 , the left hand side of eq . ( [ eq_frustration ] ) grows faster than its right hand side . this reflects the absence of frustration in this simple two - letter model as already discussed at the end of sec . [ sec_intenergies ] . one way to retain frustration is to introduce additional constraints , e.g. , the minimal hairpin length used in ref . @xcite . with this constraint , a structure with a sub - extensive number of unmatched bases can only contain a sub - extensive number of hairpins . in the mountain picture , this means that except for a sub - extensive number of steps , there is only one choice to go up or down at every step . this changes eq . ( [ eq_frustration ] ) to @xmath220 . it ensures frustration since @xmath218 for _ all _ @xmath212 . since a minimal length of @xmath221 bases is necessary in the formation of a real hairpin , real rna is certainly frustrated by this argument . the random sequence model which we study in this paper is marginal since @xmath222 and there is no constraint on the minimal hairpin length . in this case , all the prefactors on the two sides of eq . ( [ eq_frustration ] ) ( e.g. , the overcounting of sequences that support more than one structure ) must be taken into account . we will not undertake this effort here , but will verify numerically in sec . [ sec_estimatets ] that @xmath218 also in this case . in all cases with @xmath218 , it follows that there is some unique temperature @xmath223 below which the consistency condition ( [ eq_consistencycondition ] ) breaks down , implying the inconsistency of the molten phase assumption in this regime . from this we conclude that there must be a phase transition away from the molten phase at some critical temperature @xmath224 . the precise value of the bound @xmath223 depends on @xmath225 which in turn depends on the stringency of the condition we impose on the rare matching segments . for instance , if we relax the condition of exact complementarity between two segments to allow for matches within each segment , then the constant @xmath225 will be reduced from @xmath226 and the value of @xmath223 will increase . this will be discussed more in sec . [ sec_estimatets ] . the above argument does not provide any guidance on the properties of the low temperature phase itself . in order to characterize the statistics of secondary structures formed at low temperatures , we re - do the simulations reported in sec . [ sec_hightnumerics ] at @xmath227 in energy units set by @xmath107 . at this temperature , an unbound base pair is penalized with a factor @xmath228 relative to a watson - crick base pair , and a non watson - crick base pair is penalized even more . thus , only the minimal energy structures contribute ( for the sequence lengths under consideration here ) , and we may regard this effectively as at @xmath204 . as in sec . [ sec_hightnumerics ] , we average over @xmath109 realizations of the disorder for @xmath110 , over @xmath111 realizations for @xmath112 and over @xmath113 realizations for @xmath229 . [ fig_lowtnum ] shows the results for the ladder size @xmath230 of the structures for the three models of disorders . the ladder size still scales algebraically with the length of the sequences , with numerically determined exponents ranging from @xmath231 to @xmath232 for the different choices of disorder . they are clearly different from the square root behavior ( dotted line ) expected of the molten phase . thus this result reaffirms our expectation that the secondary structures of a random rna sequence at zero temperature indeed belongs to a phase that is different from the molten phase . of secondary structures in the low temperature phase with the length @xmath7 of the sequences : the plot shows data for three different choices of disorder according to eqs . ( [ eq_energychoice ] ) , ( [ eq_uncorrelateddisorder ] ) , and ( [ eq_gaussiandisorder ] ) at @xmath233 . the average system size follows a power law . however , the best fit of the data for sequence disorder at @xmath234 to a power law indicated by the dashed line leads to an exponent of @xmath232 . the corresponding fits for energy and gaussian disorder yield exponents of @xmath232 and @xmath231 , respectively . this is distinctively different from the square root behavior of the molten phase indicated by the dotted line . the comparison of this plot with its counterpart in fig . [ fig_moltennumerics](a ) suggests that the behavior of rna secondary structures at low temperatures is different from the molten phase . ] a key question in characterizing the thermodynamic properties of disordered systems is whether the zero - temperature behavior persists for a range of finite temperatures . if it does , then the system is said to have a finite - temperature glass phase . one way to address this question is to study the overlap between different replicas of the rna molecule as mentioned earlier . if a non - trivial distribution of these overlaps with significant weight on large overlaps persists into finite temperatures , then the finite - temperature glass phase exists . this approach was taken by previous numerical studies @xcite . unfortunately , the results are inconclusive and even contradictory due to the weakness of the proposed phase transition only the fourth temperature derivative of the free energy seems to show an appreciable singularity . moreover , due to limitations in the sequence lengths probed , it was difficult to get a good estimate of the asymptotic behavior of the overlap distribution . in our study , we adopt a different approach based on the droplet theory of fisher and huse @xcite . in this approach , one studies the `` large scale low energy excitations '' about the ground state . this is usually accomplished by imposing a deformation over a length scale @xmath235 and monitoring the minimal ( free ) energy cost of the deformation . this cost is expected to scale as @xmath236 for large @xmath101 . a positive exponent @xmath237 indicates that the deformation cost _ grows _ with the size . if this is the case , the thermodynamics is dominated by a few low ( free ) energy configurations in the thermodynamic limit , and the statistics of the zero - temperature behavior persists into finite temperatures . on the other hand , if the exponent @xmath237 is negative , then there are a large number of configurations which have low overlap with the ground state but whose energies are similar to the ground state energy in the thermodynamic limit . at any finite temperature @xmath170 , a finite fraction of these configurations ( i.e. , those within @xmath86 of the ground state energy ) will contribute to the thermodynamics of the system . the zero temperature behavior is clearly not stable to thermal fluctuations in this case , and no thermodynamic glass phase can exist at any finite temperature . the analysis of the previous section indicate the existence of a glass phase ; thus we expect to find the excitation energy to _ increase _ with the deformation size . it should be noted that this criterion for glassiness is purely thermodynamical in nature and does not make any statement about _ kinetics_. a system which is not glassy thermodynamically can still exhibit very large _ barriers _ between the many practically degenerate low energy configurations , leading to a _ kinetic glass_. a study of the kinetics of rna , e.g. , in terms of barrier heights , is naturally dependent on the choice of allowed dynamical pathways to transform one rna secondary structure into another one @xcite . since the latter is a highly non - trivial problem , we will restrict ourselves to thermodynamics and use the droplet picture explained above as our criterion for the existence of a glass phase . according to the criterion for glassiness just presented , our goal is to determine the value of the exponent @xmath237 for random rnas numerically . to this end , the choice of `` large scale low energy excitations '' needs some careful thoughts . as in every disordered system , there is a very large number of structures which differ from the minimal energy structure only by a few base pairs and which have an energy only slightly higher than the minimum energy structure . these structures are clearly not of interest here . instead we need to find a controlled way of generating droplet excitations of various sizes . we propose to use the pinching method introduced in sec . [ sec_observables ] as a way to generate the deformation , and regard the difference between the minimal energy pinched structure and the ground state structure as the droplet excitation . there are several desirable features about these pinch - induced deformations : first , it gives a convenient way of controlling the size of the deformation . if @xmath238 is a base pair that is bound anyways in the ground state , pinching this base pair does not have any effect and @xmath239 . if we pinch base @xmath43 with some other base @xmath240 , then we force at least a partial deformation of the ground state , for bases in the vicinity of @xmath43 , @xmath241 , and @xmath44 . this is illustrated in fig . [ fig_pinchingstep ] with the deformed region indicated by the shade . as we move the pinch further away from the ground state pairing , we systematically probe the effect of larger and larger deformations ( provided that a pinch only induces _ local _ deformation as we will show ) . second , the minimal energy or the free energy of the secondary structures subject to the pinch constraint is easily calculable numerically by the dynamic programming algorithm as shown in eqs . ( [ eq_returnprobability ] ) and ( [ eq_pinchfreeenergy ] ) . third , the pinching of the bases in a sense mimics the actual dynamics of the rna molecule at low temperatures . in order for the molecule to transform from one secondary structure to another at a temperature where all matching bases should be paired , the bases have to make local rearrangements of the secondary structures much like the way depicted in fig . [ fig_pinchingstep ] @xcite . thus , the pinching energy provides the scale of variation in the local energy landscape for such rearrangements to the pinch free energies . ] . finally , `` pinching '' of a real rna molecule can be realized in the pulling of a long molecule through a pore @xcite . and @xmath241 ( open circles ) form a base pair in the indicated minimal energy structure ( a ) . thus , forcing these two bases to be bound by pinching does not affect the structure at all . pinching of @xmath8 on the other hand will lead to a local rearrangement ( shaded region ) of the structure as shown in ( b ) . the effect of such a pinch depends on the number of base pairs of the minimal energy structure the pinch is incompatible with . as indicated by the arrow in ( a ) , this number is given by the ladder distance @xmath242 between @xmath43 and @xmath44 ; in this example @xmath243 . ] a key question to the utility of these pinch deformations is whether the deformation is confined to the local region of the pinch as depicted in fig . [ fig_pinchingstep ] or whether it involves a global rearrangement of the structure . to test this aspect , we numerically study the changes in pinch free energy as a function of the `` size '' of a pinch . here , the definition of the pinch size needs some thought . consider a specific sequence whose minimal energy structure is @xmath244 . if the binding partner of base @xmath43 is base @xmath241 in the minimal energy structure , a natural measure for the size of a pinch @xmath245 with @xmath246 would be the ladder distance @xmath242 between base @xmath43 and base @xmath44 ; see fig . [ fig_pinchingstep ] . from the mountain representation ( fig . [ fig_strreps](b ) ) , it is easy to see that this is just the difference of the respective ladder distances of base @xmath44 and base @xmath43 from base @xmath62 as defined in eq . ( [ eq_defmountainheight ] ) , i.e. , @xmath247 . to find how the excitation energy depend on the pinch size , we just need to follow how the pinching free energy @xmath248 s depend statistically on the size @xmath242 s . to do so , we choose a large number of random sequences , and determine the minimal energy structure @xmath244 for each of these sequences . then , we compute the pinch free energies @xmath248 and the pinch size @xmath242 for _ all _ possible pinches @xmath8 for each sequence . afterwards , we average over all @xmath248 s with the same pinch size @xmath242 over all of the generated random sequences to obtain the function @xmath249 the results obtained at @xmath227 for a large range of sequence sizes from @xmath250 to @xmath115 are shown in fig . [ fig_fofh](a ) . we see that the data for different @xmath7 s fall on top of each other for small @xmath251 s , with @xmath252 this behavior explicitly shows that the pinch deformation is a _ local _ deformation . in particular , we see that for small @xmath251 s , the free energy cost is _ independent _ of the overall length @xmath7 of the molecule . of minimal energy structure base pairs that are incompatible with the pinch for random sequence disorder at @xmath233 : ( a ) shows the raw data . for small @xmath253 the pinching free is independent of the length @xmath7 of the molecule and obeys a power law @xmath254 ( dashed line . ) this is consistent with the expectation that pinching at small @xmath251 leads to _ local _ rearrangements of the secondary structure . the apparent non - monotonic behavior at large @xmath251 is due to the small number of sequences in which such a value of @xmath251 is realized . ( b ) shows the same data , but the scaling of @xmath253 with @xmath7 is chosen in accordance with fig . [ fig_lowtnum ] while the scaling of the pinching free energy is chosen in accordance with the power law dependence estimated above . ] it is interesting to see at which @xmath251 the entire sequence is involved . one expects @xmath255 since @xmath230 gives the typical scale of the maximum ladder length . to test this , we normalized @xmath251 by @xmath256 and @xmath257 by @xmath258 ( such that the relation ( [ eq_fh ] ) is preserved for small @xmath251 s ) . the result is shown in fig . [ fig_fofh](b ) . we see that the data is approximately collapsed onto a single curve , indicating that pinching is indeed a good way of imposing a controlled deformation from the ground state . the scaling plot of fig . [ fig_fofh](b ) indicates strongly that the energy associated with the pinch deformation _ increases _ with the size of the deformation , i.e. , @xmath259 . however , the effective exponent involved is small , making the result very susceptible to finite size effects . in order to decide on the glassiness of the system , we want to focus on the energy scales associated with the largest pinch deformations from the ground state . assuming that there is only a single energy scale associated with large pinches , we again study the free energy @xmath103 of the largest pinch as defined in eq . ( [ eq_largestpinchfreeenergy ] ) and average this over the ensemble of sequences . this quantity yields an _ upper bound _ estimate of the energy associated with large scale pinches for each sequence length @xmath7 . we find @xmath260 and @xmath261 to have the same scaling behavior , and thus present only data for the latter . ] . the results are shown in fig . [ fig_pinch](a ) for the three models of disorders . although a weak power law dependence of @xmath261 on @xmath7 can not be excluded , the fitted exponents obtained for the three models are different from each other , ranging from @xmath262 to @xmath263 . this is a strong sign of concern , since the exponents are expected to be independent of details of the models . in fig . [ fig_pinch](b ) , the same data is plotted on a log - linear scale . the data fall reasonably on a straight line for each of the models ( especially for large @xmath7 s ) , suggesting that the pinching free energy may actually increase _ logarithmically _ with the sequence length , similar to what is expected of the behavior in the molten phase ! however in this case , the prefactor of the logarithm depends on the choice of the model and is much larger than the factor @xmath264 expected of the molten phase ; see eq . ( [ eq_moltenpinch ] ) . for example , for the numerical data obtained at @xmath227 , the prefactor is approximately @xmath265 for the sequence disorder model , while the expected slope for the molten phase is @xmath266 at this temperature . having different logarithmic prefactors for the different models is not a concern , since a prefactor is a non universal quantity . thus , our numerical results favor a logarithmically increasing pinch energy , with a prefactor much exceeding @xmath267 at low temperature . . the exponents are @xmath268 , @xmath269 , and @xmath258 for the sequence , energy , and gaussian disorder respectively . ( a ) shows the same data in a single logarithmic plot together with the best fits to a logarithmic dependence on @xmath7 . the statistical error of the data points is about the size of the symbols for large @xmath7 and smaller than that for @xmath270 . due to the apparent systematic bending of the data in the double logarithmic plot ( b ) we conclude that a logarithmic dependence fits the data better although we can not exclude a power law behavior with a very small power . ] what does this tell us about the possible glass phase of the random rna ? in order to answer this question , we should remind ourselves that rather special deformations are chosen in this study . for our choice of pinch deformations , we observe a logarithmic dependence of the gap between the ground state energy and the energy of the excited configurations on the length of the sequence or deformation . this corresponds to the marginal case of the droplet theory where the exponent @xmath237 vanishes . since the pinching free energies are increasing with length , we can not exclude a glass phase in the case @xmath271 . we can say , though , that the increase of the excitation energy with length is at most a power law with a very small exponent and most probably even less than any power law . therefore , a possible glass phase of rna has to be _ very weak_. if it turns out that the excitation energy is indeed a logarithmic function of length , with a non - vanishing prefactor as @xmath202 as our numerics suggest , then the low - temperature phase would be categorized formally as a _ marginal _ glass phase , analogous to behaviors found in some well - studied model of statistical mechanics @xcite . in any case , we should note that the actual difference in the excitation energy is only a factor of @xmath210 across two - and - a - half decades in length . thus the glassy effect will be weak for practical purposes . on the other hand , the weak dependence of the excitation energy on length may be the underlying cause of discrepancies in the literature @xcite regarding the existence of the glass phase for the random rna . now that we have studied in great detail the behavior of random rna in the low and the high temperature phase , we describe its behavior at intermediate temperatures . to this end , we again study the pinch free energies @xmath261 defined in eq . ( [ eq_largestpinchfreeenergy ] ) , but this time over a large range of temperatures . we concentrate on the sequence disorder model eq . ( [ eq_energychoice ] ) with @xmath107 , and study sequences of lengths up to @xmath114 . from secs . [ sec_scaling ] , [ sec_hightnumerics ] , and [ sec_weakness ] , we know that the pinch free energy @xmath261 depends logarithmically on the sequence length @xmath7 at both low and high temperatures . indeed , this logarithmic behavior seems to hold for _ all _ temperatures studied . the data for each temperature can easily be fitted to the form @xmath272 the prefactor @xmath273 is found to depend on temperature in a non - monotonic way as shown in fig . [ fig_logfactors ] . the figure contains values of @xmath273 extracted by fitting the data for @xmath234 to the form eq . ( [ eq_logform ] ) . the uncertainty of this fit is on the order of the size of the symbols or smaller . for high temperatures , we find @xmath274 ( dashed line in fig . [ fig_logfactors ] ) as expected for the molten phase . at low temperatures , it starts from a finite value of the order @xmath62 and decreases linearly with temperature , as @xmath275 ( dotted line in fig . [ fig_logfactors ] ) . if we identify the glass transition temperature @xmath276 as the intersection of the dashed and the dotted lines , we get @xmath277 of the logarithmic dependence of @xmath278 on @xmath7 for random rna sequences generated by the sequence disorder model : at high temperatures , the prefactor indicated by the circle is well - described by the dashed line @xmath279 expected of the molten phase . at low temperatures , it again has a linear temperature dependence and is empirically fitted by the dotted line , @xmath280 . the numerical uncertainty in @xmath273 is of the order of or smaller than the size of the symbols . ] it is interesting to compare this estimate with the lower bound @xmath223 for the glass transition temperature given in sec . [ sec_proofglass ] . according to the consistency condition ( [ eq_consistencycondition ] ) , this lower bound is defined by @xmath281=\frac{3}{2}k_b t^*\ ] ] with @xmath282 . it is necessary to determine the temperature dependence of the quantity @xmath283 on the left hand side of this equation numerically . to do this , we measure the total free energy of each sequence generated . averaging these free energies over all the sequences of a given length @xmath7 and temperature @xmath170 , and dividing the results by the respective lengths @xmath7 , we obtain an estimate @xmath284 of the free energy per length which approaches the desired @xmath165 for large @xmath7 . : the data shown is taken at @xmath227 . for small @xmath7 it is dominated by the statistical fluctuations in the number of bases according to eq . ( [ eq_smallnf0 ] ) ( the dashed line ) . at large @xmath7 , it saturates to a positive constant . ] [ fig_ump2f0ofn ] shows how these estimates depend on the sequence length @xmath7 for the lowest temperature @xmath227 studied . instead of the free energy per length itself , the figure shows the fraction of unbound bases @xmath285 . for short sequences these estimates show a clear dependence on the sequence length @xmath7 this can be understood in terms of sequence - to - sequence fluctuations in the maximum number of possible pairings , due to fluctuations in the actual number of each type of bases present in a given sequence , even if all four bases are drawn with equal probability . this effect can be quantified by assuming that there is no frustration for small @xmath7 , i.e. , for any given sequence of the four bases @xmath286 , @xmath287 , @xmath151 , and @xmath288 , a secondary structure with the maximal number of watson - crick base pairs can be formed . if we denote by @xmath289 the number of times that the base @xmath290 appears in the sequence , the maximal number @xmath291 of pairings is given by @xmath292 . the fraction of unbound bases @xmath293 due to this effect can be computed straightforwardly approximating the multinomial distribution of @xmath294 by an appropriate gaussian distribution , with the result @xmath295 we expect this effect to be responsible for the increase in @xmath296 found in fig . [ fig_ump2f0ofn ] . indeed this effect , as indicated by the dashed line in the figure , explains the @xmath7 dependence of @xmath296 well for @xmath297 . however , we also see from the figure a clear saturation effect at large @xmath7 . this saturation reflects the finite fraction of unbound bases which is a frustration effect forced upon the system through the restriction on the type of allowed pairings in a allowed secondary structure . the unbound fraction @xmath298 is _ finite _ asymptotically as expected in sec . [ sec_proofglass ] . note that this value is _ remarkably small _ , as it implies that in the ground state structure of our toy random sequence , more than @xmath299 of the maximally possible base pairs are formed . but this is an artifact of the very simple energy rule used in our toy model . this fraction certainly will become smaller if the realistic energy rules are used , making the system more frustrated hence more glassy . in order to obtain the temperature dependence of the quantity @xmath283 on the left hand side of eq . ( [ eq_defoftstar ] ) , we will use its value at @xmath114 as an estimate of its asymptotic value . the results are shown in fig . [ fig_ump2f0 ] . the behavior at low temperatures can be described by a linearly decreasing function , shown as the dotted line in fig . [ fig_ump2f0 ] . according to eq . ( [ eq_defoftstar ] ) , the temperature @xmath223 is obtained as the intersection of this curve and @xmath300 , shown as the dashed line in fig . [ fig_ump2f0 ] for @xmath178 . we find @xmath301 which is consistent with the estimate ( [ eq_tgestimate ] ) , but is a rather weak bound . improved bounds on @xmath106 can be made by relaxing the condition of perfect complementarity of the two segments imposed in sec . [ sec_proofglass ] . this leads to larger values of the prefactor @xmath302 in eq . ( [ eq_defoftstar ] ) , hence a smaller slope for the dashed line in fig . [ fig_ump2f0 ] , and a larger value of @xmath223 . while the details of improved bounds will be discussed elsewhere , let us remark here that from fig . [ fig_ump2f0 ] it is clear that no matter what the slope of the dashed line becomes , we will never have @xmath223 larger than the temperature of @xmath303 where the quantity @xmath283 goes below zero . thus , these estimates will always be consistent with the observed glass transition temperature of @xmath304 . moreover , we note that the low temperature behavior @xmath305 as indicated by the dotted line in fig . [ fig_ump2f0 ] , appears to be roughly related to the behavior of @xmath273 ( dotted line in fig . [ fig_logfactors ] ) in the same temperature range , by a single scaling factor of approximately @xmath306 . thus , it is possible that @xmath307\ ] ] if it turns out that @xmath308 for @xmath309 . if this is the case , then it means the procedure we used to estimate the pinch energy in sec . [ sec_proofglass ] is quantitatively correct , implying that the ground state of a random rna sequence indeed consists of the matching of rare segments independently at each length scale . it will be useful to pursue this analysis further using a renormalization group approach similar to what was developed for the denaturation of heterogeneous dnas by tang and chat @xcite . : the symbols show numerical estimates of the quantity @xmath283 for different temperatures . the estimates are obtained by averaging the numerically determined free energy of @xmath113 random sequences of length @xmath114 generated by the sequence disorder model eq . ( [ eq_energychoice ] ) with @xmath107 . the low temperature behavior can be described reasonably well by the expression @xmath310 ( dotted line ) . the consistency condition ( [ eq_consistencycondition ] ) for the molten phase breaks down when this line intersects @xmath311 ( the dashed line ) . this yields @xmath312 as a lower bound for the glass transition temperature of this system . ] in this manuscript , we studied the statistical properties of random rna sequences far below the denaturation transition so that bases predominantly form base pairs . we introduced several toy energy models which allowed us to perform detailed analytical and numerical studies . through a two - replica calculation , we show that sequence disorder is perturbatively irrelevant , i.e. , an rna molecule with weak sequence disorder is in a molten phase where many secondary structures with comparable total energy coexist . a numerical study of the model at high temperature recovers scaling behaviors characteristic of the molten phase . at very low temperatures , a scaling argument based on the extremal statistics of rare matches suggest the existence of a different phase . this is supported by extensive numerical results : forced deformations are introduced by pinching distant monomers along the backbone ; the resulting excitation energies are found to grow very slowly ( i.e. , logarithmically ) with the deformation size . it is likely that the low temperature phase is a marginal glass phase . the intermediate temperature range is also studied numerically . the transition between the low temperature glass phase and the high temperature molten phase is revealed by a change in the coefficient of the logarithmic excitation energy , from being disorder dominated to entropy dominated . from a theoretical perspective , it would be desirable to find an analytical characterization of the low temperature phase . if the excitation energy indeed diverges only logarithmically , one has the hope that this may be possible , e.g. , via the replica theory , as was done for another well known model of statistical physics @xcite . it should also be interesting to include the spatial degrees of freedom of the polymer backbone ( via the logarithmic loop energy ) , to see how sequence disorder affects the denaturation transition . another direction is to include sequence _ design _ which biases a specific secondary structure , e.g. , a stem - loop @xcite . from a numerical point of view , it is necessary to perform simulations with realistic energy parameters to assess the relevant temperature regimes and length scales where the glassy effect takes hold . to make potential contact with biology , one needs to find out whether a molten phase indeed exists between the high temperature denatured phase and the low temperature glass phase for a real random rna molecule , and which phase the molecule is in under normal physiological condition . finally , it will be very important to perform _ kinetic _ studies to explore the dynamical aspects of the glass phase . despite the apparent weakness of the thermodynamic glassiness , the kinetics at biologically relevant temperatures is expected to be very slow for random sequences @xcite . the authors benefitted from helpful discussions with u. gerland and d. moroz . t.h . acknowledges an earlier collaboration with d. cule which initiated this study , and is indebted to l .- h . tang for a stimulating discussion during which the simple picture of sec . [ sec_proofglass ] emerged . this work is supported by the national science foundation through grants no . dmr-9971456 and dbi-9970199 . the authors are grateful to the hospitality of the institute for theoretical physics at uc santa barbara where this work was completed . before we describe the exact solution for the two - replica problem , as defined by the partition function @xmath141 in eq . ( [ eq_gintermsofloops ] ) and the bubble weight @xmath137 in eq . ( [ eq_defq ] ) , we first provide here a heuristic derivation of the qualitative results . this mainly serves to give a flavor of the two - replica problem in the language of theoretical physics . to this end , we define the quantity @xmath313 to be the partition function over all two replica configurations of a sequence of length @xmath314 under the constraint that base @xmath62 and base @xmath314 form a common bond . it is easy to see that @xmath315 where we set @xmath316 thus , the critical behavior of @xmath317 is identical to the critical behavior of @xmath313 which we will study in the following . due to the no pseudo knot constraint of the secondary structures , @xmath313 has a very simple structure , @xmath318 as illustrated in fig . [ fig_pirecurs ] . : while the first and last of the @xmath7 bases described by @xmath319 always form a base pair this outermost base pair can be followed by a loop with @xmath320 outgoing stems . each of the stems is described by @xmath321 itself while the loop is characterized by its total length that can be split into the different pieces @xmath135 in different ways . ] to simplify the above equation , it is useful to introduce the @xmath77-transforms @xmath322&=&\sum_{n=1}^\infty \pi(n ) e^{-\mu n}\\ \qhat[y]&=&\sum_{n=1}^\infty \tworq(n ) e^{-yn}\end{aligned}\ ] ] of @xmath321 and @xmath137 . now applying the @xmath77-transform to both sides of eq . ( [ eq_pi.1 ] ) , we obtain @xmath322e^{2\mu}\!\!\!&=&\!\!\ ! ! \int\!\ ! \frac{\mathrm{d}y}{2\pi i } \qhat[y ] \khat[y,\mu ] \big [ 1\ ! + \ ! \khat[y,\mu]\pihat[\mu]e^{2\mu}e^{-y } \label{eq_pi.2}\\\nonumber & & \hspace*{25 mm } + \ ! \left(\khat[y,\mu]\pihat[\mu]e^{2\mu}e^{-y}\right)^2 \!\!\!\!+\ldots \big]\end{aligned}\ ] ] where @xmath323 = \frac{1}{e^{-y+\mu}-1},\ ] ] and the inverse transform @xmath324 e^{y\ell}$ ] was used . ( [ eq_pi.2 ] ) can be simplified greatly to the following form , @xmath325e^{2\mu } = \qhat\cdot \int\frac{\mathrm{d}y}{2\pi i } \frac{\qhat[y]}{\khat^{-1}[y,\mu]-\pihat[\mu]e^{2\mu}e^{-y}}. \label{eq_pi}\ ] ] this is reminiscent of the well - known hartree solution to the @xmath326-theory , or equivalently the self - consistent treatment of the self - interacting polymer problem @xcite , if we identify @xmath327 as the interaction parameter , @xmath328 $ ] as the propagator " . the usual form of the hartree equation @xmath325 = \qhat\cdot \int \frac{\mathrm{d}^dk}{(2\pi)^d } \ , \frac{1}{k^2 + \mu -\pihat[\mu ] } \label{eq_pi1}\ ] ] corresponds to the small-@xmath329 , small-@xmath330 limit of eq . ( [ eq_pi ] ) , with @xmath331 playing the role of the square of the wave number " @xmath332 . note that @xmath333 $ ] plays the role of the density of ( spatial ) states , i.e. @xmath334 = \frac{\mathrm{d}^dk}{(2\pi)^d}$ ] , where @xmath335 denotes the _ dimensionality _ of the embedding space " . in the context of rna , de gennes used this approach to describe the denaturation of uniformly attracting rna more than 30 years ago @xcite . recently , this approach has been extended by moroz and hwa to study the phase diagram of rna structure formation @xcite . the analysis of a self - consistent equation of the type ( [ eq_pi ] ) is well known @xcite . the analytical properties of @xmath336 $ ] depend crucially on the form of @xmath333 $ ] . let the singular part of @xmath337 be @xmath338 where @xmath339 is the position of the singularity of @xmath333 $ ] . [ note that @xmath340 by comparing the forms of the hartree equations ( [ eq_pi ] ) and ( [ eq_pi1 ] ) . ] for @xmath341 , there is only one solution for all @xmath342 , with a square root singularity in @xmath336 $ ] at some finite value of @xmath330 . for @xmath343 , there are two possible solutions depending on the value of @xmath327 . the square - root singularity exists for @xmath327 exceeding some critical value depends through @xmath337 on @xmath124 but not on @xmath125 . ] @xmath344 , while for @xmath345 , the square - root singularity disappears and @xmath336 $ ] is governed by the singularity of @xmath337 given in eq . ( [ eq_qsing ] ) . performing the inverse transform and using eq . ( [ eq_gandpi ] ) , we get @xmath346 where @xmath144 is a non - universal parameter given by the location of the singularity , while the exponent @xmath347 characterizes the phase of the system and is given by the singularity of @xmath336 $ ] : we have @xmath348 if @xmath336 $ ] is dominated by the square - root singularity and @xmath349 if @xmath336 $ ] is dominated by @xmath337 . the interpretation of the two phases with @xmath348 and @xmath349 are straightforward : the phase with @xmath350 describes the usual rna secondary structure ( see eq . ( [ eq_moltenz ] ) ) ; here the bubbles described by @xmath137 are irrelevant . in the other phase , the result that @xmath349 indicates that base pairing is not relevant and the system behaves as a single bubble . in the context of the original two - replica problem , the irrelevancy of the bubbles in the @xmath350 phase indicates that the two replicas are locked together , behaving as a single replica in this phase . in the other phase , the attraction of the common bonds is irrelevant , and the two replicas become independent of each other . as explained in sec . [ sec_replica ] , the purpose of the two - replica calculation is to determine whether the inter - replica attraction , characterized by @xmath327 here , is irrelevant , i.e. , whether the system will not yet be in the @xmath350 phase for a value of @xmath351 . this is only possible if @xmath352 . from the solution of the problem described above , this depends crucially on the singularity of @xmath337 , specifically , on whether @xmath353 . the difficulty in ascertaining the form of @xmath337 lies in the no - common - bond constraint ( i.e. , @xmath354 ) in the definition of @xmath137 ( [ eq_defq ] ) . however , we note that for @xmath355 , the two replica partition function @xmath356 is simply the square of the single replica partition function @xmath74 . thus , @xmath357 according to eq . ( [ eq_moltenz ] ) . since we just convinced ourselves that @xmath347 can take on only two possible values , namely @xmath358 and @xmath359 and since @xmath360 we conclude @xmath361 and moreover @xmath362 . thus , we do expect the phase transition to occur at @xmath363 . however , it is not clear from this calculation if the system at @xmath355 ( or @xmath364 ) is exactly at or strictly below the phase transition point . we leave it to the exact solution of the two replica problem presented in the next two appendices to establish that @xmath364 is indeed strictly below the phase transition point and that therefore disorder is perturbatively irrelevant . we note that in the context of the @xmath326-theory or the self - consistent treatment of the self - interacting polymer , the result @xmath365 implies that the embedding spatial dimension is @xmath366 . thus , the two - replica problem corresponds to the denaturation of a single rna in @xmath0 spatial dimensions . the bubbles @xmath137 s of fig . ( [ fig_combonds ] ) which originate from the branching entropy of the individual rnas play the role of spatial configurational entropy of the single - stranded rna in the denaturation problem . in this appendix , we present the exact solution of the two replica problem . while most of the details are given here , the most laborious part is further relegated to app . [ app_derivez1 ] . we start by introducing an auxiliary quantity @xmath367 . this is a restricted two - replica partition function , summing over all independent secondary structures of a pair of rnas of length @xmath314 bases in which there are exactly @xmath368 exterior bases of the common bond structure without disrespecting the no pseudo - knot constraint . ] all of which are completely unbound in both replicas . since the exterior bases form one of the bubbles of the common bond structure , the possible binding configurations of these exterior bases are described by @xmath369 . thus , the full partition function of the two replica problem can be calculated from this restricted partition function as @xmath370 now , let us formulate a recursion relation for @xmath371 by adding one additional base @xmath7 to each of the two rnas . we can separate the possible configurations of the new function @xmath372 according to the possibilities that the new base @xmath7 is either not involved in a common bond or forms a common bond with base @xmath373 . this yields the recursion relation @xmath374 for @xmath375 and @xmath376 . the applicable boundary conditions are : @xmath377 , @xmath378 , and @xmath379 for each @xmath380 and @xmath381 . at this point , it is convenient to introduce the @xmath77-transforms in order to decouple the discrete convolution in eq . ( [ eq_trrecurs ] ) . they are @xmath382 using eq . ( [ eq_trrecurs ] ) and the boundary conditions we get @xmath383 this can be solved for @xmath384 with the result @xmath385 together with the boundary condition @xmath386 , we get @xmath387^{-n}.\ ] ] if we now multiply eq . ( [ eq_gbywandq ] ) by @xmath388 and sum both sides over @xmath7 we get @xmath389 which , upon inserting eq . ( [ eq_hatw ] ) , becomes an implicit equation @xmath390 for the full partition function @xmath391 , provided that we know the function @xmath392 . since @xmath392 does not depend on @xmath125 , we can find its form using the following strategy : if @xmath355 , a common bond does not contribute any additional boltzmann factor . thus , the two replica partition function for this specific value of @xmath125 is just the square of the partition function of a single uniformly attracting rna molecule , i.e. , @xmath393 since we know @xmath74 through the exact expression ( [ eq_singlez ] ) for its @xmath77-transform @xmath394 , we can regard @xmath395 as a known function , even though a closed form expression is not available . from eq . ( [ eq_trimplicit ] ) , we have @xmath396 this is an equation for @xmath392 in terms of the known function @xmath395 . after we solve it for @xmath392 below , we can use eq . ( [ eq_trimplicit ] ) to solve for the only leftover unknown @xmath391 for arbitrary values of @xmath125 . in the thermodynamic limit , it is sufficient to consider only the singularities of the @xmath77-transform @xmath391 . from the form of @xmath397 in the vicinity of the singularity @xmath398 , the two - replica partition function @xmath317 is readily obtained by the inverse @xmath77-transform , with the result @xmath399 the result immediately yields the free energy per length , @xmath400 . more significantly , the exponent @xmath143 reveals which phase the two - replica system is in : for @xmath355 ( i.e. no disorder ) , the two - replica system is just a product of two independent single - replica systems and we must have @xmath148 as implied by the single - replica partition function @xmath74 in eq . ( [ eq_moltenz ] ) . on the other hand , for @xmath401 , the two replicas are forced to be locked together and behave as a single replica . in this case , we must have @xmath402 . as we will see , @xmath148 and @xmath403 are the only values this exponent can take on for this system ; it indicates whether or not the two replicas are locked , and hence whether or not the effect of disorder is relevant . the singularity @xmath398 of @xmath391 is given implicitly by eq . ( [ eq_trimplicit ] ) , which we now analyze in detail . we start by recalling the solution of the homogeneous single rna problem , eq . ( [ eq_moltenz ] ) . from the relation ( [ eq_g2 ] ) , we have @xmath404 for large @xmath7 , with @xmath82 , @xmath405 , and @xmath406 given in sec . [ sec_moltenpartfunc ] . hence , the @xmath77-transform @xmath407 is defined on the interval @xmath408 . it is a monotonously decreasing function of @xmath77 , terminating with a singularity at @xmath409 which produces the @xmath148 singularity in @xmath356 . due to ( [ eq_trimplicitqt1 ] ) , the same singularity must occur in @xmath410 at @xmath411 , where @xmath412 is a positive number and does not depend on anything else but @xmath124 . since @xmath413 is a smooth monotonously increasing function which maps the interval @xmath408 into the interval @xmath414 , it follows from eq . ( [ eq_trimplicitqt1 ] ) that @xmath410 is a smooth , monotonously decreasing function which maps the interval @xmath414 into the interval @xmath4150,\gatsing]$ ] . now that we have characterized @xmath410 in detail , we can proceed to study @xmath391 for arbitrary @xmath125 . clearly , according to eq . ( [ eq_trimplicit ] ) @xmath391 has a singularity leading to @xmath148 at @xmath416 , defined implicitly by @xmath417 because @xmath410 has this singularity at @xmath418 . again according to eq . ( [ eq_trimplicit ] ) , we have @xmath419 independent of @xmath125 . this leads to one of the key results @xmath420 if @xmath416 is the only singularity of @xmath391 , it would imply that there is only one phase with @xmath148 , and the free energy per length of the two - replica system is given by @xmath421\ ] ] for all values of @xmath125 . by differentiating this with respect to @xmath125 , we obtain the fraction of common contacts @xmath422 in this phase as a function of @xmath125 . for very large disorder , i.e. , for large @xmath125 , this fraction converges to one . however , since it is the fraction of bonds divided by the total number of bases and every base pair has two bases , it has to be bounded from above by @xmath423 . thus , we conclude that eq . ( [ eq_freeenergy0 ] ) can not be the free energy of the two - replica system for all @xmath125 s . at least for large @xmath125 , there must be another singularity of @xmath391 which will yield a different expression for the free energy to give physically reasonable fraction of common bonds . in order to find this other singularity , we introduce the inverse function @xmath424 of @xmath391 . from eq . ( [ eq_trimplicit ] ) , it follows that for any @xmath125 and any @xmath4250,\gatsing]$ ] , @xmath426\\ \rightarrow\quad g&= & \widehat{\tworq}[\widehat{\tworz}(g;\widetilde{q})-q^2\widetilde{q } g]\\ \rightarrow\quad \widehat{\tworq}^{-1}(g)&= & \widehat{\tworz}(g;\widetilde{q})-q^2\widetilde{q } g.\end{aligned}\ ] ] since @xmath427 does not depend on @xmath125 , we can eliminate it by evaluating the last equation above at the special value @xmath355 and write @xmath428 where @xmath429 is the inverse of the known function @xmath395 . ( [ eq_zrelation ] ) is now an explicit solution for the inverse of @xmath391 for _ arbitrary _ @xmath125 , and the singularity of @xmath430 , located at @xmath431 and @xmath432 , yields the free energy of the two - replica system , i.e. , @xmath433 , in this phase . while app . [ app_derivez1 ] derives the position of the dominant singularity present in eq . ( [ eq_zrelation ] ) rigorously , we will resort to some intuitive argument here . since @xmath434 is a monotonously decreasing , convex function , so is its inverse @xmath429 . the latter function ends at the point @xmath435 with some slope @xmath436 in a singularity which produces the @xmath148 behavior indicative of two independently fluctuating uniformly self - attracting replicas . this is shown in fig . [ fig_zhatsketch ] as the solid line . according to eq . ( [ eq_zrelation ] ) , we can obtain the corresponding function @xmath424 for arbitrary @xmath125 by simply adding a linear function @xmath437 to this function . if the slope @xmath438 of this linear function is less than the smallest slope of @xmath429 , i.e. , if @xmath439 , adding this linear function does not qualitatively change anything ( see the short dashed line in fig . [ fig_zhatsketch ] . ) the only singularity is still the one at @xmath440 , the corresponding value in @xmath77 , i.e. , @xmath441 , is trivially shifted by an amount @xmath442 from @xmath443 as @xmath125 varies . thus , the scaling behavior is characterized by @xmath148 as if @xmath355 ( i.e. , absence of disorders ) , although the free energy @xmath444 is shifted as already derived in eq . ( [ eq_freeenergy0 ] ) . of the two replica system at various values of the common bond interaction @xmath125 : the solid line shows the free system without any interaction of common bonds . in the presence of an interaction , the inverse function of the partition function can be obtained by adding a linear function to the free system function . if the interaction is not too strong ( @xmath445 , short dashed line ) adding the linear function with a small slope does not change the qualitative form of the partition function . in this case , the two replica system is controlled by the singularity at @xmath440 which is independent of @xmath125 . at stronger interactions ( @xmath446 , long dashed line ) the inverse function develops a minimum . beyond this minimum it is not invertible any more and the two replica system is then dominated by the singularity arising from this minimum . ] if the final slope @xmath447 of the right hand side of eq . ( [ eq_zrelation ] ) is _ positive _ the situation becomes much different : upon adding the linear function , @xmath424 develops a minimum at some position @xmath448 . thus , the inverse function @xmath391 has to be calculated from the left ( small @xmath449 ) branch of @xmath424 and has a _ square root singularity _ at @xmath450 . this square root singularity implies that the characteristic exponent becomes @xmath403 , consistent with the picture that in this phase the two replicas are locked together and fluctuate as one single effective rna molecule . the position @xmath451 of the minimum is determined by the root of the derivative , i.e. , by @xmath452^{-1}\!\!\!\!\!\!\!\!.\ ] ] the corresponding value @xmath453 determines the location of the square root singularity @xmath450 of @xmath391 , i.e. , the free energy per length of the two replica problem . @xmath450 can be conveniently expressed in terms of the auxiliary quantity @xmath454 defined through @xmath455 as @xmath456 comparing this expression with eq . ( [ eq_freeenergy0 ] ) which is valid for small @xmath125 , we can summarize the complete solution in terms of @xmath457\singlezsing^2,\infty[$ that ful-}\ & \widetilde{q}>\widetilde{q}_{\mathrm{c}}\\ \mbox{fils } \frac{\mathrm{d}}{\mathrm{d}z}\widehat{\tworg}(z;1 ) \!=\!\!-\!1\!/[q^2(\widetilde{q}\!-\!1)]&\\[10pt ] \singlezsing^2=(1 + 2\sqrt{q})^2&\widetilde{q}\le\widetilde{q}_{\mathrm{c } } \end{array}\right.\ ] ] where @xmath458 and @xmath459 in terms of this @xmath460 , the smallest singularity of @xmath461 is located at @xmath462 the free energy per length of the two replica system is given by @xmath463\ ] ] and the fraction of common bonds is @xmath464 this fraction of common bonds turns out to be continuous at the phase transition but it exhibits a jump in its slope at @xmath465 . in the case @xmath466 , these simplify to eqs . ( [ eq_freeenergy0 ] ) and ( [ eq_contacts0 ] ) ( or eqs . ( [ eq_moltenzinmaintext ] ) and ( [ eq_defginmaintext ] ) respectively ) , with @xmath467 independent of @xmath125 . the type of singularity of the laplace transformed partition function is the same as at @xmath355 , resulting in @xmath148 . for @xmath446 , we can not write down a closed form expression for @xmath398 any more . but it is given implicitly in terms of the solution of eq . ( [ eq_zkdef ] ) ; it involves only single replica quantities and can thus be evaluated numerically . moreover , we have seen that the dominant singularity is in this regime a square root singularity implying @xmath403 . in this appendix we give a derivation of the position of the non - trivial singularity in the laplace transform of the partition function @xmath391 . a more intuitive , graphical derivation of this result was given in app . [ app_solution ] . using eq . ( [ eq_zrelation ] ) , we start by calculating @xmath468^{-1}\\ & = & \left[\left.\frac{\mathrm{d}}{\mathrm{d}g } \right|_{g=\widehat{\tworg}(z;\widetilde{q})}\!\!\!\ ! \widehat{\tworz}(g;1)+q^2(\widetilde{q}-1 ) \right]^{-1}\\ & = & \left[\left[\left.\frac{\mathrm{d}}{\mathrm{d}z } \right|_{z=\widehat{\tworz}(\widehat{\tworg}(z;\widetilde{q});1)}\!\!\!\!\widehat{\tworg}(z;1 ) \right]^{-1}\!\!\!\!\!\!\!\!\!+q^2(\widetilde{q}-1)\right]^{-1}\!\!\!\!.\end{aligned}\ ] ] this expression obviously has a singularity at @xmath450 which is defined by @xmath469 since @xmath470 , this is only possible for @xmath471 for smaller values of @xmath125 , there is no other singularity and the free energy per length is given by eq . ( [ eq_freeenergy0 ] ) . if @xmath472 the additional singularity @xmath450 exists and is as we will see below always smaller than the singularity @xmath473 . thus , the free energy per length is given by the singularity @xmath450 in the strong coupling phase , i.e. , for @xmath472 . on the first sight , eq . ( [ eq_z1def ] ) still looks as if @xmath450 could only be calculated if the full function @xmath391 is known . however , for any @xmath472 we can define @xmath454 to be the unique solution of the equation @xmath474 this quantity depends only on the function @xmath434 . according to eq . ( [ eq_z1def ] ) , @xmath450 and @xmath454 are related by @xmath475 . this implies that @xmath476 . on the other hand , eq . ( [ eq_zrelation ] ) applied to @xmath477 yields @xmath478 which is finally an expression which involves only quantities of the non - interacting system . since @xmath479 is a monotonous function on the interval @xmath480 $ ] , we always have @xmath481 with equality if and only if @xmath467 , i.e. , for @xmath482 . therefore , the free energy per length is indeed given by @xmath483\ ] ] for any @xmath472 . 99 see , e.g. , w. saenger , _ principles of nucleic acid structure _ ( springer - verlag , new york , 1984 ) or b. alberts _ et al . _ , _ molecular biology of the cell _ ( garland publishing , new york , 1994 ) . see , e.g. , r.f . gesteland and j.f . atkins _ eds . _ , _ the rna world : the nature of modern rna suggests a prebiotic rna world _ ( cold spring harbor laboratory press , cold spring harbor , 1993 ) . k.a . dill _ et al . _ , _ protein sci . _ * 4 * , 561 ( 1995 ) . onuchic , z. luthey - schulten , and p.g . wolynes , _ annu . chem . _ * 48 * , 545 ( 1997 ) . t. garel , h. orland , and e. pitard , _ j. phys . i _ * 7 * , 1201 ( 1997 ) . shakhnovich , _ curr . op . biol . _ * 7 * , 29 ( 1997 ) . for a review see p.g . higgs , _ quart . biophys . _ * 33 * , 199253 ( 2000 ) . m. zuker and p. stiegler , _ nucl . res . _ * 9 * , 133148 ( 1981 ) . mccaskill , _ biopolymers _ * 29 * , 11051119 ( 1990 ) . hofacker , w. fontana , p.f . stadler , s. bonhoeffer , m. tacker , and p. schuster , _ monatshefte f. chemie _ * 125 * , 167188 ( 1994 ) . freier , r. kierzek , j.a . jaeger , n. sugimoto , m.h . caruthers , t. neilson , and d.h . turner , _ proc . usa _ * 83 * , 93739377 ( 1986 ) . i. tinoco jr and c. bustamante , _ biol . _ * 293 * 271281 ( 1999 ) and references therein . b. maier , d. bensimon , and v. croquette , _ proc . usa _ * 97 * , 1200212007 ( 2000 ) . de gennes , _ biopolymers _ * 6 * , 715 ( 1968 ) . higgs , _ phys . lett . _ * 76 * , 704707 ( 1996 ) . r. bundschuh and t. hwa , _ phys . lett . _ * 83 * , 14791483 ( 1999 ) . a. pagnani , g. parisi , and f. ricci - tersenghi , _ phys . * 84 * , 20262029 ( 2000 ) . hartmann , _ phys . _ * 86 * , 13821382 ( 2001 ) . a. pagnani , g. parisi , and f. ricci - tersenghi , _ phys . rev . lett . _ * 86 * , 13831383 ( 2001 ) . see , e.g. , p. young _ ed . _ , _ spin glasses and random fields _ ( world scientific , singapore , 1998 ) . d. moroz and t. hwa , in preparation . waterman , _ advan . in math . studies _ * 1 * , 167212 ( academic press , new york , 1978 ) . see , e.g. , m. mitas , _ nucl . acids res . _ * 25 * , 2245 - 2254 ( 1997 ) . see , e.g. , w. feller , _ an introduction to probability theory and its applications _ ( john wiley & sons , new york , 1950 ) . lubensky , j. isaacson , and s.p . obukhov , _ j. physique _ * 42 * , 15911601 ( 1981 ) . s.f . edwards and p.w . anderson , _ _ * f5 * , 965 ( 1975 ) . see , e.g. , m. mzard , g. parisi , and m.a . virasoro , _ spin glass theory and beyond _ ( world scientific , singapore , 1986 ) . r. arratia , l. gordon , and m. waterman , _ ann . stat . _ * 18 * , 539570 ( 1990 ) . d.a . huse and d.s . j. physique i _ * 1 * , 621625 ( 1991 ) . morgan and p.g . * 105 * , 71527157 ( 1996 ) . morgan and p.g . j. phys . a : math . gen . _ * 32 * , 31533170 ( 1998 ) . c. flamm , w. fontana , i.l . hofacker , and p. schuster , _ rna _ * 6 * , 325338 ( 2000 ) . h. isambert and e.d . siggia , _ proc . . usa _ * 97 * , 65156520 ( 2000 ) . u. gerland , r. bundschuh , and t. hwa , in preparation . cardy and s. ostlund , _ phys . b _ * 25 * , 68996909 ( 1982 ) . s.e . korshunov , _ phys . b _ * 48 * , 39693975 ( 1993 ) . t. hwa and d. fisher , _ phys . rev . b _ * 49 * , 31363154 ( 1994 ) . tang and h. chat , _ phys . lett . _ * 86 * , 830 ( 2001 ) . h. isambert , personal communication . j. zinn - justin , _ quantum field theory and critical phenomena _ ( oxford university press , united kingdom , 1989 ) .

nonlinearities are important in optical systems since they enable us to manipulate light by light . especially , nonlinearities sensitive to single photons are necessary for quantum optical applications such as optical quantum computation @xcite , entanglement generation @xcite , and quantum non - demolition measurements @xcite . in particular , strong nonlinearities are indispensable for quantum computation , since conditional operations at the single photon level are required to realize elementary gate operations . however , conventional optical media require very high photon densities to obtain significant nonlinear effects @xcite . single atom cavity quantumelectrodynamics may offer a possible solution @xcite . in particular , a cavity can focus all the light in an input beam on a single atom , effectively creating a situation where a single atom is coupled to a one - dimensional light field @xcite . such a one - dimensional atom is an ideal device for the non - linear manipulation of few photon pulses . it is therefore of great interest to investigate the response of a one - dimensional atom to one- and two - photon input pulses . there are many theories for such atom - cavity systems @xcite . however these theories usually eliminate the quantum state of the field outside the atom - cavity system and therefore the analysis of the precise spatiotemporal coherence of the input and output photon pulses can not be achieved . in our previous work , we have therefore presented a theory for one- and two - photon pulses that includes the propagation to and from the system in a bad - cavity regime @xcite , based on a one - dimensional model of light - atom interaction @xcite . in these studies , we focused on the changes of the photon - photon correlations in time @xcite and in frequency @xcite . in this paper , we study the effect of the non - linear interaction with the one - dimensional atom on the shape of one- and two - photon wavefunctions using gaussian input pulses of different pulse lengths . for this purpose , we first introduce an analysis of the interaction in terms of absorption and transmission processes in section [ sec : level2 ] . the characteristic time scale defining the response of the atom is given by the dipole relaxation time @xmath0 . in section [ sec : level3 ] , we then analyze the interaction processes for a long pulse length of @xmath1 in terms of pulse amplitude and pulse delay time . from these results , we can estimate that a maximal non - linear effect should be obtained for a pulse length equal to the dipole relaxation time of the atom . in section [ sec : level4 ] , we therefore analyze the corresponding interaction processes for a short pulse length of @xmath0 , confirming the dominance of the non - linear effect throughout the short two - photon pulse . the results of the analysis are summarized and discussed in section [ sec : level5 ] . the model of a single two - level atom in one - dimensional free space is shown in figure . [ fig : figure1](a ) . in this model , an incoming one - photon or two - photon wavepacket interacts with the atom locally , and the output wavepacket is then emitted into the output field . the physical realization of this model can be implemented by a two - level atom coupled with a single mode of a one - sided cavity in the bad cavity regime @xcite . the cavity geometry is shown in figure . [ fig : figure1 ] ( b ) . the input field of the one - dimensional free space shown in figure . [ fig : figure1 ] ( a ) corresponds to the input of the one - sided cavity in figure . [ fig : figure1 ] ( b ) . likewise , the output field in figure . [ fig : figure1 ] ( a ) corresponds to the output of the one - sided cavity in figure . [ fig : figure1 ] ( b ) . here , we assume that the spontaneous emission rate through the cavity mode is much larger than the spontaneous emission rate through non - cavity modes . in terms of the conventional cavity quantum electrodynamics parameters , this regime is characterized by @xmath2 , where @xmath3 is the cavity damping rate through the left mirror , @xmath4 is the dipole coupling between the atom and the cavity mode and @xmath5 is the rate of spontaneous emission into the non - cavity modes . since the cavity damping rate @xmath3 is much faster than the dipole coupling @xmath4 , the method of adiabatic elimination can be applied to the time evolution of the cavity field @xcite . this means that the interaction between the atom and the outside field through the cavity field can be expressed by an effective dipole relaxation rate @xmath6 . the dipole relaxation rate @xmath7 describes the dipole damping caused by emissions through the left mirror of the cavity , and the corresponding rate of spontaneous emission through the cavity is equal to @xmath8 @xcite . in our case , we assume that the rate of spontaneous emission into the non - cavity modes @xmath5 is negligible ( @xmath9 ) . nearly all emissions from the atom can then be confined to the cavity and 2@xmath7 is the total spontaneous emission rate of the excited atom in the cavity . in present cavity designs , this can be realized by exploiting the purcell effect . for example , in the case of turchette et al.s experiment @xcite , the cavity parameters indicate that about 80% of the sponaneous emission from the atom is emitted through the single cavity mode . ( 0,0 ) ( -10,160)(a ) ( 270,160)(b ) ( a ) schematic of a two - level atom in one - dimensional free space and ( b ) physical realization of the one - dimensional atom using an atom - cavity system . an incoming one - photon or two - photon wavepacket interacts with the atom locally , and the output wavepacket is then emitted into the output field . the input field of the one - dimensional free space shown in ( a ) corresponds to the input of the one - sided cavity in ( b ) . likewise , the output field of the one - dimensional free space shown in ( a ) corresponds to the output of the one - sided cavity in ( b).,title="fig:",width=340 ] ( a ) schematic of a two - level atom in one - dimensional free space and ( b ) physical realization of the one - dimensional atom using an atom - cavity system . an incoming one - photon or two - photon wavepacket interacts with the atom locally , and the output wavepacket is then emitted into the output field . the input field of the one - dimensional free space shown in ( a ) corresponds to the input of the one - sided cavity in ( b ) . likewise , the output field of the one - dimensional free space shown in ( a ) corresponds to the output of the one - sided cavity in ( b).,title="fig:",width=245 ] in the following , we investigate the effect of the one - dimensional atom on one and two - photon input pulses of different pulse length . for this purpose , we have to formulate the quantum states for a given input pulse shape for both one photon and two photon cases . if the real space representation for the one - photon state reads , @xmath10 the two - photon state can be expressed as the product state @xmath11 the light pulses characterized by the above input states are propagating at the velocity of light in the input field of the schematic shown in figure . [ fig : figure1](a ) . to simplify the expression , the spatial coordinates , @xmath12 and @xmath13 , therefore represent a coordinate system moving at the velocity of light . the indices 1 and 2 of @xmath12 and @xmath13 in eq . ( [ eq : spatialinput ] ) identify the two particles . the spatial features of the probability amplitudes given by the wavefunction @xmath14 correspond to the spatial features of the pulse mode a. thus , the wavefunction @xmath15 characterizes the single mode two - photon wavefunction , where the wavefunctions of both particles overlap perfectly . in our previous work @xcite , we derived the unitary time evolution operator in the hilbert spaces for the one photon component and the two photon component by solving the schrdinger equation of the interaction between a one - dimensional field and a single atom . the output one - photon state in the far field can then be obtained by convoluting the one - photon input wavefunction @xmath16 with the matrix description of the unitary operator in real space , @xmath17 @xmath18 the matrix elements @xmath19 are given by @xmath20 likewise , the output two - photon state in the far field can be obtained by a linear convolution , @xmath21 @xmath22 the matrix elements @xmath23 are given by @xmath24 where the contribution of the non - linear photon - photon interaction is given by @xmath25 . \label{eq : nonlineffect}\end{aligned}\ ] ] the output wavefunctions @xmath26 and @xmath27 describe the spatial feature of the output states in the far field of the atom shown in figure . [ fig : figure1 ] ( a ) . note that the spatiotemporal features of the input one - photon and two - photon wavefunctions are generally not preserved in the output due to the atomic response described by the matrix elements @xmath28 and @xmath29 . the matrix element of the time evolution @xmath28 in eq.([eq : onemat ] ) can be expanded in terms of two interaction processes , @xmath30 where the two processes are ( i ) single photon transmission without absorption , given by @xmath31 and ( ii ) single photon reemission after absorption , given by @xmath32 the output one - photon wavefunction @xmath26 can thus be interpreted as the result of quantum interference of the two processes , @xmath33 where the components are given by @xmath34 in the same way , the matrix element of the time evolution @xmath35 can be expanded in terms of three interaction processes . process ( 1 ) is the transmission of both photons without absorption . process ( 2 ) is the transmission of one photon without absorption and the absorption and reemission of the other photon . process ( 3 ) is the absorption and reemission of both photons . in the following , we will refer to process ( 1 ) as the transmission process , to process ( 2 ) as the one photon absorption process , and to process ( 3 ) as the two photon absorption process . the transformation of the two - photon wavefunction can then be written as @xmath36 the components of @xmath35 read @xmath37 note that the non - linear contribution @xmath38 only occurs in the two photon absorption process . this is because the non - linearity is due to the saturation of the two - level atom , which prevents the absorption of two photons at the same time . likewise , the output wavefunction @xmath39 can be expanded as @xmath40 where the @xmath41 are given by @xmath42 this analysis will be convenient to understand the origin of the statistical properties of the output photons . in the next section , these results are applied to the cases of gaussian one - photon and two - photon input wavepackets . the dependence of the coherent spatiotemporal correlations between the output photons on the input pulse length is analyzed numerically using the results in the previous subsection . for our analysis , we use gaussian one - photon and two - photon input wavefunctions , @xmath43 the pulse length @xmath44 is defined as twice the dispersion of the gaussian distribution of the photons in time . note that the response of the atom is governed by the time scale @xmath0 , therefore the atom responds differently to long pulses and short pulses . in the following , the term long pulse refers to a pulse whose pulse length @xmath44 is much longer than the dipole relaxation time @xmath0 . likewise , the term short pulse refers to a pulse whose pulse length @xmath44 is about equal to the dipole relaxation time @xmath0 . in the next section , we analyze the output wavefunctions for long one - photon and two - photon pulses in order to identify the characteristic quantum interference effects . [ fig : figure2 ] ( a ) shows an example of a one - photon long pulse ( broken line ) and the corresponding output wavefunction ( solid line ) . in this example , we have chosen an input pulse length of @xmath45 which is 10 times greater than the dipole relaxation time @xmath0 . [ t][t]@xmath46 [ b][b]@xmath47 [ b][b]@xmath48 ( 0,0 ) ( 0,200)(a ) gaussian input and output one - photon wave packets for an input wave packet length of @xmath45 . the horizontal axes @xmath49 represents the spatial coordinate of one photon . in ( a ) , the broken line corresponds to the input wavefunction @xmath16 and the solid line corresponds to the output wavefunction @xmath26 . in ( b ) , the dotted line corresponds to the component of the process ( i ) : single photon transmission without absorption and the thin line corresponds to the component of the process ( ii ) : single photon reemission after absorption . ] ( 0,0 ) ( 0,200)(b ) gaussian input and output one - photon wave packets for an input wave packet length of @xmath45 . the horizontal axes @xmath49 represents the spatial coordinate of one photon . in ( a ) , the broken line corresponds to the input wavefunction @xmath16 and the solid line corresponds to the output wavefunction @xmath26 . in ( b ) , the dotted line corresponds to the component of the process ( i ) : single photon transmission without absorption and the thin line corresponds to the component of the process ( ii ) : single photon reemission after absorption . ] the pulse shape of the output one - photon pulse is almost identical with the input pulse except for the @xmath50 phase flip in the output amplitudes compared to the input amplitudes , @xmath51 . furthermore the minimum of the output amplitude is delayed compared to the maximum of the input amplitudes . the delay time is about @xmath52 . let us now analyze the origin of these features . figure . [ fig : figure2 ] ( b ) shows the components corresponding to different light - atom interaction processes as given by eqs.([eq : oneoutputcomponents ] ) . the dotted line is the component of single photon transmission without absorption , @xmath53 ( see eqs.([eq : oneoutputcomponents ] ) ) . the thin line is the component of single photon absorption , @xmath54 . since the absorption process convolutes the input amplitude with an exponential function according to eqs . ( [ eq : oneoutputcomponents ] ) , this component is not a precise gaussian anymore . however , in the long pulse limit , the convolution function is so much narrower than the gaussian that the change in pulse shape is negligibly small . the component @xmath54 can then be approximated as @xmath55 as this approximation indicates , the amplitude of the absorption process is about two times as large as the amplitude of the transmission process . it is then possible to understand the @xmath50 phase flip in the output as a result of interference , @xmath56 where the negative amplitude of the absorption process is twice as high as the positive amplitude of the transmission process . however this approximation can not explain the delay of the peak of the output wavefunction in figure . 2 ( a ) . in the long pulse limit , the delay time of @xmath0 in @xmath54 can be included in the approximation given in eq.([eq : appabs ] ) by expanding @xmath54 as follows , @xmath57 the delay time of @xmath26 can then be obtained from the interference of @xmath58 and @xmath54 , @xmath59 the delay time of the output wavefunction is therefore equal to @xmath52 , as seen in figure . [ fig : figure2 ] ( a ) . let us now consider the output wavefunction for the long two photon pulse . the output wavefunction can be analyzed in the same way as the one photon case , using the three interaction processes discussed in section [ sec : level2].3 . however , there are now two photon coordinates , @xmath12 and @xmath13 , and , as shall be seen in the following , the distance @xmath60 between the two photons significantly influences the two photon absorption process . note also that the bosonic symmetry of the indistinguishable photons requires that @xmath61 . figure . [ fig : figure3 ] ( a ) shows the contour plot of a long two - photon input pulse @xmath15 . the probability amplitude increases from black to white shading . as before , the input pulse has a length of @xmath45 . figure . [ fig : figure3 ] ( b ) shows the cross - section of the input wavefunction at @xmath62 for time differences of @xmath63 between the two photons . note that the horizontal axis gives the average position of the two photons . the position of the peaks then defines the average delay of both photons at a given distance of @xmath64 between the two photons . [ t][t]@xmath65 [ b][b]@xmath66 [ t][t]@xmath67 [ ] [ ] @xmath68 [ tr][tr]20 [ tr][tr]10 [ tr][tr]0 [ tr][tr]-10 [ tr][tr]-20 ( 0,0 ) ( -5,200)(a ) input and output wavefunction for a two - photon input pulse with a pulse length of @xmath45 . ( a ) is a contour plot of the input wavefunction @xmath15 . ( b ) is the cross section of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath70 ( broken line ) and @xmath71 ( dotted line ) . the horizontal axis @xmath72 in ( b ) represents the average position of the two photons . likewise , ( c ) is a contour plot of the output wavefunction @xmath39 and ( d ) shows cross sections of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath70 ( broken line ) , and @xmath71 ( dotted line ) . ] ( 0,0 ) ( -20,200)(b ) input and output wavefunction for a two - photon input pulse with a pulse length of @xmath45 . ( a ) is a contour plot of the input wavefunction @xmath15 . ( b ) is the cross section of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath70 ( broken line ) and @xmath71 ( dotted line ) . the horizontal axis @xmath72 in ( b ) represents the average position of the two photons . likewise , ( c ) is a contour plot of the output wavefunction @xmath39 and ( d ) shows cross sections of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath70 ( broken line ) , and @xmath71 ( dotted line ) . ] [ t][t]@xmath67 [ ] [ ] @xmath73 [ tr][tr]20 [ tr][tr]10 [ tr][tr]0 [ tr][tr]-10 [ tr][tr]-20 ( 0,0 ) ( -5,200)(c ) input and output wavefunction for a two - photon input pulse with a pulse length of @xmath45 . ( a ) is a contour plot of the input wavefunction @xmath15 . ( b ) is the cross section of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath70 ( broken line ) and @xmath71 ( dotted line ) . the horizontal axis @xmath72 in ( b ) represents the average position of the two photons . likewise , ( c ) is a contour plot of the output wavefunction @xmath39 and ( d ) shows cross sections of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath70 ( broken line ) , and @xmath71 ( dotted line ) . ] ( 0,0 ) ( -20,200)(d ) input and output wavefunction for a two - photon input pulse with a pulse length of @xmath45 . ( a ) is a contour plot of the input wavefunction @xmath15 . ( b ) is the cross section of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath70 ( broken line ) and @xmath71 ( dotted line ) . the horizontal axis @xmath72 in ( b ) represents the average position of the two photons . likewise , ( c ) is a contour plot of the output wavefunction @xmath39 and ( d ) shows cross sections of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath70 ( broken line ) , and @xmath71 ( dotted line ) . ] ( 0,0 ) ( -5,35)(e ) ( -5,-15)-0.35 ( 190,-15)0.3 ( 110,-15)0 input and output wavefunction for a two - photon input pulse with a pulse length of @xmath45 . ( a ) is a contour plot of the input wavefunction @xmath15 . ( b ) is the cross section of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath70 ( broken line ) and @xmath71 ( dotted line ) . the horizontal axis @xmath72 in ( b ) represents the average position of the two photons . likewise , ( c ) is a contour plot of the output wavefunction @xmath39 and ( d ) shows cross sections of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath70 ( broken line ) , and @xmath71 ( dotted line).,height=37 ] figure . [ fig : figure3 ] ( c ) and figure . [ fig : figure3 ] ( d ) show the output wavefunction @xmath39 . a remarkable new feature of this output wavefunction is the negative amplitude region around @xmath74 . the output amplitude changes from negative to positive values at about @xmath75 . when going further away from @xmath74 , the spatial features of the output wavefunction eventually become similar to that of the input wavefunction . in order to illustrate these spatial features in detail , figure . [ fig : figure3 ] ( d ) shows the cross - sections of the output wavefunction at @xmath62 for @xmath76 . the negative amplitude of the cross section at a time difference of @xmath69 ( solid line ) is about 3 times as large as the amplitude of the input wavefunction at @xmath69 shown in figure . [ fig : figure3 ] ( b ) ( solid line ) . moreover , the peak of the output at @xmath69 is delayed by about @xmath77 compared to the peak of the input wavefunction at the same time difference . the cross - section at a time difference of @xmath70 ( broken line ) is nearly zero . this means that the probability density of detecting one photon at a time difference of @xmath70 after detecting the other photon is vanishingly small at all times . the cross - section at @xmath71 ( dotted line ) is similar to the cross - section of the input wavefunction at the same time difference between the photons . however the output at @xmath71 is delayed compared to the input wavefunction by about @xmath52 , which is equal to the delay time observed in the one photon case . let us now consider why such features appear . the two photon case can be understood as a superposition of three processes in the same way that the one photon case can be understood as a superposition of two processes . figure.[fig : figure4 ] shows the contour plots of the three components given by eq.([eq : outputcomponents ] ) . [ fig : figure4 ] ( a ) corresponds to the transmission process , figure . [ fig : figure4 ] ( b ) corresponds to the one photon absorption process , and figure . [ fig : figure4 ] ( c ) corresponds to the two photon absorption process , respectively . the component of the transmission process ( figure . [ fig : figure4 ] ( a ) ) is identical with the gaussian input wavefunction , just as in the one photon case , @xmath78 . likewise , the component of the one photon absorption process ( figure . [ fig : figure4 ] ( b ) ) , although not a precise gaussian , can be approximated by the products of single photon transmission and single photon absorption , @xmath79 this approximation indicates that the negative amplitude of the one photon absorption process in figure . [ fig : figure4 ] ( b ) is about 4 times as large as the positive amplitude of the transmission process in figure . [ fig : figure4 ] ( a ) . moreover , the wavepacket of the one photon absorption process is delayed by @xmath80 compared to the transmission process . the component of the two photon absorption process ( figure . [ fig : figure4 ] ( c ) ) includes the non - linear interaction between the two photons . for @xmath81 , it resembles the shifted gaussian expected for separate one photon absorption processes ( see eq.([eq : appabs2 ] ) ) , @xmath82 however , the saturation of the atom completely suppresses this process for @xmath83 . closer inspection shows that the transition between @xmath83 and @xmath81 can be approximated by an exponential function of @xmath60 , @xmath84 the combination of the two photon absorption process is therefore directly dependent on the time difference @xmath85 between the two photons . [ t][t]@xmath65 [ b][b]@xmath66 [ t][t]@xmath12 [ t][t]@xmath12 [ tr][tr]20 [ tr][tr]10 [ tr][tr]0 [ tr][tr]-10 [ tr][tr]-20 ( 0,0 ) ( -5,200)(a ) contour plots of the two photon processes @xmath86 . ( a ) corresponds to the transmission process , ( b ) corresponds to the one photon absorption process , ( c ) corresponds to the two photon absorption process . ] ( 0,0 ) ( -5,200)(b ) contour plots of the two photon processes @xmath86 . ( a ) corresponds to the transmission process , ( b ) corresponds to the one photon absorption process , ( c ) corresponds to the two photon absorption process . ] ( 0,0 ) ( -5,200)(c ) contour plots of the two photon processes @xmath86 . ( a ) corresponds to the transmission process , ( b ) corresponds to the one photon absorption process , ( c ) corresponds to the two photon absorption process . ] + ( 0,0 ) ( -10,35)(d ) ( -10,-10)-0.35 ( 190,-10)0.3 ( 110,-10)0 contour plots of the two photon processes @xmath86 . ( a ) corresponds to the transmission process , ( b ) corresponds to the one photon absorption process , ( c ) corresponds to the two photon absorption process.,height=37 ] in order to illustrate the dependence of the output component on the relative distance between two photons in detail , figure . [ fig : figure5 ] shows the cross - sections at @xmath62 of the output amplitude @xmath39 ( solid lines ) and the components of the transmission process @xmath87 ( dotted lines ) , the one photon absorption process @xmath88 ( thin lines ) , and the two photon absorption process @xmath89 ( broken lines ) . the time differences are @xmath90 . [ t][t]@xmath67 [ t][t]@xmath67 [ t][t]@xmath67 [ t][t]@xmath67 [ b][b ] [ b][b ] [ b][b ] [ b][b ] ( 0,0 ) ( -5,215)(a ) cross - sections of the output amplitude @xmath39 and its components @xmath86 at time differences @xmath85 of ( a ) @xmath69 , ( b ) @xmath70 and ( c ) @xmath71 . the solid lines correspond to the total output amplitude @xmath39 , the dotted lines correspond to the transmission amplitude @xmath87 , the thin lines correspond to the one photon absorption amplitude @xmath88 , and the broken lines correspond to the two photon absorption amplitude @xmath89 . ] ( 0,0 ) ( -5,215)(b ) cross - sections of the output amplitude @xmath39 and its components @xmath86 at time differences @xmath85 of ( a ) @xmath69 , ( b ) @xmath70 and ( c ) @xmath71 . the solid lines correspond to the total output amplitude @xmath39 , the dotted lines correspond to the transmission amplitude @xmath87 , the thin lines correspond to the one photon absorption amplitude @xmath88 , and the broken lines correspond to the two photon absorption amplitude @xmath89 . ] ( 0,0 ) ( -5,215)(c ) cross - sections of the output amplitude @xmath39 and its components @xmath86 at time differences @xmath85 of ( a ) @xmath69 , ( b ) @xmath70 and ( c ) @xmath71 . the solid lines correspond to the total output amplitude @xmath39 , the dotted lines correspond to the transmission amplitude @xmath87 , the thin lines correspond to the one photon absorption amplitude @xmath88 , and the broken lines correspond to the two photon absorption amplitude @xmath89 . ] for all three time differences @xmath91 , the amplitude of the component of the one photon absorption process ( thin lines ) is about 4 times as large as the amplitude of the corresponding component of the transmission process ( dotted lines ) . on the other hand , the ratio of the amplitude of the component of the two photon absorption process ( broken lines ) to the amplitude of the corresponding component of the transmission process ( dotted lines ) changes depending on the time difference @xmath91 between the two photons as discussed in the following . figure . [ fig : figure5 ] ( a ) shows the cross sections of the output components at @xmath69 . in this case , the amplitude of the component of the two photon absorption process , @xmath92 , is exactly zero ( see eq.([eq : appprocess3 ] ) ) . the output amplitude is then given by the sum of the transmission amplitude @xmath78 and the one photon absorption amplitude @xmath93 , @xmath94 the amplitude of the output at the time difference of @xmath69 ( solid line in figure . [ fig : figure5 ] ( a ) ) is therefore about 3 times as large as the corresponding amplitude of the input . figure . [ fig : figure5 ] ( b ) shows the cross sections of the output components at @xmath70 . at this special delay time , the factor of @xmath95 describing the effect of the nonlinearity in eq.([eq : appprocess3 ] ) is approximately equal to @xmath96 . therefore , the two photon absorption component in the total output can be approximated by @xmath97 . the output wavefunction is then given by @xmath98 at a delay time of @xmath70 , the positive amplitudes of the transmission process and the two photon absorption process thus compensate the negative amplitude of the one photon absorption process , resulting in a total output amplitude of nearly zero . figure . [ fig : figure5 ] ( c ) shows the cross sections of the output components at @xmath71 . in this case , the amplitude of the component of the two photon absorption process , @xmath92 , can be approximated as @xmath99 since the time difference @xmath91 is sufficiently larger than @xmath0 . the output amplitude is then given by @xmath100 the amplitude of the output at the time difference of @xmath71 ( solid line in figure . [ fig : figure5 ] ( c ) ) is thus almost identical with the amplitude of the input at the same time difference . to explain the delay of the output wavefunction , we can include the delay times in the arguments of the approximations given by eq.([eq : appprocess2 ] ) and eq.([eq : appprocess3 ] ) . for @xmath69 , the output wavefunction can then be approximated as @xmath101 the amplitude of the output wavefunction at a time difference of @xmath69 is thus delayed by about @xmath77 compared to the amplitude of the input wavefunction at the same time difference . in the case of @xmath70 , the output wavefunction is nearly zero , and no delay time can be defined . for @xmath71 , the two photon component @xmath89 of the output wavefunction can be approximated using the limit of @xmath81 given in eq.([eq : shiftedgaussian ] ) . since in this limit , the two photons are absorbed and reemitted independently of each other , the output wavefunction can then be factorized into one photon processes ( see eq.([eq : apponeoutput ] ) ) , @xmath102 the delay of the peak of the output wavefunction for @xmath81 is thus equal to the delay for the one photon output wavefunction discussed in section [ sec : level3].1 . in summary , the delay of the negative amplitude at @xmath74 is equal to @xmath103 of the delay of the positive amplitude at @xmath81 . this means that the effect of the non - linear interaction between the photons at @xmath74 reduces the total delay of the output wavefunction as compared to the delay caused by the uncorrelated transmission and absorption processes at @xmath81 . as shown in the long pulse case , the atomic nonlinearity is characterized by a negative amplitude of the output wavefunction around @xmath74 . in the long pulse case , this nonlinear effect is limited to a region of @xmath104 . in order to realize a non - linear phase change throughout the pulse , it is necessary to shorten the pulse to a pulse length comparable to @xmath0 . in the following , we therefore discuss the response of the atom to a pulse of length @xmath105 . the line plot in figure . [ fig : figure6 ] ( a ) shows the short one - photon pulse ( broken line ) and the corresponding output wavefunction ( solid line ) . the length of the input wavepacket is @xmath105 . [ t][t]@xmath46 [ b][b]@xmath106 [ b][b]@xmath48 ( 0,0 ) ( -5,200)(a ) gaussian input and output one - photon wavepackets for an input wavepacket length of @xmath105 . in ( a ) , the broken line corresponds to the input wavefunction @xmath16 and the solid line corresponds to the output wavefunction @xmath107 . in ( b ) , the dotted line corresponds to the component of the process ( i ) : single photon transmission without absorption and the thin line corresponds to the component of the process ( ii ) : single photon reemission after absorption . ] ( 0,0 ) ( -5,200)(b ) gaussian input and output one - photon wavepackets for an input wavepacket length of @xmath105 . in ( a ) , the broken line corresponds to the input wavefunction @xmath16 and the solid line corresponds to the output wavefunction @xmath107 . in ( b ) , the dotted line corresponds to the component of the process ( i ) : single photon transmission without absorption and the thin line corresponds to the component of the process ( ii ) : single photon reemission after absorption . ] the pulse shape of the output pulse is now quite different from the input pulse . specifically , the output amplitude is still positive for @xmath108 and only changes to negative values in the latter part of the pulse . let us now analyze the origin of these features . figure . [ fig : figure6 ] ( b ) shows the components corresponding to different light - atom interaction processes as given by eqs.([eq : oneoutputcomponents ] ) . the dotted line is the component of single photon transmission without absorption , @xmath58 . the thin line is that of single photon reemission after absorption , @xmath54 . as before , @xmath109 according to eqs.([eq : oneoutputcomponents ] ) . on the other hand , the absorption component @xmath54 ( thin line ) is given by the convolution with the exponential function of the dipole relaxation , given by eq.([eq : absgenerater ] ) and eqs.([eq : oneoutputcomponents ] ) . the peak of this component is shifted by about @xmath110 compared to the component of the transmission process . furthermore , the peak of the amplitude is only about 1.3 times as large as the corresponding amplitude of the component of the transmission process . therefore , the front part of the output wavefunction is dominated by the positive component of the transmission process @xmath58 , even though the major part of the output wavefunction is still dominated by the negative amplitude component of the absorption process @xmath54 . the large delay between the components and the reduced amplitude of the dominant negative absorption component thus cause the asymmetry of the output wavefunction seen in figure . [ fig : figure6 ] ( a ) . let us now consider the output wavefunction for the short two photon pulse . figure . [ fig : figure7 ] ( a ) shows the contour plot of the short two - photon input pulse @xmath15 . the probability amplitude increases from black to white shading . as before , the input pulse has a length of @xmath105 . [ t][t]@xmath65 [ b][b]@xmath66 [ t][t]@xmath67 [ b][b]@xmath111 [ tr][tr]2 [ tr][tr]1 [ tr][tr]0 [ tr][tr]-1 [ tr][tr]-2 [ tr][tr]-3 [ tr][tr]-4 ( 0,0 ) ( -5,200)(a ) input and output wavefunction for a two - photon input pulse with a pulse length of @xmath105 . ( a ) is a contour plot of the input wavefunction @xmath15 . ( b ) is the cross - section of the contour line plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath112 ( broken line ) and @xmath113 ( dotted line ) . the horizontal axis @xmath72 in ( b ) is the average position of the two photons . likewise , ( c ) is a contour plot of the output wavefunction @xmath39 and ( d ) shows cross sections of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath112 ( broken line ) and @xmath113 ( dotted line ) . ] ( 0,0 ) ( -20,200)(b ) input and output wavefunction for a two - photon input pulse with a pulse length of @xmath105 . ( a ) is a contour plot of the input wavefunction @xmath15 . ( b ) is the cross - section of the contour line plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath112 ( broken line ) and @xmath113 ( dotted line ) . the horizontal axis @xmath72 in ( b ) is the average position of the two photons . likewise , ( c ) is a contour plot of the output wavefunction @xmath39 and ( d ) shows cross sections of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath112 ( broken line ) and @xmath113 ( dotted line ) . ] [ t][t]@xmath65 [ b][b]@xmath73 [ tr][tr]2 [ tr][tr]1 [ tr][tr]0 [ tr][tr]-1 [ tr][tr]-2 [ tr][tr]-3 [ tr][tr]-4 ( 0,0 ) ( -5,200)(c ) input and output wavefunction for a two - photon input pulse with a pulse length of @xmath105 . ( a ) is a contour plot of the input wavefunction @xmath15 . ( b ) is the cross - section of the contour line plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath112 ( broken line ) and @xmath113 ( dotted line ) . the horizontal axis @xmath72 in ( b ) is the average position of the two photons . likewise , ( c ) is a contour plot of the output wavefunction @xmath39 and ( d ) shows cross sections of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath112 ( broken line ) and @xmath113 ( dotted line ) . ] ( 0,0 ) ( -20,200)(d ) input and output wavefunction for a two - photon input pulse with a pulse length of @xmath105 . ( a ) is a contour plot of the input wavefunction @xmath15 . ( b ) is the cross - section of the contour line plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath112 ( broken line ) and @xmath113 ( dotted line ) . the horizontal axis @xmath72 in ( b ) is the average position of the two photons . likewise , ( c ) is a contour plot of the output wavefunction @xmath39 and ( d ) shows cross sections of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath112 ( broken line ) and @xmath113 ( dotted line ) . ] ( 0,0 ) ( -15,25)(e ) ( -5,-10)-2.0 ( 185,-10)1.5 ( 120,-10)0 input and output wavefunction for a two - photon input pulse with a pulse length of @xmath105 . ( a ) is a contour plot of the input wavefunction @xmath15 . ( b ) is the cross - section of the contour line plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath112 ( broken line ) and @xmath113 ( dotted line ) . the horizontal axis @xmath72 in ( b ) is the average position of the two photons . likewise , ( c ) is a contour plot of the output wavefunction @xmath39 and ( d ) shows cross sections of the contour plot at @xmath62 , where the time differences are @xmath69 ( solid line ) , @xmath112 ( broken line ) and @xmath113 ( dotted line).,height=37 ] figure . [ fig : figure7 ] ( b ) shows the cross - section of the input wavefunction at @xmath62 for time differences of @xmath69 , @xmath112 , and @xmath113 between the two photons . figure . [ fig : figure7 ] ( c ) and figure . [ fig : figure7 ] ( d ) show the output wavefunction @xmath39 . the output amplitude is now negative everywhere . however , the pulse shape of the output shown in figure . [ fig : figure7 ] ( c ) is quite different from the shape of the input pulse shown in figure . [ fig : figure7 ] ( a ) . in order to illustrate the spatial features of the output wavefunction in detail , figure . [ fig : figure7 ] ( d ) shows the cross - sections of the output wavefunction for time differences of @xmath69 , @xmath112 , and @xmath113 between the two photons . note that the shape of the cross - sections of the output in figure . [ fig : figure7 ] ( d ) is similar to that of the input cross - sections in figure . [ fig : figure7 ] ( b ) . however , the amplitudes are quite different . specifically , the output amplitude at a time difference of @xmath69 ( figure . [ fig : figure7 ] ( d ) , solid line ) is larger than the input amplitude at the same time difference ( figure . [ fig : figure7 ] ( b ) , solid line ) , while the output amplitude at a time difference of @xmath113 ( figure . [ fig : figure7 ] ( d ) , dotted line ) is smaller than the input amplitude at the same time difference ( figure . [ fig : figure7 ] ( b ) , dotted line ) . this indicates that the main difference in pulse shape between the input and output pulses is that the output amplitude decreases faster than the input amplitude between @xmath69 and @xmath114 . on the other hand , the delay time of the peak of the output amplitude does not change much . its value is about @xmath115 for all time differences @xmath91 between the photons . let us now consider why such features appear . figure . [ fig : figure8 ] shows the contour plot of the three components given by eq.([eq : outputcomponents ] ) . figure . [ fig : figure8 ] ( a ) corresponds to the transmission process , figure . [ fig : figure8 ] ( b ) corresponds to the one photon absorption process and figure . [ fig : figure8 ] ( c ) corresponds to the two photon absorption process . as before , the component of the transmission process ( figure . [ fig : figure8 ] ( a ) ) is identical with the input wavefunction , @xmath116 . likewise , the component of the one photon absorption process ( figure . [ fig : figure8 ] ( b ) ) can be described as @xmath117 since the single photon transmission @xmath58 is positive and the single photon absorption @xmath54 is negative , the total amplitude of this component is negative for all @xmath118 . the component of the two photon absorption process ( figure . [ fig : figure8 ] ( c ) ) includes the non - linear interaction between the two photons . the positive amplitude of this component is therefore always smaller than the product of the single photon absorption amplitudes , @xmath119 . specifically , the two photon absorption component @xmath89 is exactly zero at @xmath74 and increases rapidly up to a maximal amplitude at about @xmath120 . for larger photon distances , @xmath121 , @xmath89 decreases as a consequence of the decrease in the amplitudes of the photon absorption processes , @xmath119 . note however that , at its maximal amplitudes near @xmath120 , the two photon absorption amplitude @xmath89 is still significantly lower than the product of the single photon absorption amplitudes , @xmath119 . this means that the non - linear interaction of the photons significantly affects the whole output pulse in the case of an input pulse length of @xmath105 . [ t][t]@xmath65 [ b][b]@xmath66 [ tr][tr]2 [ tr][tr]1 [ tr][tr]0 [ tr][tr]-1 [ tr][tr]-2 [ tr][tr]-3 [ tr][tr]-4 ( 0,0 ) ( -5,200)(a ) contour plots of the two photon processes @xmath122 . ( a ) corresponds to the transmission process , ( b ) corresponds to the one photon absorption process , ( c ) corresponds to the two photon absorption process . ] ( 0,0 ) ( -5,200)(b ) contour plots of the two photon processes @xmath122 . ( a ) corresponds to the transmission process , ( b ) corresponds to the one photon absorption process , ( c ) corresponds to the two photon absorption process . ] ( 0,0 ) ( -5,200)(c ) contour plots of the two photon processes @xmath122 . ( a ) corresponds to the transmission process , ( b ) corresponds to the one photon absorption process , ( c ) corresponds to the two photon absorption process . ] + ( 0,0 ) ( -10,35)(d ) ( -10,-10)-0.2 ( 190,-10)1.5 ( 120,-10)0 contour plots of the two photon processes @xmath122 . ( a ) corresponds to the transmission process , ( b ) corresponds to the one photon absorption process , ( c ) corresponds to the two photon absorption process.,height=37 ] in order to illustrate the dependence of the output component on the relative distance between two photons in detail , figure . [ fig : figure9 ] shows the cross - sections at @xmath62 of the output amplitude @xmath39 ( solid line ) and the components of the transmission process @xmath123 ( dotted line ) , the one photon absorption process @xmath88 ( thin line ) , and the two photon absorption process @xmath89 ( broken line ) . the time differences are @xmath69 in ( a ) , @xmath112 in ( b ) , and @xmath113 in ( c ) . in all figures , the component of the one photon absorption process ( thin line ) is always negative and significantly greater than the component of the transmission process ( dotted line ) . furthermore , the ratio of the peak of the one photon absorption amplitude ( thin line ) to the peak of the corresponding transmission amplitude does not change very much depending on the time difference @xmath91 between the two photons . on the other hand , the ratio of the peak of the two photon absorption amplitude ( broken line ) to the peak of the corresponding transmission amplitude ( dotted line ) changes significantly depending on the time difference @xmath91 between the two photons as discussed in the following . [ fig : figure9 ] ( a ) shows the cross - section of the output components at @xmath69 . in this case , the two photon absorption amplitude ( broken line ) is exactly zero . the output amplitude is then given by the sum of the transmission amplitude ( dotted line ) and the one photon absorption amplitude ( thin line ) . since the one photon absorption amplitude is about @xmath124 times as large as the transmission amplitude , the total output pulse ( solid line ) is negative and about @xmath125 times as large as the transmission amplitude . figure . [ fig : figure9 ] ( b ) shows the cross - sections of the output components at @xmath112 . the one photon absorption amplitude ( thin line ) and the transmission amplitude ( dotted line ) are not much different from @xmath69 ( figure . [ fig : figure9 ] ( a ) ) . however , the two photon absorption amplitude ( broken line ) is now about half as high as the transmission amplitude ( dotted line ) . therefore , the output amplitude is reduced to about the same height as the transmission amplitude . [ fig : figure9 ] ( c ) shows the cross - sections of the output components at @xmath113 . the two photon absorption amplitude ( broken line ) and the transmission amplitude ( dotted line ) are now significantly lower than for @xmath69 ( figure . [ fig : figure9 ] ( a ) ) . however , the negative one photon absorption amplitude is still about @xmath126 times as large as the transmission amplitude . the two photon absorption amplitude ( broken line ) has increased to about @xmath125 times the transmission amplitude , leaving a total negative output amplitude ( solid line ) of about @xmath127 times the transmission amplitude . thus the output amplitude is reduced to very low values , even though the components themselves are still very large . the fast decrease of the total output amplitude shown in figure . [ fig : figure7 ] ( c ) and [ fig : figure7 ] ( d ) can then be understood as a consequence of the rapid increase of the two photon absorption amplitude with @xmath91 . interestingly , the delay times of the pulse components are not sufficiently high to cause the same change in the sign of the amplitude observed for the one photon case ( figure . [ fig : figure6 ] ( a ) ) . no part of the two - photon output wavefunction is dominated by either the transmission or the two photon absorption amplitude . this is mostly because the two photon absorption amplitude is suppressed by the non - linearity . in fact , it can be seen in figure . [ fig : figure9 ] ( c ) that the nonlinearity even reduces the delay of the two photon absorption amplitude from a delay time of @xmath110 for @xmath119 to only about @xmath128 for @xmath89 at @xmath113 . therefore , the change of the pulse shape in the two photon pulses is actually less drastic than the change of pulse shape observed in the one photon case . [ t][t]@xmath67 [ ] [ ] [ ] [ ] [ ] [ ] ( 0,0 ) ( -5,210)(a ) cross - sections of the output amplitude @xmath39 and its components @xmath86 at time differences @xmath85 of ( a ) @xmath69 , ( b ) @xmath112 and ( c ) @xmath113 . the solid lines correspond to the total output amplitude @xmath39 , the dotted lines correspond to the transmission amplitude @xmath87 , the thin lines correspond to the one photon absorption amplitude @xmath88 , and the broken lines correspond to the two photon absorption amplitude @xmath89 . ] ( 0,0 ) ( -5,210)(b ) cross - sections of the output amplitude @xmath39 and its components @xmath86 at time differences @xmath85 of ( a ) @xmath69 , ( b ) @xmath112 and ( c ) @xmath113 . the solid lines correspond to the total output amplitude @xmath39 , the dotted lines correspond to the transmission amplitude @xmath87 , the thin lines correspond to the one photon absorption amplitude @xmath88 , and the broken lines correspond to the two photon absorption amplitude @xmath89 . ] + ( 0,0 ) ( -5,210)(c ) cross - sections of the output amplitude @xmath39 and its components @xmath86 at time differences @xmath85 of ( a ) @xmath69 , ( b ) @xmath112 and ( c ) @xmath113 . the solid lines correspond to the total output amplitude @xmath39 , the dotted lines correspond to the transmission amplitude @xmath87 , the thin lines correspond to the one photon absorption amplitude @xmath88 , and the broken lines correspond to the two photon absorption amplitude @xmath89 . ] we have studied the effect of the strong nonlinearity of a one - dimensional atom on one- and two - photon input wavepakets . the response of the atom can be characterized by transmission and absorption processes . for long pulses , the absorption components dominate in the output . however , in the two photon case , the two photon absorption component is suppressed around @xmath74 by the saturation of the two - level atom . the long two photon pulse therefore has a non - linear region of negative amplitude dominated by the one photon absorption for @xmath129 , and a linear region of positive amplitude dominated by the two photon absorption for @xmath130 . for a short pulse length of @xmath105 , the two photon absorption is strongly suppressed throughout the whole output pulse . the one photon absorption process then dominates everywhere , and the amplitude of the short two - photon output pulse is completely negative . it is therefore possible to obtain a very strong non - linear effect by adjusting the pulse length @xmath44 to the dipole relaxation time of the atom . besides the total amplitude , we have also characterized the change in delay time and pulse shape . for long pulses , the delay of the non - linear region is reduced to @xmath103 of the delay of the liner region . in the long pulse case , the interaction of the two photons thus significantly reduces the average delay time caused by the absorption and reemission processes at the atom and causes a faster reemission into the output field . in the case of a short one photon pulse , the delay between the transmission and the absorption component is so great that the pulse shape is changed significantly . in particular , the single photon output amplitude has both negative and positive regions . on the other hand , the output pulse shape for a short two photon pulse is changed much less due to the reduced delay caused by the non - linearity . it is therefore possible to obtain a negative output amplitude throughout the short output pulse . this result indicates that the atomic nonlinearity is particularly strong at a pulse length of @xmath105 . the analysis of the output in terms of components corresponding to different interaction processes thus allow us to understand the origin of the non - linear photon - photon interaction in great detail . these insights may be very useful for the development of novel non - linear optical devices operating at the few photon level . this work was partly supported by the program `` research and development on quantum communication technology '' of the ministry of public management , home affairs , posts and telecommunications of japan . 99 nielsen , m. , chuang , i. ( 2000 ) . _ quantum computation and quantum information . _ cambrige : cambridege university press . hofmann , h. , kojima , k. , takeuchi , s. , sasaki , k. ( 2003 ) . entanglement and four - wave mixing effects in the dissipation - free nonlinear interaction of two photons at a single atom . a , 68 _ , 043813 . milburn , g. , walls , d. ( 1983 ) . quantum nondemolition measurements via quadratic coupling . rev . a , 28 _ , pp . 2065 - 2070 ; n. imoto , h. a. haus , y. yamamoto . ( 1985 ) . quantum nondemolition measurement of the photon number via the optical kerr effect . rev . a , 32 _ , pp . 2287 - 2292 . mills , d. ( 1991 ) . _ nonlinear optics ( basic concepts ) . _ new york : springer - verlag . bermann , p. ( 1994 ) . _ cavity quantum electrodynamics supplement 2 to advances in atomic , molecular , and optical physics . _ san diego : academic . turchette , q. , hood , c. , lange , w. , mabuchi , h. , kimble , h. ( 1995 ) . measurement of conditional phase shifts for quantum logic . lett . , 75 _ , pp . 4710 - 4713 . hofmann , h. , kojima , k. , takeuchi , s. , sasaki , k. ( 2003 ) . optimized phase switching using a single - atom nonlinearity _ j. opt . b : quantum . semiclass . opt . , 5 _ , pp . 218 - 221 . kojima , k. , hofmann , h. , takeuchi , s. , sasaki , k. ( 2003 ) . nonlinear interaction of two photons with a one - dimensional atom : spatiotemporal quantum coherence in the emitted field . _ phys . rev . a , 68 _ , 013803 . hofmann , h. , mahler , g. ( 1995 ) . measurement models for time - resolved spectroscopy : a comment . _ quantum . semiclass . opt . 7 _ , pp . 489 - 497 . rice , p. , carmichael , h. ( 1988 ) . single - atom cavity - enhanced absorption i : photon statistics in the bad - cavity limit . _ ieee j. quantum electron . , 24 _ , pp . 1351 - 1365 . truchette , q. , thompson , r. , kimble , h. ( 1992 ) . one - dimensional atoms . b , 60 _ , pp . s1-s10 .

the rapid development and improvement of ultra - intense laser pulses have opened new areas of physics for fundamental research and have also enabled novel technological applications . relativistically intense ( @xmath0 w/@xmath1 ) laser pulses readily ionize matter converting it into a plasma . the interaction with the plasma electrons is the primary channel for the energy transfer from the laser pulse , providing the basis for a wide range of phenomena and applications . specifically , generation of copious relativistic electrons is the key to x - ray @xcite and secondary particle sources , such as energetic ions @xcite , neutrons @xcite , and positrons @xcite . it is therefore critical to understand what controls the generation of relativistic electrons in relevant regimes of laser - plasma interactions . it is well recognized that the regime in which a relativistically intense laser pulse irradiates a sub - critical plasma is optimal for generating relativistic electrons . the pulse can propagate through such a plasma , which enables an extended interaction length with the electrons . in experiments aimed at x - ray generation , this regime is deliberately achieved by using an expanding gas jet @xcite . this regime can also naturally arise in experiments with solid density targets irradiated by a powerful laser pulse due to the presence of a prepulse @xcite . the prepulse often delivers a considerable amount of energy , causing the front of the target to expand and form an extended subcritical preplasma prior to the arrival of the main pulse . the important role played by the interaction of the main pulse with the preplasma in generating an energetic electron population in experiments with initially solid - density targets was however not immediately recognized . this is in part due to the fact that in some setups the role of the prepulse can also be detrimental . it is the case in experiments aimed at ion acceleration , where the pre - expansion at the surface of a thick bulk target caused by the prepulse significantly reduces the effectiveness of ion acceleration @xcite . recently , the focus of the ion acceleration research has markedly shifted towards those regimes in which the target becomes transparent to the main relativistically intense pulse and the electron acceleration and heating can be fully utilized @xcite . such a regime is achieved by employing an initially ultra - thin target that significantly expands during the pre - pulse and forms a relativistically transparent plasma extending many wavelengths along the direction of the laser beam propagation . the same regime has also been used to generate energetic electrons that are subsequently converted into a short neutron beam @xcite . how energetic electrons are generated strongly depends on the duration of the main laser pulse . in most setups aimed at producing secondary particle sources , the laser pulse irradiates the plasma over a time period that is longer than the characteristic electron response time . this regime is the opposite to the regime used for the wakefield acceleration @xcite . as a consequence , the laser pulse establishes a quasi - steady - state structure in the plasma that slowly evolves on an ion time scale @xcite . by expelling some of the electrons radially , the laser creates a positively charged elongated channel in the plasma with quasi - static transverse and longitudinal electric fields . new electrons are continuously injected into the channel , typically through the channel opening @xcite , and then get accelerated and pushed forward by the laser pulse in the presence of the quasi - static fields . the described regime is often broadly referred to as the direct laser acceleration regime . in terms of applications , it is critical to know what controls the electron energy gain . early work on the topic @xcite indicated that the transverse static electric field can be beneficial for enhancing the electron energy gain beyond what is expected from a single electron irradiated by a plane wave in a vacuum . recently , there has been a renewed interest in the direct laser acceleration of electrons , as experimental groups shift their focus to regimes of relativistic transparency and also try to optimize the preplasma conditions using multiple laser pulses @xcite . recent simulation results have also demonstrated that the direct laser acceleration can be important in the context of the laser wakefield acceleration @xcite . in this paper , we examine the role played by transverse and longitudinal quasi - static electric fields present in a plasma channel in enhancing the electron energy gain from the laser pulse . we also address the role of electron injection into the laser beam and the limitations imposed by the super - luminosity of the laser field that is induced by the channel . this paper is based on a body of work performed by us over the last couple of years @xcite and is designed to serve in part as an overview of some novel aspects of the direct laser acceleration . here we focus on illustrating the key qualitative concepts and phenomena , while providing references to those publications where one can find more detailed technical analysis . the nature of the laser - plasma interaction strongly depends on the amplitude of the irradiating laser pulse and on the electron density of the irradiated plasma . it is convenient to use a dimensionless parameter @xmath2 to quantify the impact of a laser pulse with electric field amplitude @xmath3 and frequency @xmath4 on electron motion . here @xmath5 is the speed of light and @xmath6 and @xmath7 are the electron charge and mass . the parameter @xmath8 is often referred to as the normalized laser amplitude . it is roughly the ratio of the transverse electron momentum induced by the oscillating laser electric field to @xmath9 . therefore , a laser pulse with a normalized amplitude of @xmath10 would induce relativistic electron motion . the electron density in the plasma determines whether the laser pulse can propagate into the plasma and accelerate plasma electrons . the cut - off for a pulse with @xmath11 occurs at a critical density , @xmath12 for which the electron plasma frequency @xmath13 becomes equal to the frequency of the laser pulse . at laser amplitudes @xmath10 , the plasma can become relativistically transparent at electron densities exceeding the critical density @xmath14 . the adjusted critical density in this case depends on the amplitude of the irradiating laser pulse , because the effect is caused by the relativistic motion of electrons in the strong field of the laser . the optimal regime for generating copious relativistic electrons is then the regime in which a relativistic amplitude laser pulse @xmath15 irradiates an extended sub - critical plasma @xmath16 . in order to illustrate the key features of the laser - plasma interaction in this regime , we have performed a two - dimensional ( 2d ) particle - in - cell ( pic ) simulation whose results are shown in fig . [ fig:2d_example ] . in this simulation , a uniform sub - critical plasma with initial electron density @xmath17 is irradiated by a laser beam with wavelength @xmath18 @xmath19 m whose amplitude ramps up to @xmath20 and then remains constant . the laser pulse propagates along the @xmath21-axis and it is linearly polarized , with the laser electric field polarized in the @xmath22-plane . the ions were kept immobile in this simulation to distinguish more clearly the effect of the long laser pulse . detailed parameters of the simulation are given in appendix [ appendix_1 ] . as the pulse enters the plasma , its ponderomotive force begins to expel some of the electrons out of the laser pulse in the transverse direction producing a channel . the un - neutralized ion charge generates a counteracting force that , in the example shown in fig . [ fig:2d_example ] , prevents the channel from becoming fully evacuated . the laser pulse produces and maintains a steady - state channel if the pulse duration exceeds the characteristic electron response time . the snapshot of the electron density in fig . [ fig:2d_example ] taken at 1 ps illustrates such a channel . the positively charged elongated channel generates quasi - static transverse and longitudinal electric fields shown in the lower - left panel of fig . [ fig:2d_example ] . the fields have been averaged over ten laser periods and they are normalized to the electric field amplitude @xmath3 of a plane wave with @xmath20 , which is essentially the amplitude of the electric field in the laser pulse . the time - averaged electric fields are relatively small compared to the amplitude of the oscillating laser electric field that is also present in the channel . once the steady - state channel structure is established , new electrons are continuously injected into the channel through the opening . this is particularly clear from the snapshot of the time - averaged current density , whose absolute value is shown in the upper - right panel of fig . [ fig:2d_example ] . the injected electrons are accelerated and pushed forward by the laser pulse , producing a steady - state electron current in the channel . this current generates a quasi - static transverse magnetic field directed in and out of the plane of the simulation ( along the @xmath23-axis ) . the return current flowing outside of the channel causes for the magnetic field to be localized inside the channel , as shown in the lower - right panel of fig . [ fig:2d_example ] . the plotted time - averaged field is normalized to the magnetic field amplitude @xmath24 of a plane wave with @xmath20 . the quasi - static magnetic field is also relatively weak in this regime , as it is less that 1% of the magnetic field in the laser pulse . . color - coded is the electron longitudinal momentum . ] the presented example illustrates that , in a sub - critical plasma irradiated by a long laser pulse , electron acceleration takes place in a positively charged channel in the presence of extended transverse quasi - static electric and magnetic fields and a localized quasi - static longitudinal electric field . these fields are relatively small compared to the fields in the laser pulse , not exceeding a few percent . it is then somewhat unexpected that , for example , the transverse electric field can significantly enhance the electron energy gain @xcite . as we show in the following sections , the transverse and longitudinal electric fields can synergistically enhance the electron energy gain , with a well - pronounced threshold , well beyond the energy gain that one would expect in the absence of these fields . in order to provide the context for the discussion of the electron acceleration in the channel , we briefly review the key features of the electron motion in a vacuum where no static electric or magnetic fields are present . specifically , we consider a single electron that is initially at rest . it is irradiated by a plane electromagnetic wave whose normalized amplitude gradually increases from zero to @xmath8 . in what follows , we refer to this regime as the vacuum regime . the electron moves according to the following equations : @xmath25 , \label{main_eq:1 } \\ & & \frac{d { \bf{r}}}{d t } = \frac{c}{\gamma } \frac{{\bf{p}}}{m_e c } , \label{main_eq:2}\end{aligned}\ ] ] where @xmath26 is the relativistic factor , @xmath27 and @xmath28 are the electron position and momentum , and @xmath29 is the time . in the regime under consideration , @xmath30 and @xmath31 are the electric and magnetic fields of the wave . it is convenient to express these fields in terms of a normalized vector potential @xmath32 , @xmath33 . \label{b_wave}\end{aligned}\ ] ] in the case of a plane wave with wave - length @xmath34 propagating along the @xmath21-axis , the vector potential is only a function of a normalized phase @xmath35 without any loss of generality , we assume that the laser electric field is polarized along the @xmath36-axis . in this case , the vector potential @xmath32 only has a @xmath36-component . we consider a pulse with @xmath37 , where @xmath8 is the maximum normalized amplitude and @xmath38 is a slowly varying envelope that ramps up from zero to unity . figure [ figure_dla_vac_1 ] shows an electron trajectory for a pulse with @xmath20 obtained by solving eqs . ( [ main_eq:1 ] ) and ( [ main_eq:2 ] ) numerically . it illustrates the qualitative change that takes place at relativistic wave amplitudes of @xmath10 . at @xmath11 , the dominant force experienced by the electron is the force from the electric field of the wave . as a result , the electron oscillates across the laser pulse , while its longitudinal displacement is negligible . at @xmath10 , the momentum oscillations induced by the laser electric field become relativistic and , as a consequence of this , the lorentz force becomes important . this force causes longitudinal electron motion that leads to a trajectory shown in fig . [ figure_dla_vac_1 ] . even though the energy from the laser pulse is transferred to the transverse electron oscillations , the lorentz force converts most of this energy into the longitudinal electron motion . indeed , an analytical solution of eqs . ( [ main_eq:1 ] ) and ( [ main_eq:2 ] ) yields @xcite @xmath39 the electron moves along a parabola in the momentum space , with @xmath40 for @xmath41 . the maximum @xmath42-factor that the electron can reach is thus given by @xmath43 the change of the longitudinal electron momentum along the electron trajectory is shown in fig . [ figure_dla_vac_1 ] . -factor ( lower panel ) as functions of the phase variable @xmath44 in a pulse with @xmath20 . the color coding is the amplitude of the laser electric field . the arrows show the direction of the transverse electron velocity along the trajectory . ] the energy gain is limited by the dephasing between the electron and the wave . this aspect is illustrated in fig . [ figure_dla_vac_2 ] , where the electron trajectory and the @xmath42-factor are shown as functions of the phase variable @xmath44 . in the upper panel , the color - coding shows the amplitude of the laser electric field normalized to its maximum amplitude @xmath3 . the phase of the wave at the electron location , @xmath44 , continuously decreases , because the electron longitudinal velocity is less than the phase velocity @xmath5 . the electron , shown with a circle , is slipping with respect to the laser wave fronts , thus moving to the left in the upper panel . at @xmath45 , the electron transverse velocity ( shown with arrows ) is anti - parallel to the laser electric field , so the electron is gaining energy from the laser pulse . as the electron continues to slip , the laser electric field eventually changes its sign at @xmath46 and becomes positive . at this point , the electron transverse velocity is still positive and the electron starts to lose its energy . this positive field continues to reduce the transverse electron velocity until it becomes negative at @xmath47 , at which point the electron begins to gain energy again . the electron experiences energy gain twice every laser cycle , as evident from the plot of the @xmath42-factor in fig . [ figure_dla_vac_2 ] . it is worth pointing out that the electron comes momentarily to a complete stop also twice every laser cycle . the longitudinal distance that the electron travels between the stops to achieve @xmath48 is much longer than the laser wavelength due to the fact that the electron is moving forward and it can be estimated as @xmath49 . the discussed mechanism of electron acceleration and energy gain in the vacuum regime suggests that , in principle , there are two alternatives for further enhancing the electron energy gain . one option is to decrease the dephasing between the electron and the wave , allowing the transverse electron velocity to remain antiparallel to the laser electric field for longer in terms of the actual time . the second option is to change the oscillations of the transverse velocity , so that there is a net energy gain by the electron after a laser cycle . in what follows , we show how these scenarios can be realized utilizing relatively weak static longitudinal and transverse electric fields . as discussed in sec . [ sec - channel ] , one of the key features of the steady - state channel is the continuous injection of new electrons through the channel opening with a quasi - static longitudinal electric field . this aspect raises the question of the impact of the quasi - static longitudinal electric field on subsequent electron energy gain during acceleration by the laser pulse along the channel . in order to pinpoint the effect of the longitudinal electric field , we begin by considering electron motion in a vacuum using the same setup as in sec . [ sec - vac ] , but with an added narrow region @xmath50 of a uniform longitudinal static electric field @xmath51 . figure [ figure_dla_ez_1 ] shows numerical solutions of eqs . ( [ main_eq:1 ] ) and ( [ main_eq:2 ] ) for @xmath52 and @xmath53 , where @xmath3 is the electric field amplitude of the laser pulse . the laser pulse amplitude is again @xmath20 following a gradual initial ramp - up . the difference between the two examples shown in fig . [ figure_dla_ez_1 ] is the location of the field region along the electron trajectory . in the first example ( left panel ) , the interaction takes place when the electron momentum is close to its maximum value . the subsequent change in the maximum electron momentum after the interaction is insignificant . this result is in agreement with the expectation that the localized field we are considering is too weak to transfer significant energy to the electron . in the second example ( right panel ) , the interaction takes place as the electron comes to a stop . in stark contrast to the previous example , the maximum electron energy more than doubles during the motion in the wave after the interaction . that is @xmath53 long . the only difference between the two cases is the location of the field region . ] the key role of the longitudinal electric field in these examples is not to directly increase the electron energy , but rather to decrease the electron dephasing with the wave . we define the dephasing , @xmath54 as the rate at which the wave phase at the electron location , @xmath44 , changes with proper time , @xmath55 , defined as @xmath56 it follows directly from the definition of @xmath44 given by eq . ( [ xi ] ) that @xmath57 the quantity @xmath58 is conserved by eqs . ( [ main_eq:1 ] ) and ( [ main_eq:2 ] ) in the absence of the static electric field , which means that the dephasing defined by eq . ( [ r ] ) is constant before and after the interaction with the longitudinal field . it can also be shown that during the interaction the static field reduces the dephasing at the rate @xcite @xmath59 that is independent of the laser pulse amplitude . the reduced dephasing allows the electron to spend more time being accelerated by the wave and thus gain more energy from the laser pulse . as shown in fig . [ figure_dla_vac_2 ] of sec . [ sec - vac ] , the acceleration continues before the phase decreases by @xmath60 . a reduction in the dephasing then means that the corresponding proper time interval @xmath61 of electron acceleration is indeed longer [ see eq . ( [ r ] ) ] . the maximum @xmath42-factor that the electron with reduced dephasing achieves @xcite is @xmath62 fs from a 2d pic simulation with @xmath20 and @xmath63 . ] not all parts of the electron trajectory are equivalent in terms of the dephasing change induced by the longitudinal field . the reason for that is the strong dependence of the dephasing on the longitudinal momentum . for simplicity , let us assume that that the electric field instantaneously increases the longitudinal momentum by @xmath64 . prior to the interaction , we have @xmath65 here we used expressions ( [ vac_px ] ) and ( [ vac_py ] ) for the electron momentum in the vacuum regime . assuming that @xmath64 is smaller than @xmath66 and @xmath67 , we find from eq . ( [ r2 ] ) that the dephasing is reduced to @xmath68 evidently , the most favorable parts of the electron trajectory for reducing the dephasing are those parts where the longitudinal momentum is the lowest . this aspect is at work in the two examples shown in fig . [ figure_dla_ez_1 ] , where the dephasing drops to @xmath69 on the right and only to @xmath70 on the left . the resulting energy increase in both cases is in good agreement with eq . ( [ gamma_long ] ) . , on top of the time - averaged electron density profile . the lower panel shows the trajectory , with the color - coded dephasing rate @xmath71 , on top of the time - averaged longitudinal electric field . the field and the density are averaged over ten laser periods at 850 fs . ] in order to examine how this mechanism is realized in a plasma channel , we have performed a 2d pic simulation in which a long laser pulse with an electric field polarized perpendicular to the plane of the simulation irradiates a plasma slab with @xmath63 . detailed parameters of the simulation are given in appendix [ appendix_1 ] . the normalized laser amplitude is @xmath20 , so the @xmath42-factor in the vacuum regime would be limited by @xmath72 . however , the electron spectrum in the plasma , as seen in fig . [ figure_spectrum ] , has a pronounced energetic electron tail with @xmath73 . a trajectory of one of the energetic electrons is shown in fig . [ figure_dla_ez_2 ] together with a time - averaged electron density profile and longitudinal electric field . the electron is injected into the channel near the opening and initially moves against the laser pulse . the electron typical @xmath42-factor at this stage is comparable to @xmath8 . the quasi - static negative electric field , that is present in the channel opening , gradually turns the electron around . as the electron starts to move in the direction of the laser pulse propagation , the regime becomes similar to that considered earlier in this section . the longitudinal electric field causes the dephasing rate to decrease , as shown in the lower panel of fig . [ figure_dla_ez_2 ] . similarly to the simplified example , the significant drop in the dephasing occurs as the @xmath42-factor reaches its minimum at @xmath74 fs . the energy increase takes place as the electron continues its longitudinal motion into the channel after leaving the region with a negative quasi - static electric field . the reduced dephasing is evident not only from the color - coding in the lower panel of fig . [ figure_dla_ez_2 ] , but also from the lack of oscillations in @xmath75 in fig . [ figure_dla_ez_3 ] . the electron @xmath42-factor peaks at @xmath76 fs , roughly 20 @xmath19 m after the energy increase began . it exceeds @xmath77 by a factor of 2.3 . most of the energy comes directly from the laser electric field and not from the longitudinal field , as shown in the upper panel of fig . [ figure_dla_ez_3 ] . the laser contribution to the @xmath42-factor is calculated by integrating the work done by @xmath78 and normalizing it to @xmath79 . -factor and momentum components as functions of time for the electron trajectory shown in fig . [ figure_dla_ez_2 ] . the upper panel also shows the integrated contribution to the @xmath42-factor from the work done by the laser electric field . ] we have so far assumed that the phase velocity in the laser pulse , @xmath80 , is equal to the speed of light , @xmath5 . however , the plasma in the channel and the channel itself cause dispersion , making the phase velocity superluminal , @xmath81 . the effect of the superluminosity is faster dephasing between the electron and the wave and , as a result , lower energy gain from the wave . the superluminosity must be taken into account if it leads to a significant change in the dephasing . in general , the dephasing is determined by by the difference @xmath82 . we can then conclude that the superluminosity becomes important for @xmath83 , whereas we can set @xmath84 for @xmath85 . using the definitions @xmath86 and @xmath87 , we find that @xmath88 in the case when @xmath84 , the @xmath42-factor is given by eq . ( [ gamma_long ] ) . taking into account this estimate and the expression for @xmath77 , we find that the role of the superluminosity is negligible for @xcite @xmath89 ultimately , the superluminosity sets a limit on the energy gain from the wave . it follows from eq . ( [ v_ph_cond_2 ] ) that there is a critical value for @xmath86 , given by @xmath90 for @xmath91 , the dephasing between the electron and the wave is determined primarily by the difference @xmath92 and the difference between the electron longitudinal velocity and the speed of light is unimportant . this means that even if the longitudinal electric field lowers @xmath93 well below @xmath94 , the maximum @xmath42-factor that the electron can achieve will be approximately limited by @xmath95 which was estimated from eq . ( [ gamma_long ] ) by setting @xmath96 . is the location of the center of the window . the field is normalized to the electric field amplitude @xmath3 of a plane wave with @xmath20 . the dashed lines represent wavefronts moving with two different superluminal phase velocities @xmath80 . ] in order to examine the conditions ( [ v_ph_cond_2 ] ) and ( [ v_ph_cond_3 ] ) for the simulation shown in fig . [ figure_dla_ez_2 ] , we have plotted the wavefronts in a window moving with the speed of light along the axis of the channel . figure [ figure_v_ph ] shows the wave electric field @xmath78 on the axis of the channel in a window that is @xmath97 wide . the center of the window , located at @xmath98 , slides the entire length of the channel shown in fig . [ figure_dla_ez_2 ] ( note that @xmath18 @xmath19 m in the simulation ) . the two dashed lines indicate wavefront slopes for @xmath99 and @xmath100 . inside the channel we then approximately have @xmath99 . the corresponding critical dephasing is @xmath101 and the maximum @xmath42 that can be attained by accelerating electrons is @xmath102 , which translates into @xmath103 for @xmath20 . in the light of these estimates , a closer look at the electron data shown in figs . [ figure_dla_ez_2 ] and [ figure_dla_ez_3 ] reveals that the superluminosity likely plays a role in limiting the electron energy gain . the dephasing rate @xmath93 shown in fig . [ figure_dla_ez_2 ] falls to @xmath104 along the electron trajectory , which could lead to a factor of ten increase of the @xmath42-factor compared to @xmath77 for @xmath84 . however , our estimates show that reduction of the dephasing below @xmath101 no longer leads to an enhancement of the @xmath42-factor , capped at @xmath103 , because of the superluminosity . this is in good agreement with the @xmath42-profile shown in the upper panel of fig . [ figure_dla_ez_3 ] . finally , it is worth pointing out that heating of the bulk electrons in the channel to relativistic energies significantly reduces the superluminosity . the phase velocity of a linear electromagnetic wave propagating through a cold non - relativistic plasma with electron density @xmath105 is @xmath106 . we then should expect @xmath107 for @xmath63 used in the simulation . the corresponding wavefront should follow the right dashed line in fig . [ figure_v_ph ] . the actual @xmath92 is smaller by an order of magnitude , which clearly indicates that the critical density that determines the phase velocity is effectively lowered by an order of magnitude due to relativistic motion induced in the channel by the laser pulse . this so - called relativistic transparency enables the electron energy gain enhancement by reducing the superluminosity and thus reducing the critical dephasing given by eq . ( [ critical_dephasing ] ) . inside the channel , the electron acceleration by the laser pulse takes place in the presence of a relatively weak quasi - static transverse electric field . in order to examine the role played by such a field , we consider just a single electron irradiated by a plane electromagnetic wave in a fully evacuated cylindrical ion channel shown in fig . [ figure_cyl_setup ] . -plane . ] the electron dynamics in this model is described by eqs . ( [ main_eq:1 ] ) and ( [ main_eq:2 ] ) , where the electric and magnetic fields are given . these fields are a superposition of the fields of the wave given by eqs . ( [ e_wave ] ) and ( [ b_wave ] ) and the electric field of the channel , @xmath108 where @xmath109 . we have assumed for simplicity that the channel is an infinitely long uniform cylinder that consists of immobile singly - charged ions with density @xmath110 . it can be shown that , as the electron moves along the channel , the following quantity remains conserved @xmath111 this relation indicates that amplification of electron oscillations across the channel leads to a reduction of @xmath58 . since @xmath66 is the dominant component of the electron momentum , a significant reduction of @xmath58 implies a significant increase of @xmath66 . therefore , the relation ( [ q_def ] ) formally points at a direct connection between enhancement of transverse oscillations and a significant increase of the longitudinal electron momentum . the integral of motion ( [ q_def ] ) sets an upper limit for the amplitude of the transverse oscillations across the channel . the value of @xmath112 is determined by the initial conditions , so that @xmath113 for an on - axis electron that is initially at rest . we find from eq . ( [ q_def ] ) that the amplitude of the electron oscillations can not exceed @xmath114 for @xmath113 . the amplitude of the oscillations approaches @xmath115 as @xmath116 and @xmath117 . it is important to distinguish transverse electron oscillations in the @xmath36 and @xmath23 directions , because the motion in the @xmath36-direction is driven by the laser electric field , whereas the motion in the @xmath23-direction is only affected by the electric field of the channel . we first focus on the key aspects of the driven electron oscillations . we consider an electron that has no initial displacement or momentum in the @xmath23-direction , so that the electron trajectory driven by the laser pulse remains flat , with the electron moving only in the @xmath22-plane . the equation for the electron motion across the channel can be written in the form that resembles an equation for a driven oscillator @xcite : @xmath118 where @xmath119 and @xmath55 is the proper time defined by eq . ( [ tau_def ] ) . the wave amplitude is a function of the phase variable @xmath44 that changes at the rate @xmath120 there are two characteristic frequencies in this case : the frequency of the oscillations induced by the field of the channel electric field and the frequency of the oscillations induced by the field of the laser pulse . the former is the natural frequency @xmath121 . the latter is equal to @xmath4 in the vacuum regime , because @xmath122 and , as a result , @xmath123 . according to eq . ( [ q_def ] ) , the relation @xmath124 holds in the presence of the transverse electric field as well while the amplitude of the transverse oscillations is much smaller than @xmath125 . at low channel densities , such that @xmath126 , the channel electric field has very little impact on the electron motion . the frequency mismatch means that a resonant interaction that can lead to an amplification of the transverse oscillations is not possible . the electron motion is essentially the same as in the vacuum regime , with @xmath127 . the characteristic amplitude of the transverse oscillations , estimated as @xmath128 from eq . ( [ perp_osc ] ) , is indeed much smaller than @xmath115 for @xmath126 and @xmath127 . -factor ( lower panel ) of an electron irradiated by a plane wave with @xmath129 in a channel whose ion density corresponds to @xmath130 . the color coding is the amplitude of the laser electric field . the arrows show the direction of the transverse electron velocity along the trajectory . ] at higher channel densities for which the two characteristic frequencies become comparable , @xmath131 , the channel electric field becomes capable of significantly changing the phase of the transverse oscillations , which causes their amplitude to grow . the change of phase after significant amplification of the transverse oscillations has already taken place is illustrated in fig . [ figure_dla_vac_3 ] , where the electron trajectory and the @xmath42-factor are shown as functions of the phase variable @xmath44 . in this example , the electron is initially at rest on the axis of the channel ( @xmath132 ) whose ion density corresponds to @xmath130 . the electron is irradiated by a plane wave with @xmath133 and @xmath129 . the upper panel of fig . [ figure_dla_vac_3 ] shows that the channel electric field changes the oscillations of the transverse electron velocity , allowing it to remain anti - parallel to the laser electric field over extended segments of the electron trajectory marked with white . as a result , there is a net energy gain each laser cycle for @xmath134 that leads to a significant increase of the relativistic @xmath42-factor well above what can be achieved in the vacuum regime illustrated in fig . [ figure_dla_vac_2 ] . note that @xmath135 along this segment of the electron trajectory . the example in fig . [ figure_dla_vac_3 ] confirms that a significant increase of the @xmath42-factor goes hand in hand with the enhancement of the transverse oscillations . + in order to determine the conditions for the energy enhancement , we have performed a parameter scan solving eqs . ( [ main_eq:1 ] ) and ( [ main_eq:2 ] ) numerically for a wide range of parameters @xmath8 and @xmath136 that specify the fields acting on the electron . in all the cases , the electron is initially at rest on the axis of the channel and the laser pulse amplitude gradually ramps up as @xmath137 \sin(\xi ) , & \text{for $ \xi > \xi_0$;}\\ a_0 \sin(\xi ) , & \text{for $ \xi \leq \xi_0 $ } \end{cases } \label{a}\ ] ] where @xmath138 and @xmath139 . the maximum @xmath42-factor , @xmath140 , that the electron attains for different sets of parameters @xmath8 and @xmath136 is shown in fig . [ invited_pop_figure_8 ] , where @xmath140 is normalized to the maximum @xmath42-factor in the vacuum regime , @xmath77 , for the same laser amplitude @xmath8 . the energy enhancement has a sharp threshold , as further illustrated by the lineouts in fig . [ invited_pop_figure_8 ] for three different values of @xmath8 . the lineouts for @xmath141 and @xmath142 indicate that the threshold and the energy gain above the threshold are determined by a single dimensionless combination @xmath143 the contours of constant @xmath144 plotted in the upper panel of fig . [ invited_pop_figure_8 ] with dashed lines confirm that this is indeed a general trend for @xmath145 . the existence of the threshold that depends only on @xmath146 was first discovered in ref . [ ] and then explored in more detail in ref . the dimensionless parameter @xmath144 is the ratio of the frequencies of oscillations induced by the field of the channel and by the field of the laser prior to a significant energy enhancement . indeed , taking into account that the electron @xmath42-factor in this regime is essentially @xmath77 , we find that @xmath147 for @xmath41 . currently , the dependence of @xmath140 on @xmath144 above the threshold remains a robust numerical observation that requires an in - depth theoretical analysis . -factor attained by injected electrons . the laser amplitude in all the cases is @xmath20 . the initial injection phase defines the initial wave amplitude , @xmath148 . ] in the scan presented in fig . [ invited_pop_figure_8 ] we have considered the case where initially the electron is already in the channel , but the laser pulse has not yet reached the electron location longitudinally . on the other hand , in the steady - state channel discussed in sec . [ sec - channel ] and illustrated in fig . [ fig:2d_example ] , electrons are injected into the channel from the side with the laser beam already present in the channel . this raises the question regarding the role of electron injection in determining both the threshold and the energy gain . we test the role of injection by seeding an electron without any initial momentum into a plane wave with a given initial phase . specifically , the wave amplitude is set to @xmath149 , where @xmath150 is the phase of the laser pulse at the moment of injection defined as @xmath151 . the electron is placed onto the channel axis at the moment of injection . figure [ figure_injection ] shows how the maximum @xmath42-factor attained by the electron depends on the injection phase @xmath150 and the ion density in the channel . we show only the range of initial phases @xmath152 , because the plot for @xmath153 is identical . the energy enhancement threshold is present for all the injection phases in fig . [ figure_injection ] , but the location of the threshold is sensitive to the injection phase . the threshold is the lowest when initially the electric field of the wave is equal to zero , @xmath154 . since the wave amplitude is fixed , this means that the injection phase determines the value of the parameter @xmath144 at the threshold . it should also be pointed that the energy gain increases as the threshold becomes lower . therefore , we can conclude that proper electron injection into the laser beam can significantly enhance the electron energy gain from the wave , while lowering the corresponding threshold determined by the product @xmath146 . -plane ( shown in light - blue ) . ] we have so far deliberately limited the electron motion to the plane of the driven oscillations by placing the electron initially on the axis of the channel . small off - axis displacements in the absence of the laser pulse lead to small free oscillations across the channel since the restoring force is no longer equal zero . in what follows , we examine how the out - of - plane oscillations evolve in the presence of a laser pulse with relativistic amplitude @xmath41 . the equation for the the out - of - plane oscillations can be directly obtain from eq . ( [ perp_osc ] ) by taking into account that the channel is cylindrically symmetric . by setting the right - hand side to zero and replacing @xmath36 with @xmath23 we obtain an equation , @xmath155 that resembles an equation for an oscillator with a natural frequency @xmath121 . the free oscillations along the @xmath23-axis are coupled to the driven electron motion in the @xmath22-plane via the relativistic @xmath42-factor , @xmath156 . as long as the amplitude of the free oscillations remains small , the @xmath42-factor is determined predominantly by the driven motion . taking the @xmath42-factor in the vacuum regime shown in fig . [ figure_dla_vac_2 ] as an example of the driven motion , we can conclude that the driven motion affects the natural frequency @xmath121 in two ways : it increases the natural frequency roughly by a factor of @xmath8 for @xmath41 and it also strongly modulates the natural frequency . the first aspect has been discussed in detail earlier in this section and it is responsible for the onset of the significant energy enhancement . the modulation of the natural frequency can make the free oscillations along the @xmath23-axis parametrically unstable . the @xmath42-factor modulations effectively modulate the restoring force acting on the oscillator described by eq . ( [ free_perp_osc ] ) . it is then intuitively clear that the oscillations remain stable as long as the modulation frequency is much higher than the natural frequency @xmath121 . the oscillations become unstable when the two frequencies become comparable . as already discussed , the driven electron motion below the energy enhancement threshold can be approximated by the solution for the vacuum regime . we again consider an electron that is irradiated by a plane wave in the channel and assume that the initial displacement is much smaller than @xmath115 . in this case , @xmath157 and we then find from the @xmath42 profile shown in fig . [ figure_dla_vac_2 ] that the modulation frequency is @xmath158 . on the other hand , significantly below the threshold we have @xmath126 , as evident from fig . [ invited_pop_figure_8 ] . therefore , free oscillations should be stable below the threshold . as we cross the threshold by increasing the ion density , two important changes in the dependence of the @xmath42-factor on @xmath55 take place . the @xmath42-factor is significantly increased , which greatly increases the natural frequency . in addition to that , the modulations of the @xmath42-factor become less frequent . this is because the peaks of enhanced @xmath42 act as modulations and they take multiple laser cycles to develop . the combination of these two factors changes the relation between the two characteristic frequencies to @xmath159 . we therefore conclude that the energy enhancement changes the stability of the free oscillations , making them susceptible to the parametric instability . a parameter scan over the same range of @xmath8 and @xmath136 as in fig . [ invited_pop_figure_8 ] reveals that there is a threshold for the amplification of the free transverse oscillations @xcite . this threshold matches the energy enhancement threshold ( dashed line in fig . [ invited_pop_figure_8 ] ) , because the energy enhancement triggers the growth of the electron oscillations along the @xmath23-axis . the amplitude of the free oscillations grows until it becomes comparable to @xmath115 . the amplitude of the driven oscillations , i.e. the oscillations along the @xmath36-axis , also has an enhancement threshold that matches the energy enhancement threshold as well . we find that , in general , the electron energy enhancement is accompanied by a considerable enhancement of the transverse oscillations across the channel , making the electron trajectory considerably three - dimensional with the maximum displacement comparable to @xmath125 . to illustrate this point , we show in fig . [ figure_9 ] electron trajectories for @xmath20 and @xmath160 . these parameters are above the threshold according to fig . [ invited_pop_figure_8 ] . in the upper panel , the electron has no initial out - of - plane displacement , so its trajectory remains flat . however , an initial displacement of @xmath161 in the lower panel provides a necessary seed for the instability to develop significant out - of - plane oscillations . as a result , a non - planar electron trajectory quickly develops , accompanying the energy enhancement . we have examined a regime in which a laser pulse with relativistic intensity irradiates a sub - critical plasma over a time period much longer than the characteristic electron response time . it is shown that a steady - state channel is formed in the plasma in this case with quasi - static transverse and longitudinal electric fields . these fields , even though they are much smaller than the electric field in the laser pulse , profoundly alter the electron dynamics . the longitudinal electric field reduces the longitudinal dephasing between the electron and the wave . this allows the electron to gain significantly more energy from the wave . the energy gain in this regime is ultimately limited by the superluminosity of the wave fronts induced by the plasma in the channel , even though the relativistic transparency greatly reduces the role of the plasma . the transverse electric field alters the oscillations of the transverse electron velocity . as a result , it can remain anti - parallel to the laser electric field , which leads to significant energy gain by the electron from the wave . we showed that this process has a sharp threshold determined by a single parameter , @xmath146 . the threshold can be greatly reduced and the energy gain further enhanced by appropriately injecting electrons into the laser pulse . the threshold for the energy enhancement matches the threshold for the onset of the parametric instability in the direction perpendicular to the plane of the driven electron oscillation . therefore , the trajectories of electrons with enhanced energies eventually become non - planar if the interaction time and length with the pulse are sufficient for the instability to develop . the effect of the longitudinal electric field is akin to pre - acceleration of electrons in the longitudinal direction . analysis of pre - accelerated electrons in a channel @xcite has showed that pre - acceleration reduces the energy enhancement threshold and increases the electron energy gain . this indicates that the transverse and longitudinal electric fields can synergistically enhance the electron energy gain . further in - depth research is required to determine the energy scalings and the key parameters that determine the electron dynamics . finally , it should be pointed out that even though the presented analysis was motivated by laser - plasma interactions at significantly sub - critical plasma densities , most of the results are much more general and they are not limited just to the regime where @xmath162 . therefore , the analysis is well suited to make meaningful predictions regarding electron acceleration in near - critical and over - critical plasmas , provided that such plasmas are relativistically transparent to the incoming high - intensity laser pulse @xcite . however , wave propagation becomes a crucial aspect in this case and it should be addressed self - consistently @xcite taking into account transverse laser pulse dimensions and its polarization @xcite . ava would like to thank dr . toma toncian for constructive comments . this material is based upon work supported by the u.s . department of energy [ national nuclear security administration ] under award number de - na0002723 . was also supported by afosr contract no . fa9550 - 14 - 1 - 0045 , u.s . department of energy - national nuclear security administration cooperative agreement no . de - na0002008 , and u.s . department of energy contract no . de - fg02- 04er54742 . v.n.k . and g.s . were supported by afosr contract no . fa9550 - 14 - 1 - 0045 and u.s . department of energy contract nos . de - sc0007889 and de - sc0010622 . simulations for this paper were performed using the epoch code ( developed under uk epsrc grants ep / g054940/1 , ep / g055165/1 and ep / g056803/1 ) using hpc resources provided by the texas advanced computing center at the university of texas . the snapshots shown in figs . [ fig:2d_example ] and [ figure_dla_ez_2 ] are from two - dimensional particle - in - cell simulations performed using an open - source code epoch @xcite . in both simulations , the initial electron density profile is given by @xmath163 , & \text{for $ x < x_0$;}\\ n_0 , & \text{for $ x \geq x_0 $ } , \end{cases } \label{n_sim}\ ] ] where @xmath164 @xmath19 m and @xmath165 @xmath19 m . in fig . [ fig:2d_example ] , we set @xmath166 , whereas , in fig . [ figure_dla_ez_2 ] , we set @xmath167 . the electron population is initialized using 10 macro - particles per cell . the ion density is initially equal to the electron density and the ion density profile is initialized using 4 macro - particles per cell . the electrons are initially cold , whereas the ions are immobile throughout the simulation . the size of the domain in the simulation shown in fig . [ fig:2d_example ] is 180 @xmath19 m ( 12000 cells ) along the @xmath21-axis and 60 @xmath19 m ( 1200 cells ) along the @xmath36-axis . the size of the domain in the simulation shown in fig . [ figure_dla_ez_2 ] is 140 @xmath19 m ( 9500 cells ) along the @xmath21-axis and 60 @xmath19 m ( 1200 cells ) along the @xmath36-axis . the longitudinal resolution is chosen in agreement with the criterion outlined in ref . [ ] in order to correctly compute the electron energy gain . we use a laser pulse whose focal plane in the absence of plasma is located at @xmath168 @xmath19 m . the laser fields have a gaussian profile along the @xmath36-axis , with the electric and magnetic fields in the focal plane given by @xmath169 where @xmath170 @xmath19 m and the temporal profile is @xmath171 , & \text{for $ t < t_0$;}\\ 1 , & \text{for $ t \geq t_0 $ } , \end{cases } \label{e_sim}\ ] ] with @xmath172 fs and @xmath173 fs . in both simulations , the laser wavelength is 1 @xmath19 m and the peak laser intensity at @xmath174 @xmath19 m is @xmath175 w/@xmath1 , which corresponds to @xmath20 . the normalizing amplitudes @xmath3 and @xmath24 are maximum electric and magnetic fields in a plane wave with @xmath20 . in the simulation shown in fig . [ fig:2d_example ] , the laser electric field has @xmath21 and @xmath36 components , whereas the laser magnetic field has only a @xmath23-component . in the simulation shown in fig . [ figure_dla_ez_2 ] , the laser electric field has only a @xmath23-component , whereas the laser magnetic field has @xmath21 and @xmath36 components . 10 h .- s . park , d. m. chambers , h .- k . chung , r. j. clarke , r. eagleton , e. giraldez , t. goldsack , r. heathcote , n. izumi , m. h. key , j. a. king , j. a. koch , o. l. landen , a. nikroo , p. k. patel , d. f. price , b. a. remington , h. f. robey , r. a. snavely , d. a. steinman , r. b. stephens , c. stoeckl , m. storm , m. tabak , w. theobald , r. p. j. town , j. e. wickersham , and b. b. zhang , phys . plasmas * 13 * , 056309 ( 2006 ) . s. kneip , s. r. nagel , c. bellei , n. bourgeois , a. e. dangor , a. gopal , r. heathcote , s. p. d. mangles , j. r. marques , a. maksimchuk , p. m. nilson , k. ta phuoc , s. reed , m. tzoufras , f. s. tsung , l. willingale , w. b. mori , a. rousse , k. krushelnick , and z. najmudin , phys . * 100 * , 105006 ( 2008 ) . i. pomerantz , e. mccary , a. meadows , a. arefiev , a. bernstein , c. chester , j. cortez , m. donovan , g. dyer , e. gaul , d. hamilton , d. kuk , a. lestrade , c. wang , t. ditmire , and b. hegelich , phys . * 113 * , 184801 ( 2014 ) . m. schollmeier , a. b. sefkow , m. geissel , a. v. arefiev , k. a. flippo , s. a. gaillard , r. p. johnson , m. w. kimmel , d. t. offermann , p. k. rambo , j. schwarz , and t. shimada , phys . plasmas * 22 * , 043116 ( 2015 ) . h. chen , a. link , y. sentoku , p. audebert , f. fiuza , a. hazi , r. f. heeter , m. hill , l. hobbs , a. j. kemp , g. e. kemp , s. kerr , d. d. meyerhofer , j. myatt , s. r. nagel , j. park , r. tommasini , and g. j. williams , phys . plasmas * 22 * , 056705 ( 2015 ) . j. fuchs , c. a. cecchetti , m. borghesi , t. grismayer , e. dhumieres , p. antici , s. atzeni , p. mora , a. pipahl , l. romagnani , a. schiavi , y. sentoku , t. toncian , p. audebert , and o. willi , phys . lett . * 99 * , 015002 ( 2007 ) . h. w. powell , m. king , r. j. gray , d. a. maclellan , b. gonzalez - izquierdo , l. c. stockhausen , g. hicks , n. p. dover , d. r. rusby , d. c. carroll , h. padda , r. torres , s. kar , r. j. clarke , i. o. musgrave , z. najmudin , m. borghesi , d. neely , and p. mckenna , new j. phys . * 17 * , 103033 ( 2015 ) . d. mangles , b. r. walton , m. tzoufras , z. najmudin , r. j. clarke , a. e. dangor , r. g. evans , s. fritzler , a. gopal , c. hernandez - gomez , w. b. mori , w. rozmus , m. tatarakis , a. g. r. thomas , f. s. tsung , m. s. wei , and k. krushelnick phys . 94 * , 245001 ( 2005 ) . l. willingale , p. m. nilson , a. g. r. thomas , j. cobble , r. s. craxton , a. maksimchuk , p. a. norreys , t. c. sangster , r. h. h. scott , c. stoeckl , c. zulick , and k. krushelnick , phys . * 106 * , 105002 ( 2011 ) . l. willingale , a. g. r. thomas , p. m. nilson , h. chen , j. cobble , r. s. craxton , a. maksimchuk , p. a. norreys , t. c. sangster , r. h. h. scott , c. stoeckl , c. zulick , and k. krushelnick , new j. phys . * 15 * , 025023 ( 2013 ) . j. peebles , c. mcguffey , c. m. krauland , l. c. jarrott , a. sorokovikova , m. s. wei , j. park , h. chen , h. s. mclean , c. wagner , m. spinks , e. w. gaul , g. dyer , b. m. hegelich , m. martinez , m. donovan , t. ditmire , s. i. krasheninnikov , and f. n. beg , nucl . fusion * 56 * , 016007 ( 2016 ) . t. d. arber , k. bennett , c. s. brady , a. lawrence - douglas , m. g. ramsay , n. j. sircombe , p. gillies , r. g. evans , h. schmitz , a. r. bell , and c. p. ridgers , plasma phys . fusion * 57 * , 113001 ( 2015 ) . t. toncian , c. wang , e. mccary , a. meadows , a.v . arefiev , j. blakeney , k. serratto , d. kuk , c. chester , r. roycroft , l. gao , h. fu , x. q. yan , j. schreiber , i. pomerantz , a. bernstein , h. quevedo , g. dyer , t. ditmire , and b. m. hegelich , matter and radiation at extremes ( 2016 ) [ http://dx.doi.org/10.1016/j.mre.2015.11.001 ] . b. gonzalez - izquierdo , r. j. gray , m. king , r. j. dance , r. wilson , j. mccreadie , n. m. h. butler , r. capdessus , s. hawkes , j. s. green , m. borghesi , d. neely , and p. mckenna , nature physics ( 2016 ) [ doi:10.1038/nphys3613 ] .

symmetry is one of the basic principles that penetrate all of physics : from classical to quantum physics @xcite ; from statistical or condensed matter physics to particle physics @xcite ; and from relativity to gauge field theory . thus we see a wide range of symmetries : from external ( space - time ) to internal ones ; and from discrete to continuous ones . this is partly because geometry is an indispensable element to describing physics . and a natural algebraic language to express geometrical symmetry is group theory . @xcite classical mechanics is widely known to have close relations between symmetries and lagrangians or hamiltonians.@xcite since quantum mechanics and field theory are , in some respects , modeled and devised after classical mechanics , we see that not a few methods and notions , including symmetry , in quantum mechanics and field theory resemble those in classical mechanics . and thus symmetry plays a crucial role also in quantum physics . or we can interpret that although `` the physical world is quantum mechanical'',@xcite the quantum feature are somewhat transmitted to the classical world ; and through the latter we may try to grasp the former . of course since we are not able to capture all nature by classical analogies , there are sometimes discrepancies between quantum symmetries and classical ones . @xcite now , one of the typical mathematical tools that relate classical states with quantum ones is coherent states ( cs ) . @xcite it was originally devised by schrdinger @xcite as the states having classical `` particle '' nature . the state exhibits a wave packet whose center moves along with the classical trajectory with minimum uncertainty ; thus it shows classical nature . the original cs which is called canonical cs is , in the light of quantum optics , generated by displacing , or driving , the vacuum , i.e. the zero photon state . @xcite later cs have been developed in a wide variety of directions . viewed from a general framework , we may take up the following three subjects among the evolutions . first , cs have been extended to wider classes . a systematic way to broaden cs is constructing cs in terms of unitary irreducible representations of lie groups due to perelomov.@xcite in the approach , cs is defined by operating a unitary operator related to a physical system being considered on a `` fiducial vector ( fv ) '' , which we denote @xmath4 . from this point of view , for the canonical cs the unitary operator is a displacement operator and a fv is the ground state or vacuum . similarly , the spin cs can be constructed by operating a rotation operator on a fv ; the fv is conventionally taken as @xmath5 or @xmath6 : the highest or lowest eigenvectors of @xmath2 . we may perform the procedures to other lie groups , which automatically produces cs for the corresponding lie groups . schur s lemma coming from irreducibility always ensures the overcomplteness of cs . second , since cs enjoy overcomplete relations , `` coherent state path integrals ( cspi ) '' , i.e. , path integrals ( pi ) via cs , have been developed.@xcite such cspi have been pushing the method of pi forward strongly . and besides cspi turned out to be closely related to geometric phases @xcite ; in fact it is remarkable that geometric phases follow from the topological terms of phase space pi or cspi naturally . @xcite third , following the fruits of the above two developments , cs and cspi with arbitrary fv have been explored . @xcite in the case cs are obtained by operating a unitary operator on an _ arbitrary _ fv : for the canonical cs a fv is arbitrary superpositions of the fock number states ; and for the spin cs arbitrary superpositions of @xmath7 : a general eigenvector of @xmath2 . we found that , as mentioned in ref . , the canonical cs evolving from arbitrary fv@xcite turned out to be an arbitrary superposition of displaced number states having no classical analogues . similarly , we may regard spin cs evolving from a general fv other than the conventional one as quantum states without classical analogues . it is true that cs with the conventional fv are closest to classical states and have useful properties . @xcite however , recent technologies enable us to prepare quantum states which have no classical analogues ; the typical one is the squeezed states of light . @xcite we certainly regard the evolutions plausible since experimental developments due to high technologies have often created opportunities to reconsider nature . in this respect we may take cs evolving from a generic fv as the mathematical tools , or a sort of new language , for describing non - classical quantum states . and thus what we have done in refs . and is constructing new quantum states and investigating the dynamics : i.e. , the time evolutions of the quantum states . we can interpret the attempts as extending both cs due to schrdinger - klauder - glauber - perelomov and pi due to dirac - feynman - klauder - kuratsuji - suzuki . in the previous paper,@xcite hereafter referred to as * i * , we have developed a basic formulation of the su(2 ) , i.e. spin , cs based on arbitrary fv and of their pi . the cs and cspi are labeled by a full set of three euler angles @xmath8 . since the present paper flows directly out of * i * , we will look back the previous results concisely . in * i * we found out that the lagrangian in the action appeared in the pi expression were composed of two parts : the topological term related to geometric phases and the dynamical one originating from a hamiltonian . and the former is again split into two parts : one is the monopole type part which is the generalization of that of balachandran _ et al._@xcite and the other represents the effect of entanglements between neighboring components of a fv . such interweaving components of a fv appear in the dynamical term as well . the monopole is fictitious in that it does not represent a real physical monopole having a magnetic charge ; instead it stems from the topological or geometric phase terms . however , mathematical descriptions seem quite common to both real and fictitious monopoles . and we have confirmed the pi form by demonstrating from discrete pi to continuous ones . moreover , it has been proved that the generic spin cspi contract to the general canonical cspi in the high spin limit . at first sight it seems that we are free to choose fv ; there are no restrictions on fv and we may take an arbitrary fv . however , when a lagrangian varies at most a total derivative under a certain gauge transformation and possesses a sort of semiclassical symmetry , a full quantum state with an arbitrary fv does not always preserve the related symmetry . in the case , when a fv @xmath4 can not be reached from @xmath7 via @xmath9 , we find some strange feature : semiclassical orbits do not always represent exact quantal evolutions . it is stone who first observed that there does exist one of the central problems at the point viewed from a general framework of cs with arbitrary fv . he raised the problem in a paper@xcite commenting on the precursory version of * i * @xcite ; moreover , he went further enough to propose a criterion under which the cs capture full quantal evolutions . according to that , an arbitrary fv is not always realized and that there may be restrictions on fv so that quantum evolutions are consistent with the semiclassical ones which has an original symmetry . actually , the fv have to be identical with @xmath7 or on the orbits of @xmath7 under the action of @xmath10 . and , as stone precisely pointed out , the problem is deeply related to the charge quantization of monopoles . in this article we consider the problem posed in ref . from the general framework of spin cspi developed in * i*. as mentioned earlier , we have demonstrated the process of going from the discrete pi to the continuous pi in * i * ; we have also showed that the spin cspi contract to the canonical cspi . so the pi expressions in * i * are quite all right and are not responsible for not bringing the restriction on fv . then one might wonder from where the restriction comes . we will approach the riddle in the light of the `` gauge symmetry '' associated with the invariance of lagrangian appeared in the spin cspi in * i*. so the present article is also concerned with `` symmetry '' . it will be one of the early attempts that relate cs and cspi with symmetry , especially gauge symmetry . the plan of the paper is as follows . first , we look back the spin(su(2))cspi based on arbitrary fv in @xmath11 [ sec : spincspi ] . next , using the formulation of the spin cspi , we discuss general properties of a lagrangian in the light of gauge symmetries in @xmath11 [ sec : symlagfv ] . next , in @xmath11 [ sec : lags ] we demonstrate , using simple examples , the relations between types of fv and semiclassical as well as full quantal dynamics . we then look over several real examples of lagrangians in order to see the relations between hamiltonians , fv and gauge symmetries of the whole lagrangians . main results are presented as theorems and proved in @xmath11 [ sec : gentheorems ] . theorem [ th : typefv ] gives the central result concerning the restriction on fv in the full quantum picture . we find that gauge symmetries bring restrictions on fv and thus on the form of cs . we look into the situation much deeper by investigating the generator of the symmetry transformation in theorem [ th : gform ] . next , we revisit the gauge symmetries in the light of a new kind of isotropy subgroups due to ref . in @xmath11 [ sec : anotherviewscevolvs ] ; and we see the correspondence between the approach and that in @xmath11 [ sec : symlagfv ] and @xmath11 [ sec : lags ] . finally we summarize the results in @xmath11 [ sec : discussion ] . there we also discuss the related topics that we view in a future prospect . let us recall the results in * i*. first , we define the spin(su(2))cs , @xmath12 , evolving from an arbitrary fv @xmath4 as : @xmath13 the fv @xmath4 is represented by : @xmath14 hereafter @xmath15 stands for @xmath7 . from ( [ eqn : cs1 ] ) and ( [ eqn : spinfv ] ) we obtain @xmath16 see ( * i * - 19 ) ) denotes eq . ( @xmath17 ) in * i*. ] for the explicit form of @xmath18 which we do not need in the present paper . we called @xmath18 the `` rotated spin number state '' in * i * where we saw that it corresponded to the `` displaced number state '' in the general canonical cs . @xcite then the quantum time evolution of a physical system with a hamiltonian @xmath19 in terms of @xmath12 is given by the propagator : @xmath20 \ } \ , { \cal d } [ { \bf\omega}(t ) ] , \label{eqn : pi}\ ] ] where @xmath21 \equiv \int_{t_i}^{t_f } \bigl [ \ { { \left\langle \bf\omega\right\vert } } i \hbar \frac{\partial}{\partial t } { { \left\vert \bf\omega\right\rangle } } - h({\bf\omega } , t ) \ \bigr ] \ , d t \equiv \int_{t_i}^{t_f } l({\bf\omega } , { \bf \dot \omega},t ) \ , d t , \label{eqn : action - pol}\ ] ] with @xmath22 the explicit form of the lagrangian yields : @xmath23 - h({\bf\omega } , t ) , \label{eqn : lagpol}\ ] ] where @xmath24 the following expressions include what @xmath25 and @xmath26 mean : @xmath27 \\ a_2({\bf\omega } ; { { \ { c_m \ } } } ) = ( 1 / 2 ) \sum_{m = - s + 1}^{s } f(s , m ) \exp(i \phi ) \ { ( 1 + \cos\theta ) \exp(i \psi ) c_m^ { * } c_{m-1 } \\ \qquad \qquad - ( 1 - \cos\theta ) \exp(- i \psi ) c_m c_{m-1}^ { * } \ } \\ a_4(\psi ; { { \ { c_m \ } } } ) \equiv [ ( 1 / ( 2 i ) ] \sum_{m =- s+1}^{s } f(s , m ) [ c_m^ { * } c_{m-1 } \exp(i \psi ) - c_m c_{m-1}^ { * } \exp(- i \psi ) ] . \\ f(s , m ) = [ ( s + m)(s - m + 1)]^{1/2}. \end{array } \right . \label{eqn : defa}\ ] ] the term with the square brackets in the lagrangian ( [ eqn : lagpol ] ) : @xmath28 stemming from @xmath29 , may be called the `` topological term '' that is related to the geometric phases . for operators @xmath30 that constitute @xmath31 , we have @xmath32 where @xmath33 and @xmath34 have already been given in ( [ eqn : defa ] ) and @xmath35 . variational equations associated with the lagrangian ( [ eqn : lagpol ] ) are : @xmath36 { \dot \phi } + a_1(\psi ; { { \ { c_m \ } } } ) { \dot \psi } \ } = - ( \partial h / \partial \theta ) \\ \hbar \ { [ a_0({{\ { c_m \ } } } ) \sin\theta + a_1(\psi ; { { \ { c_m \ } } } ) \cos\theta ] { \dot \theta } - [ a_4(\psi ; { { \ { c_m \ } } } ) \sin\theta ] { \dot \psi } \ } = \partial h / \partial \phi \\ \hbar \ { [ a_4(\psi ; { { \ { c_m \ } } } ) \sin\theta ] { \dot \phi } + a_1(\psi ; { { \ { c_m \ } } } ) { \dot \theta } \ } = { \partial h / \partial \psi } , \label{eqn : vareq - pol } \end{array } \right.\ ] ] following the results in @xmath11 [ sec : spincspi ] , we now consider the relations between semiclassical time evolutions of @xmath12 and the properties of symmetries that lagrangians possess . we see from ( [ eqn : lagpol])-([eqn : defa ] ) that the @xmath1-variable does not take effect in @xmath37 provided no neighboring @xmath38 exists for any @xmath39 ; then @xmath40-terms vanishes and the topological term ( [ eqn : topol - term0 ] ) takes the form : @xmath41 the form of ( [ eqn : toptermsym ] ) is a generalization of that in ref . . taking @xmath42 in ( [ eqn : topol - term0 ] ) yields the latter form . for such a fv , if we assume that @xmath31 is linear in @xmath43 , we see from ( [ eqn : mat - pol ] ) that @xmath44 does not depend on @xmath1 . consequently the form of the variational equations ( [ eqn : vareq - pol ] ) becomes the same as that for the usual spin cspi evolving from @xmath45 , or @xmath46 with @xmath47 replaced with @xmath48 . and thus we can choose any @xmath1 as far as semiclassical dynamics is concerned . in what follows we will investigate a little deeper such fv and @xmath31 as above - mentioned that yield ( [ eqn : toptermsym ] ) and leave semiclassical dynamics invariant . from the viewpoint of the symmetry of lagrangian , we grasp the situation as follows . to begin with , in the present case we have @xmath49 under the `` gauge @xmath1-transformation '' : @xmath50 which moves the @xmath1-euler angle , a ket vector @xmath12 changes as : @xmath51 and besides we assume @xmath52 then the lagrangian ( [ eqn : weak - sym - lag ] ) behaves as : @xmath53 hence we see that the lagrangian ( [ eqn : weak - sym - lag ] ) has what is called `` weak invariance '' in refs . and : the lagrangian varies only by a total derivative term under a certain transformation . @xcite let us call it the `` weak gauge symmetry '' or the `` weak gauge @xmath1-symmetry '' here so as to stress the effect of the gauge @xmath1-transformation ( [ eqn : transfrs3 ] ) . we state them once more in the following definition : _ if a lagrangian is transformed according to ( [ eqn : transf - lag ] ) under ( [ eqn : transfrs3 ] ) , we say that the lagrangian possesses the weak gauge symmetry or the weak gauge @xmath1-symmetry . _ so far we have concentrated on the semiclassical time evolutions . we see that a fv with certain conditions meet the symmetry of semiclassical motions . however , if we consider the full quantum dynamics conformable to the @xmath1-invariance , more stringent conditions are required of fv . moreover , we will find that the conditions on fv , i.e. those on @xmath38 , for the symmetry of hamiltonians or of the whole lagrangians are changed when we take up hamiltonians that are quadratic or higher in @xmath54 . we will look into the situations more deeply using real sample lagrangians in the next @xmath11 [ sec : lags ] . in this section we will discuss the relation between fv , hamiltonians and weak symmetry of lagrangians above mentioned by demonstrating several concrete examples . the problem is closely related to semiclassical versus full quantum time evolutions first pointed by stone . @xcite first , we investigate a few simple examples to see how fv relate to symmetries and what semiclassical and full quantum time evolutions look like in @xmath11 [ sec : ex - simpie - llusts ] . this may help one to grasp various wider examples collected in table [ tab : exlagssym ] in the following @xmath11 [ sec : ex - lags ] . the examples in @xmath11 [ sec : ex - simpie - llusts ] are also included in table [ tab : exlagssym ] . in this subsection we illustrate , using simple examples of a hamiltonian and fv , the relations between semiclassical paths and exact quantal time evolutions . before going into the examples , let us recollect the general theory of cs . @xcite then we find that cs are determined in connection with fv . let @xmath55 be a lie group of our concern . consider the case in which there exists a subgroup of @xmath55 , say @xmath56 , that leaves a fv , @xmath4 , invariant : @xmath57 such a subgroup @xmath56 is called the `` isotropy subgroup '' or a `` stabilizer '' . then cs is actually defined on the coset space @xmath58 . and @xmath59 are frequently employed in place of @xmath55 and @xmath56 . in order that one does not confuse them with the hamiltonians @xmath60 , @xmath59 and the generator @xmath61 in the following sections we use calligraphic @xmath55 and @xmath56 to express general groups here . ] if there is no isotropy subgroups , we may regard @xmath62 . then @xmath58 is @xmath63 itself . hence we see that an isotropy subgroup depends upon the way in which we choose a fv , @xmath4 . consider the present su(2 ) cs case : @xmath64 . if a fv is @xmath65 , then @xmath66 and the spin cs is actually determined on the coset space @xmath67 which corresponds to a bloch sphere @xmath68 ; and thus @xmath69 . this includes what we always do in constructing the conventional spin cs with @xmath70 or @xmath71 ; and we may call the fv `` standard '' . the corresponding cs fall within what is called the `` informative '' cs in ref . . we see that cs is invariant under the transformation @xmath72 . '' after our euler angle convention . ] the case is illustrated in @xmath11 [ sec : samplefv1 ] below . next , if we take a fv other than @xmath65 , then @xmath56 is not @xmath73 no more but @xmath74 . hence @xmath75 is specified by a _ full _ set of three euler angles : @xmath76 . for this time @xmath77 phase ) . however , there are cases in which the `` little group '' @xcite that leaves semiclassical states invariant exists . the group turns out to be @xmath78 where @xmath61 is the generator of the weak symmetry transformation ( [ eqn : transfrs3 ] ) ; the explicit form will be given later in @xmath11 [ sec : explicit - g ] . see @xmath11 [ sec : sample - nonstandardfv1 ] for a sample fv . the case has something to do with another type of isotropy subgroups proposed by stone @xcite ; and we will investigate the case from a slightly different viewpoint in @xmath11 [ sec : anotherviewscevolvs ] . the case without even the weak symmetry is treated in @xmath11 [ sec : sample - nonstandardfv2 ] . now , we will treat concrete examples below . the examples , which are discussed by stone , @xcite are simplified versions of ref . . take a spin in a constant magnetic field @xmath79 . the hamiltonian is : @xmath80 in what follows we will illustrate , taking three typical types of fv , the relations between hamiltonians , fv , the @xmath1-symmetry of lagrangians . we also demonstrate how semiclassical motions concern the full quantum ones for the fv . first , let a fv @xmath65 . it is rather a standard case where @xmath81 . we have @xmath82 and the @xmath40-term is absent ; and thus the topological term has the weak @xmath1-symmetry as we saw it in @xmath11 [ sec : symlagfv ] . besides since the @xmath83- and @xmath84-term vanish , @xmath85 contains no @xmath1-variables and is symmetric under ( [ eqn : transfrs3 ] ) . consequently , the lagrangian : @xmath86 possesses the weak @xmath1-gauge symmetry ( [ eqn : transf - lag ] ) . the above situation is summarized in the column ( i ) in table [ tab : exlagssym ] ; see the following @xmath11 [ sec : ex - lags ] . the variation equations ( [ eqn : vareq - pol ] ) are : @xmath87 the third equation in ( [ eqn : vareq - pol ] ) , describing the behavior of @xmath88 , is automatically satisfied . it means that any @xmath1 works well as far as the semiclassical motions are concerned . from ( [ eqn : vareq - hs3 ] ) we obtain , setting the initial state as @xmath89 : @xmath90 therefore the semiclassical motion is given by : @xmath91 { \hat s}_3 t \ } \cdot \exp(- i \theta_0 { \hat s}_2 ) \cdot \exp[- i \psi(t ) ] { { \left\vert m\right\rangle}}. \label{eqn : vareq - hs3-sols - tot}\ ] ] on the other hand , we see that the full quantal time evolution is also described in terms of spin cs with the euler angle @xmath92 : @xmath93 \ , { \hat s}_3 \ , t \ } { { \left\vert { \bf\omega}(0)\right\rangle } } \nonumber\\ & = & \exp \{- i [ ( \mu b ) / \hbar ] \ , { \hat s}_3 \ , t \ } \cdot \exp(- i \theta_0 { \hat s}_2 ) \cdot \exp(- i \psi_0 ) { { \left\vert m\right\rangle}}. \label{eqn : time - evol - hs3-m}\end{aligned}\ ] ] for the present fv , i.e. , @xmath15 , each factor , @xmath94 $ ] in ( [ eqn : vareq - hs3-sols - tot ] ) and @xmath95 in ( [ eqn : time - evol - hs3-m ] ) , when acting on @xmath15 , yields a trivial phase factor respectively ; and thus two states ( [ eqn : vareq - hs3-sols - tot ] ) and ( [ eqn : time - evol - hs3-m ] ) belong to the same ray living on @xmath96 ; consequently , both semiclassical and genuine quantum time evolutions coincide . we may interpret that the @xmath1-gauge symmetry is preserved also in the full quantum dynamics . next , let us see the relation between the full quantum propagator and the semiclassical one . for an arbitrary final state @xmath97 at @xmath98 they are given by @xmath99 and @xmath100 respectively . then it is clear that from ( [ eqn : vareq - hs3-sols - tot ] ) and ( [ eqn : time - evol - hs3-m ] ) the relation between two propagators are : @xmath101 \}. \label{eqn : diff2props - fvm}\ ] ] the semiclassical propagator obeys the full quantal one up to @xmath1-gauge dependence . what we see here is one of the concrete examples of the dynamics for informative cs.@xcite notice that for a round trip in which the final @xmath1 differs from @xmath102 by @xmath103 or @xmath104 two propagators fully agree since @xmath105 is an integer or a half integer . second , we put fv as : @xmath106 . holds actually with the aid of the explicit form of @xmath107 . see , e.g. , messiah @xcite and references cited in * i*. ] as the previous fv only @xmath48-term survives among @xmath108 since the terms involving @xmath109 and its complex conjugate vanish . the lagrangian reads : @xmath110 it behaves like ( [ eqn : transf - lag ] ) , showing the weak @xmath1-gauge symmetry . see the column ( ii ) in table [ tab : exlagssym ] in the following @xmath11 [ sec : ex - lags ] . the variation equations are the same as ( [ eqn : vareq - hs3 ] ) and the solution gives : @xmath111 \ , { \hat s}_3 \ , t \ } \cdot \exp(- i \theta_0 { \hat s}_2 ) \cdot \begin{pmatrix } \exp[- i \psi(t ) ] \sqrt{2 / 3 } \\ 0 \\ \exp [ i \psi(t ) ] \sqrt{1 / 3 } \end{pmatrix}. \label{eqn : vareq - hs3-sols - tot - ntfv}\ ] ] since the full quantum time evolution operator is clearly the same as that in ( [ eqn : time - evol - hs3-m ] ) , the corresponding state is given by : @xmath112 \ , { \hat s}_3 \ , t \ } \cdot \exp(- i \theta_0 { \hat s}_2 ) \cdot \begin{pmatrix } \exp(- i \psi_0 ) \sqrt{2 / 3 } \\ 0 \\ \exp(i \psi_0 ) \sqrt{1 / 3 } \end{pmatrix}. \label{eqn : time - evol - hs3-ntfv1}\ ] ] we realize that this time each factor , @xmath113 $ ] in ( [ eqn : vareq - hs3-sols - tot - ntfv ] ) and @xmath114 in ( [ eqn : time - evol - hs3-ntfv1 ] ) , when acting on the present fv yields a _ non - trivial _ phase factor that changes the quantum states basically . if we have @xmath115 for all @xmath116 , two evolutions coincide . however , there seems no _ a priori _ reason to set @xmath115 ; for we know that the lagrangian ( [ eqn : lagpol ] ) has the @xmath1-gauge symmetry . and thus we conclude that the semiclassical time evolution does not agree with the genuine quantum time evolution for this _ non - trivial _ fv as pointed out by stone . @xcite notice that the relation between full quantum propagators and the semiclassical one is no longer simple as ( [ eqn : diff2props - fvm ] ) , but we have instead : @xmath117 { { \left\vert m\right\rangle}}. \label{eqn : ediff2props - nontrivialfv1}\ ] ] third , we consider the fv : @xmath118 . see the column ( v ) in table [ tab : exlagssym ] in the following @xmath11 [ sec : ex - lags ] and one will find the both @xmath40-term and @xmath59 have different properties from two above - mentioned fv cases , thus yielding the lagrangian without the @xmath1-symmetry . we have the lagrangian : @xmath119 \ } - \frac16 \mu b [ \cos\theta - { \tilde c } \sin\theta ] \cos\psi , \label{eqn : lag - non - trivialfv3}\ ] ] where @xmath120 . the variation equations ( [ eqn : vareq - pol ] ) give : @xmath121 { \dot \phi } + { \tilde c}{\dot \psi } \cos\psi \ } = - \mu b ( \sin\theta + { \tilde c}\cos\theta \cos\psi ) , \\ \hbar \ { [ \sin\theta + { \tilde c}\cos\theta \cos\psi ] { \dot \theta } - ( { \tilde c}\cos\theta \sin\psi ) { \dot \psi } \ } = 0 , \\ \hbar ( { \dot \phi } \sin\theta \sin\psi + { \dot \theta } \cos\psi ) = - \mu b \sin\theta \sin\psi . \label{eqn : vareq - nonstandardfv2 } \end{array } \right.\ ] ] one can verify that ( [ eqn : vareq - nonstandardfv2 ] ) yields a solution : @xmath122 which provides the same time evolution of the state as that for the exact quantum dynamics . for we know that the time development operator acting on a fv for the latter is the same as that in ( [ eqn : time - evol - hs3-m ] ) or ( [ eqn : time - evol - hs3-ntfv1 ] ) . when a hamiltonian @xmath31 includes terms that are quadratic or higher in @xmath54 , @xmath1-dependence of @xmath59 will be changed from that for @xmath31 linear in @xmath123 even for the same fv . and thus the gauge @xmath1-symmetry of the whole lagrangian will be also modified . we have to be careful that _ generally in cspi the combination of a hamiltonian and a fv together determines whether the corresponding lagrangian possesses the weak gauge @xmath1-symmetry or not . _ this is a crucial difference from refs . . in the point two cases are completely different . in the present subsection we take up two typical hamiltonians that are composed of @xmath43 ; they are the nmr and nqr types of hamiltonians given below@xcite : @xmath124 and @xmath125 we can obtain @xmath59 in ( [ eqn : h ] ) as @xmath126 ; we may have the explicit form with the aid of ( [ eqn : mat - pol ] ) . similarly @xmath127 is evaluated with the aid of ( * i * a.8 ) . for the present purpose , however , we do not have to know the explicit forms of @xmath128 and @xmath127 ; all that we need to grasp is the following : @xmath128 is independent of @xmath1 and hence invariant under ( [ eqn : transfrs3 ] ) if and only if no nearest neighboring @xmath38 exists for any @xmath39 ; and the condition for @xmath127 to be invariant under ( [ eqn : transfrs3 ] ) is that both no nearest neighboring @xmath38 and no next nearest neighboring @xmath38 in a given fv exist for any @xmath39 . we demonstrate various examples of combinations of hamiltonians and fv , including those illustrated in @xmath11 [ sec : ex - simpie - llusts ] , in table [ tab : exlagssym ] below ; and we indicate the semiclassical weak gauge @xmath1-symmetry of their topological term , @xmath59 and the whole lagrangian . as we put in @xmath11 [ sec : ex - simpie - llusts ] , all the examples that meet the semiclassical symmetry do not always preserve the symmetry in full quantum dynamics . we will proceed to the problem in the next section . in table [ tab : exlagssym ] notations like @xmath129 , for instance , are used . and hamiltonians ( [ eqn : h - nmr ] ) and ( [ eqn : h - nqr ] ) are referred to as nmr and nqr respectively . notice that in case ( i ) the spin actually meets @xmath130 for nqr type hamiltonians . @xcite @cccccccc@ + & ( i ) & ( ii ) & ( iii ) & ( iv ) & ( v ) & ( vi ) & ( vii ) + + + spin & arbitrary & @xmath131 & 1 & 1 & 1 & @xmath132 & @xmath133 + & & & & & & & + @xmath31 & nmr & nmr & nqr & nmr & nmr & nmr & nqr + & nqr & & & & & + + fv @xmath4 & @xmath134 & @xmath135 & @xmath135 & @xmath136 & @xmath137 & @xmath138 & @xmath138 + + @xmath48 & @xmath105 & @xmath139 & @xmath139 & @xmath140 & @xmath141 & @xmath142 & @xmath142 + + @xmath40-term & absent & absent & absent & present & present & absent & absent + + weak symmetry & & & & & & & + of the & yes & yes & yes & no & no & yes & yes + topological term + + symmetry & & & & & & & + of @xmath59 & yes & yes & no & no & no & yes & yes + + total weak & & & & & & & + symmetry & yes & yes & no & no & no & yes & yes + of @xmath143 & & & & & & & + + + [ tab : exlagssym ] we have looked over various examples of fv , hamiltonians , lagrangians and the weak symmetries in @xmath11 [ sec : lags ] . some lagrangians meet the `` semiclassical '' weak symmetry ; and some do not . do the `` full quantum '' states realize the symmetry that the former lagrangians possess ? we know from @xmath11 [ sec : ex - simpie - llusts ] that the answer is not affirmative . so there may be a kind of restriction on quantum states . this falls within the problem that stone took up more than a decade ago @xcite ; it is the problem of realizable fv in cs . in this section we treat the problem of the restriction on types of spin cs , i.e. on those of the fv , in the full quantum picture when a spin cs lagrangian has the weak semiclassical gauge @xmath1-symmetry ( [ eqn : transf - lag ] ) in the light of the formalism in @xmath11 [ sec : spincspi ] and @xmath11 [ sec : symlagfv ] . this means that we impose such restrictions on fv in spin cs that reflect the weak semiclassical gauge symmetry ( [ eqn : transf - lag ] ) . it provides one answer to the mystery on spin cspi posed by stone in ref . . or what we will perform is to see the results in ref . from a different point of view ; from the viewpoint of the gauge symmetry of the action in spin cspi first , the general theorem , theorem [ th : typefv ] , is given and proved in @xmath11 [ sec : th1 ] . second , in @xmath11 [ sec : explicit - g ] , we investigate the generator of the symmetry transformation , which yields theorem [ th : gform ] . and then we revisit theorem [ th : typefv ] via theorem [ th : gform ] . it may help us to understand what is going on concretely . we consider a class of lagrangians which we have formulated in @xmath11 [ sec : symlagfv ] and illustrated in @xmath11 [ sec : lags ] : we treat a lagrangian with the weak gauge @xmath1-symmetry in which @xmath144 as well as @xmath44 is invariant under the transformation ( [ eqn : transfrs3 ] ) . then the whole lagrangian changes at most a total derivative . among the examples in the preceding @xmath11 [ sec : ex - lags ] , ( i ) , ( ii ) , ( vi ) and ( vii ) meet the condition ; see the table [ tab : exlagssym ] . in the cases the following theorem holds . and thus actually only the fv in ( i ) survives ; fv in the form of ( ii ) , ( vi ) or ( vii ) are ruled out . the results agree with those indicated by stone @xcite from the viewpoint of two types of isotropy subgroups associated with semiclassical and full quantum dynamics . for the cases ( iii ) , ( iv ) and ( v ) the theorem will not tell anything . if a lagrangian associated with su(2 ) cs has the weak gauge symmetry related to @xmath1-variable , the fiducial vector belongs to @xmath15 or to the orbit of @xmath15 under the action of @xmath10 and the dirac condition holds for @xmath48 . [ th : typefv ] before going into the following proof , we briefly mention in which direction we will proceed . doing so is adequate for the purpose . since we concentrate on spin degrees of freedom , spin cs and fv , our lagrangian differs from refs . . the method by aitchison in ref . , however , applies also to ours ; and thus we mainly proceed along ref . in the following _ with suitable changes _ ; we particularly observe how a state vector @xmath12 behaves . that is our strategy . let us start the proof now . since we assume that the gauge @xmath1-symmetry holds for a lagrangian , the lagrangian takes the form of ( [ eqn : weak - sym - lag ] ) . and under the transformation ( [ eqn : transfrs3 ] ) a ket vector @xmath12 and the lagrangian change in the manners of ( [ eqn : transf - ketv ] ) and ( [ eqn : transf - lag ] ) respectively . now , denote @xmath61 such an operator that makes a transformation ( [ eqn : transf - ketv ] ) on @xmath12 : @xmath145 this means that @xmath61 is the generator of ( [ eqn : transf - ketv ] ) ; and also of ( [ eqn : transf - lag ] ) since we have assumed the fv and @xmath31 meet the conditions on weak symmetry . in terms of @xmath61 , the change of @xmath143 becomes : @xmath146 where we have used the expression of @xmath143 in ( [ eqn : action - pol ] ) . then we obtain from ( [ eqn : transf - lag ] ) and ( [ eqn : transfg - lag ] ) @xmath147 which is considered as a condition on semiclassical symmetry . from ( [ eqn : transfg - ket ] ) we have the following relation between operators : @xmath148 for an infinitesimal transformation we have @xmath149 the explicit form of @xmath61 , which we do not need in the present context , is given in @xmath11 [ sec : explicit - g ] . notice that ( [ eqn : transf - inf - grs3 ] ) is _ not identical with _ the corresponding expression in ref . ; the latter is @xmath150 = { \hat r}({\bf\omega } ) { \hat s}_3 $ ] . let us impose the condition on a _ full _ quantum state so that the dynamics is conformable to semiclassical @xmath1-symmetry . the definition of @xmath61 in ( [ eqn : transfg - ket ] ) , the consequent weak symmetry ( [ eqn : transf - lag ] ) and noether s theorem tell us that the state vector is subject to a certain restriction in the sense of dirac @xcite ; and thus we require a sort of gauss law @xcite : @xmath151 note that ( [ eqn : gauss ] ) is different from ( [ eqn : ga_0 ] ) . it is clear that ( [ eqn : gauss ] ) is more stringent than ( [ eqn : ga_0 ] ) . a set of @xmath152 that satisfies ( [ eqn : gauss ] ) also meets ( [ eqn : ga_0 ] ) . however , the reverse does not always hold . we will return to the relation between ( [ eqn : ga_0 ] ) and ( [ eqn : gauss ] ) in @xmath11 [ sec : anotherviewscevolvs ] . as the finite form of ( [ eqn : gauss ] ) we have @xmath153 on the other hand , with the aid of ( [ eqn : transf - grs3 ] ) , @xmath154 combining ( [ eqn : weak - inv - vec1 ] ) with ( [ eqn : weak - inv - vec2 ] ) we obtain @xmath155 only a few fv meet the condition ( [ eqn : weak - inv - vec3 ] ) ; and indeed it is possible if and only if @xmath156 the fv has only one nonzero component : the state vector becomes @xmath157 which is what we call `` rotated spin number states '' in * i*. from this we see @xmath158 since @xmath159 , eq . ( [ eqn : a_0-m ] ) indicates that the dirac condition holds for @xmath48 . let us put one more point about a fv . decompose @xmath160 and we will obtain @xmath161 from this viewpoint we may interpret @xmath162 as a fv . hence , from ( [ eqn : fvm ] ) and ( [ eqn : fvrm ] ) , we see that the fv coincides with @xmath15 , one of the eigenstates of @xmath2 , or rides on the orbit of @xmath15 under the su(2 ) rotations . the above theorem [ th : typefv ] implies the strange feature of cspi and fv : i.e. , the lagrangians , which depend upon the kinds of fv as shown in table [ tab : exlagssym ] , in turn restrict them . the restriction condition determines the forms of fv . as a result , the fv belongs to @xmath163 , or can be reached from @xmath15 by @xmath9 . mathematically , they are on orbits of @xmath15 under the action of the su(2 ) group . @xcite and the dirac condition is permitted . this gives an answer to the problem posed in ref . in the light our spin cspi formalism . notice that the reverse statement does not hold : the dirac condition does not always imply @xmath65 . this is apparent since a fv @xmath164 in ( vi ) and ( vii ) in table [ tab : exlagssym ] yields @xmath165 . however , there is clearly an exceptional case : i.e. @xmath166 case in which the fv may have several nonzero components . next , we revisit ( [ eqn : weak - inv - vec3 ] ) from another viewpoint ; for a spin @xmath167 case , @xmath168 . putting @xmath169 in ( [ eqn : weak - inv - vec3 ] ) we have @xmath170 , thus yielding @xmath171 . it is consistent with ( [ eqn : a_0-m ] ) . for a spin half - integer case , putting @xmath172 in ( [ eqn : weak - inv - vec3 ] ) , we have @xmath173 ; and thus we obtain @xmath174 integer ) . since @xmath175 in this case , the result agrees with ( [ eqn : a_0-m ] ) . we have already made a similar argument about semiclassical and full quantum propagators after ( [ eqn : diff2props - fvm ] ) . note that for a spin @xmath3 case , it is always possible to write any fv in the form of @xmath176 using a suitable euler angle @xmath177 : for spin @xmath3 any fv can be reached from @xmath178 or @xmath179 . we are able to describe any two - state system in terms of su(2 ) cs . so the condition ( [ eqn : fvrm ] ) holds automatically . what we have seen here is a cs version of magnetic charge quantization which has been known for a long time . @xcite now , it is widely known that the dirac condition is related to dirac strings . a dirac string , extending from the origin to half infinity in the @xmath180-space , corresponds to choosing @xmath181 or @xmath182 `` section '' in the topological term ( [ eqn : toptermsym ] ) . in the present case it is surely possible to prepare the above @xmath1 since the gauge @xmath1-transformation promises the freedom . and the freedom , as we saw , comes from selecting the special type of fv : @xmath65 . hence , _ looking from the present spin cspi formalism , whether dirac strings are permissible or not depends upon the types of fv . _ one may observe that conventional arguments about the dirac condition often imply that the particle interacting with a pole has spin @xmath140 or @xmath183 , which falls within the type of fv that meets theorem [ th : typefv ] . for a generic spin @xmath184 , however , we have wider possibilities on fv . and thus we may expect so also on magnetic charge . it is , of course , an open question whether we can apply our discussion to real magnetic monopoles . however , the approach presented here may provide us with a fine view of real monopoles since real and fictitious monopoles enjoy common mathematical descriptions . in the preceding subsection we used the operator @xmath61 satisfying ( [ eqn : transfg - ket ] ) or ( [ eqn : transf - grs3 ] ) . although we have not needed its explicit form , a natural question arises : what does it look like ? we find the explicit form of the generator @xmath61 easily in theorem [ th : gform ] below . this may be a by - product of theorem [ th : typefv ] . conversely , however , the form brings us to theorem [ th : typefv ] via another route again . hence theorem [ th : gform ] helps us to understand theorem [ th : typefv ] much deeper . for an operator @xmath61 to satisfy ( [ eqn : transfg - ket ] ) or ( [ eqn : transf - grs3 ] ) it is necessary and sufficient that @xmath61 is expressed as @xmath185 where @xmath186 . [ th : gform ] let us see the necessity first . from ( [ eqn : transf - grs3 ] ) we have @xmath187 , \label{eqn : transf - grs3-a}\ ] ] which , with the help of ( * i*-a.7 ) , results in ( [ eqn : g ] ) . next , that ( [ eqn : g ] ) is sufficient is obvious since we only need to cross the last equality in ( [ eqn : transf - grs3-a ] ) in reverse order . we see that the generator @xmath61 is nothing but a sort of generalized `` hopf map '' . the usual hopf map is given by @xmath188 . @xcite it is often referred to as the quantity which indicates the gauge @xmath1-symmetry ; and it is also exactly what is taken as the tool of `` gauge fixing '' in refs . and . next , we revisit theorem [ th : typefv ] via theorem [ th : gform ] . now , @xmath2 is a @xmath189 matrix whose eigenvectors are completely given by @xmath190 then , operating a non - singular matrix @xmath10 on both sides of ( [ eqn : s3-ev ] ) , we have , employing ( [ eqn : g ] ) , @xmath191 which specifies all the @xmath192 eigenvectors of @xmath61 thoroughly . comparing ( [ eqn : gom ] ) with ( [ eqn : gauss ] ) , we realize that @xmath12 must coincide with one of the @xmath193 . this leads us again to ( [ eqn : oom ] ) , from which we consequently obtain ( [ eqn : fvm ] ) and ( [ eqn : a_0-m ] ) . we thus confirm that ( [ eqn : fvm ] ) and ( [ eqn : a_0-m ] ) hold again . next , we put two additional comments . first , the form of ( [ eqn : g ] ) can be obtained also by the infinitesimal relation ( [ eqn : transf - inf - grs3 ] ) . since ( [ eqn : transf - inf - grs3 ] ) is independent of @xmath184 , we may try @xmath194 matrices : @xmath195 and @xmath196 ; see ( * i*-a.1 ) for the expression of @xmath196 . in this manner we obtain ( [ eqn : g ] ) again . it is clear that the direct evaluation of @xmath197 using ( [ eqn : g-1/2 ] ) leads to ( [ eqn : transf - grs3 ] ) as well ; see refs . and for such matrix calculations . second , we point out that @xmath48 in ( [ eqn : defa ] ) , ( [ eqn : ga_0 ] ) and ( [ eqn : gom ] ) are mutually consistent . in ref . stone characterized cs , fv and the consistency between semiclassical and full quantum dynamics by a slightly different way from ours . in order to see what they looked like , he introduced the isotropy subgroup @xmath198 for a given lie group @xmath55 that stabilized the expectation values of the lie group generators in the state of a fv in addition to the usual isotropy subgroup @xmath56 in ( [ isotropysg ] ) . he showed that semiclassical orbits lived on @xmath199 , whereas the full quantum dynamics was governed by @xmath58 . let us concentrate on @xmath198 here . for the present spin cs case it reads : @xmath200 notice that @xmath201 is used in place of @xmath4 and @xmath202 instead of @xmath56 in ref . . we take latter notations to keep harmony with the expressions in the preceding sections . one may feel that the relation between the description by @xmath198 and that by @xmath56 resembles the connection between ( [ eqn : ga_0 ] ) and ( [ eqn : gauss ] ) . one describes the semiclassical dynamics ; and the other is related to full quantum time evolution . we now revisit the framework of ref . from our point of view in @xmath11 [ sec : spincspi ] @xmath11 [ sec : gentheorems ] . let us take up three representative types of fv which are numbered ( i ) , ( ii ) , ( iv ) in table [ tab : exlagssym ] . then we obtain the following results due to ref . illustrated in table [ tab : isotropysg - fv ] : @cccc@ + & ( i ) & ( ii ) & ( iv ) + + + fv & @xmath203 & @xmath135 & @xmath204 + + @xmath56 & @xmath205 & @xmath74 & @xmath74 + + @xmath198 & @xmath205 & @xmath205 & @xmath74 + + + [ tab : isotropysg - fv ] in cases ( i ) and ( ii ) we have @xmath206 . it is clear that the effect of @xmath198 in the lagrangians amounts to that of ( [ eqn : transfrs3 ] ) which features the weak @xmath1-symmetry ; in fact we have used the convention and notation `` @xmath207 '' so that one can see the accordance easily . contrary , we have @xmath208 in the case ( iv ) ; and the case has nothing to do with the weak @xmath1-symmetry . hence we are going to investigate ( i ) and ( ii ) with the aid of ( [ eqn : isofv2sc ] ) instead of ( [ eqn : transfrs3 ] ) . consider the topological term first . for a fv with parameter dependence we have a more appropriate interpretation of ( * i*-35 ) as : @xmath209 since ( [ eqn : pol - r2 ] ) is linear in @xmath210 , we observe that under the operation @xmath198 on @xmath4 the following relation holds : @xmath211 where we have used ( [ eqn : isofv2sc ] ) . hence , the topological term @xmath212 is weakly symmetric under @xmath198 . the result may be obvious ; for it is clear that ( [ eqn : isofv2sc ] ) corresponds to ( [ eqn : transfrs3 ] ) . besides , assuming @xmath213 in ( i ) , @xmath31 is also linear in @xmath210 in the present cases ( i ) and ( ii ) in tables [ tab : exlagssym ] and [ tab : isotropysg - fv ] ; and thus @xmath214 is invariant under @xmath198 . then the whole lagrangians possesses the weak @xmath1-symmetry . so we have confirmed that the condition ( [ eqn : isofv2sc ] ) reproduces one about weak @xmath1-gauge symmetry for a lagrangian discussed in @xmath11 [ sec : symlagfv ] . if one tries to express the complete symmetry of hamiltonians that have such higher order products of @xmath210 as @xmath215 in ( [ eqn : h - nqr ] ) , more stringent conditions like @xmath216 may be required . here @xmath217 is an arbitrary @xmath218 product of @xmath54 or @xmath2 that appears in the hamiltonian . for example , we know @xmath215 gives @xmath219 ; and ( [ eqn : isofv2sc2 ] ) contains the condition @xmath220 . augmented with ( [ eqn : isofv2sc2 ] ) we see that the description due to @xmath198 brings a criteria for the weak gauge symmetry which works on all the cases in table [ tab : exlagssym ] . we have thus looked over two types of descriptions on cs , fv and semiclassical evolutions : one uses the shift of the @xmath1 variable in @xmath10 as ( [ eqn : transfrs3 ] ) and the other , i.e. , ( [ eqn : isofv2sc ] ) and ( [ eqn : isofv2sc2 ] ) , employs a transformed fv . they equally describe semiclassical dynamics well . and no matter what description we choose , both ref . and theorem [ th : typefv ] in @xmath11 [ sec : gentheorems ] tell us that if we have the gauge @xmath1-symmetry , we are led to a standard fv and informative cs in the full quantum dynamics . we have studied lagrangians having a weak gauge symmetry in the light of spin cspi with a general fv . we have set a condition on a state vector in order that the full quantum description keeps the semiclassical symmetry . this gives the restriction on fv . the types of cs get limited ; it is mandatory that the spin cs ride on the orbits of @xmath15 under the action of @xmath10 in the full quantum dynamics . and the fictitious monopole charge @xmath48 is so quantized as to bring the dirac condition . otherwise , the @xmath1-variable becomes `` anomalous '' . of course in natural sciences it is nature who gives a final decision . however , we expect that the results are all right in the case . concerning the matter , we find that the results agree with those due to stone,@xcite who first posed the problem in a general framework of cs and fv . notice that _ the rule to determine @xmath48 , i.e. the monopole charge or strength , is not built in the cspi a priori , but we impose it from physical demands gauge @xmath1-symmetry . _ this is the way in which spin cspi with general fv bring the dirac condition for fictitious monopole charges . remember that the monopole charge quantization condition is not derived by the schrdinger equation itself , but by boundary or topological conditions also in the usual wave function formalisms . @xcite now , let us see a future prospect . first , in the present paper , we have treated lagrangians with the @xmath1-gauge symmetry . if we consider a lagrangian without the symmetry , the situation looks rather different . see , for example , ( iii ) , ( iv ) and ( v ) in table [ tab : exlagssym ] . in the cases , we could not impose a subsidiary condition , i.e. gauss law , which produces the quantization of a fictitious monopole charge @xmath48 anymore . theorem [ th : typefv ] does not bring us any information for the case . and then what would happen to a fv , @xmath4 , and @xmath48 ? the problem , as well as physical applications of the present case , seems to be so intriguing for future investigations . second , it is clear that the present formalism may be extended to wider cspi cases ; among them su(1 , 1 ) and su(3 ) cases sound most probable candidates . the latter case is , of course , related to qcd . what do the restrictions on fv give physical systems described in terms of these cs ? do they bring new information other than the previous ones ? @xcite in this respect it might be as well to remember that stone @xcite actually discussed the restriction on fv for cs constructed from wider lie groups . third , we give subsidiary comments on cs and fv : one might observe a close formal analogy between cs with general fv and the ground states of many body systems or vacua in field theory . @xcite at least mathematical apparatus , a lie group @xmath55 and its coset space @xmath58 , are common to both of the cases . @xcite fv look like ground states or vacua . of course , in the many body systems the ground states themselves are expressed in terms of cs . however , such ground states are also absorbed into arbitrary fv . moreover , we may be able to prepare room for dealing symmetries associated with higher energy levels than vacua with the aid of the present arbitrary fv formalism . it is an open question as to whether there are some deeper implications behind the formal resemblance . in addition we want to indicate one more point on cs and fv . it is on the definitions of two types of isotropy subgroups @xmath198 and @xmath56 in @xmath11 [ sec : anotherviewscevolvs ] . they correspond to the relation between ( [ eqn : ga_0 ] ) and ( [ eqn : gauss ] ) . it seems natural to feel that they remind us of subsidiary conditions in covariant quantization of photons due to gupta - bleuler.@xcite in that case the expectation value of the operator describing the lorentz condition for a state vector is more crucial than the effect of the operation on the vector itself to establish the connection between quantized photon field and classical electrodynamics . and a certain gauge transformation leaves the expectation value invariant . this is also the case for the above - mentioned vacua in field theory ; there again the expectation value of a field operator plays a central role . we have not yet known whether there exists any deeper meaning of the analogies or not either . finally , we know that condensed matter systems have possibilities to simulate monopoles and gauge field theories . @xcite condensed matter physics makes it possible to examine such concepts in laboratories . the problems of cs and gauge symmetry discussed here will be of interest for a variety of realms of physics including condensed matter physics ; for symmetry is one of the basic principles that penetrate all of physics . the author is grateful to prof . m. stone for sending the draft of ref . , which has drawn his attention to the problem discussed in the present article . 99 r. p. feynman , r. b. leighton and m. sands , _ the feynman lectures on physics _ vol . 1 mainly mechanics , radiation , and heat ( addison -wesley , reading , ma , 1963 ) , chap 11 @xmath11 1 , chap 16 , chap 20 , and especially chap 52 . r. p. feynman , r. b. leighton and m. sands , _ the feynman lectures on physics _ vol . 2 mainly electromagnetism and matter ( addison -wesley , reading , ma,1964 ) , chap 30 , @xmath11 5 and @xmath11 6 . r. p. feynman , r. b. leighton and m. sands , _ the feynman lectures on physics _ vol . 3 quantum mechanics ( addison -wesley , reading , ma,1965 ) , chap 17 . l. fonda and g. c. ghirardi , _ symmetry principles in quantum physics _ , theoretical physics : text and exercise books , vol . 1 . ( m. dekker , new york , 1970 ) . l. d. landau and e. m. lifshitz , _ statistical physics _ , course of theoretical physics , vol . 5 , 3rd edn . ( butterworth - heinemann , oxford , 1984 ) . y. nambu and h. d. politzer , _ symposium on the microcosmos _ ( kosakusya , tokyo , 1979 ) ( in japanese ) . i. j. r. aitchison , _ an informal introduction to gauge field theory _ ( cambridge university press , 1982 ) . c. lanczos , _ the variational principles of mechanics _ , 4th edn . ( dover , new york , 1986 ) . w. yourgrau and s. mandelstam , _ variational principles in dynamics and quantum theory _ ( dover , new york , 1979 ) . r. p. feynman , _ int . * 21 * , 467 ( 1982 ) . r. jackiw , in _ current algebra and anomalies _ , ed . s. b. treiman _ ( wold scientific , singapole , 1985 ) , pp 211360 . r. a. bertlman , _ anomalies in quantum field theory _ ( oxford university press , oxford , 2000 ) . j. r. klauder and b .- s . skagerstam ( eds . ) , _ coherent states _ ( world scientific , singapore , 1985 ) . e. schrdinger , _ naturwiss . _ * 14 * , 664 ( 1926 ) . r. j. glauber , _ phys , rev . _ * 131 * , 2766 ( 1963 ) , reprinted in ref . . a. m. perelomov , _ generalized coherent states and their applications _ ( springer , berlin , 1986 ) . j. r. klauder , _ ann . ny _ * 11 * , 123 ( 1960 ) , reprinted in ref . . j. r. klauder , _ phys . * 19 * , 2349 ( 1979 ) , reprinted in ref . . h. kuratsuji and t. suzuki , _ j. math . * 21 * , 472 ( 1980 ) . a. inomata , h. kuratsuji and c. c. gerry , _ path integrals and coherent states of su(2 ) and su(1 , 1 ) _ ( world scientific , singapore , 1992 ) . a. shapere and f. wilczek ( eds . ) , _ geometric phases in physics _ , advanced series in math . phys . , vol . 5 ( world scientific , singapore , 1989 ) . h. kuratsuji and s. iida , prog . . phys . * 74 * , 439 ( 1985 ) , reprinted in ref . . h. kuratsuji and s. iida , _ phys . lett . _ * 56 * , 1003 ( 1986 ) . h. kuratsuji , _ phys . rev . lett . _ * 61 * , 1687 ( 1988 ) . m. matsumoto , _ phys . _ * 37 * , 3739 ( 1996 ) . m. matsumoto , _ int . b _ * 26 * , 1250143 ( 2012 ) , arxiv:1201.5258 v2 . ; this is an expanded version of the basic part of ref . . r. loudon , _ the quantum theory of light _ ( clarendon press , oxford , 2000 ) . @xmath11 5.5 - @xmath11 5.6 . a. p. balachandran , g. marmo , b .- s . skagerstam and a. stern , _ nucl . b _ * 162 * , 385 ( 1980 ) . a. p. balachandran , g. marmo , b .- s . skagerstam and a. stern , _ classical topology and quantum states _ ( world scientific , singapore , 1991 ) . i. j. r. aitchison , _ acta phys . polonica b _ * 18 * , 207 ( 1987 ) , reprinted in ref . . m. stone , _ int . b _ * 15 * , 2107 ( 2001 ) , quant - ph/0006020 . m. matsumoto , `` su(2 ) coherent state path integrals based on arbitrary fiducial vectors and geometric phases '' , quant - ph/9906100 v.2 . m. m. nieto , _ phys . a _ * 229 * , 135 ( 1997 ) , arxiv : quant - ph/9612050 . a. messiah , _ quantum mechanics _ ( dover , mineola , ny , 1999 ) ; chap 13 and appendix c. c. p. slicher _ principles of magnetic resonance _ , 3rd edn . ( springer , berlin , 1989 ) . r. tycko , _ phys . * 58 * , 2281 ( 1987 ) , reprinted in ref . . p. a. m. dirac , _ lectures on quantum mechanics _ ( dover , mineola , ny , 2004 ) . r. gilmore , _ lie groups , lie algebras , and some of their applications _ ( dover , mineola , ny , 2002 ) . f. t. arecchi , e. courtens , r. gilmore and h. thomas , _ phys . a _ * 6 * , 2211 ( 1972 ) , reprinted in ref . . p. a. m. dirac , _ proc . * a133 * , 60 ( 1931 ) . p. a. m. dirac , _ int . * 17 * , 235 ( 1978 ) . t. t. wu and c. n. yang , _ phys . * 12 * , 3845 ( 1975 ) . r. h. ryder , _ quantum field theory _ , 2nd edn . ( cambridge university press , 1996 ) , chap 8 . f. mandl and g. shaw , _ quantum field theory _ , 2nd ed . ( john wiley and sons , chichester , 2010 ) , chap 5 . j. moody , a. shapere and f. wilczek , _ phys . _ * 56 * , 893 ( 1986 ) , reprinted in ref . and references therein . m. m. salomaa and g. e. volovik , _ rev . _ * 59 * , 533 ( 1987 ) .

the study of spin dynamics in ultrathin ferromagnets is of fundamental interest , since new physics arises in these materials that has no counterpart in bulk magnetism . examples are provided by relaxation mechanisms evident in ferromagnetic resonance and brillouin light scattering studies,@xcite and also for the large wave vectors probed by spin polarized electron loss spectroscopy ( speels).@xcite of course , by now the remarkable impact of ultrathin film structures on magnetic data storage is very well known , and other applications that exploit spin dynamics in such materials are envisioned . thus these issues are important from a practical point of view as well as from that of fundamental physics . theoretical studies of the nature of spin waves in ultrathin films adsorbed on metal substrates have been carried out for some years now , along with comparison with descriptions provided with the heisenberg model.@xcite in this paper , we extend the earlier theoretical treatments to include the influence of spin orbit coupling on the spin wave spectrum of ultrathin films . this extension is motivated by a most interesting discussion of the ground state of the mn monolayer on the w(110 ) surface . a nonrelativistic theoretical study of this system predicted that the ground state would be antiferromagnetic in character.@xcite this prediction was confirmed by spin polarized scanning tunneling microscope studies of the system.@xcite however , recent experimental stm data with a more sensitive instrument showed a more complex ground state , wherein the ground state is in fact a spin density wave.@xcite one can construct the new state by beginning with the antiferromagnet , and then superimposing on this a long wavelength modulation on the direction of the moments on the lattice . the authors of ref . argued that the lack of reflection symmetry of the system in the plane of the film activates the dzyaloshinskii moriya ( dm ) interaction , and the new state has its origin in this interaction . they also presented relativistic and ab initio calculations that gave an excellent account of the new data . the reflection symmetry is broken simply by the presence of the substrate upon which the film is grown . this argument to us is most intriguing , since one can then conclude that the dm interaction must be active in any ultrathin ferromagnet ; the substrate is surely always present . the dm interaction has its origin in the spin orbit interaction , which of course is generally very weak in magnets that incorporate the 3d transition elements as the moment bearing entities . however , in the case of the mn monolayer on w(110 ) hybridization between the mn 3d and the w 5d orbitals activates the very large w spin orbit coupling , with the consequence that the strength of the dm interaction can be substantial , as illustrated by the calculations presented in ref . . one may expect to see substantial impact of the dm interaction in other ultrathin magnets grown on 5d substrates , and possibly 4d substrates as well . we have here another example of new physics present in ultrathin magnets that is not encountered in the bulk form of the material from which the ultrathin structure is fabricated . the purpose of this paper is to present our theoretical studies of spin orbit effects on spin waves and also on the dynamic susceptibility of a much studied ultrathin film / substrate combination , the fe monolayer and bilayer on w(110 ) . we find striking effects . for instance , when the magnetization is in plane , as we shall see the dm interaction introduces a term linear in wave vector in the dispersion relation of spin waves . thus the uniform spin wave mode at zero wave vector acquires a finite group velocity . we find this to be in the range of @xmath0 cm / sec for the fe monolayer on w(110 ) . furthermore , left / right asymmetries appear in the speels response functions . thus , we shall see that spin orbit coupling has clear effects on the spin excitations of transition metal ultrathin ferromagnets grown on 5d substrates . we comment briefly on the philosophy of the approach used here , and in various earlier publications.@xcite numerous authors proceed as follows . one may generate a description of the magnetic ground state of the adsorbed films by means of an electronic structure calculation based on density functional theory . it is then possible to calculate , within the framework of an adiabatic approximation , effective heisenberg exchange integrals jij between the magnetic moments in unit cell i and unit cell j. these may be entered into a heisenberg hamiltonian , and then spin wave dispersion relations may be calculated through use of spin wave theory . it has been known for decades@xcite that in the itinerant 3d magnets , effective exchange interactions calculated in such a manner have very long range in real space . thus , one must include a very large number of distant neighbors in order to obtain converged results . this is very demanding to do with high accuracy for the very numerous distant neighbors , since the exchange interactions become very small as one moves out into distant neighbor shells . at a more fundamental level , as noted briefly above , discussions in earlier publications show that in systems such as we study here , the adiabatic approximation breaks down badly , with qualitative consequences.@xcite first , spin wave modes of finite wave vector have very short lifetimes , by virtue of decay into the continuum of particle hole pairs ( stoner excitations ) even at the absolute zero of temperature@xcite whereas in heisenberg model descriptions their lifetime is infinite . in multi layer films , the earlier calculations show that as a consequence of the short lifetime , the spectrum of spin fluctuations at large wave vectors contains a single broad feature which disperses with wave vector in a manner similar to that of a spin wave ; this is consistent with speels data on an eight layer film of co on cu(100).@xcite this picture stands in contrast to that offered by the heisenberg model , in which a film of n layers has n spin wave modes for each wave vector , and each mode has infinite lifetime . the method developed earlier , and extended here to incorporate spin orbit coupling , takes due account of the breakdown of the adiabatic approximation and also circumvents the need to calculate effective exchange interactions in real space out to distant neighbor shells . we work directly in wave vector space through study of the wave vector and frequency dependent susceptibility discussed below , denoted as @xmath1 . the imaginary part of this object , evaluated for @xmath2 and considered as a function of frequency @xmath3 for fixed wave vector @xmath4 provides us with the frequency spectrum of spin fluctuations on layer @xmath5 for the wave vector chosen . spin waves appear as peaks in this function , very much as they do in speels data , and in a manner very similar to that used by experimentalist we extract a dispersion relation for spin waves by following the wave vector dependence of the peak frequency . we never need to resort to a real space summation procedure over large number of neighbors , coupled by very tiny exchange couplings . the spin wave exchange stiffness can be extracted either by fitting the small wave vector limit of the dispersion relation so determined , or alternatively by utilizing an expression derived earlier@xcite which once again does not require a summation in real space . we comment on another feature of the present study . in earlier calculations,@xcite as in the present paper , an empirical tight binding description forms the basis for our description of the electronic structure . within this approach , referred to as a multi band hubbard model , we can generate the wave vector and frequency dependent susceptibility for large systems . in the earlier papers , effective tight binding parameters were extracted from bulk electronic structure calculations . the present studies are based on tight binding parameters obtained directly from a rs - lmto - asa calculation for the fe / w(110 ) system . we also obtain tight binding parameters by fitting kkr based electronic structure calculations for the ultrathin film / substrate combinations of interest . we find that spin waves in the fe / w(110 ) system are quite sensitive to the empirical tight binding parameters which are employed , though as we shall see the various descriptions provide very similar pictures of the one electron local density of states . we note that udvardi and szunyogh@xcite have also discussed the influence of spin orbit coupling on the dispersion relation of spin waves in the fe monolayer on w(110 ) within the framework of the adiabatic approach discussed above , where exchange interactions and other magnetic parameters are calculated in real space . we shall discuss a comparison with our results and theirs below . there are differences . most particularly , we note that in fig . 3 , the authors of ref . provide two dispersion curves for propagation perpendicular to the magnetization , whereas in a film such as this with one spin per unit cell there can be only one magnon branch . additionally and very recently , bergmann and coworkers@xcite investigated within an adiabatic approach finite temperature effects on the magnon spectrum of fe / w(110 ) . in section ii , we comment on our means of introducing spin orbit coupling into the theory . the results of our calculations are summarized in section iii and concluding remarks are found in section iv . the formalism for including spin orbit coupling effects in our description of spin dynamics is quite involved , so in this section we confine our attention to an outline of the key steps , and an exposition of the overall structure of the theory . our starting point is the multi band hubbard model of the system that was employed in our earlier study of spin dynamics in ultrathin ferromagnets . the starting hamiltonian is written as@xcite @xmath6 where @xmath7 and @xmath8 are site indices , @xmath9 refer to spin , and @xmath10 to the tight binding orbitals , nine in number for each site , which are included in our treatment . the coulomb interactions operate only within the 3d orbitals on a given lattice site . the film , within which ferromagnetism is driven by the coulomb interactions , sits on a semi - infinite substrate within which the coulomb interaction is ignored . in our empirical tight binding picture , the spin orbit interaction adds a term we write as @xmath11 \label{hamilt_so}\ ] ] where @xmath12 is the angular momentum operator , @xmath13 is the local spin - orbit coupling constant , @xmath14 and @xmath15 . we assume that the spin orbit interaction , present both within the ferromagnetic film and the substrate , operates only within the 3d atomic orbitals . a convenient tabulation of matrix elements of the orbital angular momentum operators is found in ref . . information on the spin waves follows from the study of the spectral density of the transverse dynamic susceptibility @xmath1 as discussed above . from the text around eq . ( 1 ) of ref . , we see that this function describes the amplitude of the transverse spin motion ( the expectation value of the spin operator @xmath16 in the layer labeled @xmath5 ) to a fictitious transverse magnetic field of frequency @xmath3 and wave vector @xmath4 parallel to the film surface that is applied to layer @xmath17 of the sample . the spectral density , given by @xmath18 , when multiplied by the bose einstein function @xmath19^{-1}$ ] is also the amplitude of thermal spin fluctuations of wave vector @xmath4 and frequency @xmath3 in layer @xmath5 . we obtain information regarding the character ( frequency , linewidth , and amplitude in layer @xmath5 ) of spin waves from the study of this function , as discussed earlier.@xcite our previous analyses are based on the study of the dynamic susceptibility just described through use of the random phase approximation ( rpa ) of many body theory . the feynman diagrams included in this method are the same as those incorporated into time dependent density functional theory , though use of our hubbard model allows us to solve the resulting equation easily once the very large array of irreducible particle hole propagators are generated numerically . our task in the present paper is to extend the rpa treatment to incorporate spin orbit coupling . the extension is non trivial . the quantity of interest , referred to in abbreviated notation as @xmath20 , may be expressed as a commutator of the spin operators @xmath16 and @xmath21 whose precise definition is given earlier.@xcite with spin orbit coupling ignored , the rpa decoupling procedure leads to a closed equation for @xmath20 . when the rpa decoupling is carried out in its presence , we are led to a sequence of four coupled equations which include new objects we may refer to as @xmath22 , @xmath23 and @xmath24 . the number of irreducible particle hole propagators that must be computed likewise is increased by a factor of four . for a very simple version of a one band hubbard model , and for a very different purpose , fulde and luther carried out an equivalent procedure many years ago@xcite . in what follows , we provide a summary of key steps along with expressions for the final set of equations . to generate the equation of motion , we need the commutator of the operator @xmath25 with the hamiltonian . one finds @xmath26=\frac{1}{2}\sum_{\eta}\ { \lambda_{l'}l^{+}_{\nu\eta}c^\dagger_{l\mu\uparrow}c_{l'\eta\uparrow } -\lambda_l l^+_{\eta\mu}c^\dagger_{l\eta\downarrow}c_{l'\nu\downarrow } + \lambda_{l'}l^z_{\nu\eta}c^\dagger_{l\mu\uparrow}c_{l'\eta\downarrow } -\lambda_l l^z_{\eta\mu}c^\dagger_{l\eta\uparrow}c_{l'\nu\downarrow}\}. \label{comm}\ ] ] the last two terms on the right hand side of eq . [ comm ] lead to terms in the equation of motion which involve @xmath20 whereas the first two terms couple us to the entities @xmath23 and @xmath24 . when we write down the commutator of these new correlation functions with the spin orbit hamiltonian , we are led to terms which couple into the function @xmath22 which is formed from the commutator of two @xmath21 operators . in the absence of spin orbit coupling , a consequence of spin rotation invariance of the hamiltonian is that the three new functions just encountered vanish . but they do not in its presence , and they must be incorporated into the analysis . one then introduces the influence of the coulomb interaction into the equation of motion , and carries out an rpa decoupling of the resulting terms . the analysis is very lengthy , so here we summarize only the structure that results from this procedure . definitions of the various quantities that enter are given in the appendix . we express the equations of motion in terms of a @xmath27 matrix structure , where in schematic notation we let @xmath28 , @xmath29 , @xmath30 and @xmath31 . the four coupled equations then have the form @xmath32 each quantity in eq . [ eom_rpa_1 ] has attached to it four orbital indices , and four site indices . to be explicit , @xmath29 which enters eq . ( 4 ) is formed from the commutator of the operator @xmath33 with @xmath34 and in full we denote this quantity as @xmath35 . the site indices label the planes in the film , and we suppress reference to @xmath3 and @xmath4 . the products on the right hand side of eq . [ eom_rpa_1 ] are matrix multiplications that involve these various indices . for instance , the object @xmath36 is labeled by four orbital and four site indices so @xmath37_{\mu\nu,\mu'\nu'}(ll';mm ' ) = \sum_{\gamma\delta}\sum_{nn'}b^{ss'}_{\mu\nu,\gamma\delta}(ll';nn ' ) \chi^{(s')}_{\gamma\delta,\mu'\nu'}(nn';mm').\ ] ] one proceeds by writing eq . [ eom_rpa_1 ] in terms of the dynamic susceptibilities that characterize the non - interacting system . these , referred to also as the irreducible particle hole propagators , are generated by evaluating the commutators which enter into the definition of @xmath38 in the non interacting ground state . these objects , denoted by @xmath39 obey a structure similar to eq . [ eom_rpa_1 ] , @xmath40 it is then possible to relate @xmath38 to @xmath39 through the relation , using four vector notation , @xmath41 the matrix structure @xmath42 may be generated from the definition of @xmath43 , which may be obtained from the equation of motion of the non - interacting susceptibility , eq . [ eom_hf ] . then @xmath44 follows from the equation of motion of the full susceptibility , as generated in the rpa . one may solve eq . [ dyson ] @xmath45^{-1}\vec{\chi}^{(0)}(\omega ) , \label{chi_rpa}\ ] ] so our basic task is to compute the non interacting susceptibility matrix @xmath46 and then carry out the matrix inversion operation displayed in eq . [ chi_rpa ] . for this we require the single particle green s functions ( spgfs ) associated with our approach . to generate the spgfs , we set up an effective single particle hamiltonian @xmath47 by introducing a mean field approximation for the coulomb interaction . the general structure of the single particle hamiltonian is @xmath48 where the effective hopping integral @xmath49 contains the spin diagonal portion of the spin orbit interaction , along with the mean field contributions from the coulomb interaction . the form we use for the latter is stated below . the coefficients in the spin flip terms are given by @xmath50 we then have the eigenvalue equation that generates the single particle eigenvalues and eigenfunctions in the form @xmath51 ; we can write this in the explicit form @xmath52 \langle l\eta\sigma'|\phi_s\rangle = e_s\langle i\mu\sigma|\phi_s \rangle . \label{eigenvals}\ ] ] the single particle green s function may be expressed in terms of the quantities that enter eq . [ eigenvals ] . we have for this object the definition @xmath53 and one has the representation @xmath54 these functions may be constructed directly from their equations of motion , which read , after fourier transforming with respect to time , @xmath55 g_{l\eta\sigma'';j\nu\sigma ' } + \omega g_{i\mu\sigma;j\nu\sigma ' } = \delta_{\sigma\sigma'}\delta_{\mu\nu}\delta_{ij}. \label{spgf_eom}\ ] ] for the case where the substrate is semi infinite , our means of generating a numerical solution to the hierarchy of equations stated in eq . [ spgf_eom ] has been discussed earlier . what remains is to describe how the coulomb interaction enters the effective hopping integrals @xmath56 that appear in eq . [ h_sp ] , eq . [ eigenvals ] and eq . [ spgf_eom ] . there are , of course , a large number of coulomb matrix elements in the original hamiltonian , even if the coulomb interactions are confined to within the 3d shell . through the use of group theory,@xcite the complete set of coulomb matrix elements may be expressed in terms of three parameters . these are given in table i of the first cited paper in ref . . in subsequent work , we have found that a much simpler structure@xcite nicely reproduces results obtained with the full three parameter form . we use the simpler one parameter form here , for which @xmath57 . then in the mean field approximation , the coulomb contribution to the single particle hamiltonian assumes the form @xmath58 here @xmath59 is the magnitude of the moment on site @xmath7 . the coulomb interactions @xmath60 are non zero only within the ultrathin ferromagnet , and the moments @xmath59 , determined self consistently , vary from layer to layer when we consider multi layer ferromagnetic films . it should be noted that when the ansatz just described is employed in eq . [ alphas ] , the term from the coulomb interaction on the right hand side becomes proportional to the transverse component of the moment located on site @xmath7 and this vanishes identically . thus , despite the complexity introduced by the spin orbit coupling , when the simple one parameter ansatz for the coulomb matrix elements is employed , one needs no parameters beyond the moment on each layer in the self consistent loops that describe the ground state . in the present context , this is an extraordinarily large savings in computational labor , and this will allow us to address very large systems in the future . it is the case that certain off diagonal elements such as @xmath61 appear in the quantities defined in the appendix . notice , for example , the expressions in eqs . [ as ] . once the ground state single particle green s functions are determined , such expectation values are readily computed . in earlier studies of fe layers on w(110),@xcite as noted above , the electronic structure was generated through use of tight binding parameters obtained from bulk electronic structure calculations . these calculations generate effective exchange interactions comparable in magnitude to those found in the bulk transition metals,@xcite with the consequence that for both monolayer fe and bilayer fe on w(110 ) the large wave vector spin waves generated by theory are very much stiffer than found experimentally@xcite though it should be noted that for the bilayer , the calculated value of the spin wave exchange stiffness is in excellent accord with the data.@xcite subsequent calculations which construct the spin wave dispersion relation from adiabatic theory based on calculations of effective exchange integrals also generate spin waves for the monolayer substantially stiffer than found experimentally,@xcite though they are softer than in our earlier work by a factor of two or so . we remark that it has been suggested that the remarkably soft spin waves found experimentally may have origin in carbon contamination of the monolayer and bilayer.@xcite we remark here that this can be introduced during the speels measurement . we note that the magnetic properties of fe monolayers grown on carbon free w(110)@xcite differ dramatically from those grown on surfaces now known to be contaminated by carbon.@xcite in the former case , the domain walls have a thickness of 2.15 nm,@xcite whereas in the latter circumstance very narrow walls with thickness bounded from above by 0.6 nm are found.@xcite this suggests that the strength of the effective exchange is very different in the two cases , with stiffer exchange in the carbon free samples . the considerations of the previous paragraph have motivated us to carry out a series of studies of the effective exchange in the fe monolayer on w(110 ) within the framework of three different electronic structure calculations . we find that although all three give local density of states that are very similar , along with very similar energy bands when these are examined , the intersite exchange interactions vary substantially . first , we have employed the parameter set used earlier that is based on bulk electronic structures@xcite in new calculations we call case a. in case b , we have employed an approach very similar to that used in ref . , though in what follows our calculation of effective exchange integrals is non relativistic . this is the korringa kohn rostoker green s function ( kkr - gf ) method,@xcite which employs the atomic sphere approximation and makes use of the dyson equation @xmath62 as given in matrix notation . this allows us to calculate the green s function @xmath63 of an arbitrary complex system given the perturbing potential @xmath64 and the green s function @xmath65 of a reference unperturbed system . within the local spin density approximation ( lsda),@xcite we consider a slab of five monolayers of w with the experimental lattice constant on top of which an fe monolayer is deposited and relaxed by -12.9%@xcite with respect to the w interlayer distance . angular momenta up to @xmath66 were included in the green s functions with a @xmath67 mesh of 6400 points in the full two dimensional brillouin zone . the effective exchange interactions were calculated within the approximation of infinitesmal rotations@xcite that allows one to use the magnetic force theorem . this states that the energy change due to infinitesmal rotations in the moment directions can be calculated through the kohn sham eigenvalues . method c is the real space linear - muffin - tin - orbital approach as implemented , also , in the atomic sphere approximation ( rs - lmto - asa).@xcite due to its linear scaling , this method allows one to address the electronic structure of systems with a large number of atoms for which the basic eigenvalue problem is solved in real space using the haydock recursion method . the fe overlayer on the w(110 ) substrate was simulated by a large bcc slab which contained @xmath686800 atoms , arranged in 12 atomic planes parallel to the ( 110 ) surface , with the experimental lattice parameter of bulk w. one empty sphere overlayer is included , and self consistent potential parameters were obtained for the empty sphere overlayer , the fe monolayer , and the three w layers underneath using lsda.@xcite for deeper w layers we use bulk potential . nine orbitals per site ( the five 3d and 4 sp complex ) were used to describe the fe valence band and the empty sphere overlayer , and for w the fully occupied 4f orbitals were also included in the core . to evaluate the orbital moments we use a scalar relativistic ( sr ) approach and include a spin orbit coupling term @xmath69 at each variational step.@xcite in the recursion method the continued fraction has been terminated after 30 recursion levels with the beer pettifor terminator.@xcite the tb parameters so obtained are inserted into our semi - empirical scheme and this allows us to generate the non interacting susceptibilities which enter our full rpa description of the response of the structure . in order to compare the electronic structures generated by the approaches just described , we turn our attention to the local density of states for the majority and minority spins in the adsorbed fe monolayer . these are summarized in fig . [ fig1 ] . the local densities of states ( ldos ) generated by the three sets of tb parameters have approximately the same overall features , as we see from fig . the main differences appear in the majority spin band , which overlaps the 5d states in the w substrate over a larger energy range than the minority band . this is also true if we compare the ldos generated by the tight binding parameters extracted from the kkr electronic structure to the ldos obtained directly from the kkr calculations ( red dashed line in fig . [ fig1]b ) . the fe - w hopping parameters are indeed the least accurate portion of our parametrization scheme . in case a we just used the fe - fe bulk parameters to describe the fe - w hopping . in case b we extracted tb parameters for fe by fitting a kkr calculation of an unsupported fe monolayer with a lattice parameter matching that of the w substrate . for the fe - w hopping we used the fe parameters obtained from the fitting , scaled to mimic the fe - w distance relaxation . the relaxation parameter was chosen to give the correct spin magnetic moment for the adsorbed fe monolayer . in case c all parameters were directly provided by the rs - lmto - asa code , but in the dft calculations the fe - w distance was assumed to be equal to the distance between w layers . thus , the main difference between cases b and c is the treatment of the mixing between fe and w states and this is expected to affect more strongly the states that occupy the same energy range . as noted above , while the local density of states provided by the three approaches to the electronic structure are quite similar as we see from fig . [ fig1 ] ( and the same is true of the electronic energy bands themselves if these are examined ) , the exchange interactions differ substantially for the three cases . for the first , second and third neighbors we have ( in mev ) 42.5 , 3.72 and 0.46 for model a , 28.7 , -7.87 and 0.31 , for model b and 11.23 , -7.31 and 0.22 for model c. the authors of ref . find 10.84 , -3.34 and 3.64 for these exchange integrals . we now turn to our studies of spin excitations in the fe monolayer and fe bilayer on w(110 ) within the framework of the electronic structure generated through use of the approach in case c. we will discuss the influence of spin orbit coupling on both the transverse wave vector dependent susceptibility though study of the spectral density function @xmath70 discussed in section i. this function , for fixed wave vector @xmath4 , when considered as a function of frequency @xmath3 , describes the frequency spectrum of the fluctuations of wave vector @xmath4 of the transverse magnetic moment in layer @xmath5 as noted above . in the frequency regime where spin waves are encountered , this function is closely related to ( but not identical to ) the response function probed in a speels measurement . in fig . [ fig2 ] , for the fe monolayer on w(110 ) , we show the spectral density function calculated for three values of @xmath71 , for propagation across the magnetization . thus , the wave vector is directed along the short axis in the surface . this is the direction probed in speels studies of the fe monolayer on this surface.@xcite in each figure , we show three curves . the green dashed curve is calculated with spin orbit coupling set to zero . we show only a single curve for this case , because the spectral density is identical for the two directions of propagation across the magnetization , @xmath72 and @xmath73 . when spin orbit coupling is switched on , for the two directions just mentioned the response function is very different , as we see from the red and black curve in the various panels . these spin wave frequencies , deduced from the peak in the response functions as discussed in section i , differ for the two directions of propagation , and also note that the peak intensities and linewidths differ as well . it is the absence of both time reversal symmetry and reflection symmetry which renders @xmath72 and @xmath73 inequivalent for this direction of propagation . the system senses this breakdown of symmetry through the spin orbit interaction . if one considers propagation parallel to the magnetization , the asymmetries displayed in fig . [ fig2 ] are absent . the reason is that for this direction of propagation , reflection in the plane that is perpendicular to both the magnetization and the surface is a good symmetry operation of the system , but takes @xmath72 into @xmath73 thus rendering the two directions equivalent . recall , of course , that the magnetization is a pseudo vector in regard to reflections . notice how very broad the curves are for large wave vectors ; the lifetime of the spin waves is very short indeed . evaluated in the fe monolayer for three values of the wave vector in the direction perpendicular to the magnetization . we have ( a ) @xmath74 , ( b ) @xmath75 and ( c ) @xmath76 . the green curve ( dashed ) is calculated with spin orbit coupling set to zero ; the spectral density here is independent of the sign of @xmath4 . the red and black curves are calculated with spin orbit coupling turned on . now we see asymmetries for propagation across the magnetization , with the red curve @xmath4 directed from left to right and the black curve from right to left.,title="fig:",scaledwidth=45.0% ] evaluated in the fe monolayer for three values of the wave vector in the direction perpendicular to the magnetization . we have ( a ) @xmath74 , ( b ) @xmath75 and ( c ) @xmath76 . the green curve ( dashed ) is calculated with spin orbit coupling set to zero ; the spectral density here is independent of the sign of @xmath4 . the red and black curves are calculated with spin orbit coupling turned on . now we see asymmetries for propagation across the magnetization , with the red curve @xmath4 directed from left to right and the black curve from right to left.,title="fig:",scaledwidth=45.0% ] evaluated in the fe monolayer for three values of the wave vector in the direction perpendicular to the magnetization . we have ( a ) @xmath74 , ( b ) @xmath75 and ( c ) @xmath76 . the green curve ( dashed ) is calculated with spin orbit coupling set to zero ; the spectral density here is independent of the sign of @xmath4 . the red and black curves are calculated with spin orbit coupling turned on . now we see asymmetries for propagation across the magnetization , with the red curve @xmath4 directed from left to right and the black curve from right to left.,title="fig:",scaledwidth=45.0% ] as discussed in section i , we may construct a spin wave dispersion curve by plotting the maxima in spectral density plots such as those illustrated in fig . [ fig2 ] as a function of wave vector . we show dispersion relations constructed in this manner in fig . [ fig3 ] , with spin orbit coupling both present and absent . in fig . [ fig3]a , and for propagation perpendicular to the magnetization we show the dispersion curve so obtained for wave vectors throughout the surface brillouin zone , and in fig . [ fig3]b we show its behavior for small wave vectors . let is first consider fig , 3(a ) . here the dispersion curve extends throughout the two dimensional brillouin zone . at the zone boundary , quite clearly the slope of the dispersion curve does not vanish . in this direction of propagation , the nature of the point at the zone boundary does not require the slope to vanish . what is most striking , clearly , is the anomaly in the vicinity of 1 @xmath77 . this feature is evident in the calculation with spin orbit coupling absent , and for positive values of the wave vector the feature becomes much more dramatic when spin orbit coupling is switched on . anomalies rather similar to those in the black curve in fig . 3(a ) appear in the green dispersion curve found in fig . 3 of ref . , though these authors did not continue their calculation much beyond the 1 @xmath77 regime . our spin waves are very much softer than theirs in this spectral region , notice . in fig . 3 of ref . , one finds two dispersion curves , one a mirror image of the second . thus , these authors display two spin wave frequencies for each wave vector . this surely is not correct . for a structure with one atom per unit cell , there is one and only one spin wave mode for each wave vector , though as discussed above for the structure explored here symmetry allow the left / right asymmetry in the dispersion curve illustrated in our fig . [ fig3 ] . in fig . [ fig3]b , again with spin orbit coupling switched on and off , we show an expanded view of the dispersion curve for small wave vectors . with spin orbit interaction switched off , at zero wave vector we see a zero frequency spin wave mode , as required by the goldstone theorem when the underlying hamiltonian is form invariant under spin rotation . the curve is also symmetrical , and is accurately fitted by the form @xmath78 ( mev ) , with the wave vector in @xmath77 , whereas with spin orbit coupling turned on the dispersion relation is fitted by @xmath79 ( mev ) . spin orbit coupling introduces an anisotropy gap at @xmath80 , and most striking is the term linear in wave vector . this has its origin in the dzyaloshinskii moriya interaction whose presence , as argued by the authors of ref . , has its origin in the absence of both time reversal and inversion symmetry , for the adsorbed layer . at long wavelengths , one may describe spin waves by classical long wavelength phenomenology . the linear term in the dispersion curve has its origin in a term in the energy density of the spin system of the form @xmath81 here @xmath82 is a spin density , the @xmath83 plane is parallel to the surface , and the magnetization is parallel to the @xmath84 direction . one interesting feature of the spin wave mode whose dispersion relation is illustrated in fig . [ fig3]b is that at @xmath80 , the mode has a finite group velocity . the fit to the dispersion curve gives this group velocity to be @xmath85cm / s , which is in the range of acoustic phonon group velocities . we turn now to our calculations of spin waves and the response functions for the fe bilayer on w(110 ) . let us first note that experimentally the orientation of the magnetization in the bilayer appears to be dependent on the surface upon which the bilayer is grown . for instance , when the bilayer is on the stepped w(110 ) surface , it is magnetized perpendicular to the surface,@xcite a result in agreement with ab initio calculations of the anisotropy realized in the epitaxial bilayer.@xcite however , in the speels studies of spin excitations in the bilayer@xcite the magnetization is in plane . in our calculations , we find for model b the magnetization is perpendicular to the surface , whereas in model c it lies in plane , along the long axis very much as in the speels experiments . the anisotropy in the bilayer is not particularly large , on the order of 0.5 mev / fe atom , and one sees from these results that it is a property quite sensitive to the details of the electronic structure . the fact that model b and model c give the two different stable orientation of the magnetization allows us to explore spin excitations for the two different orientations of the magnetization . we first turn our attention to the case where the magnetization lies in plane . the bilayer has two spin wave modes , an acoustic mode for which the magnetization in the two planes precesses in phase , and an optical mode for which they precess 180 degrees out of phase . in fig . [ fig4 ] , we show calculations of the dynamic susceptibility in the frequency range of the acoustic mode for two values of the wave vector , @xmath86 and @xmath87 . a spin orbit induced left right asymmetry is clearly evident both in the peak frequency and the height of the feature . very recently , beautiful measurements of spin orbit asymmetries in the fe bilayer have appeared,@xcite and the results of our fig . [ fig4 ] are to be compared with fig . 3 of ref . . theory and experiment are very similar , both in regard to the intensity asymmetry and also the spin orbit induced frequency shift , though our calculated spin wave frequencies are a little stiffer than those found experimentally . ( black curve ) and @xmath88 ( red curve ) . model c has been used for the calculation . in the ground state , the magnetization lies in plane along the long axis.,scaledwidth=80.0% ] as remarked above , in fig . [ fig4 ] we show only the acoustical spin wave mode frequency regime . in fig . [ fig5 ] , for the spectral densities in the innermost layer ( upper panel ) and the outermost layer ( lower panel ) we show the spectral densities for the entire spin wave regime , including the region where the optical spin wave is found . it is clear that the spin orbit induced frequency shifts are largest for the optical mode which , unfortunately is not observed in the experiments.@xcite we show the spectral densities in the innermost fe layer ( upper panel ) and in the outermost layer ( lower panel ) for the fe bilayer on w(110 ) . the figure includes the optical spin wave feature . as in fig . [ fig4 ] , the black curve is calculated for @xmath89 positive , and the red curves are for @xmath89 negative . the calculations employ model c.,scaledwidth=80.0% ] in fig . [ fig6 ] for a sequence of wave vectors , all chosen positive , we show a sequence of spectra calculated for the entire frequency range so both the acoustic and optical spin wave feature are displayed . the black curves show the spectral density of the innermost fe layer , and the red curves are for the outer layer . the optical spin wave mode , not evident in the data , shows clearly in these figures . notice that for wave vectors greater than 1@xmath77 the acoustical mode is localized in the outer layer and the optical mode is localized on the inner layer . the optical mode is very much broader than the acoustical mode at large wave vectors , by virtue of the strong coupling to the electron hole pairs in the w 5d bands . an interesting issue is the absence of the optical mode from the speels spectra reported in refs . and . we note that these spectra are taken with only two beam energies , 4 ev and 6.75 ev . at such very low energies , the beam electron will sample both fe layers , so the speels signal will be a coherent superposition of electron waves backscattered from each layer ; the excitation process involves coherent excitation of both layers by the incident electron . as a consequence of the 180 degree phase difference in spin motions associated with the two modes it is quite possible , indeed even probable , that for energies where the acoustical mode is strong the intensity of the optical mode is weak , by virtue of quantum interference effects in the excitation scattering amplitude . in earlier studies of surface phonons , it is well documented that on surfaces where two surface phonons of different polarization exist for the same wave vector , one can be silent and one active in electron loss spectroscopy.@xcite it would require a full multiple scattering analysis of the spin wave excitation process to explore this theoretically . while earlier@xcite calculations that address speels excitation of spin waves described by the heisenberg model could be adapted for this purpose , in principle , a problem is that at such low beam energies it is necessary to take due account of image potential effects to obtain meaningful results.@xcite this is very difficult to do without considerable information on the electron reflectivity of the surface.@xcite it would be of great interest to see experimental speels studies of the fe bilayer with a wider range of beam energies to search for the optical mode , if this were possible . in fig . [ fig7 ] , we show dispersion curves for the optical and acoustic spin wave branches for the bilayer . the magnetization lies in plane , and one can see that on the scale of this figure , the spin orbit effects on the dispersion curve are rather modest compared to those in the monolayer . for small wave vectors , with spin orbit coupling present , the dispersion curve of the acoustic spin wave branch is fitted by the form @xmath90(mev ) so at long wavelengths the influence of the dzyaloshinskii moriya interaction is more than one order of magnitude smaller than it is in the monolayer . if the magnetization is perpendicular to the surface , then symmetry considerations show that there are no left / right asymmetries in the spin wave propagation characteristics . one may see this as follows . consider a wave vector @xmath4 in the plane of the surface , which also is perpendicular to the magnetization , and thus perpendicular to the long axis . the reflection @xmath91 in the plane perpendicular to the surface and which contains the magnetization simultaneously changes the sign of wave vector and the magnetization . if this is followed by the time reversal operation @xmath92 , then @xmath4 remains reversed in sign but the magnetization changes back to its original orientation . thus the product @xmath93 leaves the system invariant but transforms @xmath4 into @xmath73 . the two propagation directions are then equivalent . we illustrate this in fig . [ fig8 ] where , for @xmath94 , where it is shown that the spectral densities calculated for the two directions of propagation are identical , with spin orbit coupling switched on . model b , in which the magnetization is perpendicular to the surface , has been used in these calculations . the spectral densities calculated for the two signs of @xmath89 can not be distinguished to within the numerical precision we use . , we show spectral density function calculated for positive values of @xmath89 ( continuous lines ) and negative values of @xmath89 ( symbols ) . the black curve is the spectral density for the inner layer , and the red curve is the outermost fe layer.,scaledwidth=60.0% ] we have developed the formalism which allows one to include the influence of spin orbit coupling on the spin excitations of ultrathin ferromagnets on semi infinite metallic substrates . our approach allows us to calculate the full dynamic susceptibility of the system , so as illustrated by the calculations presented in section iii we can examine the influence of spin orbit coupling on the linewidth ( or lifetime ) of spin excitations , along with their oscillator strength . as in previous work , we can then construct effective dispersion curves by following peaks in the spectral density as a function of wave vector , without resort to calculations of large numbers of very small distant neighbor exchange interactions . the results presented in fig . [ fig4 ] are very similar to the experimental data reported in ref . , as discussed above , though we see that in the bilayer the influence of the dzyaloshinskii moriya interaction is considerably more modest than in the monolayer . we will be exploring other issues in the near future . one interest in our minds is the influence of spin orbit coupling on the spin pumping contributions to the ferromagnetic resonance linewidth , as observed in ferromagnetic resonance ( fmr ) studies of ultrathin films.@xcite it has been shown earlier@xcite that the methodology employed in the present paper ( without spin orbit coupling included ) can be applied to the description of the spin pumping contribution to the fmr linewidth , and in fact an excellent quantitative account of the data on the fe / au(100 ) system was obtained . it is possible that for films grown on 4d and 5d substrates that spin orbit coupling can influence the spin pumping relaxation rate substantially . this will require calculations directed toward much thicker films than explored here . the formalism we have developed and described in the present paper will allow such studies in the future . this research was supported by the u. s. department of energy , through grant no . de - fg03 - 84er-45083 . s. l. wishes to thank the alexander von humboldt foundation for a feodor lynen fellowship . a.t.c . and r.b.m acknowledge support from cnpq and faperj and a.b.k . was supported also by the cnpq , brazil . in this appendix we provide explicit expressions for the various quantities which enter the equations displayed in section ii . while these expressions are unfortunately lengthy , it will be useful for them to be given in full . the various expectation values in the equations above and those displayed below are calculated from the single particle green s functions once the self consistent ground state parameters are determined . then @xmath97\nonumber\\ \tilde{b}^{23}_{\mu\nu,\mu'\nu'}(ll';mm ' ) = \sum_{\eta}\left ( u_{l;\mu'\nu,\nu'\eta } \langle c^\dagger_{l\mu{\uparrow}}c_{l'\eta{\uparrow}}\rangle\delta_{l'm}\delta_{l'm'}- u_{l;\eta\mu',\mu\nu ' } \langle c^\dagger_{l\eta{\uparrow}}c_{l'\nu{\uparrow}}\rangle\delta_{lm}\delta_{l'm'}\right)\nonumber\\ \tilde{b}^{24}_{\mu\nu,\mu'\nu'}(ll';mm ' ) = -\sum_{\eta } u_{l';\mu'\nu,\eta\nu'}\langle c^\dagger_{l\mu{\uparrow}}c_{l'\eta{\downarrow}}\rangle\delta_{l'm}\delta_{l'm'}\nonumber\\ \label{bts_2}\end{aligned}\ ] ] @xmath98\nonumber\\ \tilde{b}^{34}_{\mu\nu,\mu'\nu'}(ll';mm ' ) = \sum_{\eta } u_{l;\mu'\eta,\mu\nu'}\langle c^\dagger_{l\eta{\uparrow}}c_{l'\nu{\downarrow}}\rangle\delta_{lm}\delta_{lm'}\nonumber\\ \label{bts_3}\end{aligned}\ ] ] rodrigo arias and d. l. mills , * 60 * , 7395 ( 1999 ) ; a. azevedo , a. b. oliveira , f. m. de aguiar , and s. m. rezende , * 62 * , 5331 ( 2000 ) ; d. l. mills and s. m. rezende in spin dynamics in confined structures ii , ( springer verlag , heidelberg , 2002 ) ; p. landeros , rodrigo e. arias and d. l. mills , * 77 * , 214405 , ( 2008 ) ; j. lindner , i. barsukov , c. raeder , c. hassel , o. posth , r. meckenstock , p. landeros and d. l. mills , * 80 * , 224421 ( 2009 ) . r. urban , g. woltersdorf and g. heinrich , * 87 * , 217204 ( 2001 ) , y. tserkovnyak , a. brataas and g. e. w. bauer , * 88 * , 117601 ( 2002 ) ; * 66 * , 224403 ( 2002 ) ; d. l. mills , * 68 * , 014419 ( 2003 ) ; m. zwierzycki , y. tserkovnyak , p. j. kelly , a. brataas , and g. e. w. bauer , * 71 * , 064420 ( 2005 ) ; a. t. costa , r. b. muniz and d. l. mills , * 73 * , 054426 ( 2006 ) . r. vollmer , m. etzkorn , p. s. anil kumar , h. ibach and j. r. kirschner , * 91 * , 14720 ( 2003 ) ; for a theoretical discussion of the linewidth and dispersion observed in the experiment , see a. t. costa , r. b. muniz * 70 * , 0544406 ( 2004 ) . early papers are h. tang , m. plihal and d. l. mills , journal of magnetism and magnetic materials * 187 * , 23 ( 1998 ) ; r. b. muniz and d. l. mills , * 66 * , 174417 ( 2002 ) ; a. t. costa , r. b. muniz and d. l. mills , * 68 * , 224435 ( 2003 ) . for an explicit comparison between the predictions of the heisenberg model and that of an itinerant electron description of the spin excitations , see fig . 4 and fig . 5 of the theoretical paper cited in ref . .

soft x - ray transients ( sxts ) are compact binary systems in which a low - mass secondary ( either a main - sequence star or a subgiant ) transfers mass via roche - lobe overflow onto a black hole ( bh ) or neutron star ( ns ) primary ( see reviews by tanaka & lewin 1995 ; van paradijs & mcclintock 1995 ; white , nagase & parmar 1995 ) . sxts have highly variable luminosities . they spend most of their lifetimes in a low luminosity quiescent state , but occasionally undergo dramatic outbursts during which both the optical and x - ray emission increase by several orders of magnitude ( e.g. chen , shrader & livio 1997 ; kuulkers 1998 ) . ns sxt outbursts typically occur every 1 - 10 years and last for several weeks , while bh sxt outbursts are typically separated by 10 - 50 years ( or perhaps longer ) and last for several months ( see chen et al . 1997 ) . a variety of observations ( see e.g. tanaka & shibazaki 1996 ) indicate that , near the peak of an outburst , sxts accrete matter via a standard thin disk ( shakura & sunyaev 1973 ) , so that there is little doubt that the accretion is radiatively efficient during this phase . the situation is more complex in quiescence . the spectra of quiescent bh sxts do not resemble that of a thin disk , and the accretion rates inferred from the observed x - ray luminosities disagree by orders of magnitude with the predictions of the standard disk instability model for quiescent disks ( e.g. lasota 1996 ) . narayan , mcclintock & yi ( 1996 ) and narayan , barret & mcclintock ( 1997 ) showed that the observations of quiescent bh sxts can be explained by a two - component accretion flow model consisting of an inner hot advection - dominated accretion flow ( adaf ; narayan & yi 1994 , 1995a ; abramowicz et al . 1995 ; ichimaru 1977 ; see narayan , mahadevan & quataert 1998 and kato , mineshige & fukue 1998 for reviews of adafs ) surrounded by an outer thin disk . in the recent version of this model described in narayan et al . ( 1997a ) , only the inner adaf contributes to the observed optical , uv and x - ray emission of the system . the outer thin disk acts mainly as a reservoir which accumulates mass until the next outburst is triggered . in quiescence , the emission of the disk is primarily in the infrared ; this radiation is hardly seen as it is swamped by the emission of the secondary ( see narayan et al . 1996 , 1997a and lasota , narayan & yi 1996 for details ) . a key feature of the narayan et al . ( 1996 , 1997a ) model of quiescent bh sxts is the low radiative efficiency of the adaf . in these flows , the bulk of the viscously dissipated energy is stored in the gas and advected with the flow into the black hole ( narayan & yi 1995b ; abramowicz et al . 1995 ; narayan et al . 1996 , 1997a ) . this explains the unusually low luminosities of bh sxts . by contrast , in ns sxts all the advected energy is expected to be radiated from the neutron star surface , resulting in a much higher radiative efficiency of the accretion flow even in the presence of an adaf ( narayan & yi 1995b ) . motivated by this fundamental distinction between black hole and neutron star systems , narayan , garcia & mcclintock ( 1997 ) and garcia et al . ( 1998 ) compared the outburst amplitudes of bh sxts and ns sxts as a function of their maximum luminosities and showed that the observations reveal systematically lower relative luminosities in bh sxts . they argued that this constitutes a confirmation of the presence of an event horizon in bh sxts . the argument was challenged by chen et al . ( 1998 ) . in this paper , we attempt to develop a physical understanding of the difference in quiescent luminosities between ns and bh sxts . we first show , in 2 , that there is indeed a significant difference in the observed quiescent luminosities of the two classes of objects , in contrast to the claim of chen et al . we then use binary evolution models in 3 to estimate mass transfer rates in sxts . we combine these estimates with the adaf+thin disk accretion scenario , taking into account the presence of an event horizon in bh systems and a reradiating surface in ns systems , to determine the expected x - ray luminosities in quiescence ( 4 ) . the model predictions agree well with the observations of quiescent bh sxts . however , the model substantially overestimates the luminosities of quiescent ns sxts . in 5 we show that the ns sxt data can be reconciled with the predictions of the model if we take into account the `` propeller effect '' ( illarionov & sunyaev 1975 ) , whereby the magnetosphere of a rapidly rotating neutron star prevents much of the accreting material from reaching the surface of the neutron star . in 6 we show that an adaf model with a moderate wind and a somewhat less efficient propeller ( in ns sxts ) , is also consistent with the observed quiescent luminosities of bh and ns sxts . finally , in 7 we discuss possible limitations and extensions of this work and in 8 we summarize the main results . in table [ tab : systems ] , we list key parameters of several ns and bh sxts : the orbital period @xmath0 , the distance @xmath1 , the quiescent x - ray luminosity @xmath2 in the @xmath3 kev x - ray band , and the mass @xmath4 ( in solar units ) of the compact primary . since our main interest is in the quiescent emission , we list only systems for which there exist reliable measurements of @xmath2 . the values of @xmath2 are taken from narayan et al . ( 1997b ) and garcia et al . ( 1998 ) , except for the bh sxt h1705 - 250 ( see below ) . in selecting the systems listed in table [ tab : systems ] , we were careful to avoid any possible confusion between bh and ns sxts , since this would make a comparison between the two sets of objects less reliable . thus , we limit our sample to the eight bh candidates that have firm _ dynamical _ lower limits on the mass of the primary ( e.g. mcclintock 1998 ) . ( note that this is not true for several so - called bh `` candidates '' of chen et al . 1998 , which seriously weakens their arguments . ) similarly , there is firm evidence that the five systems listed in table [ tab : systems ] as ns sxts contain ns primaries , based on the detection of type i x - ray bursts ( e.g. narayan et al . 1997b ; chen et al . 1997 ) . here we report a new and improved luminosity limit for h1705 - 250 ( = nova oph 77 ) based on the asca observation of 18 september 1996 , which we extracted from the heasarc archive . the exposure time was 31 ks , but unfortunately the target was near the chip boundaries which limits the effective area . the 4@xmath5 upper limit is @xmath6 c s@xmath7 , which corresponds to an x - ray flux @xmath8ergs s@xmath7@xmath9 . assuming a distance of 8.6 kpc , @xmath10ergs s@xmath7 . luminosities can be expressed in units of the eddington luminosity , @xmath11 where @xmath4 , the mass of the compact object in solar units , is listed in table [ tab : systems ] . for a standard radiative efficiency of @xmath12 % , the eddington luminosity corresponds to a mass accretion rate of @xmath13 in this section , we use both absolute luminosities and eddington - scaled luminosities in a system is not well known ( eq . [ [ eq : ledd ] ] ) . this is for instance the case of the bh sxt j1655 - 40 . orosz & bailyn ( 1997 ) estimate that @xmath14 in this system , while phillips et al . ( 1999 ) claim that @xmath4 could be as low as @xmath15 since it is not clear which is the more appropriate quantity for comparisons . our calculations in 3 suggest that the mass transfer rates in ns and bh sxts are similar when expressed in units of @xmath16 , especially at short orbital periods . thus , eddington - scaled luminosities might be more appropriate to compare the two classes of objects at short periods ( where most of the data lie ) . however , this argument is not very strong . therefore , we also show absolute luminosities , expressed in units of @xmath17 erg s@xmath7 ( which is roughly the luminosity expected from an object accreting at @xmath18 g s@xmath7 with a radiative efficiency of @xmath12 % ) . finally , the ratio @xmath19 ( which is independent of the distance @xmath1 ) is used for comparison with previous investigations of the luminosity difference between bh and ns sxts . the values of @xmath20 , the outburst peak luminosity , are taken from garcia et al . ( 1998 ) . in figure [ fig : lumtot ] , we show the quiescent luminosities of the bh and ns systems listed in table [ tab : systems ] as a function of their orbital periods @xmath0 . the open circles correspond to ns sxts and the filled dots correspond to bh systems . luminosity upper limits ( indicated by downward arrows ) are shown for five sxts , all of which are bh systems . the orbital period of the ns sxt h1608 - 52 is uncertain ( see table [ tab : systems ] ) and its location in figure [ fig : lumtot ] is indicated by dashed circles . among the undetected sxts , we choose to include in our sample only those systems that have been observed for more than 10 ks ( see table [ tab : systems ] herein , and also table 1 in narayan et al . 1997b ) , which therefore have flux limits @xmath21 ergs s@xmath7 with current x - ray satellites ( 0.5 - 10 kev ; rosat / asca ) . the argument for using @xmath0 along the horizontal axis of figure [ fig : lumtot ] is as follows . for any binary system with a low mass secondary , and transferring mass via roche - lobe overflow , the density of the secondary essentially determines @xmath0 . at a given @xmath0 , a bh sxt and a ns sxt will have similar secondaries , so that the mass transfer characteristics are likely to be similar . thus a reliable comparison of the radiative efficiencies of their accretion flows would be possible . in contrast , if we were to compare the quiescent luminosities of a bh sxt and a ns sxt with quite different @xmath0 , a difference in the mass transfer rates ( see 3 ) could mask actual differences in the radiative efficiencies of the accretion flows . this point , which is the motivation for the calculations in 3 , was first emphasized by lasota & hameury ( 1998 ) . figure [ fig : lumtot ] strongly suggests that bh sxts are fainter than ns sxts . this is especially true when the data are plotted in eddington units ( fig . [ fig : lumtot]a ) , but it is reasonably convincing even in absolute units ( fig . [ fig : lumtot]b ) or in terms of the ratio @xmath22 ( fig . [ fig : lumtot]c ) . note that all five of the ns sxts have been detected in quiescence and have well measured values of @xmath2 , while the majority of bh sxts have only upper limits on @xmath2 . in our view , this is significant and underlines the effect we are claiming . to make a clearer comparison between bh and ns sxts with the existing data , we isolate in figure [ fig : lumfirm ] those systems for which @xmath2 and @xmath0 are well determined ( see also the very similar fig . 3 of lasota & hameury 1998 ) . although there are only a few systems in this figure , we feel that we can conclude that ( at least in this small sample ) bh sxts in quiescence are less luminous than ns sxts . the difference is as much as two orders of magnitude ( in eddington units ) for short period systems . it is important to stress here the excellent upper limit on the quiescent luminosity of the bh sxt j0422 + 32 . this upper limit is nearly equal to the very low measured quiescent luminosity of a0620 - 00 . thus , there are now two bh sxts that are two orders of magnitude less luminous , in eddington units , than quiescent ns sxts with comparable @xmath0 . it is important to carry out more sensitive observations of those bh sxts which have only upper limits currently . narayan et al . ( 1997b ) explained the difference in quiescent luminosities of ns sxts and bh sxts as due to the presence of a hard surface in ns sxts vs an event horizon in bh sxts . the presence of a surface causes the radiative efficiency in ns sxts to be high , while in bh sxts most of the dissipated energy is advected into the event horizon of the bh , without producing observable emission . narayan et al . argued that this constitutes observational evidence for event horizons in black holes . in the remainder of the paper , we expand on this idea and construct a physical model of accretion in ns and bh systems to explain quantitatively the observations summarized in figure [ fig : lumfirm ] . in 3 we estimate the mass transfer rate in sxts , and in 4 and 5 we describe an accretion model that is able to reproduce the observed emission in the @xmath3 kev band . the theory of binary evolution relies on the roche - lobe model for the description of mass transfer ( frank , king & raine 1992 ) . depending on which mechanism drives the mass transfer , whether it is loss of orbital angular momentum through gravitational radiation and magnetic braking , or expansion of the donor as it evolves away from the main - sequence , the binaries are classified as j - driven or n - driven systems . there is a ` bifurcation ' orbital period @xmath23 separating the two classes such that for @xmath24 we find n - driven systems whose orbital periods increase with time , and for @xmath25 we find j - driven systems whose @xmath26 decrease with time . although there is no unique value of @xmath23 , because it depends on the donor mass and the strength of angular momentum losses , estimates range from @xmath27 days ( e.g. pylyser & savonije 1988 ; king , kolb & burderi 1996 ) . the exact form of the magnetic braking law is not well established ( see e.g. kalogera , kolb & king 1998 ) . moreover , menou , narayan & lasota ( 1998 ) showed that there is strong circumstantial evidence that magnetic braking ( mb ) is weak in bh sxts . in the following , for simplicity , we neglect the influence of mb in bh and ns sxts . we will see in 4.2 that any contribution to the mass transfer by mb would only strengthen the argument that ns sxts shed mass rather than accrete it . following king et al . ( 1996 ) , the mass transfer rates in j - driven systems with gravitational radiation ( gr ) and n - driven system with secondary expansion ( exp ) is given by : @xmath28 where @xmath29 is the mass of the secondary in solar units , @xmath0 is scaled in units of 1 day . the secondary in j - driven binaries is usually a main - sequence star ( see king et al . 1996 for possible complications ) filling its roche lobe . these stars obey the simple relation @xmath30 ( e.g. frank et al . 1992 ) . the prediction for @xmath31 in case of gr driven mass transfer then becomes @xmath32 which scales with the mass of the primary as @xmath33 for small @xmath29 . note this is not very different from @xmath34 , thus motivating the use of the eddington scaling of luminosity in figs . [ fig : lumtot ] and [ fig : lumfirm ] equation ( [ ev ] ) for the case of secondary expansion is the linear limit of third - order polynomial fits ( king 1988 ) to detailed evolutionary tracks of single stars computed by webbink , rappaport & savonije ( 1983 ) . along an evolutionary track , the mass @xmath29 of the expanding secondary decreases as it transfers mass onto the primary . since a variety of initial masses and orbital periods ( at the onset of mass - transfer ) are possible , @xmath29 is essentially a free parameter in equation ( [ ev ] ) . however , a careful study of evolutionary tracks shows that the band of mass transfer rates predicted by equation ( [ ev ] ) for values of @xmath29 in the range 0.5 to 1 reproduces the mass transfer rates in n - driven accreting binaries for the majority of their lifetime . this band therefore provides , in a `` population synthesis '' sense , the likely mass transfer rates one can expect in sxts with orbital periods larger than the bifurcation period . we estimate the mass transfer rate expected at a given orbital period by simply adding the contributions from gr ( eq . [ [ gr ] ] ) and secondary expansion ( a band with @xmath35 $ ] in eq . [ [ ev ] ] ) . figure [ fig : accrate ] shows that this results in a relatively short , although plausible , bifurcation period @xmath36 hr . the estimates of @xmath37 are shown separately for bh and ns sxts , in units of @xmath18 g s@xmath7 . the mass transfer rates are typically similar , in absolute units , in bh and ns sxts with long orbital periods . however , the mass transfer rate in bh sxts is larger than in ns sxts at short orbital periods ( cf eq . [ [ gr ] ] and its scaling with @xmath4 ) ; the two are in fact quite close in eddington units at these periods , where the majority of sxts are found . although this result neglects any contribution from magnetic braking , we think that it is significant enough to motivate the use of eddington - scaled luminosities in figs . [ fig : lumtot ] and [ fig : lumfirm ] . note that the comparable values of eddington - scaled @xmath31 predicted for bh sxts and ns sxts at short periods contrast with the large eddington - scaled luminosity difference of the two classes of objects ( figs . [ fig : lumtot ] and [ fig : lumfirm ] ) . this in essence is the argument of narayan et al . ( 1997b ) and garcia et al . ( 1998 ) , and we confirm their claim . the sample is , however , small . although the estimates shown in fig . [ fig : accrate ] rely on simple considerations that do not take into account the full complexity of mass transfer ( in particular , substantial fluctuations of the mass transfer rates around the secular values used here may occur ; e.g. warner 1995 ) , they probably provide better than an order - of - magnitude precision , which is sufficient for our discussion . note , however , that this simple model is not likely to give an accurate value of @xmath37 for the bh sxt j1655 - 40 since this system seems to be in a special evolutionary state ( crossing the hertzsprung gap , e.g. kolb 1998 ) . following narayan et al . ( 1996 , 1997a ) , we assume that the accretion flow is made of two components : an inner adaf and an outer thin accretion disk . since only the adaf contributes to the emission of quiescent sxts in the 0.5 - 10 kev band , we ignore the thin disk . further , since the emission of the adaf is primarily from regions close to the center , we do not concern ourselves with the precise value of the radius @xmath38 at which the transition between the adaf and the thin disk occurs . although the adaf model was initially proposed for quiescent bh sxts , there is no reason why it should not apply also to quiescent ns sxts . in particular , if the transition of the accretion flow from a thin disk configuration to an adaf configuration is triggered by the same local physics in the disk , e.g. a thermal instability in the upper layers of the disk ( shaviv & wehrse 1986 ; meyer & meyer - hofmeister 1994 ; narayan & yi 1995b ) , the process should occur independently of the nature of the central object . we note , however , that the presence of a reradiating surface in the case of a ns primary at the center of the adaf may have significant effects on the adaf . for instance , narayan & yi ( 1995b ) showed that the critical accretion rate @xmath39 in an adaf is reduced in the case of a ns ( compared to a bh ) because the additional supply of soft photons ( from the stellar surface ) causes the adaf to cool more efficiently . this is not an issue here since @xmath40 in all the quiescent systems we study is well below @xmath41 . we assume conservative mass transfer . consequently , if @xmath42 is the rate at which mass is accreted during quiescence via the adaf and @xmath43 is the rate at which mass is accumulated in the outer thin disk , the mass transfer rate satisfies @xmath44 . the proportion of mass accreted vs. mass accumulated in quiescence is , however , uncertain . menou et al . ( 1998 ) estimated the relative importance of @xmath42 and @xmath43 for most of the bh sxts listed in table [ tab : systems ] . they estimated @xmath43 by taking the integrated fluences of the outbursts of each source and assuming a radiative efficiency of @xmath45 ( which provides the total mass accreted in the outburst ) , and estimated @xmath46 by fitting the quiescent emission with an adaf model . menou et al.s work indicates that , in quiescent bh sxts , roughly half the matter that is transferred from the secondary is accreted via the adaf and the rest is accumulated in the thin disk , i.e. @xmath47 in the following models , we treat @xmath48 as a free parameter , but we expect @xmath48 to be of order unity both in bh sxts and in ns sxts . the version of the adaf model used here is described in detail by narayan et al . ( 1998a ) . it consistently includes adiabatic compressive heating of electrons in the energy equation ( nakamura et al . 1997 ) and it uses for the flow dynamics the general relativistic global solutions calculated by popham & gammie ( 1998 ) . in previous studies involving adafs , most of the model parameters were kept fixed at the following values ( see narayan et al . 1998b for a discussion ) : @xmath49 ( the viscosity parameter in the adaf ) , @xmath50 ( i.e. equipartition between the gas and the magnetic field ) , @xmath51 ( adiabatic index of the gas ) and @xmath52 ( fraction of the viscous dissipation that goes directly into heating the electrons ) . we adopt these standard values of @xmath53 , @xmath54 and @xmath55 in the present models ( see 6 for the effect of varying @xmath56 ) . we take @xmath37 to be given by the predictions of binary evolution models ( fig . [ fig : accrate ] ) and obtain @xmath46 via equation ( [ eq : g ] ) for a given assumed value of @xmath48 . we calculate the spectral energy distributions of the models and numerically integrate the spectra to obtain the luminosities in the 0.5 - 10 kev band . the results are discussed in the following sections . figure [ fig : lumbh]a shows two bands of luminosities corresponding to adaf models of bh sxts . in the upper band @xmath57 , i.e. the entire @xmath37 is accreted by the black hole via the adaf , while in the lower band @xmath58 ( or @xmath59 ) . the lower band is in satisfying agreement with the observed luminosities . the fact that we obtain agreement with a reasonable value of @xmath48 means that the modeling of the quiescent x - ray emission of bh sxts with the adaf model is consistent with the predictions of binary evolution models . the increase in the luminosity with increasing @xmath0 is also in good agreement . one should stress again , however , that the agreement for the bh sxt j1655 - 40 is unexpected if the system is crossing the hertzsprung gap ( kolb 1998 ) . the bolometric radiative efficiency in the adaf models constructed here is typically @xmath60 compared to @xmath61 for standard thin disk accretion . the x - ray radiative efficiency of these adaf models is another two or three orders of magnitude lower since much of the adaf luminosity comes out as synchrotron radiation in the optical . note that these results are sensitive to various details of the adaf models , and in particular to the choice of @xmath53 and @xmath54 . for instance , we find that , if @xmath62 , we need @xmath63 ( i.e. a fraction @xmath64 of the mass supplied by the secondary must be accreted via the adaf ) in order to obtain agreement with the observed quiescent luminosities of bh sxts . the fraction goes down further to @xmath65 if @xmath66 . the presence of a hard surface on the accreting star is the key feature of adaf models of ns sxts . the large amount of energy advected in the adaf , which is lost through the event horizon in bh sxts , will be reradiated from the ns surface ( narayan & yi 1995b ) . narayan et al . ( 1997b ) and garcia et al . ( 1998 ) argued that this additional source of luminosity is the explanation for the systematically larger luminosity of quiescent ns sxts as compared to quiescent bh sxts ( figs [ fig : lumtot ] and [ fig : lumfirm ] ) . we now investigate this quantitatively . in the following , we assume that the thermal energy stored in the adaf is reradiated as blackbody emission with a radiative efficiency @xmath67 . the luminosity in the 0.5 - 10 kev band will be dominated by the emission coming from the ns surface , so that the additional emission from the adaf can be ignored ( for simplicity ) . we also assume that the reradiation occurs from a small fraction @xmath68 of the ns . this is in agreement with observations of quiescent ns sxts ( e.g. verbunt et al . 1994 ; campana et al . 1998 ; but see rutledge et al . 1998 and the discussion in 7 ) . for now , we do not seek to justify the small emitting area of the ns . we simply assume for all our ns models that the area is @xmath69 . however , our results are not seriously affected even if the radiating area is increased or decreased by , say , an order of magnitude ( i.e as long as most of the blackbody emission is in the 0.5 - 10 kev x - ray band ) . the typical effective temperatures found are @xmath70 kev . the upper band in fig . [ fig : lumbh]b shows the luminosities predicted by binary evolution models if @xmath59 , i.e a third of the mass transferred by the secondary is accreted onto the ns ( as in bh sxts ) . we see that the luminosities predicted are much larger than those observed , by @xmath71 orders of magnitude ( except for the system exo 0748 - 676 ; see 7 for a discussion of this system ) . since any contribution to the mass transfer by magnetic braking would only increase the mass transfer rates expected ( in short orbital period systems ) , the discrepancy between the observed and predicted luminosities would be even more serious if magnetic braking were effective . one possibility for the discrepancy is that most of the quiescent flux from ns sxts is emitted outside the 0.5 - 10 kev x - ray band . this explanation seems , however , rather unlikely since this emission would be seen somewhere else in the spectrum , say in hard x - rays or soft @xmath55-rays . aql x-1 has been observed in quiescence in hard x - rays and it is clear that the energy does not come out in the 10 - 100 kev band ( campana et al . 1998 ) . in addition , although some theoretical models of boundary layers between an accretion flow and a ns predict substantial deviations from blackbody emission at higher energies ( e.g. shapiro & salpeter 1975 , turolla et al . 1994 ) , there are no models predicting that such a large fraction ( @xmath72 ) of the emission would come out in hard x - rays or @xmath55-rays . we must therefore consider the possibility that only part of the mass supplied by the secondary actually reaches the surface of the ns . the lower band in fig . [ fig : lumbh]b shows that the luminosities of quiescent ns sxts can be accounted for ( except exo 0748 - 676 ) if only @xmath73 of the transferred mass reaches the ns surface . this is of course a surprisingly small fraction . in such a disk ; e.g. cannizzo 1993 ) , we do not favor this interpretation because the propeller effect should operate in this case and prevent any mass from reaching the ns surface ( see 5 ) ] . in the following section , we argue that the propeller effect can explain the fraction if combined with accretion via an adaf , while in 6 we consider the possibility that winds from adafs could further reduce the amount of mass reaching the ns surface . the propeller effect in accreting neutron stars was initially proposed by illarionov & sunyaev ( 1975 ) to explain the existence of long - period x - ray pulsars , and it was later developed in more detail by davies and pringle ( 1981 ; see also wang & robertson 1985 ; stella et al . although the details vary , the basic mechanism of the effect relies on the presence of a centrifugal barrier at the rotating magnetosphere of a rapidly spinning ns . inside the magnetosphere , the accreting gas is forced to follow the magnetic field lines of the ns since , by definition , in this region magnetic forces dominate the flow dynamics . the fate of the gas is then determined by the relative magnitudes of the magnetospheric radius , @xmath74 , and the corotation radius , @xmath75 , defined by @xmath76 , where @xmath77 is the angular velocity of the ns ( and the magnetosphere ) . if @xmath78 , the magnetosphere rotates so fast that the centrifugal force at @xmath74 is larger than the force of gravity and hardly any of the gas can be accreted onto the ns . this is the propeller effect . the efficiency of the propeller effect is not well understood ( see davies , fabian , & pringle 1979 , and davies & pringle 1981 ) . it is also not clear whether the accreting matter is ejected from the system or is merely accumulated in a boundary layer around the magnetosphere ( wang & robertson 1985 ) . however , from the simple physical argument outlined above , it is clear that normal accretion onto the neutron star is inhibited during the propeller phase . the first direct observational evidence for the existence of the propeller effect in accreting neutron stars was reported by cui ( 1997 ) in two x - ray pulsars , gx 1 + 4 and gro j1744 - 28 . cui showed that in both systems x - ray pulsations ceased during a period of low x - ray flux , which he interpreted as the result of a decrease in the mass accretion rate and a corresponding increase in the magnetospheric radius to beyond the corotation radius . campana et al . ( 1998 ) and zhang et al . ( 1998 ) argued that the propeller effect was also seen in a ns sxt , aql x-1 , during its most recent outburst ; the system showed an abrupt decrease of the x - ray flux , accompanied by considerable hardening of the x - ray spectrum . both sets of authors proposed that the hard spectrum originates just outside the magnetosphere , where the gas becomes very hot due to the action of the propeller ( see also cui et al . 1998 ) . observations of aql x-1 show that even in quiescence the blackbody spectral component does not disappear entirely implying that a small amount of material continues to accrete onto the star , contrary to simple models of the propeller ( verbunt et al . 1994 ; stella et al . 1994 ) . to resolve this problem , zhang et al . ( 1998 ) proposed that accretion in quiescent ns sxts occurs via a quasi - spherical adaf , rather than a thin disk . this allows some material to accrete near the poles , thereby bypassing the centrifugal barrier . in this section we develop this idea quantitatively . in the following , we assume for simplicity that the ns spin axis is perpendicular to the binary orbital plane and that the ns spin and magnetic axes are aligned . in the presence of a spherical accretion flow ( e.g. bondi 1952 ) , the magnetospheric ( or alfvn ) radius , @xmath74 , is usually defined as the radius at which the magnetic pressure , @xmath79 , due to the neutron star magnetic field balances the ram pressure , @xmath80 , of the accreting gas , where @xmath81 is the radial infall velocity of the gas ( e.g. frank et al . 1992 ) . the contribution from the thermal pressure of the gas is generally neglected when considering bondi - type flows which are highly supersonic ( free - falling ) . an adaf , on the other hand , is subsonic down to a few schwarzschild radii ( e.g. narayan , kato , & honma 1997 ; chen , abramowicz & lasota 1997 ; abramowicz , chen , granath & lasota 1996 ) and the thermal pressure term must be taken into account . the polar structure of adafs was described by narayan & yi ( 1995a ) in the self - similar approximation . they showed that in the limit of strong advection ( @xmath82 ) , the gas density and temperature ( and therefore the thermal pressure of the gas ) are essentially independent of the polar angle @xmath83 ; i.e. these quantities are nearly constant on spherical shells centered on the accreting object . on the other hand , the radial infall velocity is smaller than the thermal sound speed in the equatorial plane and decreases towards the poles . clearly then , the thermal pressure of the accreting material is the dominant term in calculating the magnetospheric radius . since this pressure is constant with @xmath83 , the magnetosphere will be nearly spherical . despite the difference between a bondi spherical accretion flow and an adaf , the value of @xmath74 is roughly the same in the two cases . in the former , essentially all the gravitational energy released during accretion goes into bulk kinetic energy of the flow , while in adafs , the energy is stored as thermal energy of the gas . thus , the ram pressure in bondi flows must be of the same order as the thermal pressure in adafs . the magnetospheric radius in quiescence then takes the same form ( frank et al . 1992 ) : @xmath84 where @xmath85 is the ns surface magnetic field in units of @xmath86 g and @xmath87 is the ns radius in units of @xmath88 cm . for the remainder of the paper , @xmath89 refers to the radius in physical units and @xmath90 refers to the radius in schwarzschild units . the propeller effect in binary systems is generally discussed in the context of thin disk accretion . in this model , once the propeller becomes effective , no matter is able to reach the ns surface because the centrifugal acceleration acts equally against gravity for all the matter located in the disk orbital plane . in a spherical flow , however , the centrifugal acceleration at the magnetospheric radius acting on a parcel of gas , accreting at a polar angle @xmath83 from the spin axis , is equal to @xmath91 , where @xmath77 is the angular speed of the magnetic field lines anchored in the ns . the direction of this force is perpendicular to the spin axis of the ns ( and parallel to the orbital plane ) , so that the component of @xmath92 along the radial direction is simply @xmath93 . the maximum polar angle below which the gravitational acceleration ( @xmath94 ) wins over the centrifugal force is given by @xmath95 for @xmath96 this simplifies to @xmath97 note that the residual force acting on all parcels of gas , even those which do not overcome the centrifugal barrier , tends to direct them toward the orbital plane . in this simple formulation , we do not consider possible complications due to field line orientation effects . only matter accreting between @xmath98 and @xmath99 can overcome the centrifugal barrier and reach the surface of the ns . the mass accretion rate onto the star is then @xmath100 where @xmath101 is the gas density and @xmath102 is the radial infall velocity at angle @xmath83 ( positive outward ) , as given by the self - similar solution of narayan & yi ( 1995a ) . the factor of @xmath103 in equation ( [ eq : mdotrtheta ] ) is introduced to account for the two polar caps . the fraction of the total mass accretion rate @xmath104 in the adaf that reaches the ns surface is then @xmath105 where the last two steps have been calculated in the limit of small @xmath106 . for a neutron star with a spin period of @xmath103 ms , a magnetic field strength of @xmath107 g and accreting at @xmath108 , we find @xmath109 . note that if the radial velocity does not vary as @xmath110 but is more nearly constant with @xmath83 ( as in bondi accretion ) , then @xmath111 would be larger ( e.g. , if @xmath112 is independent of @xmath83 , @xmath113 only ) . on the other hand , if the accretion flow has a toroidal morphology with empty funnels along the rotation axis , @xmath111 would be much lower than the expression given in equation ( [ eq : facc ] ) , perhaps even zero . there is thus considerable uncertainty in the value of @xmath111 . we assume that the gas accreted during the propeller phase follows the magnetic field lines down to the ns polar caps , where its kinetic and thermal energy is converted to radiation . dipolar field lines satisfy the parametric equation ( e.g. frank et al . 1992 ) : @xmath114 where @xmath115 is constant for a given field line . thus , the field lines intersecting the magnetospheric radius at @xmath116 , emerge from the ns surface at a polar angle @xmath117 given by : @xmath118 the accreting material reaches the ns surface at @xmath119 , and therefore the fraction of the neutron star surface that re - radiates the accreted energy is @xmath120 where we have made use of the fact that @xmath121 . note the direct dependence of @xmath68 on the ns magnetic field strength @xmath122 through @xmath123 . for a neutron star with a spin period of @xmath103 ms , a magnetic field strength of @xmath107 g and accreting at @xmath124 , we find @xmath125 . equations ( [ eq : facc ] ) and ( [ eq : fsurf ] ) demonstrate that in quiescence the mass accretion rate onto a ns and the emitting area , which determine the total luminosity and spectrum of the system , depend strongly on the ns angular rotation speed @xmath77 and the field strength . although @xmath77 is not known for most ns sxts , we discuss in this section how it can be estimated using our knowledge of the accretion history of an sxt and of the interaction between the magnetosphere of the ns and the accretion flow . during the propeller phase , in quiescence , the ns ejects most of the accreting gas that reaches its magnetosphere . the ejected material leaves the system with higher specific angular momentum than it had coming in , which results in an effective spin - down of the ns . however , ns sxts experience outbursts during which the mass accretion rate onto the ns is orders of magnitude higher than in quiescence . high @xmath126 causes the magnetosphere to shrink inside the corotation radius so that all the accreting gas is able to reach the ns surface . the mass and angular momentum of the accreting gas is added to the ns , which causes it to spin - up . a ns sxt that accretes for a fairly long time is expected to reach an equilibrium spin frequency @xmath127 such that the spin up during outburst is balanced by the spin down during quiescence . at the onset of mass transfer , the spin period of the ns in an sxt is likely to be different from the equilibrium value . in standard ghosh & lamb ( 1979 ) type models ( for steady disks ) , the timescale @xmath128 on which a typical ns reaches the equilibrium spin period is usually small ( @xmath129 yr ) compared to the lifetime of the lmxb ( @xmath130 yr ; see , e.g. , henrichs 1983 for a review ) . although @xmath128 may be a little larger for transient accretion , it is reasonable to assume that most of the neutron stars in sxts have rotation rates close to @xmath131 . it is straightforward to show that @xmath77 is basically unaffected by a single outburst or a single quiescent phase , because the time spent between two outbursts ( @xmath132 a few years ) is much shorter than @xmath128 . therefore , it is not necessary to consider the small jitter in @xmath77 during the quiescence - outburst cycle . in the following , we simply assume that once a ns sxt has reached spin equilibrium , then @xmath133 both during outburst and quiescence . to determine the equilibrium spin frequency of a ns in a transient binary system we need a theoretical description of the interaction between the ns and the accretion flow . we begin with the outburst phase during which accretion occurs via a thin disk and the torque exerted by the accreting gas on the ns is relatively well understood . despite some modifications , current models describing the interaction between a ns and a magnetically threaded thin accretion disk remain essentially identical to the model initially proposed by ghosh & lamb ( 1979 ) . in this model , the spin - up torque on the ns results from two contributions . first , the gas that follows magnetic field lines and reaches the ns surface gives its angular momentum to the ns ( spin - up ) . second , the interaction between the magnetic field lines and the threaded thin disk beyond the magnetospheric radius results in a positive torque on the ns for radii where the field lines rotate slower than the local keplerian angular speed of the gas . the lines that thread the disk further out give a ( smaller ) negative torque on the ns since these field lines rotate faster than the local keplerian angular speed of the gas ( gosh & lamb 1979 ) . recent detailed numerical simulations by daumerie ( 1996 ) show that the overall contribution resulting from the interaction of the accretion flow and the magnetic field lines beyond the magnetospheric radius is nearly equal to the contribution to the spin - up from the gas reaching the ns surface . further , daumerie finds that the torque varies roughly linearly with the fastness parameter @xmath134 . using this work as a guide , we propose the following idealized formula for the torque : @xmath135 , \label{eq : idealtorque}\ ] ] where the factor of 2 appears because of the two nearly equal contributions to the spin - up mentioned above . equation ( [ eq : idealtorque ] ) assumes that the fastness parameter at equilibrium , @xmath136 , is exactly unity . this is consistent with the recent arguments of wang ( 1995 ; see also wang 1987 ) . the torque in the propeller regime , when @xmath137 , is not as well constrained . by analogy with ghosh & lamb type models , we assume that the torque arises from two nearly equal contributions . one contribution is the negative torque applied to the ns when the propeller expels gas at the magnetosphere and the other is the negative torque for all @xmath138 . ] due to the interaction of the magnetic field lines with the accretion flow beyond the magnetospheric radius . if the gas expelled leaves the magnetosphere with an angular momentum corresponding to the angular rotation speed of the ns , viz . @xmath77 ( which corresponds to an efficient propeller ) , then the negative torque on the ns in the propeller phase is @xmath139 where @xmath126 is the rate at which mass reaches the magnetosphere . ( the small positive contribution from the very small fraction of mass accreted on to the ns is neglected . ) this expression is in agreement with equation ( [ eq : idealtorque ] ) in the limit of @xmath140 . therefore , we can use the prescription given by equation ( [ eq : idealtorque ] ) to describe the interaction between the ns and the accretion flow throughout the outburst quiescence cycle . we define @xmath141 to be the integrated spin - up torque over an outburst of duration @xmath142 , and @xmath143 to be the integrated spin - down torque during a propeller phase of duration @xmath144 . the recurrence time of the sxt is @xmath145 . we assume , for simplicity , that the two phases can be approximated as bimodal , with the outburst phase occurring at a typical accretion rate @xmath146 and the quiescent phase having a typical lower accretion rate @xmath147 . these accretion rates define , in turn , the values of the magnetospheric radius in outburst , ] ) ] @xmath148 , and in quiescence , @xmath123 , used in equation ( [ eq : idealtorque ] ) to determine @xmath149 and @xmath150 . since the accretion rate from the peak of the outburst is known to decrease gradually with time , the value of @xmath146 used to determine @xmath148 should be interpreted as a mean over the outburst . by definition , the ns angular rotation speed at equilibrium @xmath131 satisfies @xmath151 . using equation ( [ eq : idealtorque ] ) , we find @xmath152}{\left [ 1 + \left ( \frac{r_{\rm mq}}{r_{\rm mo } } \right)^{2 } \tilde{g } \right ] } , \label{eq : eqomega}\ ] ] where @xmath153 is the ratio @xmath154 , i.e. the total mass reaching the magnetosphere during the adaf - propeller phase divided by the total mass accreted during outburst . this definition of @xmath153 is equivalent to the definition of @xmath48 given in equation ( [ eq : g ] ) if all the mass accumulated in the disk during quiescence is accreted onto the neutron star during outburst . ) remains valid because @xmath131 depends on the ratio @xmath155 for the remainder of the paper , we assume @xmath156 . note that the spin equilibrium period defined by equation ( [ eq : eqomega ] ) differs from the spin equilibrium period defined for _ steady _ accretion onto a ns and usually discussed in the literature ( e.g. henrichs 1983 ) . equation ( [ eq : eqomega ] ) allows us to determine the equilibrium spin period @xmath157 as a function of the parameter @xmath158 . this , in turn , allows us to estimate the quantities @xmath111 ( eq . [ [ eq : facc ] ] ) and @xmath68 ( eq . [ [ eq : fsurf ] ] ) which are relevant for determining the observational properties of quiescent ns sxts . in the following calculations , we assume for definiteness that the ns accretes at the eddington rate ( @xmath159 ) in outburst and that the mass transfer accretion rate in quiescence is @xmath160 ( cf fig . [ fig : accrate ] ) . however , our conclusions do not depend crucially on these assumptions . figures [ fig : eqfacc ] and [ fig : eqbyp ] show the variations of @xmath161 , @xmath162 and @xmath157 with @xmath48 . as @xmath46 decreases , the magnetosphere becomes more extended and the propeller effect becomes more efficient . the solid lines show the results for a ns with a magnetic field strength @xmath163 g , while the dashed and dotted lines show the cases @xmath164 and @xmath165 g , respectively . recent observations suggest that accreting neutron stars in low mass binaries have rather low magnetic fields , typically @xmath166 g ( e.g. white & zhang 1997 ) . figure [ fig : eqfacc ] can be related to fig . [ fig : lumbh]b , where we found that @xmath167 was needed in order to explain the very low observed luminosities of quiescent ns sxts . such a value of @xmath168 is predicted by the propeller model if @xmath169 . this value of @xmath48 is reasonably close to the value @xmath170 that we obtained for bh sxts . note that the dependence of @xmath168 on the magnetic field @xmath122 should cancel out through the ratios @xmath171 in equation ( [ eq : eqomega ] ) . for low magnetic field strengths , however , the inner edge of the disk in outburst is not given by half the value in equation ( [ eq : rmq ] ) but is fixed at the last stable orbit at 3 schwarzschild radii ( for instance , if @xmath172 and @xmath173 g ) . consequently , there is a residual dependence of @xmath168 on @xmath122 for small @xmath122 . figure [ fig : eqbyp]a shows that the fraction @xmath68 of the ns surface that emits in quiescence depends much more strongly on @xmath122 than @xmath111 does ( fig . [ fig : eqfacc ] ) , as expected from equation ( [ eq : fsurf ] ) . for values of @xmath169 and low magnetic fields ( @xmath166 g ) , the emitting surface is typically between @xmath174 and @xmath69 . observations of quiescent ns sxts suggest an emitting surface @xmath175 ( e.g. verbunt et al . 1994 ; campana et al . 1998 ; but see 7 ) . given that the surface could very well be underestimated by the standard techniques ( e.g. lewin , van paradijs & taam 1993 ; rajagopal & romani 1996 ; zavlin et al . 1996 ; rutledge et al . 1998 ) , values of @xmath176 appear in reasonable agreement with the observations . we note , however , that this result depends quite crucially on the assumed alignment of the ns spin and magnetic axes . this is a major uncertainty in the calculations . figure [ fig : eqbyp]b shows that @xmath157 is even more sensitive to the value of @xmath122 . for weak magnetic fields ( @xmath177 g ) , the values of the equilibrium spin period are in the range @xmath178 ms , which is in agreement with values recently discussed in the literature for accreting neutron stars in low mass binaries ( e.g. white & zhang 1997 ) . there are several sources of uncertainty in our simplified model for the propeller , which are discussed in 6.2 and 7 below . the proximity of the value @xmath170 required to explain the quiescent luminosities of bh sxts to the value @xmath169 required to explain the luminosities of ns sxts ( with a propeller ) suggests , however , that roughly the same geometry of accretion ( outer thin disk and inner adaf sharing the mass supplied by the secondary roughly equally ) can account for the properties of both classes of objects . the presence of a hard surface and a magnetic field associated with the spinning neutron star are two major ingredients of the model . the ns sxts aql x-1 ( @xmath179 g , @xmath180 ms , emitting area @xmath181 km@xmath182 ; white & zhang 1997 ; zhang et al . 1998 ; campana et al . 1998 ) and sax j1808.4 - 3658 ( @xmath183 g , @xmath184 ms ; wijnands & van der klis 1998 ; chakrabarty & morgan 1998 , hereafter cm98 ) are two interesting candidates that could allow further testing of the predictions of our adaf - propeller scenario . we note that cm98 inferred a mass transfer rate in sax j1808.4 - 3658 @xmath185 m@xmath186 yr@xmath7 , which is roughly a factor ten less than the value for a 2 hr orbital period ns sxt according to our fig . [ fig : accrate ] . these authors find that the estimate is consistent with mass transfer being driven by angular momentum losses due to gravitational radiation , if the secondary star has a mass @xmath187 . the values shown in fig . [ fig : accrate ] , however , were deduced assuming that the donor star is a main sequence star which satisfies @xmath188 ( 3 ) . this is apparently not the case in sax j1808.4 - 3658 , perhaps because of the fact that the companion star is strongly irradiated ( cm98 ) . the gas in an adaf could in principle escape to infinity via a wind because part of the flow may be unbound . narayan & yi ( 1994 , 1995a ) showed that the bernoulli constant , a measure of the energy of the gas at infinity , is positive in their self - similar solution , and narayan , kato & honma ( 1997 ) showed that this is also the case in a global solution . a simple way of taking this into account was proposed by blandford & begelman ( 1998 ; hereafter bb98 ) who derived self - similar adaf solutions that include mass loss via winds . our results in 4.1 suggest that reducing @xmath189 has only a small effect on the radiative efficiency of the accretion flow . quataert & narayan ( 1998 ) showed , however , that an efficient way of increasing the radiative efficiency of the adaf is to include winds and simultaneously increase the value of the parameter @xmath56 , the fraction of the viscously dissipated energy that directly heats the electrons . they found that models with winds and high @xmath56 are consistent with the current observations for the bh sxt v404 cyg in quiescence and the galactic center source sgr a*. note that theoretical estimates of the value of @xmath56 are highly uncertain . quataert ( 1998 ) and gruzinov ( 1998 ) argued that @xmath56 could be small if the magnetic fields in the adaf are fairly subthermal , but bisnovatyi - kogan & lovelace(1997 ; see also quataert & gruzinov 1998 ) argued that magnetic reconnection in the accretion flow could preferentially heat the electrons and lead to large values of @xmath56 ( @xmath190 ) . in 6.1 and 6.2 , we investigate the following two questions . can winds by themselves , i.e. without any propeller effect , explain the observed quiescent luminosities of ns sxts ? if not , is a combination of winds and a propeller consistent with the observations ? following bb98 , we assume that the accretion rate in the adaf scales in a self - similar way , i.e. @xmath191 where @xmath40 is the accretion rate scaled in eddington units ( @xmath192 ) , @xmath193 , and we assume that the wind is effective over the entire radial extent of the adaf out to @xmath194 , the transition radius . the efficiency of mass loss is not only determined by @xmath195 , but also by the radial extent of the adaf ( eq . [ [ eq : rmqwind ] ] ) . around a black hole , the adaf extends from the transition radius ( @xmath196 ) down to the event horizon , and so the accretion rate onto the black hole is @xmath197 . around a neutron star , however , the adaf has a minimum radius equal to the magnetospheric radius @xmath198 , so that the accretion rate at the magnetosphere is @xmath199 . if @xmath200 , as is the case for @xmath201 in quiescent ns sxts , then for a given value of @xmath195 , the presence of a magnetosphere around the ns significantly reduces the total mass loss relative to the bh case . the value of @xmath198 itself depends on the local accretion rate at the magnetosphere ( eq . [ [ eq : rmq ] ] ) . since the accretion rate at a given radius decreases with increasing @xmath195 , a larger @xmath195 implies both a larger @xmath198 and a smaller radial extent for the adaf . we solve for the magnetospheric radius by combining equations ( [ eq : rmq ] ) and ( [ eq : rmqwind ] ) and find @xmath202^{\frac{1}{7 + 2p } } , \label{eq : rmqwind2}\ ] ] where @xmath201 is expressed in schwarzschild units for a ns of 1.4 @xmath203 ( eq [ [ eq : rmqwind2 ] ] is the generalization of eq [ [ eq : rmq ] ] for non - zero @xmath195 ) . we have confirmed ( by combining eqs . [ [ eq : rmqwind ] ] and [ [ eq : rmqwind2 ] ] ) that despite the reduction of the radial extent of the adaf for larger values of @xmath195 , increasing @xmath195 leads to an effective reduction of @xmath204 for any reasonable values of @xmath195 , @xmath196 and @xmath205 . for a standard ns with low magnetic field strength ( @xmath206 , @xmath207 ) and a typical accretion rate at the outer boundary of the adaf @xmath208 ( 4.2 ; fig . [ fig : accrate ] ) , we find @xmath209 in the absence of a wind ( @xmath210 ) . for a wind with @xmath211 , @xmath212 ( `` intermediate '' wind model , @xmath213 ) , we find @xmath214 and @xmath215 , while for a wind with @xmath216 , @xmath212 ( `` strong '' wind model , @xmath217 ) , we find @xmath218 and @xmath219 . for these estimates , we have chosen @xmath220 by analogy with the values usually inferred in quiescent bh sxts ( narayan et al . 1996 ; menou et al . 1998 ) . were @xmath196 to be smaller , the overall mass loss in the wind for a given @xmath195 would be smaller . note that the adaf can not extend far beyond @xmath221 schwarzschild radii because , for a given @xmath222 , there is a maximum radius out to which the adaf can exist ( e.g. esin et al . 1997 ; menou et al . 1998 ) . following quataert & narayan ( 1998 ) , we computed two examples of adaf models assuming that the dominant effect of the wind on the structure and the emission properties of the adaf is to reduce the density in the accretion flow ; in addition , we assumed that the emission from the wind itself can be neglected relative to the emission from the accretion flow . figure [ fig : bhwind]a shows the quiescent x - ray luminosities ( 0.5 - 10 kev ) of bh sxts predicted by the adaf+wind model for which @xmath223 is equal to @xmath224 of the mass supplied by the secondary , @xmath211 , @xmath212 and @xmath225 . this model gives roughly the same x - ray luminosities as the previous no - wind adaf model ( 4.1 ) . the actual mass accreting onto the black hole is reduced here , but nevertheless the luminosities are the same because of the increase in the value of @xmath56 from @xmath226 to @xmath227 ( see quataert & narayan 1998 ) . figure [ fig : bhwind]b shows the quiescent x - ray luminosities ( 0.5 - 10 kev ) of ns sxts predicted by the same adaf+wind model , assuming that all the mass reaching the magnetosphere is accreted onto the ns surface ( i.e. no propeller effect ) . since the x - ray luminosity of the adaf is still much less than the luminosity coming from the ns surface , the main effect of including the wind is to decrease the luminosity of quiescent ns sxts in proportion to the reduction of @xmath40 at the inner edge of the adaf ( eq . [ [ eq : rmqwind ] ] ) . by comparing figures [ fig : lumbh]b and [ fig : bhwind]b , it is clear that the luminosity is smaller by roughly an order of magnitude , which is insufficient to explain the observations . thus , the intermediate wind scenario must also invoke a propeller effect , though with a somewhat lower efficiency than that described in 5 . given the uncertainties in our propeller model ( see 6.2 ) , we see no objection to such a model . only for the strongest wind models ( @xmath228 , @xmath212 ) does the luminosity difference between ns sxts and bh sxts predicted by the models reduce to the observed difference . for @xmath229 and @xmath230 ( @xmath206 , @xmath231 ) , ( [ eq : rmqwind ] ) and ( [ eq : rmqwind2 ] ) predict @xmath232 and @xmath233 , which is basically consistent with the small accretion rates required to explain the observed quiescent luminosities of ns sxts ( cf . [ fig : lumbh]b ) . but is this a reasonable model ? can the propeller effect be completely ineffective at preventing mass from reaching the ns surface ? if the neutron stars in ns sxts have spin periods of typically a few ms , as suggested by recent observations ( e.g. white & zhang 1997 ; zhang et al . 1998 ; campana et al . 1998 ; wijnands & van der klis 1998 ) , their corotation radii are a few schwarzschild radii . this is substantially smaller than the few tens of schwarzschild radii inferred for @xmath201 in 6.1 . the propeller effect is therefore expected to act quite strongly in these systems ( illarionov & sunyaev 1975 ) . note that the values of @xmath201 quoted in 6.1 , which correspond to a magnetic field strength @xmath122 of @xmath86 g , would be even larger for a larger @xmath122 ( eq . [ [ eq : rmqwind2 ] ] ) . can the propeller in such a system be so ineffective that @xmath234 ? we pointed out in 5.2 that there are large uncertainties in the values of @xmath111 , which come mainly from the strong dependence of @xmath111 on the ns rotation speed and the magnetospheric radius : @xmath235 ( the dependence on @xmath123 also affects the size of the ns emitting area , since @xmath236 ) . the latter source of uncertainty is perhaps more important , since while @xmath77 can be determined by direct observations , @xmath237 must be deduced from theoretical arguments . although the value of @xmath123 used in our work ( eq . [ [ eq : rmq ] ] ) is standard , it is based on simple dimensional arguments . given the strong dependence of @xmath111 on @xmath237 , an error , for instance , of only a factor of @xmath238 in @xmath237 translates to an error of more than an order of magnitude in @xmath111 . recent work by psaltis & chakrabarty ( 1998 ) shows that the standard scaling used for @xmath123 ( for disk - magnetosphere interactions ) is not consistent with observations in at least one weakly magnetized ns sxt ( sax j1808.4 - 3658 ) . consequently , we can not rule out the possibility that equation ( [ eq : rmq ] ) overestimates the value of @xmath123 for an adaf - magnetosphere interaction . the propeller effect would then be significantly less efficient than our previous results suggest ( 5 ) . if winds alone ( @xmath228 ) are to explain the reduced luminosity difference between bh and ns sxts in quiescence , we need a highly inefficient propeller . for example , from the no - wind ( @xmath210 ) model to the @xmath229 model described in 6.1 , @xmath201 increases by a factor @xmath239 which implies a @xmath240 times more efficient propeller according to equation ( [ eq : facc ] ) . since the @xmath210 model has a propeller with @xmath241 , the propeller would have to be @xmath242 times less efficient in the @xmath229 model than we would infer based on the theory described in 5 . in addition , in the presence of a wind , the value of @xmath111 is likely only to be smaller than the expression in equation ( [ eq : facc ] ) . this is because the wind flows out preferentially from the polar regions of the adaf ( narayan & yi 1995a ) , so there is no gas to flow down to the polar caps of the ns . even allowing for reasonable uncertainties in the scalings in equation ( [ eq : rmq ] ) , it does not seem likely that the propeller model could be that wrong . we therefore feel that the @xmath228 strong wind model is unlikely . chen et al . ( 1998 ; their fig . 3 ) argued that the segregation in luminosity swing between outburst and quiescence previously found by narayan et al . ( 1997b ) and garcia et al . ( 1998 ) disappears in a large sample . we point out in this paper that the chen et al . sample is larger only because they include numerous upper limits , many of which are based on observations with insufficient sensitivity to detect typical quiescent sxts . unless treated with care , for instance with a `` detection and bounds '' type method ( e.g. avni et al . 1980 ; schmitt 1985 ) , upper limits can obscure a true correlation , and indeed this seems to be the case with the analysis of chen et al . ( 1998 ) . our fig . [ fig : lumfirm ] shows a carefully selected and smaller sample , and we find that there is a difference of luminosity between quiescent bh and ns sxts . furthermore , we argue , based on theoretical mass transfer rates , that it is important to compare quiescent luminosities of ns sxts and bh sxts of similar orbital periods . we find that quiescent bh sxts with short orbital periods are about @xmath103 orders of magnitude fainter ( in eddington units ) than quiescent ns sxts with comparable orbital periods . the difference reduces to a little more than one order of magnitude when absolute luminosities are considered . the sample is , however , small . future observations with _ chandra x - ray observatory ( cxo ) _ and _ xmm _ will hopefully increase the sample size sufficiently to provide a more convincing case . hopefully , they will also allow separate comparisons of long and short period systems . this should help solidify the arguments advanced by narayan et al . ( 1997b ) and garcia et al . ( 1998 ) for event horizons in black holes . we find that the luminosities of quiescent ns sxts are lower than the values predicted by standard binary evolution models when combined with adaf models . following zhang et al . ( 1998 ) , we argue that the propeller effect offers a plausible explanation for the discrepancy . however , although our model for the propeller effect accounts for the luminosities and small emitting areas of quiescent ns sxts , it suffers from several sources of uncertainty that may affect the results significantly . the predictions of the model strongly depend on the spin frequency @xmath77 of the ns and therefore on the prescription chosen for the torque ( eq . [ [ eq : idealtorque ] ] ) . the magnitude of the torque is not well known , especially in the propeller regime ( e.g. davies , fabian & pringle 1979 ; henrichs 1983 ) . the results of our calculations are also sensitive to the precise values of the magnetospheric radii in quiescence and outburst and to the assumption that the fastness parameter @xmath243 . if @xmath244 , @xmath131 is reduced by a factor @xmath136 ( eq . [ [ eq : eqomega ] ] ) and more mass is accreted because @xmath111 is @xmath245 . note , however , that independently of equation [ [ eq : idealtorque ] ] and the value of @xmath136 , spin periods of a few milliseconds and low magnetic field strengths ( assuming that eq . [ [ eq : rmq ] ] is correct ) lead to predictions for luminosities and surface areas that are consistent with observations . even if the spin periods of neutron stars in ns sxts are known , the predictions of our propeller model are still somewhat uncertain because of the strong dependence of the fractions @xmath111 and @xmath68 on the magnetospheric radius in quiescence @xmath237 . if the standard expression ( eq . [ [ eq : rmq ] ] ) overestimates the value of @xmath123 , then the propeller effect could be less efficient than indicated by our results in 5 . in that case , winds from adafs could provide the additional mechanism required to explain the low quiescent luminosities of ns sxts . we have also neglected the likely misalignment between the neutron star spin axis and magnetic axis and have assumed a dipolar structure for the magnetic field . relaxing these assumptions could also modify our results , especially our estimates of the fractions @xmath111 and @xmath68 . the system exo 0748 - 676 appears unusually bright in our sample of quiescent ns sxts . one possible explanation is that magnetic braking acts more efficiently in this system because of its rather short @xmath0 , giving a larger mass transfer rate . if the mass transfer rate is higher , the propeller effect would also be less efficient , and would increase the mass falling on the ns even further . note that exo 0748 - 676 is known to be an unusual binary system , as shown by its unexplained variations of @xmath0 ( hertz , wood & cominsky 1997 ) . another possible explanation is that , for some reason ( youth ? ) , this particular system has not yet reached equilibrium and spins at a relatively slower rate than other ns sxts . this would result in a less efficient propeller and a more luminous system in quiescence . this last point is a general prediction of our propeller model : the propeller effect is less efficient if the ns in a sxt spins slowly ( @xmath246 a few ms ) . it is possible that several such ns sxts with slowly spinning nss exist in our galaxy . in these ns sxts , most of the mass accreted via the adaf would reach the ns in quiescence . the outbursts of these systems would therefore be of small amplitude , 2 - 3 orders of magnitude in x - rays rather than 5 - 6 orders of magnitude , because of their relatively high quiescent luminosities . in our model , the gas which is stopped by the propeller is neglected and its fate is left unspecified . this gas could affect both the dynamics and the emission properties of the accretion flow . for instance , it is unclear how the mass accreted via the adaf will find its way out of the quasi - spherical accretion flow after being propelled outward . in 6 , we considered adaf models with winds assuming that the efficiency @xmath195 of the wind is the same for bh and ns sxts . in principle , @xmath195 could be different for adafs around bhs and around nss ( because of the presence of the magnetosphere in nss for instance ) . having a different value of @xmath195 for bh and ns sxts would not , however , affect our conclusion that winds alone probably can not explain the observed quiescent luminosities of ns sxts . recently , brown et al . ( 1998 ) argued that nuclear reactions in the crust of neutron stars , triggered during outbursts in ns sxts , could efficiently heat up the ns cores in these systems . the energy deposited during an outburst would be reemitted during quiescence at a rate sufficient to explain the observed quiescent thermal emission . if so , the propeller effect must be even more efficient than our estimates in 5 indicate , so that accretion in quiescence does not contribute much to the quiescent emission of ns sxts . we note two criteria that bear on the relative importance of crustal heating and accretion to the quiescent luminosity of ns sxts . first , in the scenario of brown et al . , the entire surface of the ns contributes to the quiescent emission . small emitting areas have usually been inferred from the observations , by assuming that the quiescent emission is close to blackbody emission . however , rutledge et al . ( 1998 ) have shown that much larger areas are inferred if detailed hydrogen atmosphere models are used instead of idealized blackbody models . there is a large uncertainty in these results because of the poorly known hydrogen column density to the systems . nevertheless , the values of the emitting areas inferred by rutledge et al for several systems are consistent with the entire ns surface being responsible for the emission in quiescence . this success favors the crustal heating model . on the other hand , any observed rapid variability ( e.g. on timescales of a few days ) should be considered as strong evidence of the importance of accretion in quiescence , because the intrinsic thermal emission from the ns is not expected to vary on such small timescales ( brown et al . campana et al . ( 1997 ) reported that the quiescent x - ray luminosity of the prototypical ns sxt cen x-4 varied by a factor of @xmath71 over a period of @xmath247 days.this establishes that the accretion luminosity can be more than 3 times larger than any steady quiescent luminosity possibly present in this system . though very little x - ray variability data are presently available , several studies of the optical counterpart of cen x-4 in quiescence have shown a variability of order several tenths of a magnitude ( e.g. cowley et al . 1988 ; chevalier et al . 1989 ; mcclintock & remillard 1990 ) . since it is natural to relate this optical variability to the x - ray variability observed by campana et al . ( 1997 ) , both sets of observations appear consistent with accretion being the dominant mechanism for the quiescent luminosity of ns sxts . throughout this study , we assumed that accretion is responsible for the quiescent emission of both ns and bh sxts . the possibility that this emission could be due to two different processes in ns sxts certainly complicates the comparisons between the two classes of systems . we note , however , that even if the dominant contribution to the quiescent emission of ns sxts is crustal heating , the reason why quiescent ns sxts , as a class , are brighter than quiescent bh sxts remains the presence of a hard surface in ns sxts and of an event horizon in bh sxts . in this paper , we have reconsidered the luminosity difference between black hole and neutron star soft x - ray transients in quiescence , which has been used to argue for the presence of event horizons in black holes . we show that the current observational data suggest that quiescent bh sxts as a class are fainter than ns sxts . this result agrees with the previous work of narayan et al . ( 1997b ) and garcia et al . ( 1998 ) , but it disagrees with the conclusions of chen et al . ( 1998 ) who used an inappropriate sample of sxts for their comparison . we point out that , for a reliable comparison of the two classes of sxts , objects with similar orbital periods @xmath0 should be compared . otherwise , variations of the mass transfer rate @xmath37 with @xmath0 can mask the results . we find that the observed luminosities of quiescent bh sxts are consistent with the predictions of binary evolution models for @xmath248 if roughly a third of the mass supplied by the secondary is accreted by the black hole via an adaf . this estimate is for @xmath49 . the fraction goes down to @xmath249 and @xmath250 for @xmath251 and @xmath252 respectively . the observed luminosities of quiescent ns sxts suggest , on the other hand , that only a very small fraction of the mass transferred by the secondary reaches the neutron star surface . we explain the small fraction in ns sxts by invoking an efficient propeller . we have constructed a model for the propeller effect that accounts for the observed luminosities of quiescent ns sxts and their small emitting areas . in addition , the model appears to be consistent with the millisecond spin periods recently inferred from observations of accreting neutron stars in transient low mass binary systems ( aql x-1 , sax j1808.4 - 3658 ) . winds from adafs constitute an alternative explanation ( bb98 ) for the very small fraction of mass reaching the neutron star surface in quiescence . we argue that an adaf model with strong winds and no propeller can not explain the observed luminosities . however , an adaf model with a wind of low or intermediate strength and a somewhat less efficient propeller ( but still within the range of uncertainties of our propeller model ) is consistent with the observed quiescent luminosities of bh and ns sxts . in this case , the parameter @xmath56 , which measures the fraction of the viscous energy that goes directly into the electrons , has to be large @xmath253 ( quataert & narayan 1998 ) . after submitting our paper for publication , we became aware of an independent paper by asai et al . ( 1998 ) on the same subject . the conclusions of asai et al . are very similar to those presented here : ( 1 ) observational data show evidence for a difference of x - ray luminosity between quiescent bh and ns sxts , ( 2 ) only @xmath254 of the mass transferred by the secondary reaches the ns surface in quiescent ns sxts , and ( 3 ) this small accreted fraction could be due to the propeller effect in ns sxts . the quiescent fluxes and upper limits reported herein are generally in very good agreement with those listed in asai et al . ( 1998 ) , with the exception of 4u1543 - 47 . here we assume a distance of 8 kpc , while asai et al . assume 1 kpc . this accounts for most , but not all of the discrepancy between the asai et al . upper limit and that in orosz et al . we choose to use the latter limit , as it is more conservative . we are grateful to josh grindlay , mario livio , philip podsiadlowski , dimitrios psaltis and eliot quataert for useful discussions , and an anonymous referee for constructive comments . ae and rn also gratefully acknowledge their conversation with shuang - nan zhang who suggested the idea of a propeller effect in the context of a quasispherical accretion flow . this work was supported in part by nasa grant nag 5 - 2837 . km was supported by a smithsonian astrophysical observatory ( sao ) predoctoral fellowship and a french higher education ministry grant . ae was supported by a national science foundation graduate research fellowship and by nasa through axaf fellowship grant pf8 - 10002 awarded by the axaf science center , which is operated by the sao for nasa under contract nas8 - 39073 . abramowicz , m. , chen , x .- m . , granath , m. & lasota , j .- 1996 , apj , 471 , 762 . abramowicz , m. , chen , x .- m . , kato , s. , lasota , j .- p . , & regev , o. 1995 , apj lett . , 438 , l37 . asai , k. et al . 1998 , pasj , 50 , 611 . avni , y. , soltan , a. , tananbaum , h. & zamorani , g. 1980 , apj , 238 , 800 . bailyn , c.d . , jain , r.k . , coppi , p. & orosz , j.a . , 1998 . , apj , 499 , 367 . beekman , g. , et al . , 1997 , mnras , 290 , 303 . bisnovatyi - kogan , g.s . & lovelace , r.v.e . , 1997 , apj lett . , 486 , l43 . blandford , r.d . & begelman , m.c . , 1998 , mnras , submitted , astro - ph/9809083 ( bb98 ) . bondi , h. , 1952 , mnras , 112 , 195 . brown , e.f . , bildsten , l. & rutledge , r.e . , 1998 , apj lett . , 504 , l95 . campana , s. , mereghetti , s. , stella , l. & colpi , m. , 1997 , a&a , 324 , 941 . campana , s. et al . , 1998 , apj lett . , 499 , l65 . cannizzo , j.k . 1993 , in accretion disks in compact stellar systems , ed . wheeler ( singapore : world scientific ) , p. 6 . chen , w. , shrader , c.r . & livio , m. , 1997 , apj , 491 , 312 . chakrabarty , d. & morgan , e.h . , 1998 , nature , 394 , 346 ( cm98 ) . chen , w. et al . , 1998 , in accretion processes in astrophysics - some like it hot , eds . s. holt & t. kallman ( woodbury , ny : aip ) , p. 347 . chen , x.m . , abramowicz , m.a . & lasota , j .- , 1997 , apj , 476 , 61 chevalier , c. , ilovaisky , s.a . , van paradijs , j. , pedersen , h. & van der klis , m. , 1989 , a&a , 210 , 114 . cowley , a.p . et al . , 1988 , aj , 95 , 1231 . cui , w. , 1997 , apj lett . , 482 , l163 . cui , w. , barret , d. , zhang , s.n . , chen , w. , boirin , l. & swank , j. , 1998 , apj lett . , 502 , l49 . daumerie , p.r . , 1996 , ph . d. thesis , univers . illinois at urbana - champaign . davies , r. e. , fabian , a. c. & pringle , j. e. , 1979 , , 186 , 779 . davies , r. e. & pringle , j. e. , 1981 , , 196 , 209 . esin , a.a . , mcclintock , j.e . , & narayan , r. , 1997 , apj , 489 , 865 . esin , a.a . , narayan , r. , cui , w. , grove , j.e . & zhang , s .- n . , 1998 , apj , in press , astro - ph/9711167 . frank , j. , king , a. & raine , d. , 1992 , accretion power in astrophysics ( cambridge : cambridge university press ) . garcia , m.r . , mcclintock , j.e . , narayan , r. , & callanan , j. , 1998 , in the procceedings of the 13th naw on cvs , jackson hole , wy , eds . s. howell , e. kuulkers , and c. woodward , p. 506 , astro - ph/9708149 . ghosh , r. & lamb , f. k. , 1979 , , 234 , 296 . gruzinov , a. , 1998 , apj , 501 , 787 . hameury , j .- m . , lasota , j .- p . , mcclintock , j.e . , & narayan , r. , 1997 , apj , 489 , 234 . henrichs , h.f . 1983 , in accretion driven stellar x - ray sources , eds . lewin & e.p.j . van den heuvel ( cambridge : cambridge university press ) , p. 393 . hertz , p. , wood , k.s . & cominsky , l.r . , 1997 , apj , 486 , 1000 . ichimaru , s. , 1977 , apj , 214 , 840 . illarionov , a. f. & sunyaev , r. a. , 1975 , a&a , 39 , 185 . kalogera , v. , kolb , u. & king , a. , 1998 , apj , in press , astro - ph/9803288 . king , a. , 1988 , quat . soc . , 29 , 1 . king , a. , kolb , u. & burderi , l. , 1996 , apj lett . , 464 , l127 . kolb , u. , 1998 , mnras , 297 , 419 . kuulkers , e. , 1998 , new astron . , in press , astro - ph/9805031 . lasota , j .- 1996 , in compact stars in binaries ; iau symposium 165 , eds . j. van paradijs e.p.j . van den heuvel & e. kuulkers , ( dordrecht : kluwer ) , p. 43 . lasota , j .- p . & hameury , j .- m . , 1998 , in accretion processes in astrophysics - some like it hot , eds . s. holt & t. kallman ( woodbury , ny : aip ) , p. 351 . lasota , j .- p . , narayan , r. & yi , i. , 1996 , a&a , 314 , 813 . lewin , w.h.g . , van paradijs , j. & taam , r.e . , 1993 , space sci . , 62 , 223 . mcclintock , j.e . , 1998 , in accretion processes in astrophysics - some like it hot , eds . s. holt & t. kallman ( woodbury , ny : aip ) , p. 290 . mcclintock , j.e . & remillard , r.a . , 1990 , apj , 350 , 386 . menou , k. , narayan , r. & lasota , j .- p . , 1998 , apj , in press . meyer , f. & meyer - hofmeister , e. , 1994 , a&a , 288 , 175 . nakamura , k.e . , kusunose , m. , matsumoto , r. & kato , s. , 1997 , pasj , 49 , 503 . narayan , r. , barret , d. , & mcclintock , j.e . , 1997a , apj , 482 , 448 . narayan , r. , garcia , m.r . & mcclintock , j.e . , 1997b , , 478 , l79 . narayan , r. , kato , s. & honma , 1997 , apj , 476 , 49 . narayan , r. , mahadevan , r. , grindlay , j.e . , popham , r.g . & gammie , c.f . , 1998a , apj , 492 , 554 . narayan , r. , mahadevan , r. & quataert , e. , 1998b , in the theory of black hole accretion discs , eds . m. a. abramowicz , g. bjornsson , and j. e. pringle ( cambridge : cambridge university press ) , astro - ph/9803141 . narayan , r. , mcclintock , j.e . & yi , i. , 1996 , apj , 457 , 821 . narayan , r. & yi , i. , 1994 , apj lett . , 428 , l13 . narayan , r. & yi , i. , 1995a , apj , 444 , 231 . narayan , r. & yi , i. , 1995b , apj , 452 , 710 . orosz , j.a . & bailyn , c.d . , 1997 , apj , 477 , 876 . orosz , j.a . , jain , r.k . , bailyn , c.d . , mcclintock , j.e . & remillard , r.a . , 1998 , apj , 499 , 375 . phillips , s.n . , shahbaz , t. & podsiadlowski , p. 1999 , mnras , in press , astro - ph/9811474 . popham , r.g . & gammie , c.f . , 1998 , apj , in press , astro - ph/9802321 . pylyser , e. & savonije , g.j . , 1988 , a&a , 191 , 57 . quataert , e. , apj , 500 , 978 . quataert , e. & gruzinov , a. , 1998 , apj submitted , astro - ph/9803112 . quataert , e. & narayan , r. , 1998 , apj , in press . rajagopal , m. & romani , r.w . , 1996 , apj , 461 , 327 . ritter , h. & kolb , u. , 1998 , a&as , in press . rutledge , r.e . , bildsten , l. , brown , e.f . , pavlov , g.g . & zavlin , v.e . , 1998 , apj , in press . schmitt , j.h.m . , 1985 , apj , 293 , 178 . shapiro , s.l . & salpeter , e.e . , 1975 , apj , 198 , 671 . shaviv g. , wehrse r. , 1986 , a&a lett . , 159 , l5 . stella , l. , campana , s. , colpi , m. , mereghetti , s. & tavani , m. , 1994 , apj lett . , 423 , l47 . stella , l. , white , n.e . , rosner , r. , 1986 , apj , 308 , 669 . tanaka , y. & lewin w.h.g . , 1995 , in x - ray binaries , eds . lewin , j. van paradijs & e.p.j . van den heuvel , ( cambridge : cambridge university press ) , p. 126 . tanaka , y. & shibazaki , n. , 1996 , ara&a , 34 , 607 . turolla , r. , zampieri , l. , colpi , m. & treves , a. 1994 , apj lett . , 426 , l35 . van paradijs , j. , 1995 , in x - ray binaries , eds . lewin , j. van paradijs & e.p.j . van den heuvel , ( cambridge : cambridge university press ) , p. 536 . van paradijs , j. & mcclintock , j.e . , 1995 , in x - ray binaries , eds . lewin , j. van paradijs & e.p.j . van den heuvel , ( cambridge : cambridge university press ) , p. 58 . verbunt , f. , belloni , t. , johnston , h.m . , van der klis , m. & lewin , w.h.g . , 1994 , a&a , 285 , 903 . verbunt , f. & van den heuvel , e.p.j . , 1995 , in x - ray binaries , eds . lewin , j. van paradijs & e.p.j . van den heuvel , ( cambridge : cambridge university press ) , p. 457 . wang , y .- m . , 1987 , a&a , 183 , 257 . wang , y .- m . , 1995 , apj lett . , 449 , l153 . wang , y .- m . & robertson , j. a. , 1985 , a&a , 151 , 361 . warner , b. , 1995 , cataclysmic variable stars , ( cambridge : cambridge university press ) . webbink , r.f . , rappaport , s. & savonije , g.j . , 1983 , apj , 270 , 678 . white , n.e . , nagase , f. & parmar , a.n . , 1995 , in x - ray binaries , eds . lewin , j. van paradijs & e.p.j . van den heuvel , ( cambridge : cambridge university press ) , p. 1 . white , n.e . & zhang , w. , 1997 , apj lett . , 490 , l87 . wijnands , r. & van der klis , m. , 1998 , nature , 394 , 344 . zavlin , v.e . , pavlov , g.g . & shibanov , y.a . , 1996 , a&a , 331 , 821 . zhang , s.n . , yu , w. & zhang , w. , 1998 , apj lett . , 494 , l71 . lcccc + system & @xmath0 ( hr ) & d ( kpc ) & @xmath255~({\rm erg~s^{-1}})$ ] & @xmath256 + + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) + + + @xmath257 exo 0748 - 676 & @xmath258 & @xmath259 & @xmath260 & @xmath261 + @xmath257 4u 2129 + 47 & @xmath262 & @xmath263 & @xmath264&@xmath261 + @xmath257 1456 - 32 ( cen x-4 ) & @xmath265 & @xmath266 & @xmath267&@xmath261 + @xmath257 1908 + 005 ( aql x-1 ) & @xmath268 & @xmath269 & @xmath270 & @xmath261 + @xmath257 h1608 - 52 & @xmath271 or @xmath272 & @xmath273 & @xmath274&@xmath261 + + + @xmath275 gro j0422 + 32 ( xn per 92 ) & @xmath276 & @xmath277 & @xmath278&@xmath279 + @xmath275 a0620 - 00 ( xn mon 75 ) & @xmath280 & @xmath281 & @xmath282 & @xmath283 + @xmath275 gs2000 + 25 ( xn vul 88 ) & @xmath284 & @xmath285 & @xmath286 & @xmath287 + @xmath275 gs1124 - 683 ( xn mus 91 ) & @xmath288 & @xmath289 & @xmath290&@xmath291 + @xmath275 h1705 - 250 ( xn oph 77 ) & @xmath292 & @xmath293 & @xmath294 & @xmath295 + @xmath275 4u 1543 - 47 & @xmath296 & @xmath297 & @xmath298 & @xmath299 + @xmath275 j1655 - 40 ( xn sco 94 ) & @xmath300 & @xmath301 & @xmath267 & @xmath302 + @xmath275 gs2023 + 338 ( v404 cyg ) & @xmath303 & @xmath304 & @xmath305 & @xmath306 + + note . ( 1 ) @xmath257 indicates a ns primary and @xmath275 a bh primary . ( 2 ) orbital periods from mcclintock ( 1998 ) , except where indicated . distances to the systems . ( 4 ) luminosities in quiescence in the 0.5 - 10 kev band ( corrected for the revised distances ) from narayan et al . ( 1997b ) and garcia et al . ( 1998 ) , except where indicated . quiescent luminosities for 4u1543 - 47 and gro j1655 - 40 are based on 20.4 ks and 100 ks asca observations ( respectively ) , and the quiescent luminosity for gro j0422 + 32 is based on a 19 ks rosat observation . ( 5 ) primary masses . ( a ) van paradijs ( 1995 ) . ( b ) simbad cds catalog . ( c ) ritter & kolb ( 1998 ) . ( d ) chen et al . ( e ) garcia et al . ( g ) esin et al . ( h ) narayan et al . ( 1997b ) . ( i ) esin et al . ( j ) hameury et al . ( k ) beekman et al . ( l ) chen et al . ( m ) this is an arbitrary choice in the range @xmath307 given by orosz et al . ( 1998 ; see also bailyn et al . ( n ) orosz et al . ( @xmath308 ) for simplicity , all ns masses are assumed to be @xmath309 .

spiral disks form as a result of complex interactions of star formation , accretion , and gas . these processes are recorded in a galaxy s stellar component . the stellar metallicity , in particular , encodes information on a galaxy s overall chemical enrichment resulting from its history of star formation , gas accretion , and gas outflows . spiral galaxies are commonly thought to form `` inside out '' with the inner regions forming earliest ( e.g. * ? ? ? * ) . this effect has been observed by measuring a decreasing age with increasing radius in some nearby galaxies ( e.g. * ? ? ? * ; * ? ? ? a metallicity gradient is a natural consequence of inside out growth . observationally , present - day metallicity gradients can be measured using hii regions or the atmospheres of young stars . studies of the present - day metallicity gradients in the milky way and other nearby spirals do show lower metallicities at larger radii ( e.g. * ? ? ? * ; * ? ? ? * ) . to measure metallicity gradients not at the present - day , but in the past , one must analyze stellar metallicities as a function of age . the simultaneous measurement of age and metallicity is non - trivial , but can be done using either the observations of resolved stars or using integrated spectra . recently , the califa survey @xcite used integral field spectroscopic data to study both the present - day and the mean stellar metallicity gradients in a large sample of galaxies @xcite . the mean mass - weighted stellar metallicity gradient found among the califa galaxies is relatively shallow , @xmath3 dex / r@xmath4 , where r@xmath4 is the disk effective radius @xcite . the galaxies in this califa subsample are very similar to m31 in hubble type and luminosity . similar results based on photometric modeling were also found by @xcite for spiral galaxies in the virgo cluster . in both studies , there is no clear dependence on hubble type for spiral galaxies . in the milky way , the best gradients utilize measurements of 1000s of individual stars from large surveys ( e.g. sdss ) and found a gradient of @xmath5 - 0.06 dex / kpc @xcite . in addition to the smooth gradients in metallicity , distinct metallicity structures are visible within galaxies . metallicity substructure in the halos of galaxies have been linked to galactic merger events ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? within the disks of galaxies , structures with distinct metallicity signatures can arise due to bars , bulges and other dynamical structures @xcite . the ability to resolve individual stars all at the same distance makes andromeda a unique object with which to study massive galaxy disks . distance is the main source of uncertainty within the milky way and large - scale morphology observations are largely restricted to the solar neighborhood . in contrast , stars can be resolved with ease at a wide range of radii in andromeda . the panchromatic hubble andromeda treasury ( phat ; * ? ? ? * ) survey recently observed 117 million stars over a third of andromeda , providing a unique database for studying andromeda s stellar populations @xcite . in this paper , we present a detailed analysis of the metallicities of andromeda s red giant branch ( rgb ) stars using phat photometry . previous work on the rgb metallicities in andromeda have focused on the halo with limited information in the disk of the galaxy @xcite . for this study , we assume the following parameters : the center of m31 is at ra=10.68458and dec=1.2692@xcite ; m31 position angle and inclination angle are respectively 38and 74@xcite ; the distance modulus is 24.45@xmath60.05 mag or 776@xmath618 kpc ( mean distance , see @xcite for discussion ) . due to the small uncertainty in m31 s distance , and the primary dependence of the rgb metallicity on color and not luminosity , the distance uncertainty should have little effect on the analysis we present in this paper . the foreground extinction is @xmath7 0.17 mag @xcite ; m31 s effective radius is 8.9@xmath60.8 kpc @xcite . the paper is structured as follows : in section [ sec : data ] , we explain our data sources , including the rgb star photometry and dust extinction measurements . in section [ sec : principles - of - metallicities ] , we discuss our method for estimating metallicities from isochrone models and the uncertainties caused by photometric crowding and dust extinction . we present and discuss our results on andromeda s metallicity gradient and a metal rich structure associated with the bar in section [ sec : results ] . the phat survey observed approximately 117 million stars with the hubble space telescope ( hst ) in ultraviolet ( f275w and f336w ) , optical ( f475w and f814w ) , and infrared ( f110w and f160w ) filters using the wfc3/uvis , acs / wfc and wfc3/ir hst camera , respectively . the camera pointing angles divided the survey into 23 bricks each consisting of 18 overlapping ir - fields defined from the wfc3/ir camera footprint ( for details see @xcite ) . 3.4@xmath8 stars . the red clump is visible at @xmath9 and @xmath10 24.3 mag . the black dashed line indicates the rgb selection box used for this paper ( see section [ sec : completeness - limit ] ) . color indicates the metallicity . the three color lines plot padova parsec1.2s isochrones at @xmath11=$]-2.0 , 0.0 , and 0.6 dex ( in violet , orange , and yellow , respectively ) at a fiducial age of 4 gyr . the black star marks an example from our data sample at a deprojected radius of 6.1 kpc whose interpolated metallicity estimate is 0.033 dex ( see section [ sec : metallicities ] ) . the black arrow indicates 1 mag of dust extinction in @xmath12 band . extinction causes stars metallicities to be overestimated as discussed at length in section [ sec : dust - impact ] . ] two versions of the photometry are publicly available . first , there is camera - by - camera photometry derived separately on the pairs of filters in each camera producing a uv , optical , and infrared catalog that are then merged together to form a single six - filter catalogcamera - by - camera photometry can be downloaded at http://archive.stsci.edu/prepds/phat/datalist.html ] @xcite . second , the `` six - filter photometry '' is derived across all filters simultaneouslysix filter photometry can be downloaded at http://archive.stsci.edu/missions/hlsp/phat/ ] @xcite . in both versions , each star is parameterized by a position and magnitude , magnitude error , signal - to - noise ratio , crowding , and sharpness for each filter . sharpness parameterizes the shape of each source relative to the point spread function ; and , crowding parameterizes the brightness of nearby sources . in this paper , we use the optical camera - by - camera photometry to derive rgb metallicity estimates . the primary reason for using the camera - by - camera photometry over the six filter photometry is the availability of extensive artificial star tests ( also publicly available from mast archive ) . these artificial star tests allow us to understand the photometric completeness and bias ( see section [ sec : artificial - star - tests ] ) . the optical color of rgb stars also is more sensitive to the metallicity relative to the ir photometry , while the uv photometry suffers from significant incompleteness for even the brightest rgb stars . the smaller effect of dust extinction on the ir data makes it logical to include this data in our analysis , however , artificial stars are only available for the two - camera photometry in the ir , and this data is significantly shallower and crowding limited than the optical data ( see figure 11 of * ? ? ? * ) . the six filter photometry is significantly deeper in the ir due to the ability to use the optical data to determine the positions of stars , but it is not computationally feasible to run artificial star tests over the full phat footprint for the six band photometry @xcite . we create an optical only ( @xmath13 and @xmath14 ) photometric catalog with uniform coverage from the individual acs / wfc field images . this involves two primary steps : defining non - overlapping spatial regions , and filling in the chip gaps of each individual field . the boundary of each field is first defined using a grid of non - overlapping regions that roughly follow the individual wfc3/ir field footprints . we additionally crop bricks 9 and 12 , which overlap with neighboring bricks , because of orientation differences described in section 3.5 of @xcite . the phat tiling scheme allows us to use neighboring acs / wfc fields to fill in the chip gap of each acs / wfc field . thus , our final catalog has uniform spatial coverage over the full phat survey region but does not make use of multiple overlapping observations . this same catalog creation process is also used to deal with chip gaps in the the six - filter photometry as described in @xcite . we apply quality cuts on our photometry . the cuts use signal - to - noise , sharpness , and crowding in the f475w and f814w filters for each star . we use the same parameterization which defines `` gst '' ( good star ) cuts from @xcite by applying all of the following equations : @xmath15 + the variables correspond to the names given in the data files . for example , @xmath16 , @xmath17 , and @xmath18 respectively correspond to the signal - to - noise , sharpness , and crowding parameters for the @xmath13 filter . we correct the magnitudes in @xmath13 and @xmath14 for foreground extinction using @xmath70.17 mag @xcite . this extinction in @xmath19 is converted to @xmath13 and @xmath14 extinctions by multiplying with @xmath20 equal to 1.19119 and 0.60593 , respectively taken from http://stev.oapd.inaf.it/cgi-bin/cmd ] . the corrected photometry is indicated by a subscript zero ( _ e.g. _ @xmath21 ) . in addition to the `` gst '' cuts , we require that @xmath22 to remove stars with higher photometric bias and lower completeness . the foreground reddening correction is applied before this brightness cut . after applying the spatial and photometric cuts our sample contains @xmath57.0@xmath23 stars . in figure [ fig : data - selection - cmd ] , we present the optical color - magnitude diagram after quality cuts and foreground extinction corrections . this color - magnitude diagram ( cmd ) shows the stellar density prior to the magnitude cut in gray scale contours . the red - clump and bump of early - agb stars are both visible as peaks in the density below our selection region . the dashed area shows the selection region in which we determine rgb metallicities . in this paper , we estimate metallicities for rgb stars from their positions in the cmd . isochrones at a range of metallicities are shown in figure [ fig : data - selection - cmd ] , showing that stars of different metallicities are clearly separated in color ; we use this information to infer metallicities after considering a number of subtleties that can affect this measurement ( section [ sec : principles - of - metallicities ] ) . dust extinction has a significant impact on the observed photometry and thus our metallicity estimates . we discuss this in more detail in section [ sec : dust - impact ] . in this section , we describe the source of our dust extinction measurements . we use extinction measurements from @xcite ( hereafter ) . uses a novel method to directly measure the dust extinction from the ir photometry ( filters @xmath24 and @xmath25 ) of phat rgb stars . in the ir cmd , unreddened rgb stars form a tight locus that is insensitive to age and metallicity . dust extinction causes rgb stars to redden off this locus and form a second , broad rgb . models the unreddened and reddened stars using three parameters : median extinction ( ) , spread of extinctions ( ) , and the fraction of reddened stars ( ) . this latter parameter takes into account the geometry of the dust relative to the rgb stars ; 1- stars are assumed to be in front of the dust , and thus unaffected by it . maps of each parameter are created with 2 pixels across the phat survey region . note that the central region of the phat survey ( brick 1 ) is not included in the dust maps due to extreme crowding . in figure [ fig : dust - maps ] , we present maps of the and used in this paper . we rebinned and down - sampled the data into spatial bins used in our final results ( 0.01squares ) . the width is approximately 0.3 everywhere and not shown because it has little bearing on interpreting the effect of extinction . in contrast , the plays a significant role . increases in cause the reddening to be more pronounced at the same mean extinction values . for instance , in the case where @xmath26 and high , the median metallicity remains unchanged as stars in front of the thin dust layer are unaffected while those behind are reddened beyond selection bounds . one feature visible in the map of is the increase towards the lower ( northwest ) edge of the survey . in this region a higher fraction of stars appear behind the dust because of the disk inclination . increasing means more stars will be extincted . the final dust parameter , , is roughly 0.3 everywhere and is not shown here because its effect on our results is negligible . ] in this section , we explain our method for deriving metallicity estimates and the sources of systematic and random uncertainty in these estimates . this section is organized as follows . in section [ sec : isochrone - models - and - age ] , we discuss our choice of isochrone models , which we then use to interpolate the metallicity of individual stars based on their optical color and magnitude in section [ sec : metallicities ] . we then spatially bin the individual star metallicities to create a map of median metallicity . because of the large number of stars , the random sampling uncertainties are negligible . we address systematic uncertainties in the median metallicity maps due to photometric bias and completeness ( [ sec : artificial - star - tests ] ) , dust extinction ( [ sec : dust - impact ] ) , and the uncertainties in the absolute metallicity scale due to isochrone choices and age ( [ sec : absolute - metallicities ] ) . we correct the median metallicity map for the systematic effects of photometric bias , completeness and dust extinction and present results from this map in section [ sec : results ] . the intrinsic color of an rgb star depends primarily on its age and metallicity . thus , given the photometry of a star and an assumed age , we can use isochrones to estimate a star s metallicity from its color and magnitude . when estimating individual star metallicities ( section [ sec : metallicities ] ) , we break the age - metallicity degeneracy using assumptions about the age of m31 s disk motivated by literature estimates . in this section , we discuss our choice of isochrone models and the possible impact of age on our metallicity estimates ; this impact is further explored throughout the paper . for isochrone models , we use padova parsec1.2s isochrones @xcite . at ages @xmath271 gyr , these isochrone models are available at metallicities ( @xmath28 ) between 0.0001 and 0.06 using @xmath29=0.0152 ) . we use the obc bolometric corrections described in @xcite to convert the isochrones to photometric brightnesses in our optical bands . we test dependency on the isochrone models by comparing models from parsec to basti @xcite and find the later has a typical offset of approximately + 0.3 dex . this offset drops to @xmath50 for metal poor ( @xmath30 dex ) and metal rich ( @xmath31 dex ) stars . because the majority of our stars have the typical offset , the resulting median metallicity map is only affected by a change in the absolute scale of the metallicity . as we discuss in section [ sec : absolute - metallicities ] , absolute metallicity changes do not affect our results . the isochrone models for the bright rgb stars have a well known degeneracy between age and metallicity @xcite . to break this degeneracy we adopt a single age motivated by the literature . we assume that this age is either uniform throughout the disk , or that it varies linearly with radius . current constraints on the age of m31 s disk are relatively minimal . spectroscopic measurements of the age by @xcite show a mean age of @xmath512 gyr throughout the bulge , but this drops to @xmath58 gyr at their maximum radius of @xmath52 kpc ( 500 ) where the disk light starts to contribute significantly @xcite . only weak constraints on the ancient star formation history ( sfh ) can be obtained from the phat data in the inner disk , but measurements by williams et al . ( _ in prep _ ) suggest a mean sfh age of 10 gyr throughout the outer disk . we can also get some insight into the age and age gradient present in m31 s disk by examining similar galaxies . as noted in the introduction , the subsample of galaxies in the califa survey analyzed by @xcite includes a large number of galaxies similar to m31 . taking the 25 galaxies within one magnitude of m31 in @xmath32 band luminosity ( califa magnitudes taken from * and use @xmath33 for m31 ) and with hubble type sab to sbc , we find that the minimum mass - weighted age in the disks of these galaxies is 3 gyr , with typical values of @xmath57 gyr . these measurements were made over similar physical scales ( 0.5 - 2 @xmath34 ) as the region we probe with the phat data . these old mass - weighted ages of spiral disks are also consistent with previous work by @xcite . while these previous studies suggest that the disk of m31 is old , the average age of stars observed along the upper rgb is not equivalent to the mass - weighted average age of a stellar system due to varying rgb lifetimes and the difference in masses of rgb stars with age . the average age of stars along the upper rgb is normally _ younger _ than the mass - weighted average stellar age . by simulating a cmd using the star formation history derived by williams et al . ( _ in prep _ ) , we calculate the mean age of rgb stars to be @xmath54 gyr throughout the disk . for the primary median metallicity maps we assume a flat fiducial age of 4 gyr for our rgb stars . in section [ sec : fiducial - age - effect ] , we show the metallicity gradient is not dependent on the exact fiducial age because the primary effect is a scaling of the absolute metallicity which leaves the gradient unchanged . in section [ sec : radial - age - effect ] , we show that the median metallicity gradient can be changed if we assume a significant age gradient in the disk . we estimate individual star metallicities by interpolating isochrones at the distance of andromeda ( @xmath35 ) in the @xmath36 vs @xmath21 color - magnitude diagram . for interpolation , we use the linearndinterpolator function within the python package scipy . this function finds the closest three isochrone points that create a triangle containing a star s color and magnitude using delaunay triangulation . then it calculates the star s metallicity by linearly interpolating the metallicity at those three nearest isochrone points . the isochrones used in this procedure are sampled at @xmath37 over the full range of metallicities considered , so the interpolation in metallicity is very small . in figure [ fig : data - selection - cmd ] , we overplot the star metallicity on top of the stellar density contours in color and magnitude . in figure [ fig : mdf - full ] , we present the metallicity distribution function ( mdf ) of our rgb stars using an fiducial age of 4 gyr . we additionally show the mdf generated from stars in low extinction regions . from this figure , we see that the dust has a small effect on the overall mdf of our sample . mag ; see section [ sec : dust - sources ] ) . the median metallicity for both histograms is @xmath5 - 0.11 dex ( shown as a vertical line ) . we note this mdf is not corrected for the systematic biases in the metallicity discussed in section [ sec : artificial - star - tests ] and [ sec : dust - impact ] , however , after correction , the mdfs look qualitatively very similar . ] our primary goal in this paper is to examine the spatial variation in the metallicity distribution across andromeda s disk . we present a spatial map of the median metallicity of the rgb in the top panel of figure [ fig : uncorrected - med - metal ] . we derive the median metallicity in 0.01(135 pc , projected ) square bins . each bin contains between 100 and 19000 stars . the upper limit becomes 3300 after we exclude the inner regions ( @xmath38 kpc ) because of completeness ( discussed in section [ sec : completeness - limit ] ) . the typical bin contains @xmath5560 stars . in addition to the true variations in median rgb metallicity , this map shows features due to photometric bias , completeness , and dust . we attempt to correct these effects in subsequent sections . ) . _ bottom _ : model for the expected changes in the median metallicity due to dust extinction . we interpret the increases and decreases in median metallicity in section [ sec : dust - impact ] . for low extinction regions the dust causes an increase of @xmath50.03 dex in the absolute metallicity and no significant spatial variations in the median metallicity . in our results ( figure [ fig : corrected - median - metallicity - map ] ) , we subtract the two models from the raw median metallicity map to create a corrected version ( presented in figure [ fig : corrected - median - metallicity - map ] ) . note , all color bars have same range so relative differences can be assessed by eye . ] photometric bias and completeness affect a star s observed photometry and thus the inferred metallicity estimate . artificial stars allow us to evaluate and ultimately correct for these two systematics . @xcite describes the optical camera artificial star tests in detail . to summarize , @xmath39 artificial stars were distributed uniformly over each phat acs / wcs field of view , resulting in a total of @xmath54@xmath8 stars over the full survey . the input @xmath40 colors and @xmath14 magnitudes for the artificial stars were split into two halves : the first half sample the full color and magnitude range of the data , while the other half sample the local observed color - magnitude diagram with an extrapolation to the faintest observed magnitudes . each artificial star was inserted into and recovered from the data images independently using the dolphot package ( an updated version of the software presented in * ? ? ? the same photometric cuts described in section [ sec : data ] were applied to the output photometry . we use these artificial stars to measure the photometric bias in our measurement in section [ sec : bias ] , and to model the completeness in section [ sec : completeness ] . then in section [ sec : phot - corrections ] , we describe how we apply corrections for bias and completeness to our photometric metallicity estimates . unresolved and bright stars in crowded regions cause a bias in the measured photometry that becomes increasingly large in more crowded regions . this bias generally causes a star to appear brighter and with a color closer to the mean color of the overall galaxy due to blending with fainter undetected stars . in figure [ fig : cmd - bias - completeness ] , we show the average color and magnitude bias in bins across the cmd for a region of m31 at radii between 4 - 5 kpc where the bias is significant . the bias of each individual artificial star is measured from the recovered magnitude minus the input magnitude . because the star s recovered cmd position is changed from its original position , this photometric bias results in a bias in our metallicity estimates as well . examining figure [ fig : cmd - bias - completeness ] shows stars within our rgb region are typically detected to be brighter and bluer than their intrinsic cmd position . our metallicity estimates for these stars are thus biased lower than the true metallicity , with the effect being more severe for high metallicity stars which are closer to the @xmath13 detection limit due to their intrinsically redder colors . in section [ sec : phot - corrections ] , we use the measured bias to partially correct for this effect . for computational efficiency , we create a binned data set of the average bias . we bin using 4 dimensions : two spatial dimensions with step size of 0.02(270 pc ) in both projected major and minor axis of m31 ; and , two color - magnitude dimensions with step size of 0.45 mag in both dimensions ( sizes for spatial and cmd bins seen in figures [ fig : min - completeness ] and [ fig : cmd - bias - completeness ] respectively ) . the result is a median of 3.3@xmath41 stars input for every spatial pixel and 36 input stars for every 4d bin . in each bin with @xmath42 recovered artificial stars , we calculate the mean photometric bias for each filter from all the artificial stars . stars in 4d bins with @xmath43 artificial stars receive no correction value . 2 - 4 kpc , and therefore highlights the region most affected by completeness and bias that we consider in our analysis ( section [ sec : completeness - limit ] ) . our rgb selection box is indicated as a dotted line . each bin is 0.45 mag on each side . the completeness is the number of detected artificial stars divided by the number of those inserted in each cmd bin ( section [ sec : completeness ] ) . the arrows represent the magnitude and direction of bias for each bin ( section [ sec : bias ] ) . the bias generally shifts stars brighter and closer to the median color of the galaxy . we correct our median metallicity determinations for completeness and bias as discussed in section [ sec : phot - corrections ] . ] the completeness is the number of detected stars divided by the total number of stars present at a given magnitude and color . our completeness correlates with magnitude ; at fainter magnitudes we detect a smaller fraction of stars and are thus less complete . because we require stars to be detected in both filters , each filter contributes to the completeness ; for blue stars the f814w magnitude determines the completeness ( thus creating a horizontal completeness limit on our cmd ) , while for redder stars , the f475w magnitude limits the completeness , creating a diagonal completeness limit on our cmd . these completeness limits are visible in figure [ fig : cmd - bias - completeness ] where the darkest pixels represent @xmath44 completeness . the faint f475w magnitudes of metal - rich giants causes these stars to be most affected by completeness , and thus we detect a smaller fraction of the most metal - rich stars . we measure the completeness from the artificial star tests as follows . we define an artificial star as detected if it passes the photometry cuts defined in section [ sec : data ] and has an absolute magnitude change less than 0.75 mag . the completeness is then simply the fraction of artificial stars detected in a particular spatial and cmd volume . we measure the completeness using the same 4d bins described previously in section [ sec : bias ] . to correct for completeness , we divide the total number of artificial stars recovered and inserted at a given cmd position using the _ input _ magnitudes ( i.e. after the bias correction , section [ sec : phot - corrections ] ) . this is somewhat different from the standard practice where completeness is calculated using the observed magnitudes to determine the detections and the input magnitudes to determine the total number of stars inserted at a given cmd position . due to the small number of stars in each individual 4d bin , the errors on the completeness in an individual bin are large . to reduce these uncertainties , we model the completeness at a given color as a function of @xmath21 magnitude using a sigmoid function ( e.g. @xcite ) : @xmath45 our modeling process is a loop with five steps : first , we select artificial stars within a 2-dimensional spatial bin . second , we select artificial stars within a particular color bin and in the neighboring color bin(s ) . using neighboring bins increases the number of stars used for each fit . also as a result the fit parameters become covariant . this is beneficial because the covariance effectively smooths along the color dimension and makes the measurement more robust . third , we bin the color - selected stars into independent @xmath21 magnitude bins . the completeness for each 2-dimensional color - magnitude bin is the fraction of detected to input artificial stars . an error in this measurement is also estimated . fourth , we fit the completeness vs @xmath21 to equation [ eq : completeness - model ] using @xmath46 minimization and determine @xmath47 and @xmath48 for that color bin . fifth , we use the model to assign a fitted completeness to each 4d bin . at the end of this process , we have robust measurements of our completeness binned in 4d . we show a spatial map of our completeness for rgb stars in figure [ fig : min - completeness ] ; this figure is discussed further in section [ sec : completeness - limit ] . using the results from [ sec : bias ] and [ sec : completeness ] we correct our final median metallicity measurements for the photometric bias and completeness systematics . to apply this correction , we first correct the observed photometry for photometric bias . then , when aggregating into the median metallicity , we correct for completeness by weighting each star by the inverse of the recovered completeness from section [ sec : completeness ] . in this section , we further describe these corrections and the creation of a photometrically corrected median metallicity map . to correct the observed star photometry for the photometric bias in each filter , we first bin the observed stars using the same 4d bins described in section [ sec : bias ] . then , we select each spatial bin and consider only the mean bias values in the associated cmd bins . we interpolate the bias for each star in both filters using the cmd bin centers and mean bias values ( as we did for the metallicity in section [ sec : metallicities ] ) . finally , the observed star photometry is corrected for bias by subtracting interpolated bias from the observed photometry . we use the bias corrected photometry to estimate a photometric completeness for each star . the process is similar to interpolating the bias values for each star , only we now use the 4d recovered completeness data from section [ sec : completeness ] . we apply a minimum completeness limit of 0.05 so when weighting by completeness no star can dominate the calculations . this limit affects @xmath49 of our stars . for every spatial bin in our final maps , we calculate a bias and completeness corrected median metallicity . we first re - derive metallicities for the stars using the bias corrected photometry . to calculate the median metallicity , we weigh each star by the inverse of its completeness and then take the 50th percentile of the stars mdf . the middle panel of figure [ fig : uncorrected - med - metal ] shows the effect of bias and completeness on the median metallicity map ; specifically , it shows the median metallicity determined from the uncorrected photometry minus the median metallicity after we apply our bias and completeness correction . the bias and completeness correction is largest near the center where the density of stars is high and is negligible in the outer parts of the galaxy . we discuss the limits to these corrections in section [ sec : completeness - limit ] . this correction model is not exactly accurate due to uncertainties in the artificial star tests caused by stochasticity and assumptions in the prior distribution of the stars . nevertheless , the model provides a reasonable and robust correction to this known systematic . where the stellar density becomes high , photometric bias and completeness effects become large . when the completeness drops to near 0% , our bias and completeness corrections are no longer valid as they require the intrinsic distribution of stars to be at least partially detected . this limits how close to andromeda s galactic center we can measure metallicities and limits our ability to detect the most metal - rich stars . we use a minimum completeness metric to quantify exactly how close to the center our corrections are applicable . at each spatial position , we measure the completeness along the 90th percentile density contour of our rgb stars ( shown as a black contour line on figure [ fig : data - selection - cmd ] ) limited our rgb @xmath14 magnitude limit . we use the minimum completeness along the contour to create the map shown in figure [ fig : min - completeness ] . for almost all spatial positions the minimum completeness is measured at the most red rgb tip . this means the completeness along our @xmath14 magnitude limit is greater than this minimum completeness metric at practically all spatial positions . from the minimum completeness map ( figure [ fig : min - completeness ] ) , we define the limiting central region at the 50% contour which is @xmath54 kpc . at this limit , the bias and completeness correction is @xmath50.1 dex . we exclude data within from our final analysis . beyond the limit , we find the same correction has a @xmath50 effect on our metallicity graident ( section [ sec : metallicity - gradient ] ) and no bearing on our observed metal rich bar structure ( section [ sec : metal - rich - bar ] ) . we conclude this limiting region to be a reasonable choice . in all subsequent figures , we include the results at all radii but shade the excluded central region . ) . this metric is calculated by assigning a completeness to all the stars then taking the minimum from the 90th percentile in cmd density of our data . the solid lines indicate 50% , 75% , and 90% minimum completeness . in general , the stars most affected by completeness are the reddest , highest metallicity portion of the rgb . at the center of andromeda these stars are completely undetectable . as the sample becomes more incomplete it becomes impossible to correct properly for the bias and completeness ; we therefore only analyze the area with @xmath51% completeness ( see section [ sec : completeness - limit ] ) . ] dust extinction causes stars to appear dimmer and redder . for rgb stars , this causes us to overestimate the metallicity of reddened stars ( see arrow in figure [ fig : data - selection - cmd ] ) . we use information on the dust distribution ( see section [ sec : dust - sources ] ) to both identify high extinction regions and to model the effect of dust extinction on the median metallicity . to identify regions where the extinction is likely to cause biased metallicity estimates , we multiply the mean extinction by the fraction of reddened stars ( ) from within each spatial bin . we find @xmath52 mag is a good limit for identifying regions where extinction significantly biases our results . we measure the significance of dust extinction on our median metallicity measurements by forward modeling the effect of dust extinction on an unreddened fiducial cmd . to forward model our data , we first need a realistic input cmd . we take this input cmd from unreddened regions in the survey . for the unreddened fiducial cmd , we use stars in the low extinction region ( @xmath53 mag ) at @xmath5 14 kpc along the major axis ( visible in figure [ fig : dust - maps ] ) . our final result should be insensitive to low level extinction present in our fiducial cmd because we measure the relative change in median metallicity of the sample . our results are very similar using other low extinction regions within the survey ( e.g. along the ne edge ) . the primary weakness of our approach is that we do not account for the stellar population differences ( i.e. young stars ) we expect between the dust - free and dusty regions of the galaxy ; trying to account for this change would require full star formation history modeling and is beyond the scope of this paper . we expect that any young contaminants not included in our fiducial cmd would be brighter bluer stars reddened into our cmd region , and thus would likely reduce the median metallicity at moderate extinctions . we apply reddening to this cmd to measure the resulting change in median metallicity . the extinction for each individual star in the cmd is drawn from the extinction probability function from using their maps of the best fit , and dust parameters ( see figure [ fig : dust - maps ] ) . the stars are binned into 0.01spatial bins which correspond to the bin size used in our median metallicity maps . in each bin , we take the associated dust parameters and determine a random extinction for each star . then we apply that extinction to create a new reddened data set and re - evaluate the metallicity as described in section [ sec : metallicities ] . finally , we measure the change in median metallicity between new reddened data and the original unreddened sample . in figure [ fig : dmedian - vs - dust ] , we present the change in median metallicity as a function of for each spatial bin colored by . these two parameters dominate the predicted change in median metallicity estimates ; changing has a @xmath54 dex change in the median metallicity estimates . as previously noted , for a particular the change in median metallicity increases with increasing . lines of constant ( and constant = 0.3 ) are also plotted on figure [ fig : dmedian - vs - dust ] . for these constant lines , the change in median metallicity first increases then decreases with increasing . the increases are easily understood as stars becoming reddened and appearing more metal rich . the decrease is caused by stars becoming so reddened they are moved beyond our rgb selection box and the most metal rich isochrone ; the remaining unreddened stars ( which are in front of the dust distribution ) yield a very similar median metallicity to the input cmd . the individual points in figure [ fig : dmedian - vs - dust ] correspond to individual spatial bins . we also present these data as a map of the change in median metallicity ( lower panel of figure [ fig : uncorrected - med - metal ] ) . this map shows sharp changes due to dust especially along the upper ( northeastern ) edge of the survey and in the 10 kpc ring . our final results avoid these regions by imposing a cut to remove these high extinction regions ( @xmath52 mag ) . in the low extinction regions , the dust still has a low level effect . we apply a dust correction by subtracting the change in median metallicity due to dust . the changes in metallicity are small ( typically @xmath55 dex ) and more importantly distributed randomly over the survey area so the result is only a systematic on the absolute metallicity scale and not on spatial variations . systematics on the absolute metallicity scale do not heavily impact our results . [ m / h ] ) for each spatial bin in the data . the simulation uses a fiducial cmd taken from a low - dust region and applies reddening to individual stars using the dust parameters ( , , and ) from the dust maps for every spatial bin . the solid lines show changes in median metallicity at constant values of 0.1 , 0.2 , 0.3 , and 0.4 . along one of these solid lines ( and assuming = 0.3 ) . the dust increases the median metallicity until @xmath560.5 because the reddened stars receive an overestimated metallicity . the median metallicity decreases for high extinction when the reddened stars are moved beyond the edge of our rgb box and no longer affect the measurement . the points below the dashed line are low extinction regions with @xmath57 . these are the regions used in our metallicity gradient measurement . their @xmath58[m / h ] is @xmath59 dex , with typical values of 0.03 dex . details on the dust simulations are given in section [ sec : dust - impact ] ; a map of @xmath58[m / h ] is shown in the bottom panel of figure [ fig : uncorrected - med - metal ] . ] the absolute metallicities we determine have large uncertainties which we discuss in this section . _ age : _ older populations have redder rgbs than younger populations of the same metallicity . therefore , using older fiducial ages results in lower metallicity estimates . this effect is quite significant , with metallicity estimates varying by @xmath50.4 dex when changing the age from 3 to 13 gyr ( we discuss this further in section [ sec : fiducial - age - effect ] ) . overall , this is the dominant uncertainty in the absolute metallicity . because we are primarily interested in the metallicity gradient , an age gradient can significantly affect our result ; this is discussed further in section [ sec : radial - age - effect ] . _ isochrones : _ as noted in section [ sec : isochrone - models - and - age ] , use of the basti isochrones results in a median metallicity that is @xmath50.3 dex higher than the parsec 1.2s isochrones we adopt . _ dust : _ from our simulations in the previous section , we find the low level dust extinction present throughout the survey results in a slight overestimate of the intrinsic metallicity ; however this bias is only @xmath50.03 dex . _ photometric bias and completeness : _ near the center , the bias and completeness correction raise our median metallicity by more than 0.1 dex . unlike the effects above , this is primarily a bias in the relative metallicity between the center and outskirts , as the bias and completeness have no affect on the median metallicity in the outer disk regions . because of the large uncertainties in the absolute metallicities , we choose to focus on relative measurements for our results . particularly , the metallicity gradient of andromeda and metallicity substructure of the disk . after applying corrections for known systematics , we present the median metallicity map and results in the next section . in this section , we analyze the radial and spatial variation of the median rgb metallicity . our results are based on the spatial map of corrected median metallicities presented in figure [ fig : corrected - median - metallicity - map ] . our results focus on two interesting features in the corrected median metallicity map . the first is an overall gradient towards lower metallicities at larger radii ( section [ sec : metallicity - gradient ] ) . the second feature is a region of high metallicity at radii of 3 - 7 kpc lying just above ( southeast of ) the major axis , which we associate with an elongated bar ( section [ sec : metal - rich - bar ] ) . ) was divided into annular bins with @xmath60 ; within these bins the mean of the map is shown as the solid line , while the shaded region shows the standard deviation of the pixels in the bin . high extinction regions ( @xmath61 ) were excluded . the red line shows the gradient including corrections for dust extinction and photometric bias and completeness , while in cyan the metallicities have only been corrected for dust extinction . the bias and completeness correction changes the gradient by @xmath51@xmath2 , while the dust extinction correction changes the gradient by less than a 0.001@xmath2 . the black line and shaded region shows the best linear fit at deprojected radii @xmath62 kpc ( @xmath63 ; @xcite ) to the corrected metallicities , the best - fit slope is -0.020@xmath60.004 dex / kpc . the black shaded region at @xmath64 kpc where our rgb data becomes incomplete at high metallicities . ] we present the first measurement of the rgb metallicity gradient in the m31 disk from 4 to 20 kpc . we binned the median metallicity ( shown in figure [ fig : corrected - median - metallicity - map ] ) into elliptical apertures with a size of 1 kpc along the major axis , and assuming @xmath65 @xcite to account for the disk s inclination . we exclude the high extinction regions with @xmath61 to limit the effect of dust on our measurement ; this limit is shown as the white contour in the bottom panel of figure [ fig : corrected - median - metallicity - map ] . after excluding these regions , we take the average and standard deviation of the median metallicity within each elliptical aperture to create the radial metallicity profile shown in figure [ fig : feh - vs - radius ] . we fit a gradient to the measured metallicity profile at @xmath62 kpc ( @xmath63 ; @xcite ) . this radial cut excludes the regions we distrust due to low completeness ( shown as the shaded region in figures [ fig : corrected - median - metallicity - map ] and [ fig : feh - vs - radius ] ) . we fit the mean and standard deviation of the metallicity in each radial bin using a monte carlo simulation with a linear fit . for each iteration , we calculate the best fit slope and intercept using least squares optimization . we repeat this and obtain a distribution of fitted slopes and intercepts ; this procedure also allows us to examine the covariance of the slope and intercept . we show the best fit parameters and 1@xmath2 contours in figure [ fig : fiducial - age - params ] for various fiducial ages including our primary result with 4 gyr . we measure andromeda s metallicity gradient to be @xmath0 dex / kpc . this gradient uses a 4 gyr fiducial age ( see section [ sec : isochrone - models - and - age ] , fig [ fig : fiducial - age - params ] ) and our corrected metallicities . the error on the gradient is calculated using the monte carlo technique described above ; bootstrapping errors are significantly smaller than this quoted value . the photometric bias , completeness and dust corrections have little impact on producing the observed gradient : the dust corrections change the gradient by @xmath66 , and the photometric bias and completeness corrections steepen the relation somewhat , but still agree at about the 1@xmath2 level . we also measure a consistent metallicity gradient ( within 0.25@xmath2 ) when selecting stars just along the major axis which indicates the assumed inclination angle has little impact on the gradient . the intercept ( central metallicity ) of this fit is [ m / h]@xmath67 ; though caution should be taken in interpreting the absolute metallicity as explained in section [ sec : absolute - metallicities ] . to facilitate comparisons with the literature , we also present our gradient with different units . in terms of andromeda s half light radius ( 8.9@xmath60.8 kpc ; @xcite ) , the gradient is @xmath68 dex/@xmath69 . using the radius of the isophote at a bolometric b - band of 25 mag ( @xmath70 kpc ; @xcite ) , the gradient is @xmath71 dex / r25 . as discussed in section [ sec : absolute - metallicities ] , the assumed age influences the estimated metallicity of our stars . we evaluate the sensitivity of our metallicity gradient results to changes in the fiducial age by repeating our analysis using isochrone models with ages of 2 , 4 , 6 , 8 , 10 , and 12 gyr . for each age , we calculate a distribution of metallicity gradients and intercepts ; in figure [ fig : fiducial - age - params ] , we present the best fit and 1@xmath2 contour of the distribution . the metallicity gradient across all fiducial ages is consistent within 0.25@xmath2 . this is because the rgb of isochrones at different ages have similar shapes in the cmd , with the primary difference being that older ages have a lower metallicity at the same rgb color . this results in a significant change in the normalization of the metallicity gradient and the fitted intercept in metallicity , but little change in the slope . we find a change in the intercept of roughly @xmath72 dex / gyr . confidence after performing a monte carlo simulation to evaluate the slope and intercept . our primary result presented in section [ sec : metallicity - gradient ] uses a fiducial age of 4 gyr shown here as a diamond . the age gradient for all fiducial ages are within 0.25@xmath2 . changing the fiducial age does have an effect on the absolute metallicity which we discuss in section [ sec : fiducial - age - effect ] . this figure also demonstrates a covariance between the fitted slope and intercept of the metallicity gradient . ] a gradient in the age has a significant impact on our metallicity gradient . we evaluate this impact using the following experiment ; we fix the age of stars at the center of andromeda to 4 gyr . then , we calculate the age for each star using its spatial position for age gradients of -0.35 , -0.25 , -0.1 , 0.0 , and 0.1 gyr / kpc . we calculate the metallicities similar to in section [ sec : metallicities ] with age in addition to color and magnitude . in this application , the delaunay triangulation finds the nearest four color , magnitude , and age isochrone points which contain a particular star s parameters and performs the linear interpolation of the metallicity between the points . using these new metallicities , we recalculate the median metallicity map and derive the distribution of best fit metallicity gradients and intercepts as described in section [ sec : metallicity - gradient ] . in figure [ fig : feh - age - vs - radius ] , we present the results of this age gradient experiment . as mentioned , as the age gradient becomes negative , the metallicity gradient flattens , while a positive age gradient steepens the gradient . for an age gradient of -0.35 gyr / kpc the metallicity gradient is consistent with being flat . this figure can be combined with figure [ fig : fiducial - age - params ] to determine the inferred age gradient and intercept for a range of central ages and age gradients . for instance , if the mean age of stars in our rgb box is 8 gyr in the center changing by -0.1 gyr / kpc , then our measured metallicity gradient would be @xmath5 - 0.018 dex / kpc with a central metallicity of @xmath5 - 0.04 dex . we expect any age gradient in m31 to be small . this is based on results from the phat data ( williams et al . _ in prep _ ) . also , @xcite find flat mass weighted age - gradients ( @xmath73dex/@xmath34 ) for a majority of galaxies in the califa sample , with the small age gradients being both positive and negative . this translates to an age gradient less than 0.1 gyr / kpc using values for m31 and our fiducial age . however , resolved star formation histories of m33 and ngc 300 ( lower mass , later - type disk galaxies than m31 ) have shown steeper negative age gradients , with younger ages at larger radii @xcite . if such gradients exist in m31 , the younger stars at larger radii will cause a flatter metallicity gradient than using a single fiducial age at all radii . however , resolved star formation histories of m33 and ngc 300 ( lower mass , later - type disk galaxies than m31 ) have shown steeper negative age gradients , with younger ages at larger radii @xcite . if such gradients exist in m31 , the younger stars at larger radii will cause a flatter metallicity gradient than using a single fiducial age at all radii . if any new information becomes available , figure [ fig : feh - age - vs - radius ] can be used to update our metallicity gradient result based on the determined age gradient . , but with a variation of stellar age with radius . the age gradients shown have slopes of -0.35 , -0.25 , -0.1 , 0.0 , and 0.1 gyr / kpc with a central age of 10 gyr . the metallicity gradient flattens as the age gradient becomes more negative with it being consistently 0 dex / kpc when the age gradient is -0.35 gyr / kpc . based on observations of external galaxies similar to andromeda , we expect the age gradient to be @xmath74 gyr / kpc ( see section [ sec : comparison - metallicity - gradient ] ) . ] in this section , we compare our measured metallicity gradient to other metallicity gradient measurements in m31 and to metallicity gradients measured in other galaxies . we find some evidence that the metallicity gradient may become flatter at large radii , but does not appear to vary with age . the metallicity gradient of the young population of andromeda has been measured by @xcite using hii region abundances over the same radial region covered by the phat data . they find a gradient of @xmath75 dex / kpc . this gradient is consistent with our rgb metallicity gradient , suggesting that the age gradient has been stable in m31 for some time . at larger radii , @xcite used abundance measurements of planetary nebula to measure a gradient with deprojected radii of 18 to 43 kpc . these planetary nebulae have typical ages of 1 - 2 gyr , thus probing slightly younger stars than the rgb stars probed here . they find a gradient of @xmath76 dex / kpc , significantly flatter than our gradient . this may be evidence that the metallicity gradient flattens at larger radii . another , more indirect approach to measuring metallicities is the stellar population modeling of integrated photometry of m31 from the androids survey @xcite . their preliminary results seem to show a somewhat shallower gradient of @xmath5 - 0.01 dex / kpc out to @xmath530 kpc , with a possible steepening at even larger radii . these comparisons provide conflicting support for the expectations due to radial migration ( e.g. * ? ? ? on one hand , the shallower metallicity gradient in old populations at large radii could be explained if an increasing fraction of stars at large radii come from smaller radii . however , radial migration also predicts that any metallicity gradient should get shallower over time , thus the similarity of the metallicity gradients in the young and old populations in the inner disk suggests that either the metallicity gradient was steeper in the past , or significant migration has not occurred in the inner disk . higher quality stellar population modeling can be done with spectra , but published results include only the central bulge region . spectral index measurements of m31 s bulge by @xcite show the metallicity at radii between 100 pc and 2 kpc is flat and approximately solar . at smaller radii , the metallicity rises sharply , with [ m / h]@xmath50.5 in the center . at the outer most radii , these measurements include a substantial disk component . our central metallicity extrapolated inwards to 2 kpc is roughly consistent with this measurement though the absolute metallicity scale in our measurement is sensitive to many systematics ( figure [ fig : fiducial - age - params ] and section [ sec : absolute - metallicities ] ) . the recent califa survey observed @xmath77 face - on spiral galaxies using wide - field integrated field units ( ifu ) . andromeda s size , morphology , and central velocity dispersion are typical among this sample . from the califa sample , @xcite selected 62 face on galaxies and measured their stellar metallicities and ages . the mean mass - weighted metallicity gradient of these galaxies is @xmath78 dex/@xmath69 or @xmath79 dex / kpc when scaled using m31 s effective radius ( @xmath80 kpc ; @xcite ) . our metallicity gradient is slightly steeper than this mean metallicity gradient , but is well within the scatter observed of their galaxies . ) are highlighted in black . our m31 metallicity gradient ( assuming no age gradient ) is shown as the blue star . dashed lines give the median of the full sample ( gray ) , and the m31-like sample ( black ) along both axes . ] the metallicity gradient in the milky way ( mw ) disk appears to be significantly steeper than the gradient in m31 . @xcite measure the metallicity gradient as a function of height ( z ) from the disk for segue survey stars . for stars near the @xmath81 they measure @xmath82 dex / kpc which drops to @xmath83 dex / kpc for @xmath84 kpc ( see figure 8 of @xcite for metallicity gradient vs z height ) . given that the disk mass is dominated by stars below this height , the mw gradient does appear to be steeper than the m31 gradient . a similarly steep mid - plane gradient is found in other survey data as well including apogee data @xcite , rave data @xcite and the gaia - eso survey @xcite . the shallower gradient in m31 relative to the milky way is also discussed by @xcite . we observe a region of high median metallicity likely associated with the end of the bar . in figure [ fig : corrected - median - metallicity - map ] , this non - axisymmetric region lies just above the major axis between radii of @xmath53 and 7 kpc . this region contains the highest metallicities found anywhere in our maps . examination of contours of rgb density in the middle panel of figure [ fig : corrected - median - metallicity - map ] shows a similar axial asymmetry . this asymmetry coincides with m31 s bar as measured by @xcite at 8above the major axis . @xcite also noted an overdensity in agb star counts near the end of the bar . in addition to the presence of a stellar overdensity , a peculiarly high velocity dispersion ( @xmath5150 km / s ) is observed at the outer edge of this region @xcite . this high dispersion could be explained if this represents the end of a misaligned bar . this high metallicity region is not caused by dust but represents a real change in the stellar population . as previously noted in section [ sec : dust - impact ] , dust causes the median metallicity to appear more metal rich for moderate extinctions of @xmath85 mag . we observe the high metallicity region prior to the dust correction ( figure [ fig : uncorrected - med - metal ] ) and it becomes even clearer after the correction . even more noteworthy , the high metallicity is seen even in the very low extinction regions ( dust contours shown in figure [ fig : corrected - median - metallicity - map ] ) . this higher median metallicity region may be expected if either : ( 1 ) gas collects at the bar end resulting in higher enrichment there @xcite ; or , ( 2 ) the bar scatters stars from the inner part of the galaxy which have higher metallicity . an additional possibility is there could be a shift in the rgb age in the bar region ; however , to get a high metallicity region , the average age would need to drop in the bar relative to its surroundings . regardless of the mechanism for metal enrichment , this metallicity adds further evidence for a very long bar in m31 . we present an analysis of the spatial variations in the metallicity of rgb stars in andromeda s disk using the phat survey . this study is the first large study of rgb metallicities within the andromeda disk and covers a radial region between 4 and 20 kpc . metallicities are estimated for 7.0@xmath23 rgb stars using their optical photometry and isochrone models . we create a map of median metallicity and correct this map for the effects of dust extinction and photometric bias and completeness . our final results are based on the median metallicity map after correcting for these systematics ( figure [ fig : corrected - median - metallicity - map ] ) . from the corrected median metallicity map we conclude : + 1 . over the inner disk phat survey ( @xmath54 - 20 kpc ) , we measure a clear gradient in the median metallicity . the measured gradient is [ m / h]=@xmath0 dex / kpc assuming a constant fiducial age of 4 gyr ( section [ sec : metallicity - gradient ] ) . this gradient has a less than 1 sigma sensitivity to systematic effects of dust extinction , photometric bias and completeness , and fiducial age changes . the derived metallicity gradient is sensitive to any age gradient that may be present in andromeda , with negative age gradients resulting in shallower metallicity gradients . a steep age gradient of -0.35 gyr / kpc would result in no metallicity gradient ; however , spectroscopic observations of other similar galaxies typically show much shallower age gradients @xcite . we measure an enhanced metallicity region offset from the major axis . we show this region is coincident with an overdensity of rgb stars and a region of enhanced dispersion @xcite . in section [ sec : metal - rich - bar ] , we discuss how this likely is due to andromeda s bar @xcite . we thank the anonymous referee for suggestions that improved the paper . this work was supported by the space telescope science institute through go-12055 . this research made use of codes in the scipy stack , as well as astropy , a community - developed core python package for astronomy ( astropy collaboration , 2013 ) . , g. , de vaucouleurs , a. , corwin , jr . , h. g. , et al . 1991 , third reference catalogue of bright galaxies . volume i : explanations and references . volume ii : data for galaxies between 0@xmath86 and 12@xmath86 . volume iii : data for galaxies between 12@xmath86 and 24@xmath86 .

it is now well accepted that the giant planets of the solar system did not have the current orbits when the gas was removed from the protoplanetary disk . the first work showing that the planets had to migrate while interacting with the remaining planetesimal disk was @xcite . this work demonstrated that , in the planetesimal scattering process , on average saturn , uranus , and neptune gain angular momentum and thus migrate outwards , while jupiter loses angular momentum and moves inwards . since then , the process of planetesimal - driven migration was considered to be a fundamental one in shaping the current structure of the solar system . @xcite showed that planetesimal - driven migration can explain the origin of the orbit of pluto and of the resonant populations of kuiper belt objects . however , it appeared soon that planetesimal - driven migration alone is a process too smooth to explain several aspects of the kuiper belt s orbital distribution , e.g. the orbital excitation of the non - resonant objects . @xcite was the first paper showing that a dynamical instability of the giant planets could lead to a global orbital excitation of the kuiper belt , while delivering the giant planets on final orbits roughly similar to the current ones . building on these results , the so - called _ nice model _ @xcite combined the processes of planetesimal - driven migration and dynamical instability into a new global scenario . it suggested that the planets jupiter , saturn , uranus , and neptune formed in a compact configuration with jupiter around 5.45 au , saturn slightly interior to the 2:1 mean motion resonance ( mmr ) with jupiter ( @xmath3 @xmath4 8.65 au ) , and the initial semimajor axes of the ice giants ( uranus and neptune ) in the ranges 1113 au and 13.517 au . the planets were surrounded by a disk of leftover planetesimals , accounting for a total of about 35 @xmath5 , from @xmath016 to 30 au ( where @xmath6 is the mass of the earth ) . in this model the planets interacted with the planetesimal disk and , by planetesimal - driven migration , their mutual separations slowly increased . when jupiter and saturn crossed their mutual 2:1 mmr the full system of the giant planets became unstable and spread further through mutual close encounters . in particular , uranus and neptune were scattered onto orbits with relatively large eccentricities and semimajor axes , which crossed the original trans - neptunian disk . then , following a phase of dynamical friction caused by their interaction with the disk , the orbits of uranus and neptune damped in eccentricity and inclination . a residual planetesimal - driven migration placed the planets on orbits fairly similar to the current ones @xcite . the nice model had great success in explaining many features of our solar system , as : the capture of jupiter s trojans @xcite , the capture of the irregular satellites of the giant planets @xcite , the absence of regular moons beyond oberon s orbit at uranus @xcite , to cite a few . moreover , it was shown that , provided the inner edge of the trans - neptunian disk was at an appropriate distance from the original orbit of neptune , the giant planet instability could have occurred after hundreds of my of evolution , potentially explaining the origin of the so - called late heavy bombardment ( lhb ) of the terrestrial planets @xcite . however , the original version of the nice model presented a major problem ; the initial configuration of the planets was chosen in an ad hoc manner . @xcite performed hydrodynamical simulations of the evolution of the giant planets when they were still embedded in a disk of gas and found that the planets should have acquired a fully resonant configuration , with jupiter and saturn preferentially locked in their mutual 3:2 mmr with jupiter at @xmath05.4 au , saturn and uranus ( ice i ) in the 3:2 mmr , and uranus ( ice i ) and neptune ( ice ii ) in the 4:3 mmr ( although other resonant configurations are also possible ) . the authors also concluded that such a multi - resonant system is compatible with a subsequent evolution similar to that described in @xcite , although the instability in this case happens when a pair of planets leave the original resonant configuration , rather than when jupiter and saturn cross their 2:1 mmr . @xcite investigated whether , under some conditions , this initial multi - resonant configuration could still lead to a _ late _ giant planet instability , in order to explain the putative lhb as in the original @xcite paper . they found that late instabilities are possible , provided that the inner edge of the disk is several au s beyond the outermost planet and about 1,000 pluto - sized planetesimals are embedded in the disk , so to produce an important dynamical self - stirring of the planetesimals population . during the dynamical instability of the giant planets , several dynamical paths are possible , even when restricting the evolutions to those leading to final planetary orbits similar to the current ones . the actual dynamical evolution of the giant planets at the instability time can be constrained by investigating the consequences on the other components of the solar system : the terrestrial planets and the small body populations . the terrestrial planets and the asteroid belt give similar constraints , as investigated in @xcite and @xcite . if the orbits of jupiter and saturn migrate away from each other smoothly and slowly , as in planetesimal - driven migration , the @xmath7 secular frequency ( dominant in the precession of jupiter s longitude of perihelion ) decreases slowly in value and becomes very similar for some time to the @xmath8 secular frequency , dominant in the precession of mercury s perihelion . as a consequence , the orbit of mercury is destabilized @xcite . similarly , the @xmath9 secular frequency , dominant in the precession of saturn s perihelion , as it slows down becomes equal to the frequencies characterizing the precession of the asteroids at low inclination in the inner asteroid belt , so that this region becomes severely depleted @xcite . both papers concluded that the ratio of the orbital periods of jupiter and saturn @xmath10 had to evolve almost discontinuously from @xmath11 to @xmath12 ( the current value is @xmath13 ) . this happens if jupiter and saturn encounter another planet ( uranus or neptune or a third planet of comparable mass ) . the sequence of encounters causes a divergent jump in the semimajor axes of the orbits of jupiter and saturn , and therefore a jump in the @xmath14 ratio . this scenario is commonly known as the _ jumping - jupiter _ model . notice that if the giant planet instability occurred early , the terrestrial planets were possibly not yet fully formed and therefore the constraint on the evolution of @xmath7 relative to @xmath8 does not apply @xcite . however , the constraint set by the asteroid belt is still valid , supporting the need for a _ jumping - jupiter _ evolution regardless of the time of the instability @xcite . given that in a _ jumping - jupiter _ evolution the planet that encounters jupiter is often ejected onto a hyperbolic orbit , @xcite and @xcite suggested that the solar system had originally a third ice giant planet with a mass comparable to uranus or neptune . this scenario increases the probability that a _ jumping - jupiter _ evolution ends with four planets near their current orbits , once the putative fifth planet is ejected after its encounter with jupiter . then , @xcite considered many different initial multi - resonant configurations for jupiter , saturn , uranus , neptune , and the rogue fifth planet , and determined for each of them the probability that the @xmath14 ratio had a jump across the 2.12.3 range and the final planets reached orbits compatible with those of our current solar system . the successful simulations of @xcite have then been used to investigate again the capture of the trojans of jupiter @xcite , the capture of irregular satellites at jupiter @xcite , the excitation of the inclination of the moon iapetus at saturn @xcite , the evolution of the asteroid main belt and of its collisional families @xcite . they have also been shown to be compatible with the survival of the galileans satellites @xcite , despite of the encounters of jupiter with the ejected planet . however , when investigating the dynamical sculpting of the kuiper belt , @xcite and @xcite realized that , of all possible evolutions of neptune consistent with the five - planet _ jumping - jupiter _ model , only some are consistent with the current orbital structure of the trans - neptunian population . in order to get the correct inclinations of the hot population of the kuiper belt objects ( kbo ) , @xcite found that neptune should have migrated more that 5 au from its original resonance by planetesimal - driven migration before that the planets became unstable . this migration should have occurred on a timescale @xmath15 @xmath16 10 my , with the planet on a quasi - circular planar orbit ( @xmath17 0.1 and @xmath18 ) to avoid excessive orbital excitation of the cold classical belt , in the @xmath04245 au region . the planetary instability should have happened when neptune was already at 28 au @xcite in order to explain the so - called _ kernel _ of the cold kuiper - belt ( a clump of objects around 44 au ; @xcite ) . in this case the _ kernel _ would have formed from objects transported outwards in the 2:1 mmr with neptune and released from the resonance when neptune s orbit had a jump of about 0.5 au in semimajor axis , due to a close encounter with another planet . finally , in order to reproduce the observed ratio between the resonant and non - resonant kuiper belt populations , @xcite concluded that the planetesimal - driven migration of neptune should have been characterized by several small amplitude jumps , as those caused by a population of 1,0004,000 pluto - size objects in the disk . the brief review reported above , sets the context that motivated the study reported in this paper . it is clear that the work by @xcite needs to be revisited to test which initial resonant configurations are the most compatible with the new constraints on the dynamical evolution of neptune . this is done in the first part of the paper ( section [ early ] ) . in the second part of the paper ( section [ late ] ) we investigate whether the initial configurations selected in the first part of the work can lead to a _ late _ planetary instability , while still showing an evolution of neptune consistent with kuiper belt constraints . the possibility of a _ late _ instability is far from trivial . in fact , in order to exhibit a long - range planetesimal - driven migration before the planet instability , neptune needs to be embedded in the planetesimal disk . but in this case , neptune would start migrating away from the other planets from the very beginning of the simulation ( corresponding to the gas - removal time ) . this is likely to lead to an instability after just a few 10s of my . if instead the planetesimal disk is relatively far from neptune , the planet can be extracted from its initial resonance at a _ late _ time as in @xcite , but the instability is likely to happen as soon as the initial resonant configuration is broken , i.e. before that neptune can migrate far away . for this problem we will introduce a new concept : that of dust - driven migration , i.e. , migration due to the gravitational interaction of a planet with the dust produced in a distant planetesimal disk , spiraling inward by poyning - robertson drag . we will study the evolution of the system as a function of the dust properties and the numerical resolution used in modeling the planetesimal disk . finally , our results are summarized and discussed in section [ conclusions ] . @xcite characterized the evolution of the giant planets from several possible initial resonant configurations . they assumed that the solar system could have formed with 4 , 5 , or 6 giant planets . the additional planets had a total mass comparable to that of uranus or neptune . they analyzed the rate of success of each configuration by measuring the fraction of the planetary orbital evolutions that fulfilled a set of constraints , namely : the final planetary system is made of four giant planets ( criterion a ) , the orbits of the 4 remnant planets resemble the present ones ( criterion b ) , the proper mode of jupiter s eccentricity ( @xmath19 ) is greater than 0.022 in the final system , i.e. , at least half of its current value ( criterion c ) , and the ratio @xmath14 evolves from @xmath11 to @xmath12 in less than 1 my ( criterion d ) . criterion c is important because it is found that , although encounters between jupiter and an ice giant can excite @xmath19 , not all of them can do it as much as needed . criterion d is important because it is linked with the evolution of the terrestrial planets and of the asteroid belt @xcite . if jupiter and saturn migrate too slowly through the period ratio 2.12.3 , the @xmath20 and @xmath21 resonances sweep through the terrestrial planet region and the inner asteroid belt , resulting in orbital distributions inconsistent with those observed . then , from the success rates , @xcite concluded that a planetary system initially having 5 planets should be preferred . here , we revisit the @xcite work , adding a new constraint on the orbital evolution of neptune from the work of @xcite , namely that neptune should have migrated to @xmath028 au before the onset of the planetary instability . the works by @xcite and @xcite also give other constraints on the migration timescale of neptune and its eccentricity and inclination before the instability . the eccentricity and inclination constraints are met fairly easily because in this scenario neptune evolves in a planetesimal - driven migration without planetary close encounters until the instability phase . as for the migration timescale , @xcite claims that before the planetary instability , once out of resonance , neptune can migrate according to @xmath22 10 my for disks with 20 @xmath5 and according to @xmath22 50 my for disks with 15 @xmath5 . considering that those disks have the same extension , this exemplifies how sensitive the migration timescale can be with respect to the density of the disk . therefore , we decide not to consider the migration timescale as an important parameter at this point , because getting a correct timescale of the migration is probably only a matter of tuning the density and profile of the planetesimal disk . thus we add just one new criterion ( criterion e ) : at the time of the instability @xmath23 has to be in the interval 2729 au . while criteria a to d indicate that the initial solar system most probably hosted five giant planets @xcite , criterion e is a very important criterion that can be used to constrain in which resonant chain these five planets likely were at the end of the gas - disk phase . for simplicity , we consider only the cases of @xcite that involved 5 planets . we also always considered jupiter initially @xmath05.4 au , a distance between the outermost ice giant and the inner edge of the planetesimal disk @xmath24 1 and 2 au and extending to 30 au @xcite , composed by 1,000 planetesimals ( which is the resolution considered by @xcite ) , and two values of surface density ( @xmath25 ) equal to 0.02 and 0.03 @xmath26 . such configuration implies that the mass of the planetesimal disk ranges from @xmath013 @xmath5 ( in the more relaxed configuration 3:2 , 3:2 , 2:1 , 2:1 5:4 au , and the three equal - mass ice - giant planets placed beyond them ] , with @xmath27 2 au and @xmath28 0.02 @xmath26 ) to @xmath065 @xmath5 ( in the more compact configuration 3:2 , 3:2 , 4:3 , 5:4 , with @xmath27 1 au and @xmath28 0.03 @xmath26 ) . to perform our simulations , we used the hybrid integrator in the _ mercury - package _ @xcite , with a time step of half an year . table [ table1 ] gives in column 6 the rate of success for our new criterion e , after 20 runs of each initial configuration from ( * ? ? ? * table 6 in their paper ) . those 20 simulations were composed by 5 simulations for each pair @xmath29 [ 1,2 ] au , and @xmath28 [ 0.02 - 0.03 ] @xmath26 . although we have used two different values of surface density and two values of @xmath30 for the planetesimal disk , we decided to compress the results of all 20 simulations , for each configuration , in table [ table1 ] , column 6 . this is because , concerning our criterion e , only minor differences were observed with these changes . thus , when important , those minor differences will be reported . for completeness and to increase statistics , table [ table1 ] also reports the average of the rate of success for the other criteria taken from ( * ? ? ? * table 6 in their paper ) . because the simulations performed in this work are essentially the same of @xcite , the success rate for criteria a to d are consistent ( within the small number statistics ) with those reported in @xcite thus , we did not report our own results for the criteria a , b , c , and d. rather than that , we decided to check only the success rate relative to the new criterion e , and then , combine our findings with those of @xcite , which have much higher statistics , in order to determine which of the initial planetary configurations are the most compatible with the constraint on the dynamical evolution of neptune , unveiled in @xcite . lccccc 3:2 , 3:2 , 4:3 , 5:4 & 24 & 12 & 0 & 2 & 5 + 3:2 , 3:2 , 3:2 , 3:2 & 30 & 15 & 2 & 5 & 15 + 3:2 , 3:2 , 4:3 , 4:3 & 47 & 23 & 3 & 3 & 5 + * 3:2 , 3:2 , 2:1 , 3:2 * & * 32 * & * 13 * & * 4 * & * 7 * & * 65 * + 3:2 , 3:2 , 2:1 , 2:1 & 38 & 18 & 4 & 4 & 25 + 2:1 , 3:2 , 3:2 , 3:2 & 53 & 40 & 2 & 18 & 0 + 2:1 , 4:3 , 3:2 , 3:2 & 73 & 33 & 0 & 18 & 40 + our results can be interpreted as follows : : in these configurations , as it also appears in ( * ? ? ? 16 ) , it is possible that neptune leaves the resonant configuration and migrates ahead the other planets , but not always in the way required in @xcite . indeed , a configuration with initial resonances 2:1 , 3:2 , 3:2 , 3:2 never lead to an instability that fulfilled criterion e in our 20 simulations . the evolutions from the 2:1 , 4:3 , 3:2 , 3:2 configuration , on the other hand , had one of the largest rate of success for criterion e. however , such configurations present other major problem , mostly concerning the final orbital separation between jupiter and saturn , which is typically too big . the point is that , because the ratio @xmath14 has to change less than 0.5 if jupiter and saturn are initially in the 2:1 mmr , the mass of the fifth planet should be @xmath01/2 of the mass of neptune . instead , we assumed for this planet a mass equal to that of uranus - neptune , as in @xcite . a smaller mass planet , however , would not excite jupiter s eccentricity enough , so that the criterion c would not be fulfilled . moreover , hydrodynamical simulations @xcite show that in a massive gas - disk , jupiter and saturn would be locked in their mutual 3:2 mmr , rather than 2:1 . the grand tack model @xcite also implies that jupiter and saturn were in their 3:2 mmr at the disappearance of the disk of gas . so , we do nt consider that configurations with jupiter and saturn in the 2:1 mmr are very realistic . : these two configurations are too compact . only one out of 20 simulations for each configuration fulfilled criterion e. usually , the evolution of the planetary system is as in ( * ? ? ? 7 ) , where the instability takes place as soon as one planet goes out of resonance . as a result , no large - range planetesimal - driven migration occurs before the planetary instability . instead , the instability is violent and the evolution of neptune is largely different from that constrained in @xcite ( figure [ fig1 ] top left , in this work ) . nevertheless , although they are low , none of the success rates reported in table [ table1 ] for the configuration 3:2 , 3:2 , 4:3 , 4:3 are null . thus , even being very unlikely , due to the reasons presented , this configuration can not be ruled out . : this configuration is too loose . @xcite already pointed out that neptune stops farther than 30 au for whatever reasonable disk s parameters starting for this configuration . here , we found that even with the planetesimal disk of lowest mass ( @xmath31 [ 1324 ] @xmath5 ) , ranging from @xmath32 [ 25(26)30 ] au , neptune goes out of resonance and usually migrates all the way up to 30 au or more , before that the planetary instability happens ( figure [ fig1 ] top right ) . however , sometimes the instability can also occur before neptune reaches 30 au , say at @xmath028 au . therefore , although this configuration may also present some problems with the extension of the planetesimal disk , as it will be discussed in the following sections , it does have a good rate of success in fitting criterion e ( 25% success rate ) . : this configuration allows neptune to migrate away from the other planets before the instability , but usually not enough . the instability commonly happen when neptune is around 20 au ( see * ? ? ? * figs 8 and 9 , and figure [ fig1 ] bottom left in this work ) . only for high density disks neptune can migrate fast enough and reach @xmath028 au before the planetary instability . if the disk s density is low , neptune migrates slowly and ca nt go too far into the planetesimal disk . so , this configuration has a low , but non - negligible , success rate of 15% and can not be ruled out . : this is the configuration that presents the highest rate of success , considering all criteria together , so that it is highlighted in bold in table [ table1 ] . this is also the most successful configuration relative to criterion e , with a success rate of 65% independently of the disk setup . the evolution of neptune and of the rest of the planetary system is usually similar to that shown in ( * ? ? ? * figs 12 , 13 , and 14 ) , and satisfies the constraint identified in @xcite ( figure [ fig1 ] bottom right in this work ) . figure [ fig1 ] shows examples of the evolution of the systems corresponding to the descriptions given for each of the resonant chains discussed above ( with the exception of the 2:1 jupiter - saturn resonance , due to the considerations previously made ) . almost all evolutions depicted in figure [ fig1 ] fulfill the criteria a , b , c , and d. criterion c was not fulfilled only for the configuration 3:2 , 3:2 , 4:3 , 4:3 ( top left ) , where , although the eccentricity of jupiter was excited during the instability , it was quickly damped afterwards ( @xmath33 0.0075 ) due to the strong dynamical friction provided by the planetesimal disk . only the configuration 3:2 , 3:2 , 2:1 , 3:2 ( bottom right ) fulfilled all constraint together , including our new criterion e. summarizing , when the new constraints on the evolution of neptune from @xcite are considered , the most probable initial configuration of the giant planets was 3:2 , 3:2 , 2:1 , 3:2 . the configurations 3:2 , 3:2 , 2:1 , 2:1 , and 3:2 , 3:2 , 3:2 , 3:2 , or 3:2 , 3:2 , 4:3 , 4;3 also appear possible , but less probable . this seems to be a consequence of the more relaxed orbital separation between the first and second ice giants . therefore , one would expect that any configuration having the first and the second ice giant in distance from the sun in a 2:1 mmr or similarly relaxed could perhaps also work . however , for simplicity , we decide to restrict ourselves only to the cases of @xcite , which we know the rate of success in filling criteria a , b , c , and d. the analysis performed in this section does not take into account the timing of the planetary instability . with the location of the inner edge of the disk that we adopted , all evolutions are relatively fast , and lead to instabilities after a few 10s my . in the next section we analyze whether it is possible to get a late instability out of the 3:2 , 3:2 , 2:1 , 3:2 configuration , under some assumption on the geometry of the disk , still satisfying criterion e on the evolution of neptune . since @xcite presented the _ nice model _ , the existence of a late heavy bombardment of the terrestrial planets has been associated to the possibility that the giant planet instability occurred after hundreds of my of evolution @xcite . actually , the nice model , both in its initial version and from generic initial resonant configurations @xcite , is compatible with both early and late instabilities , depending on the distance of the inner edge of the original trans - neptunian disk and other disk properties ( mass , excitation , etc . ) . thus , the nice model does not prove the existence of a late instability . the timing of the instability can only be determined by looking at other constraints , such as the lunar impact chronology . unfortunately , the latter is still under debate . however , it is not obvious that the nice model can still be compatible with a late instability , if we fix a specific initial resonant configuration and impose a specific evolution of neptune . as explained in the introduction , if the disk is sufficiently close to neptune at the beginning , it is relatively easy that neptune migrates several au away before that the instability happens , but in this case the full dynamical evolution is rapid , as in the previous section . if the initial location of the disk is sufficiently far from neptune , the instability may happen late , but neptune is unlikely to migrate significantly before that the instability happens . moreover , in the case of the most relaxed resonant configurations ( e.g. 3:2 , 3:2 , 2:1 , 2:1 , which has neptune initially around @xmath024.5 au ) , the requirement that the inner edge of the disk is several au beyond neptune while the outer edge is at @xmath030 au reduces the disk to a ring . this seems implausible , particularly when considering the constraint that the disk should have hosted 1,000 - 4,000 pluto - size bodies @xcite , which would probably be problematic for the late instability as well , because the spreading of the narrow disk would not permit a delay . if one could show that the most developed and constrained version of the nice model is incompatible with a late instability , this would be very important to clarify the chronology of events in the solar system . thus , in this section we try hard to find whether the instability could have happened late and under which conditions . we let the reader judge whether these conditions are realistic or plausible , with our considerations summarized in sect . [ conclusions ] . for this goal , we proceed in steps . first , in sect . [ dr ] , we determine two critical values for the distance of the inner edge of the disk from neptune . one , @xmath34 , is the maximal distance that allows neptune to have a long - range planetesimal - driven migration away from the other planets , as in the previous section . the other one , @xmath35 is the minimal distance of the disk for which the five - planet system remains stable in resonance for all times . clearly @xmath36 and @xmath34 is larger than or equal to the neptune - disk distances considered in the previous section . the problem is that both @xmath35 and @xmath34 depend on disk resolution ( number of planetesimals and individual masses that are used to model the disk ) and that realistic values can not be used in practical simulations . so , we need to estimate trends to evaluate the values of the critical distances . in a second step , in sect [ de ] , we study how a flux of dust from the planetesimal disk , due to collisional planetesimal grinding and poynting - robertson drag , influences planet migration . this depends on total mass flux and individual dust size . the idea is that , under an appropriate dust - flux , neptune may migrate away from its initial location , so that its distance from the planetesimal disk decreases from @xmath37 to @xmath38 . at this point planetesimal - driven migration starts , as in the previous section . if the dust - driven migration is slow enough , the planet instability can occur late , even though it occurs only a few millions of years after the beginning of planetesimal - driven migration . this concept is then proven in section [ dde ] , where we combine in a single simulation dust - driven migration and planetesimal - driven migration , and we obtain an instability after 300 my , with neptune s evolution satisfying criterion e. this simulation is of course performed with an unrealistically low disk - resolution , but is a proof of concept that , for whatever disk resolution , an appropriate dust flux can be found . we finally discuss what a realistic dust flux could be in sect . [ dgl ] the interaction of a planetary system with a planetesimal disk can be very resolution - dependent , i.e. , sensitive on the number of planetesimals used to described the disk in the simulation and to their individual mass . in some cases , as in those of section [ early ] , this problem of sensitivity on disk - resolution is not strong because the distance between the outermost planet and the inner edge of the disk is small , so that there are from the very beginning many planetesimals on unstable orbits that encounter the planet , force its migration and , consequently , bring many new planetesimals to planet - crossing orbit . this is the reason for which @xcite found similar results for test simulations conducted with 100 , 300 , 1,000 , 3,000 , and 10,000 planetesimals in the disk ( same total disk masses ) , so that they decided to use only 1,000 planetesimals in their final simulations . of course , even in these cases , the disk resolution has an influence on how grainy or smooth the planet migration will be , but does nt have a major influence on the amplitude and the timescale of the migration . on the other hand , the resolution of the planetesimal disk becomes a real issue when the inner edge of the disk is far enough from the outermost planet . in this situation , the fraction of the disk that is unstable is small , mostly limited to the volume of the strongest mean motion resonances with the planet . the resonant planetesimals develop planet - crossing eccentricities and , with close encounters , force the planet to move a bit outwards . this moves outwards the resonances as well , destabilizing new planetesimals ( those which were not in resonance at the beginning but are swept by the resonant motion ) . depending on whether the mass of the new unstable planetesimals is larger or smaller than the mass of the original unstable planetesimals , the planet migration will accelerate or damp @xcite . in the damped migration mode , the planet eventually stops migrating while the resonances become empty . if the number of planetesimals is small , the planetesimals are individually more massive to represent a disk with the same total mass . thus , the migration of the planet is more grainy because each encounter between the planet and a planetesimal causes a larger jump in the planet s semimajor axis . of course , there are fewer jumps , so that the planet migrates , on average , by the same amount . but the more jumpy migration of the planet causes the resonances to jump around as well , thus destabilizing more mass from the disk than in the case of a smoother planet migration with the same averaged range . because more mass is transferred from the disk to planet - crossing orbit , the migration of the planet can change , in some cases , from the damped to the accelerated mode . with this issue in mind , we proceed in our study looking for two critical disk distances , for a series of disk resolutions . first , we look for the minimum distance ( @xmath1 ) that the inner edge of the planetesimal disk has to have from neptune ( the outermost planet for our case ) in order to maintain the planets stable for at least 400 my . second , we search for the maximal distance ( @xmath2 ) that neptune can have from the inner edge of the planetesimal disk to evolve as in section [ early ] . from these results we will extrapolate the values of @xmath1 and @xmath2 to realistic almost infinite resolution . once again we consider disks with fixed surface densities ( scaling as @xmath40 ) , because in this case we guarantee always the same amount of mass in the planets mmrs , independently of the resolution of the disk ( apart from stochastic sampling ) . the initial distribution of eccentricity and inclination of the disk was such that @xmath41 and @xmath42 ( the same initial distribution for all disks considered in this work ) . this is the simplest configuration that represents a cold planetesimal disk emerging from the end of the dissipation of the gas nebula . even if the disk was initially more dynamical excited not much differences in the overall evolution are expected because the mmrs inside the disk are very efficient in exciting the disk s eccentricities to planet - crosser values . so , in a very short timescale an initially cold disk would appear like an initially excited one . finally , we consider disks with resolution of 1,000 , 5,000 , 10,000 , 15,000 , and 20,000 planetesimals . the outer edge of the disk is fixed at 30 au . the distance @xmath30 between the inner edge of the disk and the planet ranges from 1 to 7 au , with increments of 1 au . given the great demand of cpu - time required to simulate high - resolution disks , we only considered the most successful initial configuration in section [ early ] , with planets in the 3:2 , 3:2 , 2:1 and 3:2 mmrs . for this configuration , neptune is initially @xmath020 au . for each value of @xmath30 , we considered six different values of @xmath43 . .@xmath1 as a function of @xmath43 for the less resolved planetesimal disk ( 1,000 planetesimals ) . @xmath1 : minimum distance that the inner edge of the planetesimal disk has to have from neptune ( the outermost planet for our case ) in order to maintain the planets stable for at least 400 my . @xmath43 : density of the planetesimal disk , scaled as @xmath40 . the mass of the planetesimal disk , @xmath44 , in earth masses , for each pair @xmath43@xmath1 for the initial configuration 3:2 , 3:2 , 2:1 , 3:2 , with neptune at @xmath020 au and outer disk edge at 30 au is also shown for reference . [ cols="^,^,^,^,^,^,^,^,^",options="header " , ] the results of the proof of concept " simulation are shown in figures [ fig4 ] and [ fig5 ] . figure [ fig4 ] shows the evolution of the semimajor axis ( @xmath45 ) , the perihelion distance ( @xmath46 ) and the aphelion distance ( @xmath47 ) for all planets as a function of the time . figure [ fig5 ] brings snapshots of the ( @xmath45 @xmath39 @xmath48 ) and ( @xmath45 @xmath39 @xmath49 ) of the whole system for different times of the evolution ] . the evolution shown in these figures has an instability after about 300 my and the evolution of the planets is in a considerably good agreement with that deduced in @xcite from kuiper belt constraints . particularly , the instability happens when neptune is at @xmath50 au as required . moreover , as we can see from these figures , the evolution of neptune before the onset of the instability , follows closely the assumptions / predictions of tables [ table2 ] and [ table3 ] , and the dust mechanism of section [ de ] . the instability did not happen precisely at the time we targeted , due to the approximations inherent in our linear combination of different processes , which in reality have a non - linear interplay . nevertheless , the instability happens late enough to suggest that the association between a planetary instability and the trigger of the late heavy bombardment is indeed possible . to assess whether the dust fluxes that are needed to migrate neptune into the disk ( see table [ table3 ] ) are reasonable , we turned to the simulation presented in the supplementary material ( s4 ) of @xcite . that simulation considered a disk with initially 60 earth masses , extended from 21 to 34 au the disk was quite excited in eccentricity and inclination within 27 au and dynamically cold beyond it . the planetesimals were assumed to have a size - frequency distribution ( sfd ) similar to that of the hot kuiper belt population in the inner part of the disk ( r @xmath4 27 au ) and similar to that of the cold kuiper belt population beyond it . both sfds were scaled up in number so that the masses in the inner disk and outer disk were respectively 1/3 and 2/3 of the total . the collision probabilities and impact velocities within particles in the inner part of the disk , the outer part or in different parts were evaluated to be @xmath51 km@xmath52y@xmath53 and 0.83 km / s , @xmath54 km@xmath52y@xmath53 and 0.25 km / s , @xmath55 km@xmath52y@xmath53 and 0.68 km / s , respectively . the collisional evolution simulations were conducted with the code boulder , developed in @xcite . the specific impact energy for collisional disruption was assumed to be 1/3 of that of strong ice @xcite . according to these simulations , the disk produced 32 earth masses of dust in 400 my . a total mass of 32 earth masses is big compared to the numbers in table [ table3 ] . on the other hand , not all this dust would survive all the way to neptune - crossing orbit , because the collisional cascade in the dust population comminute grains until they reach the ejection limit @xmath56 . this part of the collisional cascade is not modeled in the boulder code . also , the total amount of dust produced could be reduced by truncating the initial sfd to objects larger than some limit size ; in a new simulation where we retained only bodies with @xmath57 km in the initial sfd the total dust production is reduced to 15 earth masses . another possibility is to increase the specific energy of disruption for the bodies . in fact , @xcite revised the previous estimates of @xcite and concluded that the specific energy for disruption should be similar to that for strong ice in @xcite . a simulation with an initial truncated sfd but with a specific disruption energy 3 times larger produced only 7 earth masses of dust . even more simply , reducing the initial population in the disk would reduce the amount of dust produced , given that the dust production scales as the square of the number of planetesimals in the disk . in conclusion , we think that the dust fluxes listed in table [ table3 ] , required to cause a significant migration of neptune , are reasonable and in the good ballpark of what a massive planetesimal disk can produce . a more relevant problem is the dust production timeline . in the boulder experiments described above , the dust production decays as @xmath58 . in the boulder simulations , the collisional evolution of the dust is not tracked . but the dust collisional lifetime should scale as the square root of the dust production . consequently , the total amount of dust of a given size ( or @xmath59 ) should evolve with time as @xmath60 . because the migration of the planet is linear in the amount of dust that it encounters , the evolution of its semimajor axis should be proportional to @xmath61 . this means that , of the total migration range spanned in -say- 400 my , 1/2 of it is covered in the first 100 my and 13% in the last 100 my . in other words , the time to reach @xmath2 is proportional to @xmath62 , instead of being proportional to @xmath63 as in the case of a linear evolution ( where @xmath64 is @xmath65 au ) . thus , small changes in the flux can make the curve @xmath66 to reach @xmath2 distance early or very late . obtaining a late instability around 400 my requires a more delicate tuning of the dust production than that done in sect . [ dde ] , where we considered that the dust production is constant and the collisional lifetime of dust infinite . it is very likely that , at the disappearance of the protoplanetary disk of gas , the giant planets of the solar system were all in mean motion resonances with each other , because this is the typical outcome of gas - driven planet migration @xcite . @xcite investigated which resonant configurations are more likely to reproduce the current orbital configuration of the giant planets via a phase of dynamical instability , as that described in the nice model . they considered four criteria of success and found a number of initial configurations that meet all of them in a non - negligible fraction of the simulations . all these configurations require the existence of a fifth giant planet , with a mass comparable to those of uranus or neptune , eventually ejected during the instability . more recently , @xcite showed that explaining and reproducing the current orbital structure of the kuiper belt requires that neptune had a specific evolution , reaching a heliocentric distance of @xmath028 au by planetesimal - driven migration before that the instability happened . thus , in section [ early ] of this paper , we have revisited the work of @xcite , adding a new criterion of success from the results of @xcite . of the configurations selected by @xcite , the one denoted 3:2 , 3:2 , 2:1 , 3:2 ( each integer ratio referring to the mean motion resonance between a pair of adjacent planets ) stands out as the more likely to fulfill the new criterion of success . other configurations ( the 3:2 , 3:2 , 2:1 , 2:1 , the 3:2 , 3:2 , 3:2 , 3:2 , and the 3:2 , 3:2 , 4:3 , 4:3 ) can not be excluded , but their success rate relative to the new criterion is significantly lower . an open issue with the nice model is the timing of the giant planet instability . if one wants to explain the late heavy bombardment of the moon as a spike in the impact crater rate @xcite , the instability has to have happened late @xcite , unlike in the simulations of section [ early ] , in which the instability occurred after a few 10s of my . it is not obvious that a 5-planet resonant configuration can remain stable for long time and become unstable then , particularly with a specific evolution of neptune as that required to sculpt the kuiper belt . thus , in section [ late ] we investigated whether a late instability from the 3:2 , 3:2 , 2:1 , 3:2 is possible and under which conditions . first , we looked for characteristics of the planetesimal disk that can maintain the planets in stable orbits for 400 my . we found that the evolution and stability of the planets are very sensitive to the resolution of the planetesimal disk . thus , we determined the minimum distance that the inner edge of the planetesimal disk has be away from neptune to keep the planets in stable orbit for 400 my as a function of the surface density of the disk and its numerical resolution . we named this distance @xmath1 ( table [ table2 ] ) . in a second step , we found a characteristic neptune - disk distance , which we named @xmath2 , from where neptune starts a long - range planetesimal - driven migration . thus , for the system to exhibit a late instability together with an evolution of neptune like the one that is needed to sculpt the kuiper belt , neptune needs to be extracted from the 3:2 , 3:2 , 2:1 , 3:2 resonance chain , and to migrate outwards over hundreds of mys until it reaches the distance @xmath2 . to understand how this could be possible , we introduced the concept of dust - driven migration , i.e. , migration due to the gravitational interaction of a planet with the dust produced in a distant planetesimal disk , spiraling inward by poyning - robertson drag . we have shown that a late instability can be achieved when considering a flux of dust ( @xmath67 ) ranging from 7 to 22 @xmath5 per 400 my , depending on the value of @xmath1 ( table [ table3 ] ) . continuing , in section [ dgl ] we evaluated how much dust could be produced from the disk - like configurations that we assumed and found good agreement with the dust flux described above . indeed , we found that the total amount of dust produced in 400 my may vary from @xmath07 @xmath5 to a maximum of about 32 @xmath5 depending on the size frequency distribution ( sfd ) and catastrophic disruption energies assumed for the planetesimal disk and its constituents . finally , we also found that the dust production rate expected for a self - grinding disk is not constant , but it decays as @xmath58 . the abundance of dust , which is also limited by the dust collisional lifetime , should decay as @xmath60 . thus , because the migration of the planet is linear in the amount of dust that it encounters , the evolution of its semimajor axis should be proportional to @xmath61 , which implies that small changes in the flux of dust can result in neptune reaching @xmath2 distance early or very late . therefore , obtaining a late instability around 400 my requires a more delicate tuning of the dust production . summarizing , based on the work of @xcite , and considering the evolution of neptune required to explain the structure of the kuiper belt @xcite , our results indicate that the preferred initial configuration of the giant planets at the time of the dissipation of the gas nebula is the 3:2 , 3:2 , 2:1 , 3:2 , and that , although an early instability reproduces more easily the evolution of neptune required to explain the structure of the kuiper belt , such evolution is also compatible with a late instability . we thank an anonymous reviewer for useful comments and suggestions on the submitted manuscript . r.d . acknowledges support provided by grants @xmath682015/18682 - 6 and @xmath682014/02013 - 5 , so paulo research foundation ( fapesp ) and capes . a.m. acknowledges support by the french anr , project number anr-1313-bs05 - 0003 - 01 projet mojo ( modeling the origin of jovian planets ) . r.s.g . acknowledges his grant no . 307009/2014 - 9 from cnpq , conselho nacional de desenvolvimento cientfico e tecnolgico , brazil . the work of d.n . was supported by the nasa emerging worlds program . 99 batygin , k. , brown , m. e. , & betts , h. 2012 , , 744 , l3 benz , w. , & asphaug , e. 1999 , , 142 , 5 bottke , w. f. , vokrouhlick , d. , minton , d. , et al . 2012 , , 485 , 78 brasil , p. i. o. , roig , f. , nesvorn , d. , et al . 2016 , , 266 , 142 brasser , r. , morbidelli , a. , gomes , r. , tsiganis , k. , & levison , f. h. 2009 , 134 , 1790 burns , j. a. , lamy , p. l. , & soter , s. 1979 , , 40 , 1 chambers , j. e. 1999 , , 304 , 793 deienno , r. , yokoyama , t. , nogueira , e. c. , callegari , n. jr . , & santos , m. t. 2011 , , 536 , a57 deienno , r. , nesvorn , d. , vokrouhlick , d. , & yokoyama , t. 2014 , , 148 , 25 deienno , r. , gomes , r. s. , walsh , k. j. , morbidelli , a. , & nesvorn , d. 2016 , , 272 , 114 fernandez , j. a. , & ip , w .- h . 1984 , , 58 , 109 fernandez , j. a. , & ip , w. h. , 1996 , , 44 , 431 gomes , r. s. , morbidelli , a. , & levison , h. f. 2004 , , 170 , 492 gomes , r. s. , tsiganis , k. , morbidelli , a. , & levison , h. f. 2005 , , 435 , 466 izidoro , a. , morbidelli , a. , raymond , s. n. , hersant , f. , & pierens , a. 2015 , , 582 , a99 kaib , n. a. , & chambers , j. e. 2016 , , 455 , 3561 leinhardt , z. m. , & stewart , s. t. 2009 , , 199 , 542 leinhardt , z. m. , & stewart , s. t. 2012 , , 745 , 79 levison , h. f. , morbidelli , a. , gomes , r. , & backman , d. 2007 , protostars and planets v , 669 levison , h. f. , morbidelli , a. , vanlaerhoven , c. , gomes , r. , & tsiganis , k. 2008 , , 196 , 258 levison , h. f. , bottke , w. f. , gounelle , m. , et al . 2009 , , 460 , 364 levison , h. f. , morbidelli , a. , tsiganis , k. , nesvorn , d. , & gomes , r. 2011 , , 142 , 152 nesvorn , d. , vokrouhlick , d. , & morbidelli , a. 2007 , , 133 , 1962 nesvorn , d. 2011 , , 742 , 22 nesvorn , n. & morbidelli , a. 2012 , , 144 , 117 nesvorn , d. , vokrouhlick , d. , & morbidelli , a. 2013 , , 768 , 45 nesvorn , d. , vokrouhlick , d. , & deienno , r. 2014 , , 784 , 22 nesvorn , d. , vokrouhlick , d. , deienno , r. , & walsh , k. j. 2014 , , 148 , 52 nesvorn , d. 2015 , , 150 , 73 nesvorn , d. 2015 , , 150 , 68 nesvorn , d. , & vokrouhlick , d. 2016 , , 825 , 94 malhotra , r. 1993 , , 365 , 819 malhotra , r. 1995 , , 110 , 420 morbidelli , a. , levison , h. f. , tsiganis , k. , & gomes , r. 2005 , , 435 , 462 morbidelli , a. , tsiganis , k. , crida , a. , levison , f. h. , & gomes , r. 2007 , , 134 , 1790 morbidelli , a. , bottke , w. f. , nesvorn , d. , & levison , h. f. 2009 , , 204 , 558 morbidelli , a. , brasser , r. , gomes , r. , levison , h. f. , & tsiganis , k. 2010 , , 140 , 1391 moro - martin , a. 2013 , planets , stars and stellar systems . volume 3 : solar and stellar planetary systems , 431 petit , j .- m . , kavelaars , j. j. , gladman , b. j. , et al . 2011 , , 142 , 131 . pierens , a. , & nelson , r. p. 2008 , , 482 , 333 roig , f. , & nesvorn , d. 2015 , , 150 , 186 tera , f. , papanastassiou , d. a. , & wasserburg , g. j. 1974 , earth and planetary science letters , 22 , 1 thommes , e. w. , duncan , m. j. , & levison , h. f. 1999 , , 402 , 635 toliou , a. , morbidelli , a. , & tsiganis , k. 2016 , , 592 , a72 tsiganis , k. , gomes , r. s. , morbidelli , a. , & levison , h. f. 2005 , , 435 , 459 walsh , k. j. , & morbidelli , a. 2011 , , 526 , a126 walsh , k. j. , morbidelli , a. , raymond , s. n. , obrien , d. p. & mandell , a. m. 2011 , , 475 , 206

macroscopic objects in contact with an ionized gas acquire a negative charge because the influx of electrons from the plasma outruns the influx of ions . the collection of electrons at the wall ( boundary of the object ) gives rise to a repulsive coulomb potential which reduces the electron influx until the wall charge reaches a quasi - stationary value . as a consequence of the electron accumulation at the wall an electron depleted region , the plasma sheath , is formed adjacent to the wall . most of the voltage driving the discharge drops across the sheath . wall charges may however not only affect the spatial structure of the plasma but also surface - supported elementary process such as electron - ion recombination and secondary electron emission , which are particularly important in dusty plasmas,@xcite dielectric barrier discharges,@xcite and solid - state based micro - discharges.@xcite a macroscopic description of the plasma - induced wall charge , sufficient for the modeling of the plasma sheath , @xcite is clearly insufficient for quantifying the effect wall charges might have on these processes . a microscopic description of the plasma - induced wall charge and the potential across the plasma - wall interface it leads to is required . traditionally , plasma walls are treated as perfect absorbers.@xcite irrespective of the microscopic interaction , all electrons and ions impinging on the wall are assumed to recombine instantly . from this model only the wall potential just outside the wall can be obtained . this is the potential that balances the electron and ion influx at the wall . a first , qualitative step going beyond this model was taken by emeleus and coulter,@xcite who envisaged the wall charge as a two - dimensional surface plasma coupled to the bulk plasma via electron - ion wall recombination . no attempt was however made to put this appealing idea onto a formal basis . later the notion of a two - dimensional surface charge was developed further by behnke and coworkers@xcite utilizing phenomenological rate equations for the electron and ion surface densities . in these equations , the microphysics at the wall is encapsulated in surface parameters , such as , electron and ion sticking coefficients , electron and ion desorption times , and an electron - ion wall recombination coefficient . in principle these parameters can be calculated . assuming , for instance , plasma electrons to ad- and desorb in the long - range image potential of the wall we calculated in our previous work electron sticking coefficients and electron desorption times for uncharged metallic @xcite and dielectric surfaces . @xcite we also made a first attempt to estimate these two quantities for charged dielectric plasma walls @xcite and proposed a physisorption - inspired microscopic charging model for dust particles in a gas discharge . @xcite in this work , we shift gears and focus on the potential and charge distribution across the plasma - wall interface after the quasi - stationary wall charge ( electron adsorbate in the notation of our previous work @xcite ) has been established . in other words , we extend the modeling of the plasma sheath @xcite to the region inside the solid and calculate the plasma - induced modifications of the potential and charge distribution of the surface . although knowing the potential and charge distribution across the interface may not be of particular importance for present day technological plasmas , it is of fundamental interest from an interface physics point of view . in addition , considering the plasma wall as an integral part of the plasma sheath may become critical when the miniaturization of solid - state based plasma devices @xcite continues . in the model outlined below we specifically consider a dielectric wall and treat the plasma - induced quasi - stationary wall charge , that is , the surplus charge on top of the charge distribution of the bare , free - standing surface , as an electron surface layer of a certain extent , which is trapped by and in thermal equilibrium with the wall . in order to determine the chemical potential , width , and spatial position ( relative to the crystallographic interface ) of the electron surface layer , which depend on surface as well as plasma parameters , we employ a one - dimensional model for a graded interface between a collisionless plasma sheath and a dielectric surface which is assumed to be a perfect absorber , that is , the wall potential balances at a certain distance from the crystallographic interface the electron and the ion influx from the plasma . the model of a graded interface encompasses two important ingredients : the surface dipole of the bare surface responsible for the offset of the conduction band minimum to the potential just outside the dielectric and the long - range image potential . the former accounts for the charge re - distribution of the free - standing , uncharged surface arising from the truncation of the crystal and the latter supports polarization - induced external surface states ( image states ) , first predicted for liquid helium @xcite and later studied for metallic and dielectric surface with negative electron affinity @xcite , which may trap the electron surface layer in front of the crystallographic interface . originally proposed by stern for the interface between two dielectrics,@xcite and later on used by others for semiconductor heterojunctions @xcite and electron trapping in nanopores,@xcite the graded interface model also guarantees continuity of the electrostatic potential across the plasma - wall interface . the model thus allows us to study the spatial distribution of the plasma - induced wall charge across the interface . to insert the surplus charge into the interface we follow tkharev and danilyuk @xcite and minimize , in the spirit of density functional theory , @xcite the grand canonical potential of wall - thermalized excess electrons we also investigate how the electron surface layer merges with the neutral bulk of the dielectric which we describe with the model of an intrinsic semiconductor . various improvements of the model are conceivable but the increased mathematical complexity would mask the general ideas we would like to convey . for instance , the model of a collisionless sheath could be replaced by more realistic models . @xcite going beyond the perfect absorber model , on the other hand , is an unsolved problem . it would require the inclusion of electron desorption , electron sticking , and electron - ion recombination , with the respective coefficients to be self - consistently calculated for the quasi - stationary electron adsorbate at the wall . replacing the graded interface by an ab - initio theory for the surface , for instance , along the lines given in refs . @xcite , possibly taking ad - layers of the host gas atoms or molecules as well as impurities and imperfections into account , is desirable but at the present stage of the investigation impractical . it would require an expensive atomistic characterization of the plasma - wall interface , either experimentally via various surface diagnostics or theoretically via ab - initio simulations . as long as the atomistic details affect however only the off - sets of the dielectric constant , the electron affinity , and the effective mass , the graded interface model already incorporates these details by a suitable parameterization . what is not well described is the non - universal region a few atomic units below and above the crystallographic interface . in particular , intrinsic surface states ( shockley and tamm states @xcite ) and additional surface states which may arise from the short - range surface potential due to impurities , imperfections , and ad - layers are not included . if unoccupied these states could trap the electron surface layer in the vicinity of the interface even for surfaces with positive electron affinity where image states are absent and can not trap the surplus charge in front of the surface . the remaining paper is structured as follows . in sec . [ electron surface layermain ] we first construct a crude model for the plasma - induced electron surface layer at the interface between a plasma and a dielectric wall . it does not account for the merging of the electron surface layer with the bulk of the dielectric . as long as the primary interest is in the region close to the crystallographic interface and the band gap of the dielectric is large enough , the crude model is sufficient . section [ negativescr ] describes a refinement of the model which enables one to also investigate the cross - over of the electron surface layer to the bulk of the dielectric . this is particularly important for dielectrics with small energy gaps . numerical results for the potential and the electron distribution are given in section [ results ] and a short summary is formulated in sec . [ summary ] . as depicted in fig . [ figure1 ] , we consider an ideal , planar interface at @xmath2 with the dielectric occupying the half - space @xmath3 and the discharge occupying the half - space @xmath4 . chemical contamination and structural damage due to the burning gas discharge are discarded . at the moment we focus on the physical principles controlling the electronic properties of the plasma - wall interface . in the model we propose the plasma - induced wall charge to be treated as an electron surface layer ( esl ) which is an interface specific electron distribution on top of the charge re - distribution due to the truncation of the solid . the esl is assumed to be thermalized with the solid and to stretch from the plasma sheath over the crystallographic interface to the bulk of the dielectric . the boundary between the esl and the plasma sheath is located in front of the surface at @xmath5 . it is the position where the attractive force due to the surface potential @xmath6 and the repulsive force due to the sheath potential @xmath7 balance each other . thus , @xmath8 is given by @xmath9 it gives the position of an effective wall for plasma electrons and ions at which , for instance , the flux balance condition of the perfect absorber model , @xmath10 , with @xmath11 and @xmath12 , respectively , the electron and ion flux towards the dielectric surface , has to be fulfilled . for @xmath13 an electron is attracted to the surface and thus contained in the esl while for @xmath14 it is repelled back into the plasma . on the solid side , for @xmath3 , the esl is bounded because of the shallow potential well formed by the restoring force from the positive charge in the plasma sheath . in this section we will outline the essential building blocks of the esl model . putting together concepts from plasma as well as surface physics a detailed , self - contained account seems to be helpful . in the traditional view , electrons missing in the positive space charge region in front of the plasma wall accumulate on the wall and give rise to a wall potential . for the construction of our one - dimensional interface model we need the total number per unit area of missing sheath electrons ( that is , the total surface density of missing sheath electrons ) as a function of the wall potential because it is this number of electrons which can be distributed across the esl . hence , we require a model for the plasma sheath . for simplicity , we use a collisionless sheath @xcite , more realistic sheath models @xcite make no difference in principle . in the collisionless sheath electrons are thermalized , that is , the electron density @xmath15 , with @xmath16 the potential , @xmath17 the plasma density and @xmath18 the electron temperature . the ions enter the sheath with a directed velocity @xmath19 and satisfy a source - free continuity equation , @xmath20 , implying @xmath21 , and an equation of motion @xmath22 , with @xmath23 the ion density , and @xmath24 the ion mass . the potential @xmath16 satisfies poisson s equation @xmath25 . thus , the governing equations for the collisionless plasma sheath are @xcite @xmath26 \text { . } \label{sheath2}\end{aligned}\ ] ] using dimensionless variables @xmath27 where @xmath28 equations ( [ sheath1 ] ) and ( [ sheath2 ] ) become @xmath29 in the esl model the plasma occupies not the whole half space @xmath4 but only the portion @xmath14 ( see fig.[figure1 ] ) . the integration of the first equation gives , where @xmath30 is the reduced velocity of ions entering the sheath , so that the second equation becomes @xmath31 using the boundary condition that the potential and the field vanish far inside the plasma , that is , @xmath32 and @xmath33 for @xmath34 , eq . ( [ clsheatheq ] ) can be integrated once and we obtain @xmath35 for ions entering the sheath with the bohm velocity @xmath36 . the field at the wall as a function of the wall potential @xmath37 is then given by @xmath38 the total surface density of electrons in the esl equals the total surface density of missing sheath electrons , in other words , the total surplus surface density of positive ions in the sheath @xmath39 which can be calculated from the electric field at the wall . integrating poisson s equation yields @xmath40 combing eqs . ( [ neq ] ) and ( [ etawprime ] ) gives the total surface density of electrons to be inserted into the esl as a function of the wall potential . the wall potential itself is determined by the flux balance condition , @xmath10 , which , in the esl model , is assumed to be fulfilled at @xmath5 . using the bohm flux for the ions and the thermal flux for the electrons , @xmath41 the wall potential is given by @xcite @xmath42 that is , @xmath43 in the collisionless sheath model the wall potential depends only on the electron temperature and the ion to electron mass ratio . we now turn to the interface region in which the missing sheath electrons will be inserted . this region is absent in the traditional modeling of plasma walls . in our model it is an extended region surrounding an ideal dielectric surface . in comparison to the electrons responsible for the chemical binding within the dielectric the additional electrons coming from the plasma are only a few . the electronic structure of the surface , in particular , the charge re - distribution due to truncation of the solid and the offset of the energy bands in the bulk with respect to the potential outside the dielectric will not be changed significantly by the presence of the surplus electrons comprising the wall charge . in order to quantify the above statement let us first consider the electrostatic potential and the electronic structure of a free - standing , uncharged dielectric surface . according to tung,@xcite it has to minimize the thermodynamical potential and satisfy poisson s equation implying that the potential is continuous across the surface . strictly speaking , the continuity of the potential only applies to the microscopic crystal potential which has to merge continuously with the surface potential outside the crystal . the averaged long - range potential , in contrast , can be discontinuous at the interface . it is this offset which is encoded in the surface dipole . the energy of an electronic state in the bulk of the dielectric can be referenced to the vacuum level @xmath44 , that is , the potential far outside the crystal , in the following way,@xcite @xmath45 where @xmath46 is the quantum - mechanical contribution to the energy , @xmath47 is the averaged potential of a cell due to the charge distribution within the same cell , and @xmath48 is the long range potential due to the surface dipole , space charges , and external fields . in the simple two - band model depicted in fig . [ figure1 ] , @xmath49 . @xmath50 contains the surface dipole arising from the truncation of the solid and responsible for the potential offset at the surface and a slowly varying component due to external fields and internal and plasma - induced space charges . external fields and internal space charges will be neglected in the following and plasma - induced space charges will be accounted for by poisson s equation ( see below ) . in order to judge whether the surplus charge arising from the plasma affects the surface dipole it is useful to consider first the typical strength of the surface dipole of a free - standing , uncharged dielectric surface . it results from a charge double - layer in immediate proximity to the surface . depending on the material it can have various origins . for an ionic crystal , for example , it is the lattice relaxation at the surface which makes anions or cations to protrude and the other species to retract ( e.g. protruding oxygen and retracted cations for magnesium oxide@xcite ) , while for semiconductors it is the regrouping of covalent bonds which leads to charge re - distribution at the surface . even in the absence of these effects the minimization of the thermodynamic potential of the surface s electrons leads already to an electron density leaking out into the vacuum . this is particularly important for metals . as a result a charge double - layer is formed over a length on the order of a lattice constant . the dipole layer is usually characterized by a dipole strength @xmath51 where @xmath52 is the limit of the long range potential just inside the crystal at the surface position @xmath53 and @xmath54 is the limit of the potential at that position just outside the surface . usually these two potentials , which characterize the discontinuity of the long - range potential at the surface , are termed the potential just inside and the potential just outside , respectively.@xcite here , just outside denotes a distance small compared to variations of the long - range potential due to external fields or space charges but large compared with the width of the charge double - layer . note also that in the definition of the potential just outside the image potential is assumed to have already decreased to zero . @xcite this point will be important later . the strength of the dipole layer is a microscopic property of the surface which is relatively insensitive against the additional charges from the plasma . the reason for this lies in the small number of additional electrons from the plasma compared to the number of displaced electrons involved in the formation of the dipole layer . to prove this statement we give a simple estimate . typical surface dipoles @xmath55 are on the order of electron volts . for a double - layer of one @xmath56 to each side of the crystallographic interface a potential difference of one volt requires a surface charge density of @xmath57 . the surface charge density at the wall of a helium discharge with plasma density @xmath58 and electron temperature @xmath59 amounts , however , only to @xmath60 . for typical plasma densities , the number of additional electrons is thus far too small to lead to a change of the surface dipole . even for semiconductor - based microdischarges @xcite , which can have much higher plasma densities , we expect the surface dipole of the plasma wall not to be modified by the plasma . in view of the just given estimate , we have to revise an assumption in our previous work,@xcite where we assumed the surface charge accumulating on the wall would increase the dipole energy @xmath55 by @xmath61 , leading to the image states being pushed from the band gap into the energy region of the conduction band . the numbers given in the previous paragraph indicate , however , that the band line - up of the conduction band and the potential just outside the solid , will not be affected much by the wall charge . hence , if a negative electron affinity supports image states in front of the uncharged surface , these states remain in the band gap for the charged surface . electron trapping as investigated in refs . @xcite is thus even possible for charged plasma walls . instead of the dipole strength @xmath55 which can not be measured directly , it is more convenient to characterize the dipole layer by the electron affinity @xmath62 which is a measurable quantity for a dielectric surface . @xcite the electron affinity is the energy released when an electron is moved from just outside the surface to the bottom of the conduction band . it accounts for charge re - distribution in the vicinity of the surface due to the truncation of the crystal . while many surfaces have positive electron affinity such as al@xmath0o@xmath1 or sio@xmath0 , there are also materials with negative electron affinity , for instance , diamond @xcite , boron nitride@xcite , or the alkaline earth oxides . @xcite the electron affinity depends also on ad - atoms . in some cases this is even used to control the electron affinity of a surface . terminating , for instance , a surface with weakly electronegative elements such as hydrogen induces a negative electron affinity , @xcite while termination with strongly electronegative elements can lead to a positive electron affinity . @xcite from eq . ( [ energyreference ] ) it is clear that @xmath62 equals @xmath55 plus a bulk contribution , @xmath63 where @xmath64 denotes the minimum of the conduction band . we can thus use @xmath62 to characterize the potential offset at the surface . there is however a caveat . the long - range potential inside the solid is only specified up to a constant.@xcite typical choices are the cell - averaged potential or the inter - sphere potential of the muffin - tin approximation . for our purpose it will be however more convenient to take the conduction band minimum as the long - range potential inside the solid . this choice is motivated as follows . we are considering a dielectric with a large energy gap . the valence band is thus fully occupied and the conduction band is essentially empty . hence , only the conduction band can be populated by additional electrons coming from the plasma and referencing the electrostatic potential inside the solid to the conduction band minimum allows us to relate the total surplus electron density in the interface region to the potential in the interface region in analogy to what we have done for the plasma sheath in the previous subsection . adopting the above discussion to the one - dimensional model shown in fig . [ figure1 ] and assuming a quadratic dispersion for the conduction band , the energy of an electron in the conduction band is given by @xmath65 where @xmath66 is the total surface potential to be determined in the next subsection and the offset of the electrostatic potential at the surface , @xmath67 encompasses the surface dipole as well as the unspecified bulk contribution . the surface potential of the bare , uncharged surface comprises at least the surface dipole and a long - range contribution , the image potential , resulting from the mismatch of the dielectric constants at the surface . far away from the surface the image potential is given by@xcite @xmath68 but this expression can not be employed for our purpose because the singularity at @xmath2 prohibits a smooth electron distribution across the interface . in reality the image potential has to continuously merge with the crystal potential . equation ( [ imageclassical ] ) is thus also unphysical . to obtain a realistic image potential without performing an atomistically accurate calculation we employ the model of a graded interface . it also has the virtue to be parameterizable with experimentally measured values for the electron affinity , the dielectric mismatch , and the mismatch between effective electron masses . the model incorporates therefore important properties of a surface , most importantly , it accounts for the charge re - distribution due to the truncation of the solid . initially proposed by stern @xcite to remove the unphysical singularity of the image potential at the interface of two dielectrics , the graded interface model assumes the dielectric constant @xmath69 to vary over a distance on the order of a lattice constant . later the model was extended to variations of other physical quantities and applied to semiconductor heterostructures and nanopores . @xcite clearly , because of the interpolation the model can not account for effects associated with intrinsic surface states ( shockley and tamm states @xcite ) and additional surface states which may arise from the short - range surface potential . nevertheless , the graded interface model is a reasonable description of a surface . in the spirit of a graded interface , we assume the dielectric constant @xmath69 , the electron mass @xmath70 , and the potential offset at the surface to vary smoothly according to the grading function @xmath71 where @xmath72 is the half width of the graded interface and @xmath73 stands for the quantity that varies across the interface . we use the value @xmath74 which is an estimate used in previous applications of the graded interface.@xcite while the grading parameter @xmath72 is not based on definite experimental or theoretical results it is motivated by the assumption that the bonding and the electron density at the surface change over one to two lattice constants implying a transition layer for the effective potential and the dielectric constant that is somewhat larger . hence , across the interface the electron mass , the dielectric constant and the offset potential are given by @xmath75 and @xmath76 respectively , with @xmath77 the effective mass of the conduction band . within the model of the graded interface , the image potential is the change in the selfenergy of an electron due to the proximity of the dielectric mismatch . positioning the electron at @xmath78 it is given by @xcite @xmath79~,\end{aligned}\ ] ] where @xmath80 is the potential in the medium with dielectric mismatch arising from the electron at @xmath78 and @xmath81 is the same quantity in a homogeneous medium with dielectric constant @xmath82 . hence , @xmath80 is the solution of @xmath83 while @xmath81 is the solution of @xmath84 to solve eqs . ( [ potmismatch ] ) and ( [ potnomismatch ] ) we follow stern@xcite and make the ansatz @xmath85 placing the electron on the @xmath86 axis , @xmath87 which implies @xmath88 and @xmath89 for @xmath90 . hence , we need to keep only the @xmath91 term , so that @xmath92 where @xmath93 is given by @xmath94 and @xmath95 is the solution of @xmath96 which has to be obtained numerically . the image potential is then given by @xmath97 in contrast to eq . ( [ imageclassical ] ) it is now smoothly varying across the interface with a deep well on the low-@xmath69 side and a small bump on the high-@xmath69 side . the total surface potential comprises the graded offset potential ( [ gradedoffsetpotential ] ) and the graded image potential . hence , @xmath98 it is continuous across the crystallographic interface at @xmath2 and enables us thereby to also calculate a smoothly varying electron distribution in the esl . the band structure and the total surface potential at the graded interface are visualized in the lower panel of fig . [ figure1 ] . using eq . ( [ z0condition ] ) we can now determine the position @xmath8 of the effective wall , that is , the maximum extent of the esl on the plasma side . the derivative of the bare surface potential is @xmath99 . due to the relatively weak field in the sheath compared to the image force , the boundary @xmath8 will be so far away from the interface that @xmath100 vanishes and the image potential obeys ( [ imageclassical ] ) . thus , the boundary between the esl and the plasma sheath is given by @xmath101 with @xmath102 and @xmath103 given by ( [ etawprime ] ) . the plasma - induced wall charge is assumed to be in thermal equilibrium with the wall . hence , the distribution of the excess electrons in the esl has to minimize the excess electron s grand canonical potential in the external potential due to the surface . the coupling to the sheath is maintained by the constraint that only as many electrons can be filled into the esl as are missing in the sheath and the boundary conditions to the poisson equation which links the electron distribution in the esl to the ( internal ) electrostatic potential . to minimize the grand canonical potential of the surplus electrons we follow tkharev and danilyuk @xcite and apply density functional theory @xcite to the graded interface . while more refined schemes of density functional theory@xcite could , in principle , be employed , we will use for the purpose of this exploratory calculation density functional theory in the local approximation . quite generally , the grand canonical potential of an electron system in an external potential @xmath104 is given in the local approximation by @xmath105- \mu \int n(\vec{r } ) \mathrm{d}\vec{r } \text { , } \end{aligned}\ ] ] where @xmath106 $ ] is the grand canonical potential of the homogeneous system with density @xmath107 and the coulomb potential is determined by @xmath108 the ground state electron density minimizes @xmath109 , that is , it satisfies @xmath110 where @xmath111/\delta n$ ] is the chemical potential for the homogeneous system . specifically for the excess electrons in the one - dimensional graded interface eq . ( [ dft ] ) reduces to @xmath112 where @xmath113 and the electrostatic potential , @xmath114 consists of the potential of the bare surface given by eq . ( [ totalphi ] ) and the internal coulomb potential which satisfies poisson s equation , @xmath115 with the graded dielectric constant @xmath116 given by eq . ( [ graded1 ] ) and the boundary conditions @xmath117 and @xmath118 to guarantee continuity of the potential at @xmath8 and to include the restoring force from the positive charge in the sheath . note that the coulomb potential derived from this equation includes the attraction of an electron to the image of the charge distribution . for the functional relation @xmath113 we take the expression adequate for a homogeneous , non - interacting , non - degenerate electron gas , @xmath119 this is justified because the density of the excess electrons is rather low and the temperature of the surface is rather high , typically a few hundred kelvins . in order to calculate the quasi - stationary distribution of the surplus electrons , eqs . ( [ vareq ] ) and ( [ poiseq ] ) have to be solved self - consistently with the additional constraint that the total electron surface density in the esl equals the total surface density of electrons missing in the sheath , that is , @xmath120 with @xmath39 given by eq . ( [ neq ] ) . in the above equation we introduced a cut - off @xmath121 at which the esl terminates inside the dielectric . as long as @xmath122 is chosen large enough it does not affect the numerical results close to the surface . an improved treatment of the esl , avoiding the ad - hoc cut - off , is given in the next section . within the crude esl model developed in this section the computation is performed iteratively in the interval @xmath123 according to the following scheme : ( i ) we start with the potential of the empty surface given by eq . ( [ potdft ] ) with @xmath124 obtained from eq . ( [ poiseq ] ) with @xmath125 but with the boundary conditions at @xmath8 as specified . ( ii ) we integrate both sides of eq . ( [ neq ] ) over @xmath86 with @xmath126 given by eq . ( [ vareq ] ) . enforcing the constraint ( [ nconstraint ] ) determines @xmath127 . ( iii ) using @xmath127 we calculate from eq . ( [ vareq ] ) a new @xmath126 which gives with eq . ( [ neq ] ) a new electron density @xmath128 . lastly , we determine from eq . ( [ poiseq ] ) the electrostatic potential associated with the up - dated @xmath128 . steps ( ii ) - ( iv ) are iterated until @xmath127 , which is far below the conduction band edge because of the non - degeneracy of the excess electrons , converges . in the previous section we have taken into account only the electron concentration in the conduction band of the dielectric due to the electrons coming from the plasma . for wide band gap materials this is justified , especially near the surface , as their concentration is much larger than the intrinsic carrier concentration . deep inside the dielectric , however , charge neutrality is not enabled by a vanishing electron density , but by the electron density decreasing to its intrinsic value , which is then balanced by the intrinsic hole concentration in the valence band . to take this effect into account , which is particularly important when the additional electrons accumulate deep inside the bulk of the dielectric , we divide the esl into two regions : a very narrow interface - specific region ( isr ) and a wide space charge region ( scr ) in the bulk of the dielectric . the parameter @xmath129 denotes now no longer an ad - hoc cut - off but the boundary between the two regions . it has to be chosen so that the isr includes the major effect of the image potential in the dielectric implying @xmath130 . the electron distribution and the potential in the isr are calculated using the density functional approach outlined in the previous section . in the scr we use for simplicity the model of an intrinsic semiconductor to describe electron and hole densities as well as the long - range potential . as the energy bands in the dielectric follow the long - range potential the refined esl also captures the band bending which might be induced by the presence of the wall charge . it is however only significant when most of the excess electrons are trapped in the scr and not in the isr . figure [ figure1a ] schematically shows the electron and hole densities for the refined esl model . the boundary between the plasma sheath and the esl is still located at @xmath8 . as our model does not encompass the electron and ion flux from the plasma for @xmath13 , the densities @xmath131 and @xmath23 are discontinuous at @xmath8 . this is obvious for the ions which are not allowed to enter the solid . the discontinuity of the electron density , in contrast , arises because we ensure only the total number of missing sheath electrons per unit area to be conserved . this global constraint can not guarantee continuity of the electron density at @xmath8 . at the boundary between the isr ( @xmath123 ) and the scr ( @xmath132 ) the electron density and the potential are continuous . in principle , also the hole density @xmath133 should be continuous . as @xmath134 for the materials we are considering , we can however neglect holes in the isr . for the modeling of the scr it is convenient to use @xmath135 for the long range potential , which vanishes for charge neutrality in the bulk . here , @xmath136 ( see below for an explicit relation for @xmath137 ) . then , poisson s equation is given by , @xmath138 where the electron and hole densities for an intrinsic semiconductor with parabolic bands whose extremal points are , respectively , @xmath139 and @xmath140 are given by @xcite @xmath141 from a comparison of the exponents in eq . ( [ compneq ] ) and eq . ( [ neq ] ) , where @xmath142 is given by eq . ( [ vareq ] ) we find @xmath143 far from the surface , @xmath144 and @xmath145 . this gives the chemical potential @xmath146 and the bulk carrier concentration @xmath147 @xmath148 where @xmath149 . hence , poisson s equation becomes @xmath150 and using dimensionless variables @xmath151 with @xmath152 we obtain @xmath153 this equation can be integrated once , which gives @xmath154 the boundary conditions in the bulk @xmath155 and @xmath156 for @xmath157 imply @xmath158 so that eq . ( [ first_integral ] ) becomes @xmath159 integration with the boundary condition at @xmath129 , that is , at @xmath160 , @xmath161 and requiring @xmath32 for @xmath162 gives @xmath163 \end{aligned}\ ] ] with @xmath164 \text { , } \end{aligned}\ ] ] where the upper sign is for @xmath165 and the lower sign for @xmath166 . in analogy to what we have done at the boundary of the esl with the plasma sheath at @xmath5 we relate the potential @xmath167 to the total electron surface density in the space charge region . from poisson s equation we obtain for the total electron surface density in the scr @xmath168 where @xmath169 is given by eq . ( [ etaprime ] ) , so that @xmath170 or , @xmath171 for a negative space charge @xmath165 , so that the potential is given by @xmath172 and the electron and hole densities are given by @xmath173 and @xmath174 , respectively . the relation between @xmath175 and @xmath16 is given by @xmath176 . since @xmath177 we obtain @xmath178 . now , quite generally , the excess electrons in the esl are distributed over the isr and scr according to @xmath179 where @xmath180 is the surface density of electrons in the isr , @xmath127 is the chemical potential in both regions , and @xmath39 is the total surface density of missing sheath electrons given by eq . ( [ neq ] ) . the total surface density in the isr is given by @xmath181 , where @xmath128 is calculated with the density functional approach for the graded interface . requiring continuity of the electron density at @xmath129 , @xmath182 gives @xmath167 as a function of @xmath127 . from @xmath167 we finally obtain using eq . ( [ nsc ] ) @xmath183 . for the calculation of the electron distribution and the potential in the refined esl we use the iteration cycle described in the last section with one modification . in step ( ii ) we solve eq . ( [ ndistrib ] ) instead of eq . ( [ nconstraint ] ) to fix @xmath127 . from @xmath127 we obtain using eq . ( [ ncontatzs ] ) @xmath167 which in turn determines the electron distribution and the potential in the scr . this gives for each iteration step a continuous potential and electron distribution at @xmath129 . as before , the steps ( ii ) -(iv ) are iterated until @xmath127 converges . at the end of this section let us finally mention two simplifications of the refined esl model which could be used , respectively , for large band gap dielectrics irrespective of the electron affinity and dielectrics with small band gap and positive electron affinity . in the former case the intrinsic carrier concentration @xmath147 is very low and the merger with the bulk occurs very deep in the dielectric . almost all surplus electrons are however much closer to the surface where the holes can be neglected . this can be seen from the differential equation for @xmath184 . for small @xmath147 eq . ( [ et0eq ] ) gives a large @xmath167 . as @xmath184 satisfies a highly nonlinear differential equation ( [ diffeqetabb ] ) , a large @xmath167 implies a steeper gradient of @xmath184 near the surface so that almost all electrons are concentrated close to the surface where neglecting the holes has little effect . hence , for large band gap dielectrics surplus electrons can be filled into a sufficiently large isr for which the crude esl model of the previous section will be sufficient provided the cut - off @xmath129 is large enough . the merger with the bulk is of course not correctly captured by such an approach . for dielectrics or semiconductors with small energy gaps and positive electron affinity , on the other hand , almost all surplus electrons are deep inside the material . it is thus a good approximation to neglect the isr and to fill all electrons in a scr . neglecting the surface potential has little effect in this case and using the scr already for @xmath185 gives a good description of the electron distribution inside the esl . the electron density and potential at the surface and in front of it can of course not be captured by such an approach . as before @xmath186 with @xmath187 where @xmath188 is now the limit of the long range potential just inside the dielectric given by @xmath189 . .material parameters for the dielectrics considered in this work : dielectric constant @xmath190 , electron affinity @xmath62 , conduction band effective mass @xmath77 , valence band effective mass @xmath191 , band gap @xmath192 . [ cols="<,<,<,<,<,<",options="header " , ] [ kttable ] let us now turn to the discussion of the influence of the electron temperature @xmath193 and the plasma density @xmath17 on the properties of the esl . these two parameters enter through the total surface density of electrons @xmath39 depleting the sheath and accumulating in the esl . how @xmath193 and @xmath17 affect the interface depends therefore on the sheath model and the model used for the interaction between plasma particles and the surface . for simplicity we have used a collisionless sheath model and assumed the surface to be a perfect absorber for plasma electrons and ions . the results for the properties of the esl as a function of the plasma parameters are thus to be taken only indicative . at a sio@xmath0 surface in contact with a helium discharge as a function of the electron temperature ( upper panel , @xmath194 ) and the plasma density ( lower panel , @xmath59 ) for @xmath195 . the refined esl model without an isr was employed to produce the data . ] the effect of a variation of @xmath17 and @xmath196 is most significant for surfaces with positive electron affinity . n0table ] shows the effect of the plasma density @xmath17 for a sio@xmath0 surface . if @xmath17 increases , the boundary @xmath8 between sheath and esl moves closer to the surface this , however , does not affect the charge distribution much as most of the electrons occupy the scr inside the dielectric ( as shown in fig . [ figure3 ] for al@xmath0o@xmath1 ) . more important is that an increase in @xmath17 leads to an increase of the total surface electron density @xmath39 . this entails a stronger restoring force from the plasma sheath so that the potential well confining the space charge inside the dielectric becomes steeper and the electrons in the scr of the esl are shifted towards the surface , in other words , the @xmath197 value increases with @xmath17 . mathematically , the steeple - like shape of the electron distribution arises because a larger @xmath39 leads through eq . ( [ et0eq ] ) to a larger @xmath167 which makes the potential steeper at the surface so that the electron distribution is more peaked there . this trend can be seen in the lower panel for fig . [ figure4 ] . and @xmath59 ) . the data were obtained from the crude esl model . ] a variation of the electron temperature @xmath193 has similar effects as the variation of the plasma density . if @xmath196 increases , the total surface density of electrons increases also , as can be seen from table [ kttable ] . as shown in the upper panel of fig . [ figure4 ] , this leads again to a steeple - like electron distribution which is the more concentrated at the surface the higher the electron temperature is . for a surface with negative electron affinity ( mgo ) the surplus electrons are strongly bound in the image potential . while a variation of @xmath196 or @xmath17 changes the total number of surplus electrons per unit area in the same way as for a surface with positive electron affinity , the distribution of the surplus electrons within the esl is not affected significantly because of the strong image interaction . so far we have shown the potential and the electron distribution in the esl . now , we will compare potential and charge distribution in the esl with the ones in the plasma sheath . the electron distribution at the interface is the quasi - stationary electron gas on top of the charge re - distribution due to the truncation of the solid that guarantees flux equality at the sheath - esl boundary @xmath8 . as already mentioned not included in this simple model is the flux of plasma electrons and ions in the esl before the electrons are trapped at the surface and the ions recombine with the negative wall charge . the electron and ion densities in this model are thus discontinuous at @xmath8 . the potential , however , which has been obtained from the integration of poisson s equation is continuous and differentiable everywhere . between the well of the image potential and @xmath8 the electron and ion flux from the sheath would be important . the neglect of the charge densities associated with these fluxes does however not affect the potential because they are too small to have a significant effect . surface in contact with a helium discharge ( @xmath194 and @xmath59 ) . the refined esl model without an isr was employed to produce the data . ] figure [ figure6 ] shows the esl and the plasma sheath in front of a mgo surface . due to the negative electron affinity , the plasma - supplied surface electrons are bound by the image potential in front of the surface . in fig . [ figure6 ] , we plot the electron and ion density ( upper panel ) , as well as the electric potential ( lower panel ) over the distance from the surface @xmath86 . far from the surface , the potential approaches the bulk plasma value chosen to be zero . in the sheath the potential develops a coulomb barrier and reaches the wall potential @xmath198 at @xmath8 , the distance where the sheath merges with the surface layer ( vertical dotted line ) . the wall potential is the potential just outside to which the energies of the bulk states are referenced . closer to the surface the potential follows the attractive image potential while at the surface the repulsive potential due to the negative electron affinity prevents the electron from entering the dielectric ( only scarcely seen on the scale of the figure ) . in fig . [ figure6a ] we finally plot the electron and ion densities ( upper panel ) as well as the electric potential ( lower panel ) for sio@xmath0 . note the linear z - axis in contrast to the logarithmic z - axis of fig . [ figure6 ] . due to the positive electron affinity , the excess electrons constituting the wall charge penetrate deep into the dielectric and occupy therefore the scr of the esl . compared to the variation of the electric potential in the sheath the band bending in the dielectric induced by the wall charge is rather small as indicated by the variation of @xmath16 inside the dielectric . this is because @xmath69 is large and the width of the scr is narrow on the scale of the sheath . only on the scale of the isr ( a vertical line at @xmath2 ) the scr of the esl is wide . we have studied the potential and the charge distribution across the interface of a plasma and a dielectric wall treating the plasma - induced wall charge as a quasi - stationary electron gas trapped by and in thermal equilibrium with the dielectric . our approach is based on a model for a graded surface including the offset between the potential just outside the dielectric and the conduction band minimum arising from the re - distribution of charge due to the truncation of the solid as well as the image potential due to the dielectric mismatch at the boundary . the missing electrons from the sheath populate the interface potential and thereby form an electron surface layer ( esl ) which minimizes the grand canonical potential of wall - thermalized excess electrons and satisfies poisson s equation . within this model the boundary between the plasma sheath and the esl is given by the distance from the crystallographic interface where the potential for the excess electrons turns from the repulsive sheath potential into the attractive surface potential . this distance is typically on the order of a micron . it gives the position of an effective wall for plasma electrons and ions and thus the portion of the esl which lays in front of the surface . most of the surplus electrons trapped in the esl , that is , the plasma - induced wall charge , will be , however , much closer to the surface or even inside the dielectric depending on the electron affinity . we presented numerical results for the potential and the distribution of the plasma - supplied surplus electrons at the interface between a helium discharge and the surfaces of mgo , al@xmath0o@xmath1 , and sio@xmath0 , respectively . the electron distribution within the esl strongly depends on the electron affinity . for negative electron affinity , the conduction band minimum is above the potential just outside the dielectric . hence , it is energetically unfavorable for electrons to penetrate into the bulk and the surface electrons are bound in the image potential in front of the surface . in this case , their spatial profile changes little over a variation of the surface temperature or the plasma parameters . for positive electron affinity the conduction band minimum is below the potential just outside the dielectric and the surface - bound electrons accumulate inside the wall . the space charge in the bulk broadens if the surface temperature is increased and becomes more peaked if the total surface density of the electrons missing in the sheath is raised through either an increase in the plasma density or the electron temperature . separating the esl into an interface specific and a space charge region and modeling the bulk of the dielectric as an intrinsic semiconductor we also investigated how the esl merges with the bulk of the dielectric . this is particularly important for dielectrics with small energy gaps and positive electron affinities where excess electrons coming from the sheath accumulate not in the image potential in front of the surface but deep inside the wall . in this case the wall charge may also induce a significant band bending . whereas the crude esl model we proposed neglects the space charge deep inside the bulk of the wall and is thus only applicable to large band gap dielectrics with negative electron affinity where basically the whole plasma - induced wall charge is trapped in the image potential in front of the surface , the refined esl model keeping the interface specific as well as the space charge region of the esl provides a quantitative description of the whole spatial structure of the extended charge double - layer which forms at a dielectric plasma wall as a result of the electrons in the esl and the positive space charge in the plasma sheath . the esl can be regarded as that part of the plasma sheath which is inside the plasma wall . it is thus the ultimate boundary of a bounded gas discharge and constitutes , depending on the electron affinity either a quasi two - dimensional electron plasma in front of the wall or an electron(-hole ) plasma inside the wall .

the quantization of polarized gravitational cylindrical waves has received a lot of attention in recent years @xcite . this is partly due to the fact that this system provides a tractable , yet non - trivial , reduction of full general relativity and hence is an ideal framework to explore several issues involved in the quantization of gravity . some intriguing phenomena , related to the existence of large quantum gravity effects , have been discussed by studying precisely this model @xcite . it has also been argued that some manifestations of quantum gravity , such as the smearing of light cones are , indeed , present and can be understood in this simplified setting @xcite . one of the interesting points behind the obtained results is the realization of the fact that the physical hamiltonian is a function of the free field hamiltonian for a 2 + 1 dimensional , axially symmetric , massless scalar field evolving in an auxiliary minkowski background @xcite . as we show in the first section of the paper , this free hamiltonian naturally appears when one linearizes the system , thus suggesting that , in a precise sense , it can be considered as the free part of an interacting model . however the full interacting hamiltonian is obtained by adding a very specific type of terms to the free part , namely , just functions of it . here we plan to explore the consequences of this functional relation between the two physically relevant hamiltonians for the system and explore how this affects the causal structure of quantum spacetime . to this end we will pay attention to the smearing of the light cones due to quantum gravity effects within the framework of linearly polarized cylindrical waves . instead of considering the full information encoded in the metric tensor we will concentrate on the causal structure provided by light cones . an interesting , albeit somewhat indirect way , to look at this structure is to study the commutators of field operators at different spacetime points . these are the basic objects to discuss the commutativity of observables and the microcausality of the model ; conventionally , a physical model should be such that observables commute for space - like intervals . this has been discussed in the standard perturbative quantum field theory framework for simple examples such as scalar or fermion fields ( see e.g. @xcite ) . in fact , microcausality is one of the key conditions to prove such important results as the spin - statistics theorem @xcite . here we will use the commutator of the scalar field that describes linearly polarized cylindrical waves as a way to get information about the causal structure of quantum spacetime . as we show later it is possible to give exact expressions for this commutator both for the evolution provided by the free and the full physical hamiltonians . we will use these expressions to study in a quantitative way the smearing of the light cones as a function of the three - dimensional gravitational constant and explore some physical issues , in particular the appearance of singularities as a consequence of having a hamiltonian bounded from above @xcite . the rest of the paper is structured as follows . in sec . [ lingrav ] we discuss how the free hamiltonian is derived from the linearized cylindrical wave model . [ evolutions ] deals with the classical and quantum dynamics of cylindrical gravitational waves under the evolution provided both by the free hamiltonian and the physical hamiltonian . we will pause here to discuss and compare on a familiar example ( the harmonic oscillator ) the main features of the time evolution defined by functionally related hamiltonians , both from the classical and the quantum points of view . this will provide valuable insights for the problem considered in this work . [ microcausality ] is devoted to the study of microcausality . we will look at the main features of the field commutators and study the smearing of the light cones due to quantum gravitational effects . we end the paper with a discussion of the main results and perspectives for future work . linearly polarized cylindrical waves in general relativity can be described by the spacetime metric @xcite : @xmath0 where @xmath1 is the coordinate of the symmetry axis and @xmath2 is the three - metric @xmath3 from this three - dimensional point of view , @xmath4 and @xmath5 correspond to polar coordinates , @xmath6 is the radial component of the shift vector and @xmath7 is the lapse function . all metric functions ( @xmath8 , @xmath9 , @xmath7 , and @xmath6 ) depend only on the time and radial coordinates , @xmath10 and @xmath11 . unless otherwise stated , we adopt a system of units such that @xmath12 , where @xmath13 is the speed of light , @xmath14 is planck constant , and @xmath15 is the effective newton constant per unit length in the direction of the symmetry axis @xcite . in these units , the gravitational action of the system has the form @xcite @xmath16,\ ] ] where the dot denotes the derivative with respect to @xmath10 , the @xmath17 s are the momenta canonically conjugate to the metric variables and @xmath18 is the total hamiltonian @xmath19.\end{aligned}\ ] ] the first term is a boundary contribution at infinity [ @xmath20 and the second term is a linear combination of the hamiltonian constraint @xmath21 and the ( radial ) diffeomorphisms constraint @xmath22 : @xmath23 , \nonumber\\ c^r&=&e^{-\gamma}\left(-2p^{\prime}_{\gamma}+ p_{\gamma}\gamma^{\prime}+p_rr^{\:\prime}+p_{\psi}\psi^{\prime } \right).\end{aligned}\ ] ] the prime denotes the derivative with respect to @xmath11 . the gauge freedom associated with these constraints can be removed by imposing , respectively , the gauge fixing conditions @xcite @xmath24 on the other hand , the lagrangian form of the action can be obtained from the relations between momenta and time derivatives of the metric provided by the hamilton equations , @xmath25 from a three - dimensional perspective , the system describes an axially symmetric model consisting of a scalar field @xmath26 coupled to gravity @xcite , the line element being ( [ 3met ] ) . a particular classical solution is a vanishing scalar field in three - dimensional minkowski spacetime or , equivalently , minkowski spacetime in four dimensions . in this solution , @xmath27 and @xmath28 , whereas the rest of metric fields and momenta vanish ( i.e. @xmath29 ) . in this section , we will consider this solution as a background and discuss first - order perturbations around it . in other words , we will analyze the linearized theory of gravity around this minkowski spacetime , as it is usually done in the perturbative , quantum field theory approach to gravity . in order to expand the metric fields around the classical solution , let us call @xmath30 up to first - order terms in the fields , the expression of the three - metric becomes @xmath31 while the four - dimensional metric is given by @xmath32 here , it is understood that the product of @xmath26 with any other metric field vanishes in the perturbative order considered . on the other hand , regularity on the axis of symmetry imposes the following conditions : @xmath33 in order to discuss the linearized gravitational system , we must keep up to quadratic terms in the fields in the action ( [ act ] ) . a straightforward calculation leads to the result @xmath34,\nonumber\\ \bar{\cal h}&=&\int_0^\infty dr \left [ \frac{p_{\psi}^2}{2r}+\frac{r\,(\psi^{\prime})^2}{2}-p_{\gamma } p_{\bar{r}}+\bar{n}\bar{c}+n^r\bar{c}^r\right]\nonumber\\ & & + ( 2-\gamma_{\infty})\,\bar{r}^{\,\prime}(r\rightarrow\infty).\end{aligned}\ ] ] here , @xmath35 is the momentum canonically conjugate to @xmath36 , and the linearized constraints are : @xmath37 the diffeomorphisms gauge freedom can be fixed just like in the cylindrical reduction of general relativity , namely , by demanding that @xmath38 coincides with the radial coordinate @xcite . we thus impose the gauge fixing condition @xmath39 . it is easily checked that the poisson brackets @xmath40 of this condition with the linearized constraint do not vanish , so that the gauge fixing is admissible . dynamical consistency of the gauge fixing procedure requires , in addition , @xmath41 hence , the shift must be chosen as @xmath42 . finally , the momentum conjugate to @xmath36 is fixed by solving the diffeomorphisms constraint : @xmath43 . in this way , the canonical pair @xmath44 is removed from the set of degrees of freedom . the action of the resulting reduced model is @xmath45.\end{aligned}\ ] ] here , @xmath46 is the hamiltonian constraint of the reduced linearized system , and @xmath47.\ ] ] remarkably , the condition employed to eliminate the hamiltonian gauge freedom in full cylindrical gravity @xcite can be used as well to fix the corresponding gauge in the linearized theory . the gauge fixing @xmath48 is acceptable , because the poisson brackets of @xmath49 and @xmath50 differ from zero . in addition , consistency of the chosen gauge demands the vanishing of @xmath51 therefore , @xmath52 has to be independent of the radial coordinate . actually , we can set @xmath53 by demanding that the total lapse equals the unity at spatial infinity . on the other hand , taking into account the regularity condition ( [ reg ] ) , the solution to the hamiltonian constraint is simply @xmath54 . this allows us to remove the canonical pair @xmath55 from the system and arrive at a constraint - free model in linearized gravity . the degrees of freedom of this system are the field @xmath26 and its momentum . the reduced action is @xmath56.\end{aligned}\ ] ] note that @xmath57 , given in eq . ( [ ham0 ] ) , is the hamiltonian of a massless scalar field with axial symmetry in three - dimensions . furthermore , in the gauge that we have selected , the three - dimensional metric of the linearized gravitational theory is exactly that of minkowski spacetime and contains no physical degrees of freedom . the scalar field @xmath26 determines the norm of the killing vector @xmath58 , and appears in the four - dimensional metric of the gauge - fixed , linearized model in the form ( [ metlin ] ) , but with @xmath59 substituted by the flat metric @xmath60 where we have renamed @xmath61 the time coordinate of the reduced system . summarizing , the perturbative description provided in linearized gravity for cylindrical waves with linear polarization around four - dimensional minkowski spacetime is equivalent to a massless scalar field with axial symmetry in a three - dimensional flat background . the dynamics of this field is dictated by the free hamiltonian @xmath57 , which generates the evolution in the minkowskian time @xmath61 . it is worth noticing that the action and metric of the gauge - fixed model in linearized gravity reproduce in fact the results that one would obtain from the gauge - fixed model in full cylindrical gravity by working just in the first perturbative order , i.e. , by keeping in the action and metric , respectively , up to quadratic and linear terms in the field @xmath26 and its momentum . in this sense , the gauge fixing and linearization procedures commute . one of the significant features of gravitational cylindrical waves is the existence of two distinct , physically relevant hamiltonians ( or equivalently two distinct time coordinates ) to define both the classical and quantum evolution . one is the hamiltonian @xmath57 [ given in eq . ( [ ham0 ] ) ] that generates the dynamics in the linearized gravitational theory ; the other is the hamiltonian @xmath62 that provides the energy per unit length along the symmetry axis in general relativity @xcite . in fact , they are functionally dependent , since @xmath63 . in order to gain insight into the relation that can be established between the dynamics associated with these two different hamiltonians , we open this section by discussing a similar situation in a very simple example provided by the harmonic oscillator . the usual description of the harmonic oscillator in a phase space coordinatized by @xmath64 comes from its standard hamiltonian @xmath65 . the dynamics is given by the hamilton equations @xmath66 the general solution can be written as @xmath67 where @xmath68 and its complex conjugate @xmath69 are fixed by the initial conditions . consider next a phase space @xmath70 with hamiltonian @xmath71 , i.e. a function of the standard hamiltonian for the harmonic oscillator . for instance , the case @xmath72 arises in the context of quantum optics , in relation with the propagation of light in non - linear kerr media @xcite . the equations of motion now read @xmath73 where @xmath74 denotes the derivative of @xmath75 with respect to its argument . these new equations can be easily solved by means of a change of time . specifically , making use of the time independence of the hamiltonian @xmath76 ( @xmath77 on solutions to the equations of motion ) , we can introduce the new time parameter @xmath78 . then , the functions @xmath79 $ ] , @xmath80 $ ] and the new time @xmath61 serve us to transform the hamilton equations into the standard ones corresponding to the harmonic oscillator . so , the classical solutions are @xmath81\!=\!\frac{1}{\sqrt{2\omega}}\!\left[ae^{\ ! -i\omega f'(\omega a^\dagger\!a)t}\!+\!a^\dagger e^{i\omega f'(\omega a^\dagger\!a)t}\right]\!,\\ p(t)\!=\!p_0[t(t)]\!=\!\frac{-i\omega}{\sqrt{2\omega}}\ ! \left[ae^{\!-i\omega f'(\omega a^\dagger\!a)t}\!-a^\dagger e^{i\omega f'(\omega a^\dagger\!a)t}\right]\!.\end{aligned}\ ] ] what we find is an energy dependent redefinition of time that induces a different time change for each solution to the equations of motion . the situation in the quantum theory is quite different . the reason lies on the fact than in quantum mechanics a physical state does not need to have a definite energy . then , a `` change of time '' of the form @xmath82 has non - trivial consequences for the dynamics . the usual quantum theory for the harmonic oscillator can be described by introducing a fock space with creation and annihilation operators @xmath69 , @xmath68 . every initial state can be expressed as @xmath83 , where @xmath84 are fourier coefficients ( with the convenient normalization ) and @xmath85 are energy eigenvectors ( that is , @xmath86 ) . in the schrdinger picture , evolving with @xmath87 , we find @xmath88 however , if the same state @xmath89 evolves in time according to the evolution generated by @xmath90 , we get and @xmath90 act on the same hilbert space . moreover , if we demand the function @xmath75 to have a unique absolute minimum at @xmath91 , the two hamiltonians have also the same vacuum . ] @xmath92 hence we do not recover an analogous situation to that found in the classical system by replacing ( formally ) the time @xmath61 in @xmath93 with @xmath82 , because @xmath94 except for linear homogeneous functions @xmath75 . moreover , the properties of the states @xmath93 and @xmath95 are quite different . for example , if we consider the bounded hamiltonian @xmath96 it is obvious that the high energy contributions to @xmath95 are essentially frozen in time with respect to the evolution of the low energy ones , in the sense that the phase of the former type of contributions remains practically coherent in time . let us turn now to the discussion of our gravitational system . we will deal with the einstein - rosen family of linearly polarized gravitational waves @xcite . these waves display the so - called whole cylindrical symmetry @xcite , namely , they correspond to topologically trivial spacetimes which possess two linearly independent , commuting , spacelike , and hypersurface orthogonal killing vector fields . it is well known @xcite that these spacetimes admit coordinates @xmath97 such that the metric is given by @xmath98 where @xmath26 and @xmath8 are functions of only @xmath11 and @xmath61 . when the einstein equations are imposed , @xmath26 encodes the physical degrees of freedom and satisfies the usual wave equation for an axially symmetric massless scalar field in three - dimensions : @xmath99 the metric function @xmath8 can be expressed in terms of this field on the classical solutions @xcite . one gets @xmath100,\nonumber\\ \label{gaminf } \gamma_\infty&=&\frac{1}{2}\int^\infty_0 dr\ , r\ , \left [ ( \partial_t\psi)^2+(\partial_r\psi)^2\right].\end{aligned}\ ] ] note that @xmath101 and @xmath102 are the energy of the scalar field in a ball of radius @xmath11 and in the whole of the two - dimensional flat space , respectively . furthermore , @xmath103 coincides with the hamiltonian @xmath57 given in eq . ( [ ham0 ] ) @xcite . nevertheless , to reach a unit asymptotic timelike killing vector field in the actual four - dimensional spacetime , with respect to which one can truly introduce a physical notion of energy ( per unit length ) @xcite , one must make use of a different system of coordinates , namely @xmath104 where @xmath105 . in these new coordinates the metric has the form @xmath106 assuming as a boundary condition that the metric function @xmath26 falls off sufficiently fast as @xmath107 , the above metric describes asymptotically flat spacetimes with a , generally non - zero , deficit angle . in this asymptotic region @xmath108 is a unit timelike vector . the einstein field equations can be obtained from a hamiltonian action principle @xcite where the on - shell hamiltonian is given in terms of that for the free scalar field by @xmath109 owing to these reasons , in the following we will refer to @xmath10 as the physical time and to @xmath62 as the physical hamiltonian . when we use the @xmath61-time and impose regularity at the origin @xmath110 @xcite , the classical solutions for the field @xmath26 can be expanded in the form @xmath111.\nonumber\end{aligned}\ ] ] @xmath112 and @xmath113 are fixed by the initial conditions and are complex conjugate to each other , because @xmath26 and @xmath114 ( the zeroth - order bessel function of the first kind ) are real . from eq . ( [ gaminf ] ) , we then obtain @xmath115 using this formula , we can express the field in the @xmath10-frame as @xmath116.\;\nonumber\end{aligned}\ ] ] notice that @xmath117 . in principle , the quantization of the field @xmath26 can be carried out in a standard way . we can introduce a fock space in which @xmath118 , the quantum counterpart of @xmath119 , is an operator - valued distribution @xcite . its action is determined by those of @xmath120 and @xmath121 , the usual annihilation and creation operators , whose only non - vanishing commutators are @xmath122=\delta(k_1,k_2).\ ] ] explicitly , @xmath123\!. \label{psi0}\ ] ] in the schrdinger picture , operators do not evolve , and the problem of the time evolution is transferred to the physical states . we will return later to this issue . in the heisenberg picture , on the contrary , operators change in time and states remain fixed . in this case , the value of the quantum field @xmath124 at any time can be obtained from its value at @xmath125 , evolving in one of the two times that we have at hand . one is the physical time @xmath10 , with associated hamiltonian @xmath62 given in eq . ( [ newham ] ) . the other is the time @xmath61 of the auxiliary three - dimensional minkowski spacetime , its hamiltonian being that of a massless scalar field , @xmath57 . from our discussion in sec . ii , this time can be identified with the perturbative time ( denoted also by @xmath61 ) that arises in the study of the einstein - rosen waves in linearized gravity . in the perturbative time @xmath61 , the evolution is provided by the unitary operator @xmath126 where @xmath127 is the quantum hamiltonian of an axially symmetric scalar field in three dimensions . then @xmath128\!. \hspace*{.3cm}\end{aligned}\ ] ] the quantization procedure in this case is very simple : we have substituted the initial conditions @xmath112 and @xmath113 in the classical solution by the corresponding quantum operators . the situation changes when we choose the physical time @xmath10 as the time parameter . the quantum hamiltonian can be defined as @xmath129 . we can then reach a unitary evolution by means of @xmath130 . this leads to the following time evolved operators @xmath131\!\hat{a}(k),\nonumber\\ \label{crean } \hat{a}_e^\dagger(k , t)\!\!&=&\!\hat{a}^\dagger(k)\,\exp\!\left[ite(k ) e^{-\hat{h}_0/2}\right],\end{aligned}\ ] ] where @xmath132 . it is important to realize that the quantum evolution in the physical @xmath10-frame is not obtained by changing @xmath112 , @xmath113 , and @xmath103 by their direct quantum counterparts . in fact , by restoring the value of the dimensionful constants @xmath14 and @xmath15 , we can write @xmath133 with @xmath134 , so that @xmath135 and we can expand @xmath136 and @xmath137 in powers of @xmath14 , @xmath138 ( here , we have expressed @xmath139 with dimensions of an inverse length ) setting again @xmath140 , the expansion above clearly shows that the quantum evolution of the creation and annihilation variables in the physical time differs from the classical evolution " in higher - order quantum corrections . this unusual behavior can be partially corrected in the sense that one can actually find an operator for @xmath62 such that the quantum evolution in the @xmath10-time is similar to the classical one . in fact , if one considers the normal ordered hamiltonian @xmath141 and its associated unitary evolution operator @xmath142 , it is easy to prove that @xmath143 therefore , the quantum evolution in @xmath10 given by @xmath144 parallels that encountered in the classical picture . unfortunately , there is a severe problem with this quantum dynamics that renders it physically unacceptable : the hamiltonian @xmath144 is unbounded both from above and below @xcite . obviously , the different possibilities for the unitary time evolution of the field @xmath124 considered here are closely related . specifically , there exists a unitary mapping between both types of evolution . this is a consequence of the fact that they are both unitary and coincide with the identity in the same hilbert space at @xmath125 . so , if we denote the operators evolved from @xmath145 in the times @xmath61 and @xmath10 by @xmath146 and @xmath147 , respectively , we obtain @xmath148 then , we can go from one evolution to the other by means of the unitary operator @xmath149 . let us close this section with a few comments about the fock space on which the field @xmath124 acts . in the heisenberg picture the states do not depend on time . they are constructed by successive actions of the time independent creation operators on the vacuum of the theory . it is important to point out that both @xmath139 and @xmath150 act on the same hilbert space and have the same vacuum , that will be referred to as @xmath151 . explicitly , given any square integrable complex function @xmath152 ( with the convenient normalization ) we can write a @xmath153-particle state in the form @xmath154 according to the usual interpretation of quantum mechanics , the measurable physical quantities correspond to expectation values of observables . we can go over to the schrdinger picture by assigning the time evolution to the quantum states @xmath155 defining @xmath156 , @xmath157 , and noticing that the @xmath158-operators satisfy @xmath159 it is straightforward to see that the evolved states are solutions to the schrdinger equations @xmath160 as in the heisenberg picture , the unitary operator @xmath161 provides the bridge between the two kinds of quantum evolution , @xmath162 particularizing the above results to the case of @xmath153-particle states we readily get that , for the @xmath61-time , @xmath163 where @xmath164 . notice that @xmath165 is a superposition of eigenvectors of @xmath139 with eigenvalues equal to @xmath166 . on the other hand , if we evolve the states with @xmath167 , it is not difficult to check that @xmath168 in other words , @xmath169 is a superposition of eigenvectors of @xmath150 , each of them with energy @xmath170 . finally , it is worth pointing out that the @xmath150-energy is not additive : @xmath171 this property is directly related to the existence of an upper bound for the physical hamiltonian . microcausality plays a crucial role in perturbative quantum field theory ; in fact it is a crucial ingredient in such important issues as the spin - statistics theorem @xcite . the point of view that we will develop in this section is the idea that field commutators of the scalar field that encodes the physical information in linearly polarized cylindrical waves can be used as an alternative to the metric operator to extract physical information about quantum spacetime . some relevant information may be lost , but the availability of explicit exact expressions for commutators [ even under the evolution given by the hamiltonian ( [ newham ] ) ] opens up the possibility of getting precise information about the quantum causal structure of spacetime . in particular we will see how the light cones get smeared by quantum corrections in a precise and quantitative way . as is well known , the study of causality in perturbative quantum field theory requires the consideration of measurements of observables at different spacetime points and their mutual influence . the key question is whether measurements taken at spatial separations commute or not . the relevant commutators can be seen to be proportional to those of the quantum fields at two different spacetime points @xmath172 and @xmath173 . for a free massless scalar field described by the standard lagrangian , this commutator is a ( c - number ) function of @xmath172 and @xmath173 that is exactly zero when @xmath174 is space - like @xcite . this means that observables at points separated by spatial intervals commute . something similar happens for fermions described by the usual lagrangians if , instead of commutators , one takes anticommutators of the fields ( observables for fermion fields can be written as even powers of them and the relevant commutators can be written in terms of anticommutators of the basic fields ) @xcite . in the case that we are considering in this work the only physical local degree of freedom that we have is given by the scalar field @xmath26 . what we will do in the following is discussing microcausality by considering the different time evolutions introduced in the first part of the paper . we start by computing @xmath175 $ ] for the field operators ( [ hpsit ] ) obtained in the heisenberg picture by evolving the field at time @xmath176 with the hamiltonian @xmath139 given in eq . ( [ qham0 ] ) . it is straightforward to get ( see @xcite for a somewhat related computation ) @xmath177=\hspace{4.cm}\nonumber\\ \label{s2 - 004 } i\int_0^\infty\!\!\!dk j_0(r_1k)j_0(r_2k ) \sin[(t_2-t_1 ) k ] .\hspace*{.3cm}\end{aligned}\ ] ] let us discuss the main features of this commutation function , that we will refer to in the following as the @xmath57-commutator . to begin with we can easily see that , as it happens for the familiar free field theories , the commutator ( [ s2 - 004 ] ) is a c - number , i.e. it is proportional to the identity in the fock space . in addition , for @xmath178 fixed , let us call regions i , ii , and iii the regions of the @xmath179 plane defined , respectively , by @xmath180 , @xmath181 , and @xmath182 . then , it can be shown that the @xmath57-commutator vanishes in region i , whereas in region ii it can be written as @xcite @xmath177=\hspace*{4.4 cm } & & \nonumber \\ \label{propii } \frac{i}{\pi}\frac{1}{\sqrt { r_1 r_2}}\,k\left(\sqrt{\frac{(t_2\!-\!t_1)^2\!-\!(r_2\!-\!r_1)^2 } { 4r_1r_2}}\right).\hspace*{1.cm}&&\end{aligned}\ ] ] finally , its value in region iii is @xcite @xmath183}}&&\hspace*{-.4 cm } k\!\!\left(\!\!\sqrt{\!\frac{4r_1r_2}{(t_2\!-\!\!t_1)^2\!\!- \!(r_2\!-\!\!r_1)^2}}\right)\!.\nonumber\end{aligned}\ ] ] here , @xmath184 denotes the complete elliptic integral of the first kind , @xmath185 ( alternatively , the integral in eq . ( [ s2 - 004 ] ) can be written in terms of the associated legendre functions @xmath186 and @xmath187 @xcite ) . plane for the @xmath57-commutator . region i corresponds to @xmath188 , region ii to @xmath181 , and region iii to @xmath189 . the singularity of the @xmath57-commutation function lies in the boundary between regions ii and iii , whereas the singularity for the @xmath190-commutator appears for @xmath191.,width=321 ] several plots of the absolute value of this function ( in fact , of its imaginary part ) for fixed values of @xmath178 and @xmath192 are shown in fig . [ fig : freeprop2 ] , and a three - dimensional plot for a fixed value of @xmath178 can be found in fig . [ fig : freeprop ] . of the commutator of the scalar field @xmath175 $ ] for @xmath193 and several values of @xmath192 as a function of @xmath194.,width=321 ] of the commutator of the scalar field @xmath175 $ ] for @xmath193 as a function of @xmath194 and @xmath192 . the light cone structure and the singularity at @xmath195 can be readily seen . the plot for negative values of @xmath194 is obtained by reflecting with respect to the @xmath192 axis.,width=321 ] some interesting properties can be read off from the previous expressions . first of all we see that the commutator is in fact identically zero in the region labeled i ; outside it differs from zero . second we see that it is singular in the line @xmath195 . it is easy to check that the singularity is logarithmic by using the explicit form of the propagator ( [ propii ] ) and expanding around @xmath195 . this singularity can be understood as a consequence of the presence of the symmetry axis ; this is supported by the fact that the @xmath57-commutation function satisfies @xmath196=0\,,\nonumber\\ & & \left(\!\partial_t^2-\frac{1}{r}\partial_r(r\partial_r)\!\right)\!\cdot \theta(t - t_2)[\hat{\psi}(r , t),\hat{\psi}(r_2,t_2)]=\nonumber\\ & & \hspace{3.cm}\frac{i}{r_2}\delta^{(2)}\!(r - r_2,t - t_2)\,,\nonumber\end{aligned}\ ] ] which shows that the full commutator is an axially symmetric solution to the 2 + 1 dimensional wave equation and the future part of it ( obtained by multiplying by a step function ) is an axially symmetric green function for the same equation . it is important to point out that even though there is a minkowskian background metric , the presence of a center of symmetry breaks lorentz invariance ; in fact , the only symmetries of the 2 + 1 axially symmetric wave equation @xmath197 that transform solutions @xmath198 into solutions are \i ) translations @xmath199 \ii ) dilatations @xmath200 \iii ) inversions @xmath201 } { \sqrt{1\!-\!2t\epsilon\!+\!(t^2\!-\!r^2)\epsilon^2}}\!\quad(\epsilon\textrm { small enough})\end{aligned}\ ] ] with @xmath202 these can be obtained systematically by using the general theory of symmetry groups for partial differential equations ( see @xcite ) . let us consider now the commutator of the field operators obtained in the heisenberg picture with the quantum hamiltonian @xmath203 . here @xmath204 , and we have restored the values of the planck constant and the three - dimensional gravitational constant ( but kept @xmath205 ) to have the possibility of discussing the semiclassical limit . recall that the hamiltonian @xmath139 is given by eq . ( [ qham0 ] ) , where we choose the normalization of the creation and annihilation operators so that the commutation relation ( [ comma ] ) is satisfied ( with no @xmath14 in the r.h.s . ) . notice that , with this convention , @xmath120 , @xmath139 , and @xmath150 have formally the same dimensions as @xmath206 ( or @xmath207 ) , @xmath208 , and @xmath209 ( or @xmath210 ) , respectively . to distinguish the field operators evolved with the new hamiltonian @xmath150 , we will denote them by @xmath211 . in order to compute the commutator @xmath212 $ ] , we make use of the expressions ( [ crean ] ) for the creation and annihilation operators evolved in @xmath10 , which can also be written as @xmath213 , \nonumber\\ \label{crean2 } \hat{a}^\dagger_e(k , t)\!&=&\!\hat{a}^\dagger(k)\exp\left[it \left(e(\hat{h}_0+k)-e(\hat{h}_0)\right)\right].\hspace*{.9cm}\end{aligned}\ ] ] substituting then the relation @xmath214,\end{aligned}\ ] ] we get the relevant field commutator @xmath215\!\!=\!\!\ ! \int_0^\infty\!\!\!\!\!\!\!dk_1\!\!\int_0^\infty\!\!\!\ ! \!\!\!dk_2j_0(r_1k_1)j_0(r_2k_2)\\ \times 4\bar{g } \bigg\{\!\big[\hat{a}_e(k_1,\!t_1),\hat{a}_e(k_2,\!t_2)\big]\!+\ ! \big[\hat{a}^\dagger_e(k_1,\!t_1),\hat{a}^\dagger_e(k_2,\!t_2)\big ] \\ + \big[\hat{a}_e(k_1,\!t_1),\hat{a}^\dagger_e(k_2,\!t_2)\big]\!+\ ! \big[\hat{a}^\dagger_e(k_1,\!t_1),\hat{a}_e(k_2,\!t_2)\big ] \bigg\}.\end{aligned}\ ] ] the commutators involving @xmath136 and @xmath216 can be found in appendix [ a1 ] . note that , in contrast with the evolution given by @xmath57 , for which the commutation function is a c - number , the situation now is more complicated because @xmath217 $ ] is an operator , as it happens in interacting theories . we will analyze two types of matrix elements for it , namely , the expectation value on the vacuum and on one - particle states . in addition , we will briefly comment on the expectation value on coherent states . @xmath218 vacuum expectation value . @xmath219\,|0\rangle=\hspace*{3.3 cm } & & \nonumber\\ i8\bar{g}\!\!\int_0^{\infty}\!\!\!\!\!\!dk\,j_0(r_1 k)j_0(r_2 k)\sin\left[\frac{t_2-t_1}{4\bar{g}}(1\!-\!e^{-4\bar{g}k})\right]\!. \hspace*{.4cm}&&\label{s2 - 005}\end{aligned}\ ] ] in the following we will discuss the differences and similarities of this matrix element and the @xmath57-commutation function . ) of the @xmath57-commutator and the vacuum expectation value of the @xmath190-commutator for two different values of @xmath220 , plotted for @xmath193.,width=321 ] we point out that the factor @xmath221 $ ] that appears in the integrand of the @xmath57-commutator is substituted now by @xmath222 $ ] . they coincide for @xmath223 , but the former of these functions oscillates for all values of @xmath208 whereas the latter approaches a constant value when @xmath224 . this changes the convergence properties of the integral . in particular it is straightforward to see that the integral ( [ s2 - 005 ] ) diverges whenever @xmath191 ( except if @xmath225=0 $ ] ) but converges otherwise . therefore , the vacuum expectation value has a singularity structure that differs from the one given by eq . ( [ s2 - 004 ] ) . this has some interesting physical consequences . first we see that the singularity that originates in the axis in the @xmath57 case is not present when the evolution is dictated by @xmath190 ; this can be interpreted as a blurring of the axis due to quantum corrections . second we see that a completely different kind of singularity pops up when the evolution is generated by @xmath190 . from a mathematical point of view its origin is clearly related to the fact that the energy is bounded from above and , hence , for large values of @xmath208 the integrand is just a product of two @xmath114 bessel functions ( which give a divergent integral if their arguments coincide ) . physically , the emergence of the singularity can be understood in an intuitive manner by writing a state as a superposition of vectors of the form @xmath226 , @xmath227 and projecting onto @xmath228 . the @xmath229 factor ( for @xmath230 or 2 ) goes to a phase that depends only on @xmath231 as @xmath224 . so , if @xmath191 , the coefficients of the linear superposition defining , respectively , @xmath232 and @xmath233 differ only by a constant phase for large values of @xmath208 . this means that , in the sector of large @xmath208 , these two states have coherent phases and therefore a constructive interference . since each of them has an infinite norm , their scalar product diverges . a similar effect can be found in the quantum dynamics of a free particle with an energy given by a function @xmath234 with @xmath235 as @xmath224 . if one builds a wave packet as @xmath236},\ ] ] with @xmath237 peaked around a large value of @xmath208 , @xmath238 , the group velocity becomes almost zero and @xmath239 stays essentially the same at every @xmath172 for long periods of time . the integral in eq . ( [ s2 - 005 ] ) can be written explicitly as a convergent series ( when @xmath240 ) by expanding the sine function as a power series of @xmath241 and computing the resulting integrals involving two bessel functions and an exponential @xcite @xmath219\,|0\rangle= \hspace*{3.3cm}&&\nonumber \label{s2 - 006}\\ \frac{i8\bar{g}}{\pi\sqrt{r_1r_2}}\left\{\!\sin(\delta t)\!\!\sum_{n=0}^{\infty } \!\frac{\!(-1)^n(\delta t)^{2n}\!\ ! } { ( 2n)!}q_{-1/2}\left[\sigma_{2n}(r_1,r_2)\right]\right . & & \nonumber\\ -\left.\!\cos(\delta t)\!\sum_{n=0}^{\infty } \!\frac{\!(-1)^n(\delta t)^{2n+1}\!\!\ ! } { ( 2n+1)!}q_{-1/2}\!\left[\sigma_{2n+1}(r_1,r_2)\right]\!\right\}\!. \hspace*{.3cm}&&\end{aligned}\ ] ] here , we have introduced the notation @xmath242 besides , @xmath243 [ with @xmath244 is the associated legendre function of the second kind @xcite . we recall that the function @xmath245 grows without bound as the argument approaches @xmath246 and falls off to zero as @xmath247 when @xmath248 . when @xmath191 the singularity in eq . ( [ s2 - 006 ] ) comes just from the @xmath249 term in the first series of the expansion , that is given by @xmath250 a series of plots of @xmath251 $ ] and @xmath252 \,|0\rangle$ ] ( both over @xmath253 ) is shown in fig . [ fig4:compare ] for fixed values of @xmath178 and @xmath254 as a function of @xmath192 , with several choices for @xmath220 . we choose @xmath254 small enough to guarantee the rapid convergence of the series in eq . ( [ s2 - 006 ] ) and leave a discussion of the behavior of the integral ( [ s2 - 005 ] ) when @xmath255 for future work . as we can see @xmath256\,|0\rangle$ ] seems to approach @xmath251 $ ] at least in a certain average sense when @xmath220 is sufficiently small ( though not vanishing ) . it falls off to zero quite quickly outside the light cone defined by the free commutator and the auxiliary minkowski metric , and displays an oscillatory behavior within this light cone . the characteristic length of this oscillation decreases with @xmath220 , as well as close to the @xmath191 singularity . the approximation obtained by truncating the series expansion ( [ s2 - 006 ] ) , keeping a sufficiently large number of terms , compares well with the results of numerically computing expression ( [ s2 - 005 ] ) , at least for low enough values of @xmath254 . @xmath218 expectation value on one - particle states . we consider now states of the form @xmath257 where the function @xmath258 satisfies @xmath259 . we then have @xmath260 } \cos{\left[\omega_f(k_1,k_2)\right ] } \nonumber\\ & & -j_0(r_2k_1)|f(k_2)|^2\sin{\left[(t_2-t_1)e(k_1 ) e^{-4\bar{g}k_2 } \right]}\bigg\},\hspace*{.9 cm } \label{s2 - 007}\end{aligned}\ ] ] where @xmath261 a complete discussion of the meaning of the previous expression is beyond the scope of this paper . nevertheless , some features already present in the vacuum expectation value are also present here ; in particular the @xmath191 singularity . this can be seen by considering the last term in eq . ( [ s2 - 007 ] ) : the integral in @xmath262 is @xmath263},\ ] ] which takes in general a non - vanishing constant value ( depending on @xmath254 and @xmath220 ) as @xmath264 , thus rendering the remaining integral in @xmath265 divergent . as the first term in eq . ( [ s2 - 007 ] ) leads to a convergent integral , we conclude that the expectation value is singular when @xmath191 . it is not difficult to obtain as well an explicit expression for the expectation value of the @xmath190-commutator on the coherent states of the field @xmath26 . these diagonal matrix elements are calculated in appendix b. for our discussion in this work , let us only comment that the result is actually divergent when @xmath191 . this supplies further support to the claim that the considered singularity is indeed a generic feature of the system . linearly polarized cylindrical waves can be studied in great detail both from the classical and quantum points of view . as we have seen , there are two relevant hamiltonians for the study of the system . we have shown that the action and the metric of the gauge - fixed model in linearized gravity reproduce the results obtained by considering full cylindrical gravity and working to the first perturbative order . we get in this way a free hamiltonian . the hamiltonian governing the dynamics of the full system , on the other hand , is different from the free one , but turns out to be a function of it and presents certain features with deep physical consequences , like e.g. the existence of an upper bound . we have studied the similarities and differences of these two admissible kinds of evolution ; in particular , we have discussed how the emergence of an upper bound for the energy affects the causal structure of the model and the spreading of the light cones . the field commutator for the free hamiltonian is a c - number and shows the typical light cone structure found in standard perturbative quantum field theories . the commutator for the physical hamiltonian , as it usually happens for interacting theories , is no longer a c - number , so one has to consider its matrix elements . by concentrating on the vacuum expectation value we have been able to see several interesting phenomena : a spreading of the light cone as a function of the gravitational constant , the disappearance of the singularity present in the free case due to the smearing of the symmetry axis and the appearance of a new type of singularity associated with the fact that the energy is bounded from above . this new singularity is also present in the other expectation values discussed in the paper , namely for one - particle states and coherent states , and appears to be a generic feature of the model . there are several open questions that we plan to address in future work . in particular , it would be desirable to reach a better understanding of the behavior of the field commutator in the limit in which the length scale provided by @xmath220 goes to zero . the expectation values of the @xmath190-commutator discussed here resemble those derived from the free hamiltonian @xmath57 at least in a certain average sense . however it is not obvious how precisely and up to what extent they actually relate to each other . this is partly so because of the different singularity structure found in both cases . further research on this subject will concentrate on the properties of the model in the semiclassical limit @xmath266 . we will also pay detailed attention to matrix elements of the field commutator other than the vacuum expectation value , with the aim at discussing how the smearing of the light cones depends on the energy . in this appendix , we compute the commutators of the creation and annihilation operators @xmath136 and @xmath216 , obtained from the corresponding operators @xmath120 and @xmath267 via the unitary evolution generated by @xmath268 , where @xmath269 . employing relations ( [ crean2 ] ) and the basic commutators ( [ comma ] ) , it is possible to show that @xmath270&= & \hat{a}(k_1)\hat{a}(k_2 ) \exp\left[it_1e(\hat{h}_0-k_1-k_2)+i(t_2-t_1)e(\hat{h}_0-k_2 ) -it_2e(\hat{h}_0)\right ] \\&-&\hat{a}(k_1)\hat{a}(k_2)\exp\left[it_2e(\hat{h}_0-k_1-k_2 ) + i(t_1-t_2)e(\hat{h}_0-k_1)-it_1e(\hat{h}_0)\right]\ , , \\ \bullet \ , \big[\hat{a}^\dagger_e(k_1,t_1),\hat{a}^\dagger_e(k_2,t_2)\big]&= & \hat{a}^\dagger(k_1)\hat{a}^\dagger(k_2 ) \exp\left[it_1e(\hat{h}_0+k_1+k_2)+i(t_2-t_1)e(\hat{h}_0+k_2 ) -it_2e(\hat{h}_0)\right ] \\ & -&\hat{a}^\dagger(k_1)\hat{a}^\dagger(k_2 ) \exp\left[it_2e(\hat{h}_0+k_1+k_2)+i(t_1-t_2)e(\hat{h}_0+k_1 ) -it_1e(\hat{h}_0)\right]\ , , \\ \bullet \ , \big[\hat{a}_e(k_1,t_1),\hat{a}^\dagger_e(k_2,t_2)\big]&= & \hat{a}(k_1)\hat{a}^\dagger(k_2 ) \exp\left[it_1e(\hat{h}_0+k_2-k_1)+i(t_2-t_1)e(\hat{h}_0+k_2 ) -it_2e(\hat{h}_0)\right ] \\&-&\hat{a}^\dagger(k_2)\hat{a}(k_1)\exp\left[it_2 e(\hat{h}_0+k_2-k_1 ) + i(t_1-t_2)e(\hat{h}_0-k_1)-it_1e(\hat{h}_0)\right]\ , , \\ \bullet \ , \big[\hat{a}^\dagger_e(k_1,t_1),a_e(k_2,t_2)\big]&= & \hat{a}^\dagger(k_1)\hat{a}(k_2 ) \exp\left[it_1e(\hat{h}_0+k_1-k_2)+i(t_2-t_1)e(\hat{h}_0-k_2 ) -it_2e(\hat{h}_0)\right]\\&-&\hat{a}(k_2)\hat{a}^\dagger(k_1 ) \exp\left[it_2e(\hat{h}_0+k_1-k_2 ) + i(t_1-t_2)e(\hat{h}_0+k_1)-it_1e(\hat{h}_0)\right].\end{aligned}\ ] ] we consider coherent states of the field @xmath26 , given by @xmath271 where @xmath237 is a square integrable function and @xmath272 is a normalization constant satisfying @xmath273 the expectation value of the @xmath190-commutator is @xmath274 \int_0^\infty\!\frac{dk_1}{2}\!\!\int_0^\infty \!\!\!\!\!\!dk_2 j_0(r_1k_1)j_0(r_2k_2)\sum_{s=0}^\infty\frac{i_s(c ) } { s!}g_s(k_1,k_2),\end{aligned}\ ] ] where @xmath275^s \!\!-\![b(k_2,-k_1)]^s\!\bigg\}\\+ \overline{c}(k_2)c(k_1)\bigg\{[b(k_1,-k_2)]^s\!- \![b(-k_2,k_1)]^s\bigg\}e^{-4\bar{g}s k_1}\hspace*{.2cm}\\ + \overline{c}(k_1)c(k_2)\bigg\{[b(-k_1,k_2)]^s\ ! -\![b(k_2,-k_1)]^s\bigg\}e^{-4\bar{g}s k_2}\hspace*{.2 cm } \\ + c(k_1)c(k_2 ) \bigg\{[b(k_1,k_2)]^s\!-\![b(k_2,k_1)]^s\bigg\}e^{-4\bar{g}s ( k_1+k_2 ) } \\ + \overline{c}(k_1)\overline{c}(k_2 ) \bigg\{[b(-k_1,-k_2)]^s\!-\![b(-k_2,-k_1)]^s\bigg\ } , \hspace*{.7cm}\end{aligned}\ ] ] and we have employed the notation @xmath276.\end{aligned}\ ] ] note that , when @xmath191 , the delta in the expression of @xmath277 leads to the divergent integral @xmath278^{\!2s+1}\!\!\!\!\!\!\!\!.\end{aligned}\ ] ] the authors wish to thank a. ashtekar , g. date , and l. garay for interesting discussions . they are especially grateful to m. varadarajan for suggesting the subject and sharing enlightening conversations and insight . is supported by a spanish ministry of education and culture fellowship co - financed by the european social fund . this work was supported by the spanish mcyt under the research projects bfm2001 - 0213 and bfm2002 - 04031-c02 - 02 .

whenever a group of autonomous software agents or robots needs to decide on a joint course of action in a fair and satisfactory way , they need to aggregate their preferences . a common way to achieve this is to use voting rules . voting rules are studied in detail in social choice theory and are coming under increasing scrutiny from computer scientists who are interested in their computational properties or want to utilize them in multiagent systems ( see , * ? ? ? * ; * ? ? ? * ) . in social choice theory , voting rules are usually compared using desirable properties ( so - called axioms ) that they may or may not satisfy . there are a number of well - known impossibility theorems among which arrow s impossibility is arguably the most famous which state that certain axioms are incompatible with each other . these results , which show the non - existence of voting rules that satisfy a given set of axioms , are important because they clearly define the boundary of what can be achieved at all . this applies to the explicitly stated axioms as well as implicit ones such as boundaries on the number of voters or alternatives . for instance , if there are only two alternatives , arrow s theorem does not apply and there are many voting rules , including majority rule , that satisfy the conditions used in arrow s theorem . one impossibility that requires unusually high bounds on the number of voters and alternatives is moulin s _ no - show paradox _ @xcite , which states that the axioms of condorcet - consistency and participation are incompatible whenever there are at least 4 alternatives and 25 voters . moulin proves that the bound on the number of alternatives is tight by showing that the maximin voting rule ( with lexicographic tie - breaking ) satisfies the desired properties when there are at most 3 alternatives . the tightness of the more restrictive condition on the number of voters was left open , however . the goal of this paper is to give tight bounds on the number of voters required for moulin s theorem and related theorems that appear in the literature . to achieve this , we encode these problems as formulas in propositional logic and then use sat solvers to decide their satisfiability and extract minimal unsatisfiable sets ( muses ) in the case of unsatisfiability this approach is based on previous work by @xcite , @xcite , @xcite , and @xcite . however , it turned out that a straightforward application of this methodology is insufficient to deal with the magnitude of the problems we considered . several novel techniques were necessary to achieve our results . in particular , we extracted knowledge from computer - generated proofs of weaker statements and then used this information to guide the search for proofs of more general statements . as mentioned above , moulin s theorem uses the axioms of condorcet - consistency and participation . condorcet - consistency goes back to one of the most influential notions in social choice theory , namely that of a _ condorcet winner_. a condorcet winner is an alternative that is preferred to any other alternative by a majority of voters . the marquis de condorcet , after whom this concept is named , argued that , whenever a condorcet winner exists , it should be elected @xcite . a voting rule satisfying this condition is called _ condorcet - consistent_. apart from the intuitive appeal of this condition , condorcet - consistent rules are more robust to changes in the of feasible alternatives and less susceptible to strategic manipulation than other voting rules ( such as borda s rule ) ( see , * ? ? ? * ; * ? ? ? * ) . while the desirability of condorcet - consistency as that of any other axiom has been subject to criticism , many scholars agree that it is very appealing if not indispensable and a large part of the social choice literature deals exclusively with condorcet - consistent voting rules @xcite . _ participation _ was first considered by @xcite and requires that no voter should be worse off by joining an electorate , or alternatively that no voter should benefit by abstaining from an election . the desirability of this axiom in any context with voluntary participation is evident . all the more surprisingly , @xcite have shown that single transferable vote ( stv ) , a common voting rule , violates participation and referred to this phenomenon as the _ no - show paradox_. @xcite , perhaps even more startlingly , proved that no condorcet - consistent voting rule satisfies participation when there are at least 25 voters . we leverage sat solving to obtain an elegant human - readable proof of moulin s result that requires only 12 voters . while computer - aided solving techniques allow us to tackle difficult combinatorial problems , they usually do not provide additional insight into these problems . somewhat surprisingly , the computer - aided proofs we found possess a certain kind of symmetry that has not been exploited in previous proofs . moreover , the sat solver is able to construct a condorcet - consistent voting rule that satisfies participation as well as a number of other desirable properties for up to 11 voters , proving the optimality of the above bound . this computer - generated voting rule is compatible with the maximin voting rule in 99.8% of all cases and , in contrast to maximin , only selects alternatives from the top cycle . as a practical consequence of our theorem , strategic abstention need not be a concern for condorcet - consistent voting rules when there are at most 4 alternatives and 11 voters , for instance when voting in a committee . we also use our techniques to provide optimal bounds for related results about set - valued and probabilistic voting rules @xcite . in particular , we give a tight bound of 17 voters for the optimistic preference extension , 14 voters for the pessimistic extension , and 12 voters for the stochastic dominance preference extension . these results are substantial improvements of previous results . for example , the previous statement for the pessimistic extension requires an additional axiom , at least 5 alternatives , and at least 971 voters @xcite . our results are summarized in . lccccccccccccccccccccccccc & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 & 10 & 11 & 12 & 13 & 14 & 15 & 16 & 17 & 18 & 19 & 20 & 21 & 22 & 23 & 24 & 25 + condorcet & & & & + maximin & & + kemeny & & + optimistic & & + pessimistic & & + strong sd & & + + [ cols= " < , < , < , < , < " , ] the no - show paradox was first observed by @xcite for the stv voting rule . @xcite and @xcite investigate how frequently this phenomenon occurs in practice . the main theorem addressed in this paper is due to @xcite and requires at least 25 voters . this bound was recently brought down to 21 voters by @xcite . simplified proofs of moulin s theorem are given by @xcite and @xcite . @xcite and @xcite strengthen moulin s theorem by weakening condorcet - consistency and participation , respectively . @xcite prove variants of moulin s theorem for set - valued voting rules based on the optimistic and the pessimistic preference extension . @xcite defines a weaker notion of participation in the context of set - valued voting rules and shows that all common condorcet extensions except the maximin rule and young s rule violate this property . @xcite notes that `` a practical question , which has not been dealt with here , refers to the number of candidates and voters that are necessary to invoke the studied paradoxes '' @xcite . when assuming that voters have incomplete preferences over sets or lotteries , participation and condorcet - consistency can be satisfies simultaneously and positive results for common condorcet - consistent voting rules have been obtained by @xcite and @xcite . abstention in slightly different contexts than the one studied in this paper recently caught the attention of computer scientists working on voting equilibria and campaigning @xcite . the computer - aided techniques in this paper are inspired by @xcite , who reduced well - known impossibility results from social choice theory such as arrow s theorem to finite instances , which can then be checked by a sat solver . this methodology has been extended and applied to new problems by @xcite , @xcite , and @xcite . the results obtained by computer - aided theorem proving have already found attention in the social choice community @xcite . more generally , sat solvers have also proven to be quite effective for other problems in economics . a prominent example is the ongoing work by @xcite in which sat solvers are used for the development and execution of the fcc s upcoming reverse spectrum auction . in some respects , our approach also bears some similarities to _ automated mechanism design _ ( see , * ? ? ? * ) , where desirable properties are encoded and mechanisms are computed to fit specific problem instances . let @xmath0 be a set of @xmath1 alternatives and @xmath2 be a set of @xmath3 voters . whether no - show paradoxes occur depends on the exact values of @xmath1 and @xmath3 . by @xmath4 we denote the set of _ electorates _ , non - empty subsets of @xmath2 . for many of our results , we will take @xmath5 , and we use the labels @xmath6 for arbitrary elements of @xmath0 . a _ ( strict ) preference relation _ is a complete , antisymmetric , and transitive binary relation on @xmath0 . the preference relation of voter @xmath7 is denoted by @xmath8 . the set of all preference relations over @xmath0 is denoted by @xmath9 . for brevity , we denote by @xmath10 the preference relation @xmath11 , eliding the identity of voter @xmath7 , and similarly for other preferences . a _ preference profile _ @xmath12 is a function from an electorate @xmath13 to the set of preference relations @xmath9 . the set of all preference profiles is thus given by @xmath14 . for the sake of adding and deleting voters , we define for any preference profile @xmath15 with @xmath16 , and @xmath17 , @xmath18 @xmath19 if the identity of the voter is clear or irrelevant we sometimes , in slight abuse of notation , refer to @xmath20 by @xmath21 , and write @xmath22 instead of @xmath23 . if @xmath24 voters with the same preferences @xmath8 are to be added or removed , we write @xmath25 and @xmath26 , respectively . the _ majority margin _ of @xmath12 is the map @xmath27 with @xmath28 . the majority margin can be viewed as the adjacency matrix of a _ weighted tournament _ @xmath29 . we say that a preference profile @xmath12 _ induces _ the weighted tournament @xmath29 . an alternative @xmath30 is called _ condorcet winner _ if it wins against any other alternative in a majority contest , if @xmath31 for all @xmath32 . conversely , an alternative @xmath30 is a _ condorcet loser _ if @xmath33 for all @xmath32 . our central object of study are _ voting rules _ , functions that assigns every preference profile a socially preferred alternative . thus , a voting rule is a function @xmath34 . in this paper , we study voting rules that satisfy _ condorcet - consistency _ and _ participation_. a _ condorcet extension _ is a voting rule that selects the condorcet winner whenever it exists . thus , @xmath35 is a condorcet extension if for every preference profile @xmath12 that admits a condorcet winner @xmath30 , we have @xmath36 . we say that @xmath35 is _ condorcet - consistent_. a voting rule @xmath35 satisfies _ participation _ if all voters always weakly prefer voting to not voting , if @xmath37 for all @xmath15 and @xmath38 with @xmath13 . equivalently , participation requires that for all preference profiles @xmath12 not including voter @xmath39 , we have @xmath40 . the proofs of both positive and negative results to come were obtained through automated techniques that we describe in . to become familiar with the kind of arguments produced in this way , we will now study a more restricted setting which is of independent interest . specifically , let us consider voting rules that select winners in accordance with the popular _ maximin _ and _ kemeny _ rules . for a preference profile @xmath12 , an alternative @xmath30 is a _ maximin winner _ if it maximizes @xmath41 ; thus , @xmath30 never gets defeated too badly in pairwise comparisons . an alternative @xmath30 is a _ kemeny winner _ if it is ranked first in some _ kemeny ranking_. a kemeny ranking is a preference relation @xmath42 maximizing agreement with voters individual preferences , it maximizes the quantity @xmath43 . we call a voting rule a _ maximin extension _ _ kemeny extension _ ) if it always selects a maximin winner ( resp . kemeny winner ) . since a condorcet winner , if it exists , is always the unique maximin and kemeny winner of a preference profile , any such voting rule is also a condorcet extension . we can now prove an easy version of moulin s theorem for these more restricted voting rules . to this end , we first prove a useful lemma allowing us to extend impossibility proofs for @xmath44 alternatives to also apply if there are more than @xmath44 alternatives . its proof gives a first hint on how condorcet - consistency and participation interact . [ lem : avoid - last ] suppose that @xmath35 is a condorcet extension satisfying participation . let @xmath12 be a preference profile and @xmath45 a set of _ bad _ alternatives such that each voter ranks every @xmath46 below every @xmath47 . then @xmath48 . by induction on the number of voters @xmath49 in @xmath12 . if @xmath12 consists of a single voter @xmath7 , then , since @xmath35 is a condorcet extension , @xmath50 must return @xmath7 s top choice , which is not bad . if @xmath12 consists of at least 2 voters , and @xmath38 , then by participation @xmath37 . if @xmath50 were bad , then so would be @xmath51 , contradicting the inductive hypothesis . the following computer - aided proofs , just like the more complicated proofs to follow , can be understood solely by carefully examining the corresponding ` proof diagram ' . an arrow such as indicates that profile @xmath52 is obtained from @xmath12 by adding a voter @xmath10 , and is read as `` if one of the bold green alternatives ( here @xmath53 ) is selected at @xmath12 , then one of them is selected at @xmath52 '' ( by participation ) . a circled node indicates a profile admitting condorcet winner @xmath54 , although in the proofs of theorems [ thm : maximin ] and [ thm : kemeny ] , we use it to refer to maximin and kemeny winners , respectively . [ thm : maximin ] there is no maximin extension that satisfies participation for @xmath55 and @xmath56 . ( for @xmath57 and @xmath58 , such a maximin extension exists . ) let @xmath35 be a maximin extension which satisfies participation . consider the following 6-voter profile @xmath12 : @xmath59 \draw node [ inner sep=5pt ] ( fakeroot ) at ( 0 , 0 ) { \begin{array}[t]{c@{\quad}c@{\quad}c@{\quad}c } \toprule 1&2&2&1\\ \midrule { \texttt a}&{\texttt b}&{\texttt c}&{\texttt d}\\ { \texttt b}&{\texttt d}&{\texttt a}&{\texttt c}\\ { \texttt d}&{\texttt c}&{\texttt b}&{\texttt a}\\ { \texttt c}&{\texttt a}&{\texttt d}&{\texttt b}\\ \bottomrule\\[-5pt ] \multicolumn{4}{c}{r } \end{array } } ; \draw node [ rectangle , minimum width=3 cm , minimum height=1.25 cm ] ( root ) at ( 0 , 0.75 cm ) { } child { node[condorcet ] ( alpha ) { $ { \texttt c}$ } edge from parent node[lbl , xshift=-0.15 cm ] { $ { } + { { \color{black!70!green}{\fontseries{b}\selectfont{\texttt a}{\texttt b}}}}{{\color{black!70}\fontseries{l}\selectfont { \texttt c}{\texttt d}}}$ } } child { node[condorcet ] ( alphaprime ) { $ { \texttt b}$ } edge from parent node[lbl ] { $ { } + { { \color{black!70!green}{\fontseries{b}\selectfont{\texttt d}{\texttt c}}}}{{\color{black!70}\fontseries{l}\selectfont { \texttt b}{\texttt a}}}$ } } ; \ifshowtournaments \begin{scope}[shift={(alpha.south)},shift={(-0.65cm,-0.5 cm ) } , scale=1.3,vertex/.style={fill = white , circle , draw , minimum size=13.9,inner sep=2 } ] \draw ( 0,0 ) node[vertex ] ( a ) { $ { \texttt a}$ } ( 1,0 ) node[vertex ] ( b ) { $ { \texttt b}$ } ( 1,-1 ) node[vertex , very thick ] ( c ) { $ { \texttt c}$ } ( 0,-1 ) node[vertex ] ( d ) { $ { \texttt d}$ } ; \draw [ -latex ] ( b ) to node [ white , right ] { 1 } ( c ) ; \draw [ -latex ] ( c ) to node [ above , sloped , near start ] { 3 } ( a ) ; \draw [ -latex ] ( d ) to node [ white , above ] { 1 } ( c ) ; \draw [ -latex ] ( a ) to node [ above ] { 3 } ( b ) ; \draw [ -latex ] ( b ) to node [ sloped , above , near end ] { 5 } ( d ) ; \draw [ -latex ] ( a ) to node [ white , left ] { 1 } ( d ) ; \end{scope } \begin{scope}[shift={(alphaprime.south)},shift={(-0.65cm,-0.5 cm ) } , scale=1.3,vertex/.style={fill = white , circle , draw , minimum size=13.9,inner sep=2 } ] \draw ( 0,0 ) node[vertex ] ( a ) { $ { \texttt a}$ } ( 1,0 ) node[vertex , very thick ] ( b ) { $ { \texttt b}$ } ( 1,-1 ) node[vertex ] ( c ) { $ { \texttt c}$ } ( 0,-1 ) node[vertex ] ( d ) { $ { \texttt d}$ } ; \draw [ -latex ] ( c ) to node [ white , above ] { 1 } ( b ) ; \draw [ -latex ] ( c ) to node [ above , sloped , near start ] { 5 } ( a ) ; \draw [ -latex ] ( d ) to node [ below ] { 3 } ( c ) ; \draw [ -latex ] ( a ) to node [ white , above ] { 1 } ( b ) ; \draw [ -latex ] ( b ) to node [ sloped , above , near end ] { 3 } ( d ) ; \draw [ -latex ] ( d ) to node [ white , left ] { 1 } ( a ) ; \end{scope } \fi \end{tikzpicture } \vspace{-10pt}\ ] ] suppose @xmath60 . adding an @xmath10 vote leads to a weighted tournament in which alternative @xmath61 is the unique maximin winner . but this contradicts participation since the added voter would benefit from abstaining the election . symmetrically , if @xmath62 , then adding a @xmath63 vote leads to a weighted tournament in which @xmath64 is the maximin winner , again contradicting participation . the symmetry of the argument is due to an automorphism of @xmath12 , namely the relabelling of alternatives according to @xmath65 . if @xmath66 , we add new bad alternatives @xmath67 to the bottom of @xmath12 and of the additional voters . by , @xmath35 chooses from @xmath68 at each step , completing the proof . the existence result for @xmath69 is established by the methods described in . for 3 alternatives , @xcite proved that the voting rule that chooses the lexicographically first maximin winner satisfies participation . shows that this is not the case for @xmath44 alternatives , even if there are only @xmath70 voters and no matter how we pick among maximin winners . [ thm : kemeny ] there is no kemeny extension that satisfies participation for @xmath55 and @xmath71 . ( for @xmath57 and @xmath72 , such a kemeny extension exists . ) let @xmath35 be a kemeny extension which satisfies participation . consider the following 4-voter profile @xmath12 : @xmath73 \draw node [ inner sep=5pt ] ( fakeroot ) at ( 0 , 0 ) { \begin{array}[t]{c@{\quad}c@{\quad}c@{\quad}c } \toprule 1&1&1&1\\ \midrule { \texttt a}&{\texttt b}&{\texttt c}&{\texttt d}\\ { \texttt d}&{\texttt a}&{\texttt b}&{\texttt c}\\ { \texttt c}&{\texttt d}&{\texttt a}&{\texttt b}\\ { \texttt b}&{\texttt c}&{\texttt d}&{\texttt a}\\ \bottomrule\\[-5pt ] \multicolumn{4}{c}{r } \end{array } } ; \draw node [ rectangle , minimum width=2.5 cm , minimum height=1.4 cm ] ( root ) at ( 0 , 0.7 cm ) { } \ifshowtournaments child { node[condorcet ] ( 1 ) { $ { \texttt a}$ } edge from parent node[lbl ] { $ { } - { { \color{black!70}\fontseries{l}\selectfont { \texttt c}{\texttt b}{\texttt a}}}{{\color{black!70!green}{\fontseries{b}\selectfont{\texttt d}}}}$ } } child { node[condorcet ] ( 2 ) { $ { \texttt b}$ } edge from parent node[lbl ] { $ { } - { { \color{black!70}\fontseries{l}\selectfont { \texttt d}{\texttt c}{\texttt b}}}{{\color{black!70!green}{\fontseries{b}\selectfont{\texttt a}}}}$ } } child { node[condorcet ] ( 3 ) { $ { \texttt c}$ } edge from parent node[lbl ] { $ { } - { { \color{black!70}\fontseries{l}\selectfont { \texttt a}{\texttt d}{\texttt c}}}{{\color{black!70!green}{\fontseries{b}\selectfont{\texttt b}}}}$ } } child { node[condorcet ] ( 4 ) { $ { \texttt d}$ } edge from parent node[lbl ] { $ { } - { { \color{black!70}\fontseries{l}\selectfont { \texttt b}{\texttt a}{\texttt d}}}{{\color{black!70!green}{\fontseries{b}\selectfont{\texttt c}}}}$ } } ; \begin{scope}[shift={(1.south)},shift={(-0.65cm,-0.5 cm ) } , scale=1.2,vertex/.style={fill = white , circle , draw , minimum size=13.9,inner sep=2 } ] \draw ( 0,0 ) node[vertex , very thick ] ( a ) { $ { \texttt a}$ } ( 1,0 ) node[vertex ] ( b ) { $ { \texttt b}$ } ( 1,-1 ) node[vertex ] ( c ) { $ { \texttt c}$ } ( 0,-1 ) node[vertex ] ( d ) { $ { \texttt d}$ } ; \draw [ -latex ] ( c ) to node [ white , right ] { 1 } ( b ) ; \draw [ -latex ] ( d ) to node [ above ] { 3 } ( c ) ; \draw [ -latex ] ( b ) to node [ white , above ] { 1 } ( a ) ; \draw [ -latex ] ( a ) to node [ white , left ] { 1 } ( d ) ; \draw [ -latex ] ( a ) to node [ white , above , sloped , near start ] { 1 } ( c ) ; \draw [ -latex ] ( d ) to node [ white , sloped , above , near end ] { 1 } ( b ) ; \end{scope } \begin{scope}[shift={(2.south)},shift={(-0.65cm,-0.5 cm ) } , scale=1.2,vertex/.style={fill = white , circle , draw , minimum size=13.9,inner sep=2 } ] \draw ( 0,0 ) node[vertex ] ( a ) { $ { \texttt a}$ } ( 1,0 ) node[vertex , very thick ] ( b ) { $ { \texttt b}$ } ( 1,-1 ) node[vertex ] ( c ) { $ { \texttt c}$ } ( 0,-1 ) node[vertex ] ( d ) { $ { \texttt d}$ } ; \draw [ -latex ] ( c ) to node [ white , right ] { 1 } ( b ) ; \draw [ -latex ] ( d ) to node [ white , above ] { 1 } ( c ) ; \draw [ -latex ] ( b ) to node [ white , above ] { 1 } ( a ) ; \draw [ -latex ] ( a ) to node [ left ] { 3 } ( d ) ; \draw [ -latex ] ( a ) to node [ white , above , sloped , near start ] { 1 } ( c ) ; \draw [ -latex ] ( b ) to node [ white , sloped , above , near end ] { 1 } ( d ) ; \end{scope } \begin{scope}[shift={(3.south)},shift={(-0.65cm,-0.5 cm ) } , scale=1.2,vertex/.style={fill = white , circle , draw , minimum size=13.9,inner sep=2 } ] \draw ( 0,0 ) node[vertex ] ( a ) { $ { \texttt a}$ } ( 1,0 ) node[vertex ] ( b ) { $ { \texttt b}$ } ( 1,-1 ) node[vertex , very thick ] ( c ) { $ { \texttt c}$ } ( 0,-1 ) node[vertex ] ( d ) { $ { \texttt d}$ } ; \draw [ -latex ] ( c ) to node [ white , right ] { 1 } ( b ) ; \draw [ -latex ] ( d ) to node [ white , above ] { 1 } ( c ) ; \draw [ -latex ] ( b ) to node [ above ] { 3 } ( a ) ; \draw [ -latex ] ( a ) to node [ white , left ] { 1 } ( d ) ; \draw [ -latex ] ( c ) to node [ white , above , sloped , near start ] { 1 } ( a ) ; \draw [ -latex ] ( b ) to node [ white , sloped , above , near end ] { 1 } ( d ) ; \end{scope } \begin{scope}[shift={(4.south)},shift={(-0.65cm,-0.5 cm ) } , scale=1.2,vertex/.style={fill = white , circle , draw , minimum size=13.9,inner sep=2 } ] \draw ( 0,0 ) node[vertex ] ( a ) { $ { \texttt a}$ } ( 1,0 ) node[vertex ] ( b ) { $ { \texttt b}$ } ( 1,-1 ) node[vertex ] ( c ) { $ { \texttt c}$ } ( 0,-1 ) node[vertex , very thick ] ( d ) { $ { \texttt d}$ } ; \draw [ -latex ] ( c ) to node [ right ] { 3 } ( b ) ; \draw [ -latex ] ( d ) to node [ white , above ] { 1 } ( c ) ; \draw [ -latex ] ( b ) to node [ white , above ] { 1 } ( a ) ; \draw [ -latex ] ( a ) to node [ white , left ] { 1 } ( d ) ; \draw [ -latex ] ( c ) to node [ white , above , sloped , near start ] { 1 } ( a ) ; \draw [ -latex ] ( d ) to node [ white , sloped , above , near end ] { 1 } ( b ) ; \end{scope } \else child { node[condorcet ] ( 1 ) at ( -0.4 cm , 1 cm ) { $ { \texttt a}$ } edge from parent node[lbl ] { $ { } - { { \color{black!70}\fontseries{l}\selectfont { \texttt c}{\texttt b}{\texttt a}}}{{\color{black!70!green}{\fontseries{b}\selectfont{\texttt d}}}}$ } } child { node[condorcet ] ( 2 ) at ( -1.5 cm , 0.3 cm ) { $ { \texttt b}$ } edge from parent node[lbl ] { $ { } - { { \color{black!70}\fontseries{l}\selectfont { \texttt d}{\texttt c}{\texttt b}}}{{\color{black!70!green}{\fontseries{b}\selectfont{\texttt a}}}}$ } } child { node[condorcet ] ( 3 ) at ( 1.5 cm , 0.3 cm ) { $ { \texttt c}$ } edge from parent node[lbl ] { $ { } - { { \color{black!70}\fontseries{l}\selectfont { \texttt a}{\texttt d}{\texttt c}}}{{\color{black!70!green}{\fontseries{b}\selectfont{\texttt b}}}}$ } } child { node[condorcet ] ( 4 ) at ( 0.4 cm , 1 cm ) { $ { \texttt d}$ } edge from parent node[lbl ] { $ { } - { { \color{black!70}\fontseries{l}\selectfont { \texttt b}{\texttt a}{\texttt d}}}{{\color{black!70!green}{\fontseries{b}\selectfont{\texttt c}}}}$ } } ; \fi \end{tikzpicture } \vspace{-10pt}\ ] ] suppose @xmath74 . then removing @xmath75 from @xmath12 yields a weighted tournament in which @xmath54 is the ( unique ) kemeny winner , which contradicts participation . analogously , we can exclude the other three possible outcomes for @xmath12 by letting a voter abstain , which always leads to a unique kemeny winner and a contradiction with participation . the arguments are identical because @xmath12 is completely symmetric in the sense that for any pair of alternatives @xmath30 and @xmath76 , there is an automorphism of @xmath12 that maps @xmath30 to @xmath76 . just like for , if @xmath66 , we add new bad alternatives @xmath67 to the bottom of @xmath12 and of the additional voters . by , @xmath35 chooses from @xmath68 at each step , completing the proof . one remarkable and unexpected aspect of the computer - aided proofs above is that their simplicity is due to automorphisms of the underlying preference profiles . similar automorphisms will also be used in the proofs of the stronger theorems in sections [ sec : results ] , [ sec : set - valued ] , and [ sec : probabilistic ] . we emphasize that these symmetries are not hard - coded into our problem specification and , to the best of our knowledge , have not been exploited in previous proofs of similar statements . the bounds in this paper were obtained with the aid of a computer . in this section , we describe the method that we employed . the main tool in our approach is an encoding of our problems into propositional logic . we then use sat solvers to decide whether ( in a chosen setting ) there exists a condorcet extension satisfying participation . if the answer is yes , the solver returns an explicit such voting rule . if the answer is no , we use a process called _ mus extraction _ to find a short certificate of this fact which can be translated into a human - readable proof . by successively proving stronger theorems and using the insights obtained through mus extraction , we arrived at results as presented in their full generality in this paper . `` for @xmath3 voters and @xmath44 alternatives , is there a voting rule @xmath35 that satisfies condorcet - consistency and participation ? '' a natural encoding of this question into propositional logic proceeds like this : generate all profiles over 4 alternatives with at most @xmath3 voters . for each such profile @xmath12 , introduce 4 propositional variables @xmath77 , @xmath78 , @xmath79 , @xmath80 , where the intended meaning of @xmath77 is @xmath81 we then add clauses requiring that for each profile @xmath12 , @xmath50 takes _ exactly one _ value , and we add clauses requiring @xmath35 to be condorcet - consistent and satisfy participation . sadly , the encoding sketched above is not tractable for the values of @xmath3 that we are interested in : the number of variables and clauses used grows as @xmath82 , because there are @xmath83 possible preference relations over 4 alternatives and thus @xmath84 profiles with @xmath3 voters . for @xmath85 , this leads to more than 400 billion variables , and for @xmath86 we exceed @xmath87 variables . to escape this combinatorial explosion , we will temporarily restrict our attention to _ pairwise _ voting rules . this means that we assign an outcome alternative @xmath88 to every weighted tournament @xmath89 . we then define a voting rule that assigns the outcome @xmath90 to each preference profile @xmath12 , where @xmath29 is the weighted tournament induced by @xmath12 . the number of tournaments induced by profiles with @xmath3 voters grows much slower than the number of profiles our computer enumeration suggests a growth of order about @xmath91 . this much more manageable ( yet still exponential ) growth allows us to consider problem instances up to @xmath92 which turns out to be _ just _ enough . other than referring to ( weighted ) tournaments instead of profiles , our encoding into logic now proceeds exactly like before . for each tournament @xmath89 , we introduce the variables @xmath93 , @xmath94 , @xmath95 , @xmath96 and define the formulas @xmath97 with our intended interpretation of the variables @xmath98 , all models of @xmath99 are functions from tournaments into @xmath68 . ( the word @xmath100 abbreviates ` mutual exclusion ' and corresponds to the voting rule selecting a unique winner . ) since we are interested in voting rules that satisfy participation , we also need to encode this property . to this end , let @xmath101 be a tournament induced by @xmath12 and let @xmath102 be a preference relation . define @xmath103 . ( the tournament @xmath104 is independent of the choice of @xmath12 . ) we define @xmath105 requiring @xmath35 to be condorcet - consistent is straightforward : if tournament @xmath89 admits @xmath64 as the condorcet winner , we add @xmath106 and we add similar formulas for each tournament that admits a condorcet winner . then the models of the conjunction of all the @xmath107 , @xmath100 , @xmath108 , and @xmath109 formulas are precisely the pairwise voting rules satisfying condorcet - consistency and participation . by adapting the @xmath109 formulas , we can impose more stringent conditions on @xmath35this is how our results for maximin and kemeny extensions are obtained . we can also use this to exclude pareto - dominated alternatives , and to require @xmath35 to always pick from the top cycle . for some purposes it will be useful not to include the @xmath100 clauses in our final formula . models of this formula then correspond to _ set - valued _ voting rules that satisfy participation interpreted according to the optimistic preference extension . see for results in this setting . the formulas we have obtained above are all given in _ conjunctive normal form _ ( cnf ) , and thus can be evaluated without further transformations by any off - the - shelf sat solver . in order to physically produce a cnf formula as described , we employ a straightforward python script that performs a breadth - first search to discover all weighted tournaments with up to @xmath3 voters ( see for a schematic overview of the program ) . the script outputs a cnf formula in the standard dimacs format , and also outputs a file that , for each variable @xmath98 , records the tournament @xmath89 and alternative @xmath30 it denotes . this is necessary because the dimacs format uses uninformative variable descriptors ( consecutive integers ) and memorizing variable meanings allows us to interpret the output of the sat solver . @xmath110weighted tournament on @xmath68 with all edges having weight 0@xmath111 . @xmath112 calculate @xmath113 add @xmath114 to @xmath115 write @xmath116 write @xmath117 [ alg : write_formula ] as an example , the output formula for @xmath118 in dimacs format has a size of about 7 gb and uses 50 million variables and 2 billion clauses , taking 6.5 hours to write . plingeling @xcite , a popular sat solver that we used for all results in this paper , solves this formula in 50 minutes of wall clock time , half of which is spent parsing the formula . in case a given instance is satisfiable , the solver returns a satisfying assignment , giving us an existence proof and a concrete example for a voting rule satisfying participation ( and any further requirements imposed ) . in case a given instance in unsatisfiable , we would like to have short certificate of this fact as well . one possibility for this is having the sat solver output a resolution proof ( in drup format , say ) . this yields a machine - checkable proof , but has two major drawbacks : the generated proofs can be uncomfortably large @xcite , and they do not give human - readable insights about _ why _ the formula is unsatisfiable . we handle this problem by computing a _ minimal unsatisfiable subset ( mus ) _ of the unsatisfiable cnf formula . an mus is a subset of the clauses of the original formula which itself is unsatisfiable , and is minimally so : removing any clause from it yields a satisfiable formula . we used the tools muser2 @xcite and marco @xcite to extract muses . if an unsatisfiable formula admits a very small mus , it is often possible to obtain a human - readable proof of unsatisfiability from it @xcite . note that for purposes of extracting human - readable proofs , it is desirable for the mus to be as small as possible , and also to refer to as few different tournaments as possible . the first issue can be addressed by running the mus extractor repeatedly , instructing it to order clauses randomly ( note that clause sets of different cardinalities can be minimally unsatisfiable with respect to set inclusion ) ; similarly , we can use marco to enumerate all muses and look for small ones . the second issue can be addressed by computing a _ group mus _ : here , we partition the clauses of the cnf formula into _ groups _ , and we are looking for a minimal set of groups that together are unsatisfiable . in our case , the clauses referring to a given tournament @xmath89 form a group . in practice , finding a group mus first and then finding a standard ( clause - level ) mus within the group mus yielded sets of size much smaller than muses returned without the intermediate group - step ( often by a factor of 10 ) . to translate an mus into a human - readable proof , we created another program that visualized the mus in a convenient form . clause that occurs in the mus , and marks the nodes for which @xmath119 clauses appear in the mus . ] indeed , this program outputs the ` proof diagrams ' like that appear throughout this paper ( though we re - typeset them ) . we think that interpreting these diagrams is quite natural ( and is perhaps even easier than reading a textual translation ) . more importantly , the automatically produced graphs allowed us to quickly judge the quality of an extracted mus . [ - > , level distance=9 mm , sibling distance=55 mm , level 2/.style = level distance=7mm , sibling distance=23 mm , level 3/.style = level distance=11 mm , level 4/.style = level distance=7 mm , level 5/.style = level distance=11 mm , xscale=1.65 , ne/.style = inner sep=2pt , empty/.style = circle , draw = black!75,fill = black!40 , inner sep=1.5pt , condorcet/.style = circle , draw , inner sep=1.5pt , lbl/.style = fill = white , inner sep=1pt ] node [ inner sep=5pt ] ( fake root ) at ( 0 , 0 ) c@c@c@c 2&3&3&2 + ` a`&`b`&c&`d ` + ` b`&`d`&`a ` & + ` d`&c&`b`&`a ` + c&`a`&`d`&`b ` + + ; node [ rectangle , minimum width=2.5 cm , minimum height=1 cm ] ( root ) at ( 0 , 0.6 cm ) child node ( alpha ) @xmath120 child node[empty ] ( 1 ) child node[ne ] ( beta ) @xmath121 child node[condorcet ] ( ca ) @xmath54 edge from parent node[lbl ] @xmath122 child node[empty ] ( 2 ) child node[condorcet ] ( cb ) @xmath64 edge from parent node[lbl ] @xmath123 edge from parent node[lbl ] @xmath124 edge from parent node[lbl ] @xmath125 edge from parent node[lbl ] @xmath126 child node[empty ] ( r2 ) child node[condorcet ] ( cc ) @xmath61 edge from parent node[lbl ] @xmath127 edge from parent node[lbl ] @xmath128 edge from parent node[lbl ] @xmath129 child node ( alphaprime ) @xmath130 child node[empty ] ( r4 ) child node[condorcet ] ( cbprime ) @xmath64 edge from parent node[lbl ] @xmath131 edge from parent node[lbl ] @xmath132 child node[empty ] ( 1prime ) child node[ne ] ( betaprime ) @xmath133 child node[empty ] ( 2prime ) child node[condorcet ] ( cc ) @xmath61 edge from parent node[lbl ] @xmath134 edge from parent node[lbl ] @xmath135 child node[condorcet ] ( cd ) @xmath136 edge from parent node[lbl ] @xmath137 edge from parent node[lbl ] @xmath138 edge from parent node[lbl ] @xmath139 edge from parent node[lbl ] @xmath140 ; the sat encoding described in only concerns pairwise voting rules , yet none of the ( negative ) results in this paper require or use this assumption . this is the product of multiple rounds of generating and evaluating sat formulas , extracting muses , and using the insights generated by this as ` educated guesses ' to solve more general problems . following the process as described so far led to a proof that for @xmath44 alternatives and @xmath141 voters , there is no pairwise condorcet extension that satisfies participation . that proof used the assumption of pairwiseness , it assumed that the voting rule returns the same alternative on profiles inducing the same weighted tournament . however , intriguingly , the preference profiles mentioned in the proof did not contain all @xmath83 possible preference relations over @xmath68 . in fact , it only used 10 of the possible orders . further , each profile included @xmath142 as a subprofile . as we argued at the start of , it is intractable to search over the entire space of preference profiles . on the other hand , it is much easier to merely search over all extensions of @xmath143 that contain at most @xmath144 voters and only contain copies of the 10 orders previously identified . the sat formula produced by doing exactly this turned out to be unsatisfiable , and a small mus extracted from it gave rise to . the proof of for 17 voters was obtained by running with @xmath145 initialized to the weighted tournament induced by the initial profile @xmath12 used in the proof of . before finding this tournament , we tried various other tournaments as @xmath145 , including ones featuring in moulin s original proof , and ones occurring at other steps in the proof of , but @xmath12 turned out to give the best ( and indeed a tight ) bound , and additionally exhibits a lot of symmetry that was also present in the mus we extracted . we are now in a position to state and prove our main claim that condorcet extensions can not avoid the no - show paradox for @xmath141 or more voters ( when there are at least @xmath44 alternatives ) , and that this result is optimal . [ thm : anonymous ] there is no condorcet extension that satisfies participation for @xmath55 and @xmath146 . the proof follows the structure depicted in . let @xmath12 be the preference profile shown there . since @xmath12 remains fixed after relabelling alternatives according to @xmath65 , we may assume without loss of generality that @xmath147 . ( an explicit proof in case @xmath148 is indicated in . ) by participation , it follows from @xmath147 that also @xmath149 since the voters with preferences @xmath10 can not be worse off by joining the electorate . if @xmath150 , again by participation , removing @xmath151 voters with preferences @xmath152 does not change the winning alternative ( so @xmath153 ) , and neither does adding @xmath154 , so @xmath155 , which , however , is in conflict with @xmath156 having a condorcet winner , @xmath61 . thus we must have @xmath157 , which implies that @xmath158 , and thus @xmath159 . we again proceed by cases : if @xmath160 , we can add a voter @xmath161 to obtain a profile with condorcet winner @xmath54 , which contradicts participation . but then , if @xmath162 , we get that @xmath163 and , by another application of participation , that @xmath164 in contrast to the existence of condorcet winner @xmath64 , a contradiction . if @xmath66 , we add bad alternatives @xmath67 to the bottom of @xmath12 and all other voters . by , @xmath35 chooses from @xmath68 at each step , completing the proof . the following result establishes that our bound on the number of voters is tight . a very useful feature of our computer - aided approach is that we can easily add additional requirements for the desired voting rule . two common requirements for voting rules are that they should only return alternatives that are _ pareto - optimal _ and contained in the _ top cycle _ ( also known as the _ smith set _ ) ( see , * ? ? ? [ thm : positive - pairwise ] there is a condorcet extension @xmath35 that satisfies participation for @xmath57 and @xmath165 . moreover , @xmath35 is pairwise , pareto - optimal , and a refinement of the top cycle . the condorcet extension @xmath35 is given as a look - up table , which is derived from the output of a sat solver . the look - up table lists all @xmath166 weighted tournaments inducible by up to 11 voters and assigns each an output alternative ( see for an excerpt ) . the relevant text file has a size of 28 mb ( gzipped 4.5 mb ) . comparing this voting rule with known voting rules , it turns out that it picks a maximin winner in 99.8% and a kemeny winner in 98% of all weighted tournaments . moreover , the rule agrees with the maximin rule with lexicographic tie - breaking on 95% of weighted tournaments . the similarity with the maximin rule is interesting insofar as a well - documented flaw of the maximin rule is that it fails to be a refinement of the top cycle ( and may even return condorcet losers ) . our computer - generated rule always picks from the top cycle while it remains very close to the maximin rule . 80% of the considered weighted tournaments admit a condorcet winner , which uniquely determines the output of the rule ; this can be used to reduce the size of the look - up table . a,#1,(1,1,1,1,1,1 ) + a,#1,(1,1,1,1,1,-1 ) + a,#1,(1,1,1,-1,1,1 ) + a,#1,(1,1,1,-1,-1,1 ) + a,#1,(1,1,1,1,-1,-1 ) + a,#1,(1,1,1,-1,-1,-1 ) + b,#1,(-1,1,1,1,1,1 ) + b,#1,(-1,1,1,1,1,-1 ) + b,#1,(-1,-1,1,1,1,1 ) + b,#1,(-1,-1,-1,1,1,1 ) + b,#1,(-1,1,-1,1,1,-1 ) + b,#1,(-1,-1,-1,1,1,-1 ) + c,#1,(1,-1,1,-1,1,1 ) + c,#1,(1,-1,1,-1,-1,1 ) + a,#11,(11,9,3,7,1,-9 ) + c,#11,(5,-9,-1,-11,-1,7 ) + c,#11,(5,-9,-1,-11,-1,5 ) + c,#11,(3,-11,-1,-9,1,7 ) + c,#11,(3,-11,-3,-9,1,7 ) + c,#11,(3,-11,-3,-11,-1,7 ) + b,#11,(-1,3,-5,-3,5,-3 ) + b,#11,(-3,3,-7,-3,5,-3 ) + b,#11,(-3,1,-7,-3,5,-3 ) + c,#11,(-3,1,-5,-5,5,-1 ) + a,#11,(3,7,11,-3,9,11 ) + a,#11,(3,7,11,-3,9,9 ) + a,#11,(3,7,11,-5,9,11 ) a drawback of voting rules , as we defined them so far , is that that the requirement to always return a single winner is in conflict with basic fairness conditions , namely anonymity and neutrality . a large part of the social choice literature therefore deals with set - valued voting rules , where ties are usually assumed to be eventually broken by some tie - breaking mechanism . a _ set - valued voting rule _ ( sometimes known as a voting _ correspondence _ or as an _ irresolute _ voting rule ) is a function @xmath167 that assigns each preference profile @xmath12 a non - empty set of alternatives . the function @xmath168 is a _ ( set - valued ) condorcet extension _ if for every preference profile @xmath12 that admits a condorcet winner @xmath30 , we have @xmath169 . following the work of @xcite and @xcite , we seek to study the occurrence of the no - show paradox in this setting . to do so , we need to define appropriate notions of participation , and for this we will need to specify agents preferences over _ sets _ of alternatives . here , we use the _ optimistic _ and _ pessimistic preference extensions_. an optimist prefers sets with better most - preferred alternative , while a pessimist prefers sets with better least - preferred alternative . for example , if @xmath170 and @xmath171 , then an optimist with preferences @xmath10 prefers @xmath172 to @xmath173 , while a pessimist prefers @xmath173 to @xmath172 . with these notions , we extend the participation property to set - valued voting rules . [ def : set - valued - participation ] a set - valued voting rule @xmath168 satisfies _ optimistic participation _ if @xmath174 . a set - valued voting rule @xmath168 satisfies _ pessimistic participation _ if @xmath175 . a set - valued voting rule @xmath168 is called _ resolute _ if it always selects a single alternative , so that for all @xmath12 we have @xmath176 . a ( single - valued ) voting rule @xmath35 is naturally identified with a resolute set - valued voting rule @xmath168 ; if @xmath35 satisfies participation , then this @xmath168 satisfies both optimistic and pessimistic participation . hence , by , there is a ( resolute ) set - valued condorcet extension @xmath168 that satisfies both optimistic and pessimistic participation . however , there might be hope that allowing voting rules to be irresolute also allows for participation to be attainable for more voters , and indeed this is the case . [ thm : positive - optimist ] there is a set - valued condorcet extension f that satisfies optimistic participation for @xmath177 and @xmath178 , and also is pareto - optimal and a refinement of the top cycle . the sat solver indicates that no such set - valued voting rule is pairwise . is optimal in the sense that optimistic participation can not be achieved if we allow for one more voter . [ thm : optimist ] there is no set - valued condorcet extension that satisfies optimistic participation for @xmath55 and @xmath179 . let @xmath168 be such a function , and consider the following 10-voter profile @xmath12 : @xmath180 \draw node [ inner sep=5pt , scale=0.9 ] ( fakeroot ) at ( 0 , 0 ) { \begin{array}[t]{c@{\quad}c@{\quad}c@{\quad}c } \toprule 2&3&3&2\\ \midrule { \texttt a}&{\texttt b}&{\texttt c}&{\texttt d}\\ { \texttt b}&{\texttt d}&{\texttt a}&{\texttt c}\\ { \texttt d}&{\texttt c}&{\texttt b}&{\texttt a}\\ { \texttt c}&{\texttt a}&{\texttt d}&{\texttt b}\\ \bottomrule\\[-5pt ] \multicolumn{4}{c}{r } \end{array } } ; \draw node [ rectangle , minimum width=3 cm , minimum height=1.25 cm ] ( root ) at ( 0 , 0.6 cm ) { } child { node ( alpha ) { $ r_\alpha$ } child { node[condorcet ] ( 1 ) { $ { \texttt a}$ } edge from parent node[lbl , xshift=-0.2 cm ] { $ { } + 5\cdot { { \color{black!70!green}{\fontseries{b}\selectfont{\texttt b}}}}{{\color{black!70}\fontseries{l}\selectfont { \texttt a}{\texttt c}{\texttt d}}}$ } } child { node[condorcet ] ( r2 ) { $ { \texttt c}$ } edge from parent node[lbl ] { $ { } + 3\cdot { { \color{black!70!green}{\fontseries{b}\selectfont{\texttt a}}}}{{\color{black!70}\fontseries{l}\selectfont { \texttt c}{\texttt b}{\texttt d}}}$ } } edge from parent node[lbl , xshift=-0.3 cm ] { $ { } + 2\cdot { { \color{black!70!green}{\fontseries{b}\selectfont{\texttt a}{\texttt b}}}}{{\color{black!70}\fontseries{l}\selectfont { \texttt c}{\texttt d}}}$ } } child { node ( alphaprime ) { $ r_\alpha'$ } child { node[condorcet ] ( r4 ) { $ { \texttt b}$ } edge from parent node[lbl , xshift=-0.2 cm ] { $ { } + 3\cdot { { \color{black!70!green}{\fontseries{b}\selectfont{\texttt d}}}}{{\color{black!70}\fontseries{l}\selectfont { \texttt b}{\texttt c}{\texttt a}}}$ } } child { node[condorcet ] ( 1prime ) { $ { \texttt d}$ } edge from parent node[lbl ] { $ { } + 5\cdot { { \color{black!70!green}{\fontseries{b}\selectfont{\texttt c}}}}{{\color{black!70}\fontseries{l}\selectfont { \texttt d}{\texttt b}{\texttt a}}}$ } } edge from parent node[lbl , xshift=0.2 cm ] { $ { } + 2\cdot { { \color{black!70!green}{\fontseries{b}\selectfont{\texttt d}{\texttt c}}}}{{\color{black!70}\fontseries{l}\selectfont { \texttt b}{\texttt a}}}$ } } ; \end{tikzpicture}\ ] ] suppose that either @xmath181 or @xmath182 . ( the case of @xmath183 or @xmath184 is symmetric . ) then let @xmath185 . by optimistic participation , we then have either @xmath186 or @xmath187 . if we had @xmath186 , then also @xmath188 but this profile has condorcet winner @xmath61 , and if @xmath187 then also @xmath189 but this profile has condorcet winner @xmath54 . this is a contradiction . this argument extends to more than @xmath44 alternatives by appealing to a set - valued analogue of . inspecting moulin s original proof @xcite shows that it also establishes an impossibility for optimistic participation ( for @xmath190 voters ) . apparently unaware of this , @xcite explicitly establish such a result for @xmath191 voters and @xmath192 alternatives . it is worth observing that each step of the proof of involves _ adding _ voters to the current profile , and we never remove voters . in light of , this is the reason why the proof establishes a result for optimistic participation . if we restrict ourselves to deleting voters , we obtain a result for pessimistic participation . [ thm : pessimist ] there is no set - valued condorcet extension that satisfies pessimistic participation for @xmath55 and @xmath193 . on the other hand , for @xmath177 and @xmath194 , there exists such a set - valued voting rule . the proof has a similar structure to the proof of , displayed in . the initial profile of this proof is @xmath195 , taking @xmath12 to be the profile of . we further replace proof steps in which voters are added by similar ones where voters are deleted , and invoke pessimistic participation at each such step to obtain a contradiction . this result strengthens a result of @xcite , who show that for @xmath196 no set - valued condorcet extension satisfying a property called `` weak translation invariance '' can also satisfy pessimistic participation . our proof does not need the extra assumption , already works for @xmath177 alternatives , and uses just 14 instead of 971 voters . voters to the preference profile under consideration . ] as previously observed , adding voters in our impossibility proofs corresponds to optimistic participation , while removing voters corresponds to pessimistic participation . in the proof of , we use both operations , which allows for a tighter bound of just @xmath141 voters . in the set - valued setting , we can formulate this result in a slightly stronger way . [ thm : em - participation ] there is no set - valued condorcet extension that satisfies optimistic and pessimistic participation simultaneously for @xmath55 and @xmath146 . on the other hand , for @xmath177 and @xmath165 such a set - valued rule exists ( and also is pareto - optimal and a refinement of the top cycle ) . use the proof of , invoking optimistic participation for edges labelled with the addition of a voter ( @xmath197 ) , and invoking pessimistic participation for edges labelled with removal of a voter ( @xmath198 ) . on the other hand , the ( single - valued ) voting rule of clearly satisfies both optimistic and pessimistic participation . the preference extension combining the optimistic and pessimistic preference extension is also known as the _ egli - milner extension_. a _ probabilistic voting rule _ ( also known as a _ social decision scheme _ ) assigns to each preference profile @xmath12 a probability distribution ( or _ lottery _ ) over @xmath0 . thus , a probabilistic voting rule might assign a fair coin flip between @xmath54 and @xmath64 as the outcome of an election . formally , let @xmath199 : \sum_{{\mathtt x}\in a } \mathbf p ( { \mathtt x } ) = 1 \}$ ] be the set of lotteries over @xmath0 ; a lottery @xmath200 assigns probability @xmath201 to alternative @xmath30 . a probabilistic voting rule @xmath35 is a function @xmath202 . in this context , we say that @xmath35 is a _ condorcet extension _ if @xmath50 puts probability 1 on the condorcet winner of @xmath12 whenever it exists : if @xmath12 admits @xmath30 as the condorcet winner , then @xmath203 . as in the set - valued case , we need a notion of comparing outcomes in order to extend the definition of participation . here , we use the concept of _ stochastic dominance ( sd)_. let @xmath204 be a preference relation over @xmath0 , and let @xmath205 be lotteries . then @xmath206 is ( weakly ) sd - preferred over @xmath207 by @xmath102 if for each alternative @xmath30 , we have @xmath208 in this case , we write @xmath209 . for example , the lottery is sd - preferred to the lottery @xmath210 by a voter with preferences @xmath10 . a voter with preferences @xmath211 will feel the other way around . the main appeal of stochastic dominance stems from the following equivalence : @xmath209 if and only if @xmath206 yields at least as much von - neumann - morgenstern utility as @xmath207 under any utility function that is consistent with the ordinal preferences @xmath102 . using this notion of comparing lotteries , we can define participation analogously to the previous settings . a probabilistic voting rule @xmath35 satisfies _ strong sd - participation _ if @xmath212 for all @xmath15 and @xmath38 with @xmath13 . any ( single - valued ) voting rule @xmath35 can be seen as a probabilistic voting rule that puts probability 1 on its chosen alternative . if @xmath35 satisfies participation , then this derived probabilistic voting rule is easily seen to satisfy strong sd - participation . hence gives us a probabilistic condorcet extension that satisfies strong sd - participation for @xmath165 voters and @xmath177 alternatives . we now establish a connection between strong sd - participation and the set - valued notions of participation that we considered in . this connection will allow us to translate the impossibility results we obtained there to the probabilistic setting . to set up this connection , let us define the _ support _ of a lottery @xmath200 to be @xmath213 . let @xmath35 be a probabilistic voting rule satisfying strong sd - participation . let @xmath214 be the support of @xmath35 , @xmath215 for all profiles @xmath12 . then @xmath168 satisfies both optimistic and pessimistic participation . we only verify optimistic participation ; the pessimistic case is similar . let @xmath12 be a preference profile with electorate @xmath216 , and let @xmath217 be a voter with preferences @xmath8 . let @xmath218 , and let @xmath219 . we need to show that @xmath220 , by finding an alternative @xmath221 that is in the support of @xmath222 . but since @xmath35 satisfies strong sd - participation , we have @xmath223 where the strict inequality follows from the definition of the support and of @xmath30 . hence some alternative from @xmath172 is in the support of @xmath222 , as required . putting these results together with the impossibility result of , we obtain the following . [ thm : sd - participation ] there is no probabilistic condorcet extension that satisfies strong sd - participation for @xmath146 and @xmath55 . on the other hand , for @xmath177 and @xmath165 , such a probabilistic voting rule exists . resolves an open problem mentioned by ( * ? ? ? we have given tight results delineating in which situations no - show paradoxes must occur . as such , our results nicely complement recent advances to satisfy condorcet - consistency and participation by exploiting uncertainties of the voters about their preferences or about the voting rule s tie - breaking mechanism @xcite . due to unmanageable branching factors when there are @xmath192 alternatives ( and hence @xmath224 possible preference relations ) , we were unable to check using our approach whether no - show paradoxes occur with even less voters when the number of alternatives grows . it would be interesting to gain a deeper understanding of the computer - generated condorcet extension that satisfies participation for up to @xmath225 voters . so far , we only know that it ( slightly ) differs from all condorcet extensions that are usually considered in the literature . as a first step , it would be desirable to obtain a representation of this rule that is more concise than a look - up table . another interesting topic for future research is to find optimal bounds for a variant of the no - show paradox due to @xcite , in which participation is weakened to half - way monotonicity . their proof requires @xmath226 voters . christian geist is supported by deutsche forschungsgemeinschaft under grant br 2312/9 - 1 . dominik peters is supported by epsrc . part of this work was conducted while dominik peters visited tum , supported by the cost action ic1205 on computational social choice . f. brandl , f. brandt , c. geist , and j. hofbauer . strategic abstention based on preference extensions : positive results and computer - generated impossibilities . in _ proc . of 24th ijcai _ , pages 1824 . aaai press , 2015 . m. de condorcet . _ essai sur lapplication de lanalyse la probabilit des dcisions rendues la pluralit des voix_. imprimerie royale , 1785 . facsimile published in 1972 by chelsea publishing company , new york .

ever since the foundation of modern cosmology based on general relativity , there have been attempts to determine the cosmological parameters of our universe . gravitational lensing has been recognized as one of several such tools . for example , the results of radio surveys for gravitational lenses have recently been used to place significant constraints on cosmological models ( falco et al . it has also been realised that observations of supernovae at high redshift can be used for this purpose ( goobar & perlmutter @xcite ) , in particular for determining the value of the cosmological constant . in fact , several collaborations with this in mind are in progress , and the first sets of data show an intriguing hint of a non - vanishing cosmological constant in a universe consistent with having a flat geometry ( see , e.g. , riess et al . @xcite ; perlmutter et al . although the two groups which have published results mutually agree on the best - fit parameters , it is important to note that the effects of geometry are small ( on the order of half a magnitude ) , and the need to go to even higher redshift to get larger effects is obvious . when observing such distant sources , at redshift greater than unity , it is necessary to estimate the effects of lensing due to inhomogeneities in the matter distribution . this will be of growing importance when supernovae at very high redshifts become accessible , e.g. , with ngst , the next generation space telescope ( miralda - escud & rees @xcite ; dahln & fransson @xcite ) or with the dedicated supernova search satellite , snap ( perlmutter et al . , private communication ) . the literature on gravitational lensing is quite rich . much of the history and developments up to the early 1990s can be found in the excellent textbook by schneider , ehlers & falco ( @xcite ) . a new method for examining lensing effects has recently been proposed by holz & wald ( @xcite ; hw ) , one which has the virtue of lending itself easily to numerical calculations . in that work it is also shown that given a large and deep enough sample of standard candle supernovae , one may in principle distinguish between different matter distributions in the galactic halos responsible for the lensing . in particular the case of point masses gives a quite different distribution of magnifications ( or de - magnifications ) compared to the case of singular isothermal spheres ( see also metcalf & silk @xcite ) . the method of hw can be summarized as follows : first , a friedmann - lematre ( fl ) background geometry is selected . inhomogeneities are accounted for by specifying a matter distribution in cells which have an average energy density equal to that of the underlying fl model . following the history of observed light from a distant source , a light ray is traced backwards to the desired redshift by being sent through a series of cells , each time with a randomly selected impact parameter with respect to the matter distribution in the cell . between each cell , the fl background is used to update the scale factor and the expansion rate . by using monte carlo techniques to trace a large number of such light rays , statistics for the apparent luminosity of an ensemble of sources at a given redshift is obtained . the purpose of this paper is to generalize the method of hw in a number of ways and to investigate some issues related to the different lensing signatures in the luminosity distribution of supernovae caused by different mass distributions of the intervening lens population . the outline is as follows . in sect . [ sec : cosmodel ] it is demonstrated how the hw method can be generalized from matter in dust form to perfect fluids with pressure . [ sec : area ] is concerned with the interpretation of results as probability distributions . [ sec : devi ] presents the geodesic deviation and in sect . [ sec : hwsummary ] , we summarize and discuss some of the conceptual issues in the analysis of hw . in sect . [ sec : nfw ] , we include the possibility of using the more realistic halo matter distribution proposed by navarro et al . ( @xcite ; nfw ) based on their n - body simulation results , and in sect . [ sec : massdist ] we discuss the mass distribution of lensing galaxies . by using realistic density profiles and mass distributions of dark matter halos , we can obtain high accuracy results on gravitational lensing without the need of using the extensive full data sets of n - body simulations , as in some other methods ( wambsganss et al . @xcite ; premadi et al . @xcite ; jain et al . one strength of our ray - tracing monte carlo method is further that it can be continuously refined as more observational information , e.g. , on galaxy distributions is obtained . also , effects such as absorption by dust and other possibly @xmath0-dependent effects can be straightforwardly added to the algorithm . in sect . [ sec : results ] , some results from the generalizations of the method of hw are presented as luminosity distributions . in sect . [ sec : limit ] , we investigate some limitations of the numerical implementation of the method and in sect . [ sec : mi ] , we discuss multiple images . [ sec : cc ] investigates different consistency checks , sect . [ sec : formula ] contains convenient analytical fits to our numerical results and the paper is concluded with a discussion in sect . [ sec : disc ] . the starting point of hw is a newtonianly perturbed fl universe . the line element of robertson - walker form is then @xmath1 where @xmath2 is the scale factor , and @xmath3 is a constant - curvature three - geometry , for which @xmath4 indicates the sign of the curvature . ( we use geometrized units such that @xmath5 ) . the perturbation @xmath6 is assumed to satisfy a number of properties that will be employed when determining dominant terms . firstly , @xmath7 , reflecting that @xmath6 indeed is a perturbation . secondly , inhomogeneities are assumed to be significant , so that @xmath8 is small compared to spatial derivatives . thirdly , second - order spatial derivatives will be assumed to dominate over products of first - order ones ( e.g. , @xmath9 ) . in this paper , we have generalized the treatment of hw so that the matter content of the model can be taken to be a number of perfect fluids : @xmath10 where @xmath11 is the stress - energy tensor , @xmath12 is the energy density , @xmath13 the pressure of fluid @xmath14 , @xmath15 is the ( normalized ) fluid four - velocity , and @xmath16 is the metric corresponding to the line element ( [ eq : line - element ] ) . cosmologically reasonable equations of state are linear barotropic ones : @xmath17 . in this paper , the @xmath18 s are constants with @xmath19 . however , it is straightforward to allow for the equations of state to evolve with redshift , i.e. , @xmath20 . hw considered dust models with a cosmological constant , which corresponds to a two - fluid model , the first fluid having @xmath21 , and the second fluid homogeneously distributed with @xmath22 and @xmath23 . the field equations , corresponding to eqs . ( hw7,8 ) , become + _ the friedmann equation : _ @xmath24 _ the raychaudhuri equation : _ @xmath25 where the definitions of the hubble scalar @xmath26 and deceleration parameter . ] @xmath27 have been made explicit , and where @xmath28 denotes the spatial projection of contracted covariant derivatives . the pressure does not enter the friedmann equation , i.e. , the hubble scalar is entirely given by the energy densities and the curvature . the deceleration , on the other hand , is affected by the presence of pressure . comparing the above field equations with the field equations for the corresponding fl background ( with averaged densities and pressures @xmath29 , @xmath30 ) it can be shown [ eq . ( hw11 ) ] that @xmath6 satisfies a poisson equation @xmath31 , where @xmath32 , while @xmath33 vanishes . thus , the pressures result in homogeneous contributions , and the specification of an equation of state only makes sense in terms of the averaged energy density : @xmath34 . from the field equations for the background model follows + _ the energy conservation equation : _ @xmath35 if the fluids are assumed to couple only to gravity , they will be separately conserved , and each component in the sum of eq . ( [ eq : ec ] ) vanishes . note that it is only for dust ( @xmath36 ) that the mass within a volume co - moving with the hubble flow is constant ( @xmath37 ) . for other types of matter , the mass varies with time because of the work done by the pressure . also note that @xmath38 results in a constant energy density . thus , a homogeneously distributed fluid with @xmath38 is equivalent to a cosmological constant , or in terms of quantum field models , to vacuum energy . it is convenient to introduce density parameters @xmath39 which depend on time and therefore on redshift . assuming the above discussed equations of state the time dependence is governed by the following set of conservation equations ( see , e.g. , goliath & ellis @xcite ) @xmath40\omega_ih .\ ] ] hw proceed by performing a number of coordinate transformations to co - moving coordinates on isotropic form , eq . the inclusion of pressure results in a modification of this line element : @xmath41 dt^2 \nonumber \\ & + & \left[1 - 2\phi\right](dx^2+dy^2+dz^2 ) , \end{aligned}\ ] ] where @xmath42 , and the effective potential @xmath43 satisfies the poisson equation @xmath44 the lookback time of an event at redshift @xmath45 is the time difference ( with respect to the coordinate @xmath46 ) between the event and the present . by using @xmath47 , we get a relation between the lookback time and the redshift for a fl universe : @xmath48 to obtain @xmath49 , the friedmann equation , eq . ( [ eq : fe ] ) , is expressed in terms of redshift and the present densities @xmath50 . to do so , first note that the energy conservation equation , eq . ( [ eq : ec ] ) , together with the equation of state gives @xmath51 the hubble scalar can then be expressed @xmath52 where we introduced the curvature density parameter @xmath53 the luminosity distance to the event at redshift @xmath45 is given by a similar expression ( see , e.g. , bergstrm & goobar @xcite ) @xmath54 note that , when @xmath55 , the present scale factor can be expressed [ using ( [ eq : ook ] ) ] as @xmath56 . consider a narrow ( strictly speaking ; infinitesimal ) ray bundle connecting a source and an observer as depicted in fig . [ fig : focusing ] . the ray bundle can be focused either at the source or the observer . the relevant quantity when determining luminosity distances is the ratio of the area of a beam and the solid angle it subtends . of course , the smaller the distance , the brighter the source . in the method of hw , a beam focused at the observer with infinitesimal @xmath57 is integrated backwards using the geodesic deviation equation , and the resulting area at the source sphere is considered . in the lower panel of fig . [ fig : focusing ] , the beam has undergone additional focusing . a finite aperture at the observer will detect light from a larger @xmath57 of an isotropically emitting source , and a finite source will subtend a larger solid angle on the sky . since surface brightness is conserved , the source will be magnified . the magnification , @xmath58 , is defined as the ratio of the _ observed _ flux and the flux of the corresponding image in absence of any additional focusing . @xmath59 where @xmath60 denotes the solid angle subtended by the source in absence of lenses . hence the magnification will be proportional to the inverse of the area of the beam . in absence of any lenses , the ray bundle will have an area corresponding to the so called _ empty - beam _ area ( see sect . [ sec : devi ] ) , and we will therefore define the magnification as @xmath61 for further reference , see schneider , ehlers & falco ( @xcite ) . as noted in hw , some care is needed when interpreting results as probability distributions . this is due to the fact that individual ray bundles do not correspond to random source positions . this is illustrated in the left panel of fig . [ fig : focusing3 ] where we consider the situation of a ray bundle originating from an observer at @xmath62 at the center of a source sphere at some fixed redshift . from the right panel of fig . [ fig : focusing3 ] we see that focused beams sample a smaller fraction of the source sphere . since the fraction of the source sphere being sampled by a beam is proportional to the area of the beam , i.e. , inversely proportional to the magnification , magnified sources will be overrepresented if not compensated for . in hw , this is done by convoluting the obtained probability distribution , @xmath63 , with a factor proportional to @xmath64 , thus obtaining a new distribution @xmath65 . the method we use is very similar in that we designate a probability , proportional to the area of the ray bundle , to keep every event . the advantage of this rejection method is that each selected event represents a statistically `` real '' event . further discussions of the statistical weight given to random line - of - sights can be found in ehlers & schneider ( @xcite ) . the geodesic deviation equation for a light ray is ( see , e.g. , wald @xcite ) @xmath66 where @xmath67 is the deviation vector and @xmath68 is a null vector tangent to the light ray . the affine parameter @xmath69 is a parametrization along the light ray given by @xmath70 , where @xmath71 is the frequency of the photons in the beam . it is instructive to calculate the deviation in an exact ( homogeneous ) fl universe , analogous to eq . ( hw21 ) : @xmath72 thus , the pressure will cause additional focusing of the light beam . note that a cosmological constant , which corresponds to @xmath38 , does not affect the deviation . to derive the deviation experienced when propagating through a cell in the inhomogeneous model , hw start by writing the deviation equation in terms of a @xmath73 matrix @xmath74 , such that @xmath75 where @xmath76 ( hw28 ) . the properties of the geodesic deviation are contained in @xmath74 . for example , the area ( with orientation ) of the beam is given by @xmath77 . in the case of the homogeneous background friedmann - lematre cosmology , we denote the corresponding reference area @xmath78 , which will of course depend on the redshift , but also on the cosmological parameters of the fl background model . also , the area of a light beam that propagates without encountering any matter will be denoted @xmath79 . sometimes @xmath78 and @xmath79 are referred to as the _ filled - beam _ and _ empty - beam _ areas , respectively . the deviation equation , eq . ( [ eq : devia ] ) , can now be written as a set of difference equations ( hw35,36 ) @xmath80 with @xmath81 where the integration is along a straight line through the cell . next , @xmath82 is calculated from the line element eq . ( [ eq : lineco ] ) @xmath83 \!\eta_{\alpha\beta}\!\right\ } , \ ] ] where @xmath84 is the minkowski metric associated with the coordinates @xmath85 . thus , with non - vanishing pressure , there will be an additional contribution in @xmath86 that takes the form @xmath87 with @xmath88 where @xmath89 is an impact parameter , and @xmath90 and @xmath91 are the radius and volume of the cell , respectively . assuming linear barotropic equations of state , @xmath92 , eq . ( [ eq : jp ] ) becomes @xmath93 where @xmath94 is the mass of matter component @xmath14 in the cell . let us now examine the above result for the filled - beam case . ( hw39 ) gives the quantity @xmath86 for matter uniformly distributed in a ball of radius @xmath95 at the center of a cell of radius @xmath90 . setting @xmath96 in this equation , and adding our derived pressure contribution @xmath97 to this yields @xmath98 from which it is seen that a cosmological constant ( @xmath99 ) does not contribute , in consistency with eq . ( [ eq : fldev ] ) . as an interesting example , consider a flat cosmological model in which ordinary matter ( @xmath100 ) constitutes only 10 % of the total matter content , i.e. , @xmath101 , while the rest is made up of some homogeneously distributed `` x - matter '' with @xmath102 . the ordinary matter is assumed to be in the form of point masses ( we will return later to more realistic mass distributions ) . for this cosmology , garnavich et al . ( @xcite ) have put an upper limit on the parameter of the equation of state for an x - matter component using type ia supernova data . their limit is @xmath103 at 95 % confidence level . similarly , the supernova cosmology project ( perlmutter et al . @xcite ) has deduced a limit of @xmath104 ( 95 % cl ) for any value of @xmath105 . in fig . [ fig : xmatter ] , we compare the cases of @xmath106 , corresponding to a cosmological constant , @xmath107 , which may either arise from `` quintessence '' ( i.e. , a slowly evolving scalar field ) ( wang et al . @xcite ) or from a domain wall network ( battye et al . @xcite ) . for reference , we also show the result for @xmath108 , corresponding to dust ( i.e. , equivalent to an einstein de sitter model with @xmath109 , of which 90 % is homogeneously distributed ) . the choice of @xmath110 is made to allow for a direct comparison of the curve corresponding to @xmath106 in fig . [ fig : xmatter ] with fig . 7 in hw , with which it agrees well . note that the normalization differs between the three cases since @xmath78 depends on the cosmology . nevertheless , from the slope of the curves , one can conclude that lensing effects become more prominent for large negative @xmath111 . this is _ not _ what one would expect from studying the geodesic deviation equation , eq . ( [ eq : fldev ] ) , since negative pressure causes less focusing of a beam . however , for negative @xmath111 , distances will get larger for a fixed redshift , which means a larger number of lenses between the source and the observer ( see , e.g. , turner @xcite ) . the method of holz & wald ( @xcite ) which we adopt has a number of properties , which we summarize and comment upon here : * * local nature of lensing * + if @xmath6 is small within distances much smaller than the hubble radius @xmath112 , then the curvature ( and thereby the lensing effects ) is determined locally within @xmath113 . likewise , if we assume that there is no strong correlation on scales greater than a clustering scale @xmath114 , it follows that the curvature is determined locally within @xmath114 . * * relevant scales * + the question now is what relevant clustering scale to use . hw find it convenient to employ the typical separation between galaxies , and argue that galaxy clustering should have a negligible effect on magnification , as shown in some of their simulations . they also argue that the smallest relevant scale for matter clumping is of the order @xmath115 , since objects smaller than that would not give a coherent lensing effect over the angular size of most of the interesting sources . * * generality * + for the case of point masses , hw demonstrate that their results are fairly insensitive to the size of individual point masses . furthermore , their results are rather insensitive to clustering . this is tested by cutting out a tube in a uniform matter distribution and distributing `` stars '' within this tube . * * largest effects * + based upon their results , hw conjecture that randomly distributed point masses should cause the largest lensing effects of all possible mass distributions for a given average mass density . note that the point - mass case corresponds to the limit in which all matter clumps lie well within their einstein radii [ see eq . ( [ eq : re ] ) ] . * comment on relevant scales : * + the einstein radius is a characteristic length scale in the lens plane given by @xmath116 where @xmath117 is the schwarzschild radius of the lens and @xmath118 , @xmath119 and @xmath120 are angular diameter distances between lens and source , observer and lens , and observer and source , respectively . the corresponding length scale in the source plane , @xmath121 , is @xmath122 for lensing to be important , the source size , @xmath123 , has to be smaller than the length scale @xmath121 introduced by a point mass lens , @xmath124 . the source size of a type i supernova is @xmath125 cm at maximum luminosity . putting the source at @xmath126 and the lens at @xmath127 in a flat universe with @xmath128 and the dimensionless hubble parameter @xmath129 , we obtain @xmath130 i.e. , only compact objects with @xmath131 should be of importance through microlensing of supernovae . the work of hw has as a first treatment been concerned with the @xmath86 expressions for idealised cases like point sources ; singular , truncated isothermal spheres ( sis ) ; uniform spheres ; and uniform cylinders . however , another often - used matter distribution is the one based on the results of detailed n - body simulations of structure formation given by navarro et al . ( @xcite ) . they parametrize the density profile by @xmath132^{(\beta-\gamma)\delta } } , \ ] ] where @xmath133 is the critical density , @xmath134 is a dimensionless density parameter and @xmath135 is a characteristic radius . nfw found that halos found in n - body simulations ranging in mass from dwarf galaxies to rich galaxy clusters can be fit by eq . ( [ eq : nfw ] ) with @xmath136 , @xmath137 , @xmath138 . the potential for this density profile is given by @xmath139 where @xmath140 . the matrix @xmath86 can then be obtained analytically , see [ app : jnfw ] . the mass inside radius @xmath141 of a nfw halo is given by @xmath142,\ ] ] since we want to have the average mass density in each cell equal to that of the background cosmology , we get the relation @xmath143 setting @xmath144 , we obtain @xmath145},\ ] ] where @xmath146 . that is , for a given mass @xmath147 , @xmath148 is a function of @xmath149 . from the numerical simulations of nfw we also get a relation between @xmath148 and @xmath149 this relation is computed numerically by a slight modification of a fortran routine kindly supplied by julio navarro . of course , one wants to find an @xmath149 which gives both the required average density in each cell and compatibility with the numerical results of nfw . generally , @xmath149 will be a function of mass @xmath147 , the hubble parameter @xmath150 , @xmath105 , @xmath151 and @xmath0 . however , we use the result from del popolo ( @xcite ) and bullock et al . ( @xcite ) that @xmath149 is approximately constant with redshift . a comparison of luminosity distributions between point masses , isothermal spheres ( sis ) and the nfw matter distribution is presented in sect . [ sec : results ] . as discussed above , hw have shown that when studying luminosity distributions and other quantities where the null geodesics only differ by infinitesimal amounts , their method is insensitive to the individual masses and clustering of point mass lenses . for point mass lenses , the mass distribution and number density will therefore be of small importance . however , realistic modeling of galaxies does call for realistic mass distributions and number densities , i.e. , one has to allow for the possibility of @xmath90 to reflect the actual distances between galaxies , and for galaxy masses to vary in agreement with observations . one of the advantages of the method of hw is that any mass distribution and number density , including possible redshift dependencies , can easily be implemented and used as long as the average energy density in each cell agrees with the underlying fl model . to exemplify this , we derive a galaxy mass distribution , @xmath152 , by combining the schechter luminosity function ( see , e.g. , peebles @xcite , eq . 5.129 ) @xmath153 with the mass - to - luminosity ratio ( see , e.g. , peebles @xcite , eq . 3.39 ) @xmath154 normalizing to a `` characteristic '' galaxy with @xmath155 and @xmath156 , this can be written @xmath157 using eq . ( [ eq : schechter ] ) , we find that @xmath158 in fig . [ fig : mdistr ] , mass distributions for various values of @xmath159 are displayed . assuming that the entire mass of the universe resides in galaxy halos to @xmath160 at the truncation radius . ] we can write @xmath161 using the schechter luminosity function and the mass - to - luminosity ratio we get @xmath162 thus , by supplying values for @xmath163 , reasonably well - determined by observations , and @xmath164 and @xmath165 , on which the dependence of @xmath166 is weak , together with parameters @xmath159 and @xmath167 we can obtain an @xmath166 consistent with @xmath105 . note that assuming that the schechter luminosity function and the mass - to - luminosity ratio is valid over the entire luminosity range ( i.e. , putting @xmath168 and @xmath169 ) , we get @xmath170 where @xmath171 . a compilation of parameter values can be found in [ app : parametervalues ] . for @xmath172 , @xmath173 and @xmath174 , eq . ( [ eq : mref2 ] ) can be written @xmath175 the sis model is built on the assumption that stars ( and other mass components ) behave like particles in an ideal gas , confined by their spherically symmetric gravitational potential . the density profile is given by @xmath176 where @xmath177 is the line - of - sight velocity dispersion of the mass particles . [ note that this corresponds to eq . ( [ eq : nfw ] ) with @xmath178 , @xmath179 and @xmath180 . ] the mass of a sis halo truncated at radius @xmath95 is then given by @xmath181 we want this to be equal to the mass given by the schechter distribution @xmath147 , @xmath182 where , in addition to @xmath183 , we have introduced a characteristic velocity dispersion @xmath184 . combining the faber - jackson relation @xmath185 where @xmath186 is defined in eq . ( [ eq : schechter ] ) , with the mass - to - luminosity ratio , eq . ( [ eq : masstolum ] ) , we can substitute for @xmath177 in eq . ( [ eq : d ] ) , @xmath187 for values of @xmath184 and @xmath69 , see [ app : parametervalues ] . using eq . ( [ eq : mrefestimate ] ) , we can write the truncation radius for a halo with mass @xmath188 as @xmath189 a comparison between point masses and isothermal spheres for sources at @xmath190 in a @xmath109 , @xmath191 universe is presented in fig . [ fig : iso - nfw ] , where the full line corresponds to matter in the form of point masses . this plot agrees well with fig . 5 of hw . the long - dashed line corresponds to matter clumped in isothermal spheres with cut - off radius @xmath192 kpc . we choose this fix @xmath95 in order to be able to compare directly with fig . both lines agree very well with the results in hw . finally , the short - dashed line corresponds to isothermal spheres for which the cut - off radius @xmath95 has been calculated according to eq . ( [ eq : dcalc ] ) ( see sect . [ sec : siscutoff ] ) . for the presented model , this results in a typical cut - off of @xmath193 mpc , which explains why this case comes closer to the filled - beam value . in figs . [ fig : model-1][fig : model-9 ] , we compare the luminosity distributions obtained for nfw halos with the point - mass and sis matter distributions . comparing fig . [ fig : model-1 ] with fig . 18 and 20 of hw , we see that the point - mass cases agree very well . the sis luminosity distributions differs in that our results is shifted towards the filled beam value . this is due to the fact that the truncation radii used [ see eq . ( [ eq : destimate ] ) ] are an order of magnitude larger than the ( constant ) value of @xmath192 kpc used in hw . lensing effects are governed by the lensing potential of the galaxy model used , and the number of lenses between the source and the observer . as noted in the discussion of fig . [ fig : xmatter ] , a larger number of lenses will tend to make lensing effects for a given galaxy model more prominent . for a fixed source redshift and a fixed number density of lenses , the number of lenses will only be a function of the cosmology . we get the largest number of lenses in the @xmath194 cosmology , followed by the @xmath195 case , and the smallest number of lenses for @xmath196 . for point masses , the lensing potential will be unaffected by the cosmology . as discussed in sect . [ sec : hwsummary ] , point masses cause the largest lensing effects for a given average mass density . for a sis halo , the truncation radius , @xmath95 , is proportional to @xmath105 [ see eq . ( [ eq : mref ] ) and eq . ( [ eq : dcalc ] ) ] . since the potential for a truncated sis halo is equal to the point - mass potential for @xmath197 ( see fig . [ fig : sispotentials ] ) , the sis luminosity distribution will be shifted towards the point - mass distribution when lowering @xmath105 . in fact , the agreement between our results and those in hw is better for lower @xmath105 , as can be seen by comparing the curve corresponding to the sis luminosity distribution in fig . [ fig : model-2 ] with fig . taking also the number of lenses into consideration , we expect the largest lensing effects for the @xmath194 cosmology , followed by the @xmath195 case , and the smallest effects in the @xmath196 cosmology . for nfw halos , the value of the characteristic radius @xmath135 will depend on the cosmology used , the largest value obtained with the @xmath195 cosmology , followed by @xmath194 and the @xmath196 case . now , it can be shown that for a fixed halo mass , @xmath147 , the nfw potential will approach the point - mass potential when lowering @xmath135 , see fig . [ fig : nfwpotentials ] , @xmath198 . ] . therefore , we expect the closest resemblance to the point - mass case for @xmath196 , followed by @xmath194 and @xmath195 . the method of hw , or rather the numerical implementation of it , has some limitations due to the approximations used . since the area of the light beam when exiting a cell is computed by linear interpolation from the value before entering the cell [ see eq . ( [ eq : ada ] ) ] , we need very small cells or very many cells to get robust results . small cells means shorter interpolations while a large number of cells tends to average out the error in every interpolation . in order to quantify this effect we have made some tests with empty and filled beams using cell sizes obtained from the schechter matter distribution . in fig . [ fig : error ] , we have plotted the mean fractional error in area vs. redshift . to get a mean fractional error less than one percent , one would need to have @xmath199 . note that this is the average error of individual measurements ; the average value will be very close to the real value even at very small redshifts . for the cosmology used ( @xmath200 ) , a redshift of @xmath201 corresponds to @xmath202 cells another drawback with the hw method is that there is no way to keep track of multiple images from the same source . this poses no problem when the time delay between a primary image and secondary images is big enough to allow observational separation of the images . however , if image separation is not observationally feasible , we have to compensate for multiple images . this will be the case when considering microlensing , see sect . ( [ subsec.ola ] ) . a typical case of multiple imaging is shown in fig . [ fig : focusing4 ] . note that the secondary image has undergone a caustic crossing and that its parity has been reversed ( this corresponds to @xmath203 ) . also , the secondary image is demagnified compared to the primary image ( remember that it is the relation between @xmath57 and the area that determines the magnification ) . since the quantity @xmath57 is more or less constant in our method , areas of secondary images often becomes very big , i.e. , very dim . for a point mass lens , the time delay between two images with flux ratio @xmath141 is given by @xmath204 where @xmath205 for a lens at redshift @xmath127 and a flux ratio of two , we get @xmath206 as an illustration , to get a time delay of the order of weeks , we need @xmath207 . the image separation in the same case ( assuming a source at @xmath126 ) is given by @xmath208 i.e. , a mass of @xmath207 will give an image separation of @xmath209 . in the realistic case that a galaxy is causing the lensing , the surface brightness of the lens may make the detection of secondary images impossible ( porciani & madau @xcite ) . now , let us discuss what kind of information we can obtain from secondary images . in fig . [ fig : points ] we see what a source would look like in a universe filled with point - mass lenses . since the deflection angle from a point mass can be very large , there will , apart from the primary image , be one secondary image next to every point mass lying in the vicinity of the line - of - sight to the source ( not to mention all images with two or more caustic crossings ) .. ] most of these images will be _ very _ dim , i.e. not observable . in our implementation of the hw method , we can happen to follow any one of these . however , the vast majority of the flux will almost always be in the primary image , the only exception being the case where we have a point mass very close to the line - of - sight ( see fig . [ fig : points2 ] ) . in this case , the primary and the brightest secondary image will have almost equal flux . if we want to consider the total flux from a source we need to correct for these images . of course , the correction would have to be larger in the case of fig . [ fig : points2 ] . multiple - image corrections are discussed in sects . [ subsec.ola ] and [ subsec.olsis ] below . note that if we happen to follow any of the weak secondary images in fig . [ fig : points2 ] , we do not have any information on the flux of the primary image , i.e. , we would not be able to discriminate between the case of fig . [ fig : points ] and fig . [ fig : points2 ] . from this we conclude that weak secondary images does not give any valuable information concerning the total flux of a source . filling the universe with sis - type galaxies , the situation would be quite different . no longer can we have very large deflection angles , and multiple images will therefore be less common . , where @xmath177 is the velocity dispersion of the lens and @xmath210 is the impact parameter , i.e. , the magnitude of the deflection does not depend on @xmath210 . ] since we are typically dealing with galaxy masses , images will be separable and we need only to consider secondary images bright enough to be observable . since for ( almost ) every source , there will be at most one secondary image , it is a more straightforward procedure to associate to every secondary image a primary image . here , we will discuss reasonable corrections when images are non - separable . since we have a very large number of images with two or more caustic crossings contributing to the flux of the total image , we can not correct for each one of them . instead , we will make a first correction for the brightest secondary image using the approximation that lensing effects from a single lens ( cell ) is dominant . this is the so called one - lens approximation . possible tests of the validity of this assumption is discussed in sect . [ sec : cc ] . its plausibility is also reinforced by figs . 8 and 9 in hw . we can now use standard analytical expressions for one - lens systems for this correction : @xmath211 in fact , this formula is valid also for secondary images but one then has to include the sign in @xmath58 , i.e. , @xmath212 . however , eq . ( [ eq : mucorr ] ) is only valid for the brightest secondary image , and one should therefore not apply it to very weak , secondary images . also note , that to make the correction self - consistent , the magnification should be defined as the ratio of the expected area if the most prominent cell is empty and the real area . it is easy to see that the most important effect of this correction is to make the high magnification tail more pronounced . it is as straightforward to compute the total magnification using the one - lens approximation with a sis - lens . for a primary image we get @xmath213 for a secondary image we get @xmath214 since images are separable in the majority of cases , this is perhaps most interesting to use as a consistency check for the total luminosity . in order to validate our implementation of the method , we have performed a number of tests . * * filled and empty beam distances * + first in order to get an independent check we have compared results from our implementation of the hw method using empty cells and cells with a homogeneous dust component with analytical empty - beam and filled - beam results ( using the angsiz routine , see kayser et al . this has been done for a variety of cosmologies and the agreement is good even at very high @xmath0 ( less than 1 % discrepancy up to @xmath215 ) . * * comparison with hw * + second , we have reproduced most of the plots presented by hw . we have also tested that primary images satisfy @xmath216 , see fig . [ fig : areasum ] . * * average luminosities * + imagine a source situated at the center of the sphere in the lower panel of fig . [ fig : focusing3 ] and observers sitting on the sphere at some constant redshift . since the number of photons emitted by the source , as well as the area of the observers sphere is constant , i.e. independent of any details of the distribution of matter as long as the average matter density is the same , observers will on average receive the same flux , regardless of the matter distribution . this average will be equal to the fl value , @xmath217 when computing @xmath58 for a source , one has to consider the total flux if there are multiple images . if we consider primary images only , without any corrections , we would expect to obtain a @xmath58 slightly lower than the fl value ( since we lose some flux in secondary images ) . + note that this does not mean that the average area should equal the fl area . instead , @xmath218 i.e. , the fl area is the harmonic mean of the event areas , not the arithmetic ( remember that @xmath219 ) . since we lose some flux in secondary images , we have ( considering primary images only ) @xmath220 see fig . [ fig : lumtest ] for a plot of the average magnification using point mass lenses and nfw lenses . the result for sis halos is not plotted since the result is indistinguishable from the nfw result . note that since the deviation from the filled - beam value is due to the absence of the flux contained in secondary images , fig . [ fig : lumtest ] indicates that the flux in secondary images is very low when using nfw or sis halos , even at very high redshifts . + the relationship between this test and the area test of hw is discussed in [ app : meanarea ] . since the numerical work involved in computing the results presented in this paper is quite extensive , it may be convenient to use simple analytical fitting formulas for the probability distributions for the different cosmologies , which we now give . the probability distribution of the deviations from filled - beam magnitudes induced by sis and nfw lenses , @xmath221mag(filled beam)@xmath222mag(lensed ) , can be parametrized by an exponential function folded with a gaussian distribution : @xmath223 thus , for any combination of cosmological parameters , the corrections due to gravitational lensing can be described by three parameters : the mean of the gaussian distribution , @xmath58 , the standard deviation @xmath224 , and the magnification tail , @xmath225 . the expression in eq . ( [ eq : gausexp1 ] ) can be rewritten so that ( after normalization ) a simpler integral is invoked : @xmath226 where @xmath227 in fig . [ fig : sisdis ] the fits to the magnification distributions are shown for sis lenses for source locations @xmath126 , @xmath228 and @xmath229 and for three different cosmologies . in fig . [ fig : nfwdis ] the same source locations and cosmologies are used but the lenses are of nfw type . the parametrization of eqs . ( [ eq : gausexp1 ] ) and ( [ eq : gausexp2 ] ) may be used to study the shape of the magnification distribution as a function of redshift and cosmological parameters , as shown in figs . [ fig : sispar ] and [ fig : nfwpar ] . the estimates of the cosmological parameters may be further refined and the understanding of the mass distribution in galaxy halos could be deduced by fitting the parameters @xmath225 , @xmath224 and @xmath58 to the residual magnitude distributions in future high - redshift supernova searches , such as the snapsat project or at ngst . the accuracy of this method will be discussed in a forthcoming paper ( bergstrm et al . , in preparation ) . in this paper , the method of holz & wald ( @xcite ) has been generalized to include the effect of non - vanishing pressure . a motivation for this exercise is to use this method as part of a large package for a detailed simulation of most aspects high - redshift supernova determination of the geometry of the universe ( bergstrm et al . , in preparation ) . the possibility to allow for various matter contributions is then desirable . another improvement has been to allow for a more realistic mass profile of the galaxy halos , the navarro - frenk - white distribution , where we have given analytical formulas for the geodesic deviation integrals . with this improvement , and by adding matter in our model universes according to a schechter function with parameters fixed by observation , we think we have achieved a sufficient degree of sophistication to make realistic predictions of the lensing effects of present and future deep supernova searches . our results show , in agreement with previous work ( holz & wald @xcite ; metcalf & silk @xcite ) that the ( unlikely ) case of having most of the dark matter distributed in the form of point masses can be quite easily distinguished from more realistic halo models , if a sufficient number of supernovae can be detected at large redshifts . it will be more difficult to distinguish between a singular isothermal sphere model and the n - body results of navarro , frenk and white , unless @xmath105 is large and one can measure supernovae out to @xmath230 ( see fig . [ fig : model-7 ] ) . however , this also means that our method of determining the properties of the lens population by matching the measured schechter function with the required @xmath105 gives quite robust results for the predicted magnifications of type ia supernovae over a broad range of reasonable halo models . this will be helpful when making realistic predictions for future deep supernova searches . the authors would like to thank daniel holz and bob wald for helpful comments , and julio navarro for providing his numerical code relating the parameters of the nfw model . in this appendix , we give analytic expressions for the integrals ( [ eq : jab ] ) needed for the evolution of the optical matrix for a bundle of light which passes near a mass distribution of the nfw form . we assume a spherically symmetric matter distribution , the impact parameter is @xmath89 , the characteristic radius of the nfw model is @xmath135 and the cell radius is @xmath90 . when @xmath231 , it follows that @xmath232 \nonumber \\ & & -\frac{\sqrt{r_c^2-b^2}(2r_c^2+b^2)}{b^2r_c^3 } \ln\left(1+\frac{r_c}{r_s}\right ) \nonumber \\ & & \left . + \frac{2r_s(r_s^2 - 2b^2)}{b^2|r_s^2-b^2|^{3/2}}f(b , r_s , r_c)\right\ } , \end{aligned}\ ] ] @xmath233 \nonumber \\ & & -\frac{\sqrt{r_c^2-b^2}(2r_c^2+b^2)}{b^2r_c^3 } \ln\left(1+\frac{r_c}{r_s}\right ) \nonumber \\ & & \left . + \frac{2r_s\,{\rm sgn}(r_s - b)}{b^2|r_s^2-b^2|^{1/2}}f(b , r_s , r_c)\right\ } , \end{aligned}\ ] ] @xmath234 and the limit @xmath235 is : @xmath236 \right.\nonumber\\ & & -\frac{\sqrt{r_c^2-b^2}(2r_c^2+b^2)}{b^2r_c^3 } \ln\left(1+\frac{r_c}{b}\right)- \nonumber\\ & & \left.\left(\frac{r_c - b}{r_c+b}\right)^{3/2 } \frac{r_c^2 + 3(r_c+b)^2}{3r_c^2b^2 } \right\ } , \nonumber\\ ( j^{\rm nfw})^y\!_y & = & -2 c \left\ { \frac{\sqrt{r_c^2-b^2}}{r_c^2(r_c+b ) } \right . \nonumber \\ & & + \frac{2}{b^2 } \ln\left[\frac{r_c}{b}+\sqrt{\left(\frac{r_c}{b}\right)^2 - 1}\right ] \nonumber \\ & & -\frac{\sqrt{r_c^2-b^2}(2r_c^2+b^2)}{b^2r_c^3 } \ln\left(1+\frac{r_c}{b}\right ) \nonumber \\ & & \left . -2\frac{\sqrt{r_c^2-b^2}}{b^2(r_c+b ) } \right\ } .\end{aligned}\ ] ] the constant @xmath237 is given by @xmath238 where @xmath239 . here , we have compiled various parameter values , relevant for the mass distribution discussed in sect . [ sec : massdist ] . in what follows , absolute magnitudes are denoted by @xmath240 , while masses are denoted by @xmath147 . * lin et al . * : : see eq . ( 32 ) of ( lin et al . @xcite ) . + * @xmath241 * @xmath242 * @xmath243 + lin et al . gives its range of validity as @xmath244 , which is about @xmath245 magnitudes around their characteristic magnitude @xmath246 . this translates into a luminosity range @xmath247 , and a mass range @xmath248 , which , with @xmath249 , gives @xmath250 $ ] . * kirshner et al . * : : ( kirshner et al . @xcite ; ki ) + * @xmath251 ( ki19 ) + @xmath252 ( ki22 ) * @xmath253 ( ki18 ) + @xmath254 ( ki21 ) * @xmath255 ( see binney & tremaine @xcite ) . * @xmath256 ( ki23 ) * peebles * : : ( peebles @xcite ; pe ) + * @xmath174 ( pe3.39 ) * @xmath257 ( pe5.130 ) * @xmath258 in the @xmath259 magnitude system where @xmath260 . ( pe5.139,5.140 ) * @xmath261 ( pe5.141 ) * @xmath262 ( pe5.142 ) * @xmath263 ( pe3.35 ) * @xmath264 ( pe3.35 ) in the numerical work in this paper , we have used * @xmath174 * @xmath173 * @xmath265 * @xmath266 * @xmath267 * @xmath263 * @xmath264 holz and wald ( @xcite ) use the fact that the area of the boundary of the past of an observer should be very nearly equal to the area of the past light cone in the underlying fl model . from this they deduce that adding up all areas counting beams with an even number of caustics as positive and beams with an odd number as negative one should very nearly obtain the fl result , i.e. , @xmath268 , where the zero - subscript denote unweighted quantities . here we investigate whether this claim is consistent with the fact that the average magnification should be equal to the fl magnification . imagine that , without weighting , we get an area distribution @xmath63 , where we assume that lensing effects are weak enough to only give primary images . according to hw , we then have @xmath269 in our simulations , the area distribution will be a convolution of @xmath63 and a probability distribution @xmath270 , @xmath271 since we want the harmonic mean of areas to be equal to the fl value , we get @xmath272 that is , the two consistency checks at least show consistency in the case where we only have primary images . the inclusion of non - primary images is highly non - trivial since the concept of negative areas is difficult to interpret as luminosities . one possibility is to combine this approach with the one - lens approximation method for multiple image correcting . however , this is difficult to do analytically since one would then need the form of @xmath63 .

massive multiple - input multiple - output ( mimo ) ( also known as large - scale mimo ) is considered as one of the most promising technologies for next generation wireless communication systems @xcite , since it can provide very high spectral and energy efficiencies . besides , for massive mimo systems , the effects of small - scale fading and interference can be significantly reduced with linear signal processing , such as maximal - ratio transmission ( mrt ) and zero - forcing ( zf ) @xcite . the performance of massive mimo single / multi - cell downlink systems has been widely studied in the literature . in @xcite , the asymptotic signal - to - interference - plus - noise - ratio ( sinr ) for linear precoding in single - cell _ frequency - division - duplex _ ( fdd ) systems was derived based on which the optimization of the user number was investigated . in @xcite , the power scaling , i.e. , how fast the transmission power can decrease with the antenna number while maintaining a certain sinr level , for multi - cell _ time - division - duplex _ ( tdd ) systems was given based on asymptotic sinr analysis , where the pilot contamination was shown to be the only performance limit . in @xcite , capacity lower bounds and energy efficiency of mrt and zf in single - cell tdd systems were studied and compared with each other . in the parallel field of uplink systems , similar results were derived in @xcite , @xcite . these works mainly focus on the derivations of asymptotically deterministic expression for the sinr , and the performance analysis only consider some special cases , e.g. , the constant or linearly increasing user number with respect to the base station ( bs ) antenna number , the power scaling for perfect or imperfect channel state information ( csi ) with equal training and transmission power and constant user number . recently , along this route , the performance improvement for using coherent detection based on the dedicated downlink channel estimation was shown in @xcite , especially for large but finite antenna numbers . meanwhile , the impact of other practical factors such as channel aging @xcite , hardware limits @xcite , and rician fading @xcite on the performance of massive mimo systems was studied . however , a comprehensive scaling law analysis for multi - cell multi - user massive mimo is still missing in existing works . another important problem in the performance analysis of massive mimo is when the derived asymptotic results are applicable in practical systems . while many works derived deterministic sinr approximations via calculating the expected value of the sinr or its compenents , the analysis on whether the instantaneous sinr becomes asymptotically deterministic and converges to the derived expressions is sometimes missing . in @xcite , this problem was studied through the so called massive mimo efficiency for the ideal situation that the noise , channel estimation error , and interference are negligible compared with pilot contamination , where the dependence of the massive mimo efficiency on the ratio of spatial degrees of freedom / antenna number to the user number and the product of antenna number and transmit power was shown by simulations . in @xcite , via simulations , the convergence speed of the strong law of large numbers with respect to the antenna number and the convergence of the sinr of mrt and zf to their expected values were both analyzed for fixed ratio of antenna number to user number . it was shown that the sinr of mrt has slower convergence than that of zf , while the explicit relationship between the convergence speed and important parameters , i.e. , antenna number , user number , transmit power , and channel training energy ( csi quality ) , is missing . in this work , by drawing lessons from the analyzing method and the mean square convergence definition in @xcite , we provide a comprehensive performance scaling law analysis for multi - cell multi - user tdd massive mimo downlink systems with channel spatial correlation , csi error , and pilot contamination and investigate the applicability of the derived asymptotic scaling law in practical systems with large but finite antenna numbers . extensive simulations are also conducted to validate the analytical results . our main new contributions are summarized as follows . * we consider a multi - cell multi - user massive mimo network with correlated channel model , csi error , and pilot contamination . in dealing with pilot contamination , recent advancements in pilot contamination elimination ( pce ) @xcite is also considered , while in existing works it is treated as a constant bottleneck of massive mimo . an approximate model for the pce effect is proposed . * a lower bound on the system sum - rate is derived based on which a general performance scaling law is obtained . the result reveals the effect and joint interaction of extensive network parameters , i.e , the channel training energy , the transmission power , and the user number . in the scaling - law analysis , a general model is used where the parameters have continuous scaling exponents with respect to the number of bs antennas , while in most existing work , only a few discrete values for the power scaling exponent , e.g. , @xmath0 ; @xmath1 ; @xmath2 , are allowed . * to understand the applicability of the derived scaling law in practical systems with large but finite antenna number , the effect of antenna number on its accuracy is analyzed and evaluated by simulations . * we conduct quantitative analysis on the asymptotically deterministic property of random instantaneous sinr using the mean square convergence . sufficient conditions for the instantaneous sinr to be asymptotically deterministic are given , from which the tradeoff among different parameter scales is discovered . in @xcite , the performance scaling law for multi - user massive mimo relay networks was extensively studied . while our work adopts the mean square convergence idea and the sinr analysis method from @xcite , it is fundamentally different in both network model and channel model . we consider the multi - cell multi - user massive mimo downlink networks with a general spatially correlated channel model . furthermore , as crucial issues for multi - cell systems , the pilot contamination and pce are taken into account in our work . as a result , the corresponding derivation procedures and analytical results are significantly different . in addition , the analysis on the effectiveness of the derived scaling law for practically large but finite antenna numbers is completely new . the remaining of the paper is organized as follows . in section ii , the channel model and system model including both the multi - cell channel estimation with pilot contamination and the downlink transmission with mrt are introduced . section iii shows the performance scaling law result , discussions on typical network scenarios , and the analysis on the applicability to practical systems with large but finite antenna numbers . in section iv , sufficient conditions for the sinr to be asymptotically deterministic are derived . section v shows simulations and conclusions are provided in section vi . in this paper , bold upper case letters and bold lower case letters are used to denote matrices and vectors , respectively . for a matrix @xmath3 , its conjugate transpose , transpose , and trace are denoted by @xmath4 , @xmath5 and @xmath6 , respectively . @xmath7 and @xmath8 are the mean and variance operators . @xmath9 means equivalence in probabilistic distribution . the notation @xmath10 means that when @xmath11 , @xmath12 and @xmath13 have the same scaling with respect to @xmath14 , in other words , for large @xmath14 , there exists positive constants @xmath15 and @xmath16 , such that @xmath17 . consider a multi - cell multi - user massive mimo network consisting of @xmath18 cells with one @xmath14-antennas bs and @xmath19 scheduled single - antenna users in each cell . the bss and users are assumed to operate a tdd protocol with universal frequency reuse . the uplink spatially correlated channel from user @xmath20 in cell @xmath21 to the antennas of bs @xmath22 in cell @xmath22 can be written as @xmath23 where @xmath24 is the channel correlation matrix and @xmath25 is the independent fast - fading channel vector . the following assumptions on @xmath24 are made by taking into consideration the pilot contamination , pce , and tractability analysis . the set @xmath26 denotes the index set of all cells having pilot contamination to cell @xmath22 while @xmath27 denotes the index set of remaining cells whose pilot contamination is eliminated . we assume that for all @xmath28 and @xmath29 , @xmath30 where the channel direction matrix @xmath31 is an @xmath32 unitary matrix and @xmath33 with @xmath34 $ ] indicates the channel correlation level . for example , elements of the channel vector become independent and identically distributed ( i.i.d . ) when @xmath35 . also , @xmath36 is the pass - loss fading factor where @xmath37 and @xmath38 . the parameter @xmath39 denotes the large - scale fading between different cells and the intra - cell one is normalized to @xmath1 . for @xmath40 , the pilot contamination from cell @xmath21 to cell @xmath22 is eliminated by using the spatial information in channel correlation matrix . thus we have @xmath41 this channel model considers both the channel correlation and the inter - cell pass - loss fading . it is similar to the one used in ( * ? ? ? * section iii ) where motivations and justifications for this model can be referred to . new feature of the modeling in and is that the possible effect of pce , i.e. , the bss schedule users with relatively less pilot contamination for channel training and transmission , can be embodied . in @xcite , several efficient methods by exploiting the statistical channel information have been proposed to alleviate the pilot contamination due to the pilot reuse among @xmath42 cells @xcite , @xcite . generally speaking , the overlap among users channel directions can be in any degree leading to different pilot contamination level @xcite . for explicit analytical results , in , we assume that channels from both users in cell @xmath22 and users in other cells with pilot contamination to cell @xmath22 have the same @xmath3 matrix , while in other cells , pilot contamination is eliminated by having the orthogonal channel directions as shown in . this is a discretization approximation to show the average effect of pce and the extension to the case of more general channel direction distribution is left for future research . in the uplink training phase , users in cell @xmath21 transmit mutually orthogonal pilot sequences which are reused by users in other cells . denote the length of the pilot sequence as @xmath43 ( @xmath44 for reliable performance @xcite ) . assume that all @xmath19 users in each cell use the same training power @xmath45 . define @xmath46 which is the channel training energy . by correlating the received training signals with the pilot sequence of user @xmath47 , bs @xmath21 has the observation @xmath48 as @xmath49 where @xmath50 is the local noise . with the minimum - mean - squared - error ( mmse ) estimation @xcite , the estimate @xmath51 is given as @xmath52 we assume that all the @xmath42 cells have the same pilot contamination level , i.e. , @xmath53 $ ] for all @xmath21 . by defining @xmath54 and utilizing , , we have @xmath55 , for @xmath56 where @xmath57 from , it can be shown that @xmath58 this will be utilized to simplify the derivations in later parts . define @xmath59 , which is the estimation error . due to the feature of mmse estimation and gaussian distribution , @xmath60 and @xmath61 are mutual independent and @xmath62 where @xmath63 note that @xmath64 denotes the csi quality , which is dependent on both channel training energy and pilot contamination . when @xmath65 , we have @xmath66 and the csi quality is only affected by pilot contamination . if further perfect pilot contamination elimination is achieved , we have @xmath67 , @xmath68 and perfect csi can be obtained . the mrt precoding is considered due to its computational simplicity , robustness , and high asymptotic performance @xcite . by assuming channel reciprocity , the received signal @xmath69 of user @xmath20 in cell @xmath21 is given as @xmath70 where @xmath71 denotes the average downlink transmission power or the signal - to - noise - ratio ( snr ) , @xmath72\in\mathbb{c}^{m \times k}$ ] is the mrt precoding matrix , @xmath73^t\in\mathbb{c}^{k}\sim\mathcal{cn}({\bf0},{\bf i}_k)$ ] contains the data symbols for the @xmath19 users in cell @xmath22 , and @xmath74 is the receiver noise . further , @xmath69 can be rewritten as @xmath75 in this formulation , the instantaneous effective gains of the estimated channel vector , i.e. , @xmath76 , are assumed to be known at the user side for coherent detection . in several recent works , downlink training for the estimation of the instantaneous effective gains is considered @xcite . in this work , we use @xmath77 as an approximate estimate of @xmath78 to indicate the effect of instantaneous effective channel gain and at the same time to delete the insignificant details of the downlink training to better focus on the main purpose of this paper . this assumption has been used in recent works on tdd massive mimo relay networks @xcite . since the bs contaminated by one user will create interference to it in the downlink transmission , we define @xmath79 and @xmath80 which are the index sets of cells or bss which are contaminated and not contaminated by users in cell @xmath21 in the uplink training period , respectively . for simplicity , we assume @xmath81 , for all @xmath21 which is reasonable in the average sense . define @xmath82which are respectively the normalized power of the desired signal , intra - cell interference , inter - cell interference , and effective noise created by unknown csi error . their different normalization factors are used to guarantee that the mean of each term is bounded as shown in the next section . in the derivations of and we utilize @xmath83 for @xmath84 based on , and . with the results in - , the sinr of user @xmath85 in cell @xmath21 can be written as : @xmath86 the achievable ergodic rate is thus @xmath87 this section studies the general performance scaling law of the massive mimo network . first , the means and scvs of all sinr components are calculated . consequently , a sum - rate lower bound based on jensens inequality is derived via exploiting the asymptotically deterministic property of the desired signal power . then , the general performance scaling law is obtained and typical network scenarios for the sinr to be non - decreasing are discussed . finally , conditions for the scaling law to be applicable for large but finite antenna numbers and extended discussions on the possible effect of pce are both given . different from directly using the almost sure convergence based lemmas @xcite , @xcite , @xcite to calculate the deterministic equivalence of , we consider the mean square convergence @xcite and use the scv ( defined as the square of the ratio of the standard deviation over the mean of a random variable ) in the analysis @xcite . specifically , for a random variable sequence @xmath88 with a bounded mean , it converges in mean square to a deterministic value @xmath89 , denoted as , @xmath90 , if @xmath91 . further , @xmath88 is said to be asymptotically deterministic in the sense of mean square convergence if @xmath92 @xcite . to utilize the mean square convergence , we first calculate the means and scvs of all sinr components . the following lemma is obtained . [ lemma1 ] @xmath93 see appendix [ appendix a ] . [ remark 1 ] by noticing that @xmath94 $ ] , and @xmath95 $ ] , it can be seen that the random variables @xmath96 , @xmath97 and @xmath98 , @xmath99 , all have bounded means . moreover , apparently the scv of @xmath100 approaches 0 as @xmath11 with a fast linear convergence rate . thus for large @xmath14 , @xmath96 can be approximated with its mean . however , for @xmath97 , @xmath98 and @xmath101 , their scvs depend on the scaling of @xmath19 , @xmath102 and @xmath103 which are not necessarily @xmath104 . thus , at this point , we do not use their mean values for further derivations , but keep their random nature for more careful analysis . therefore , the sinr expression in becomes @xmath105 based on which the following result on the achievable rate can be obtained . [ lemma 2 ] the achievable rate of user @xmath20 in cell @xmath21 in the massive mimo network has the following lower bound : @xmath106 where @xmath107 as @xmath108 is a convex function of @xmath89 @xcite , according to jensen s inequality , we have @xmath109 by applying the sinr approximation in , it can be shown that @xmath110 by defining the effective asymptotic sinr as @xmath111 and directly using the results in - , the lower bound in can be obtained . in existing works @xcite , closed - form approximations of the ergodic rate were derived by using ( * ? ? ? * lemma 1 ) , i.e. , @xmath112 . however , in lemma [ lemma 2 ] we utilize the definite asymptotically deterministic property of the desired signal power term only in the sinr expression to derive the lower bound of the ergodic rate . due to , understanding the scaling law of the achievable sum - rate is transformed to understanding the scaling law of @xmath113 where the subscript of user index and cell index in is omitted due to homogeneous network assumptions . in this subsection , the scaling law of the asymptotically effective sinr is analysed to show how the system performance is affected by network parameters and pilot contamination level . for all system parameters , including the number of users @xmath19 , the transmission power @xmath71 and the channel training energy @xmath102 , a general scaling model with respect to the bs antenna number @xmath14 is used . note that other system parameters , i.e. , the channel spatial correlation metric @xmath114 , the normalized inter - cell large scale fading @xmath115 are constant with respect to @xmath14 . assume that @xmath116 where the exponents @xmath117 , @xmath118 , and @xmath119 represent the scales of @xmath19 , @xmath120 , and @xmath121 with respect to @xmath14 ; and @xmath122 represents the scale of the effective sinr . for practical ranges of the network parameters , it is assumed that @xmath123 . the reasons are given as follows . firstly , in typical applications of massive mimo , @xmath19 either increases with @xmath14 or keeps constant . thus @xmath124 . on the other hand , @xmath19 can not exceed @xmath14 since the maximum multiplexing gain is @xmath14 . thus , @xmath125 . secondly , due to the high energy efficiency requirement of massive mimo@xcite , the transmission power @xmath71 should not increase with @xmath14 . but it can decrease as @xmath14 increases with the condition that its decreasing rate is lower than the increasing rate of @xmath14 . this is because that the maximum achievable array gain for compensating the loss of receiving energy is @xmath14 . thus @xmath126 . for the same reasons , @xmath127 is considered . another important parameter for the performance analysis is @xmath103 which quantifies the pilot contamination level . existing works @xcite , @xcite show that pilot contamination does not vanish as @xmath14 grows for i.i.d . channel @xcite or spatially correlated channel with uniform channel directions @xcite . for more practical channel models , by exploiting the large spatial degrees of freedom provided by the massive antenna array at the bs , several efficient methods have been proposed to reduce pilot contamination @xcite . specifically , the difference among users channel directions @xcite or locations @xcite and smart pilot allocation based on them @xcite were utilized to avoid pilot contamination , and corresponding simulations showed that as @xmath14 increases , lower pilot contamination level can be achieved . in our work , the effect of pce is reflected by the @xmath128 parameter in our modeling , which is assumed to be a constant . the special case of @xmath129 represents perfect pce , i.e. , highly correlated channels . ] , while @xmath130 represents that pilot contamination is not fully annihilated and on average there are @xmath128 cells with pilot contamination to the cell of interest . with the above scaling model , the general performance scaling law is given as follows . [ theorem 1 ] for the massive mimo network with mmse channel estimation and mrt , if perfect pce is achieved , the performance scaling law is @xmath131 for imperfect pce , the performance scaling law is @xmath132 the scaling exponent of @xmath133 in is determined by the maximal scaling exponent of the terms in its denominator . directly from , it can be seen that given aforementioned parameter range , either @xmath134 or @xmath135 or both have the highest order with respect to @xmath14 . further , since @xmath64 and @xmath102 have the same scaling from , ( [ eq30 ] ) and ( [ eq31 ] ) can be obtained from ( [ eq28 ] ) . since having decreasing performance with respect to @xmath14 contradicts the motivations of massive mimo , we present the condition for _ non - decreasing _ sinr ( i.e. , @xmath136 ) in the following corollary . [ corollary 1 ] for both perfect and imperfect pce , the necessary and sufficient condition for the massive mimo network with mmse channel estimation and mrt to have non - decreasing sinr is @xmath137\ ] ] the result is a straightforward extension from theorem [ theorem 1 ] . the scaling law in ( [ eq30 ] ) and ( [ eq31 ] ) illustrates quantitatively the relationship between the performance scaling and scalings of all important parameters . the condition in ( [ eq32 ] ) , on the other hand , provides useful guidelines for the design of the massive mimo network . equation ( [ eq30 ] ) shows that with perfect pce , the scaling of the sinr @xmath122 is determined by the scaling exponent of the channel training energy @xmath119 and per - user transmission power @xmath138 . moreover , ( [ eq30 ] ) also shows that the scaling of the sinr is a decreasing function of both . thus higher per - user transmission power and training energy result in improved performance , and one can compensate for the other in performance . ( [ eq31 ] ) shows that with imperfect pce , @xmath122 is upper bounded by zero due to the pilot contamination bottleneck and it can be seen from that @xmath139 . this means that the asymptotic performance will be constant with respect to @xmath14 , while the concrete value is dependent on the efficiency of pce , which is more general than the result in @xcite . in this subsection , the scaling law in ( [ eq30 ] ) and ( [ eq31 ] ) and the condition for non - decreasing sinr in ( [ eq32 ] ) are elaborated for typical network settings . * [ d1 ] first , we consider the case of @xmath140 , i.e. , the training energy @xmath102 is constant . for perfect pce , @xmath141 . for imperfect pce , @xmath142 . to have non - decreasing sinr , @xmath143 is needed . this shows a tradeoff between the scaling of the user number @xmath19 and the transmission power @xmath71 . the most power - saving design is to make the per - user transmission power decrease linearly with @xmath14 , i.e. , @xmath144 . with a larger @xmath14 , the network can serve more users or have less power consumption , while maintaining certain performance . however , improvements in the two aspects both have a limit : 1 ) @xmath145 , @xmath146 and 2 ) @xmath147 , @xmath148 . case 1 ) means that when @xmath19 increases linearly with @xmath14 , to achieve non - decreasing sinr , @xmath71 must remain constant , and thus the goal of reducing @xmath71 can not be achieved . case 2 ) means that when @xmath71 is scaled inversely proportional to @xmath14 , the goal of serving more users can not be achieved . the latter case is the major power scaling scenario considered in the literature @xcite . obviously , our results cover this case , and shows more insights for general scalings of @xmath19 and @xmath71 . * [ d2 ] then we consider the case of @xmath149 , i.e. , @xmath150 . for perfect pce , @xmath151 . for imperfect pce , @xmath152 . to have non - decreasing sinr , @xmath153 is needed , i.e. , the transmission power and the user number should both remain constant . this shows that the training energy is key to the performance , i.e. , with low training energy , the promising features of massive mimo can not be achieved . * [ d3 ] for the general case where @xmath154 , non - decreasing sinr requires @xmath155 . that is , the scale of the per - user transmission power should be no less than @xmath156 . this shows the trade - off between the training phase and the data transmission phase . * [ d5 ] in previous discussions , the scaling of the training energy @xmath119 is treated as a free parameter , next , the special but commonly used training case where @xmath157 and @xmath158 is considered , i.e. , each user uses the same power in training as that for data transmission . since in this case , @xmath159 , we have @xmath160 for perfect pce and @xmath161 for imperfect pce . non - decreasing sinr requires @xmath162 . if further @xmath19 is constant , @xmath163 is needed , i.e. , the transmission power should scale as @xmath164 or higher . this is the same as the results in @xcite , @xcite . * [ d6 ] another typical setting from the perspective of supporting a large number of users is to have @xmath19 increase linearly with @xmath14 , i.e. , @xmath165 . for this case , @xmath166 for perfect pce and @xmath167 for imperfect pce . non - decreasing sinr requires @xmath168 . thus , to support such a number of users , the training energy must be constant or higher and the transmission power can not decrease with @xmath14 simultaneously . the performance scaling law is derived for @xmath169 . to be practically useful , its effectiveness for large but finite antenna numbers needs to be understood . in our scaling law derivation , the term with the highest scaling exponent in the denominator of is kept while other lower order terms are neglected . thus for the interested range of antenna number , the result is said to be applicable if the term with the highest order is dominant , in other words , significantly larger ( denoted by the symbol `` @xmath170 '' ) than the rest is said to be dominated by another positive number @xmath171 , i.e. , @xmath172 , when @xmath171 is at least one - order high than @xmath173 , or in other words , @xmath174 . this common practice is used in our work . ] . in the following , we derive conditions for the scaling law to be applicable in practical systems with large but finite antenna numbers . [ corollary 2 ] for networks with perfect pce , the scaling law in ( [ eq30 ] ) is applicable for large but finite @xmath14 when @xmath175 . in this case , @xmath129 and the term with the highest order with respect to @xmath14 is @xmath176 . for it to be dominant in the denominate of ( [ eq25 ] ) , it can be seen straightforwardly that @xmath175 is sufficient . [ corollary 3 ] for networks with imperfect pce , the scaling law in ( [ eq31 ] ) is applicable for large but finite @xmath14 when the following holds : @xmath177 where @xmath178 . the derivation is almost the same as the one for corollary [ corollary 2 ] . thus it is omitted . [ remark 5 ] corollary [ corollary 2 ] shows that for large but finite @xmath14 , the scaling law in ( [ eq30 ] ) for perfect pce is more accurate for less correlated channel ( larger @xmath114 ) and/or lower transmission power . corollary [ corollary 3 ] shows that for large but finite @xmath14 , the scaling law in ( [ eq31 ] ) for imperfect pce is also more accurate for less correlated channel , while the effect of the transmission power on the scaling law accuracy is unclear . the sufficient conditions . thus , another case for the scaling law in ( [ eq30 ] ) to be applicable is @xmath146 and @xmath179 . ] in corollaries [ corollary 2 ] and [ corollary 3 ] for the scaling laws to be applicable provide guidance for using the scaling law and in practical system design with large but finite antenna number . for systems that satisfy the conditions in corollaries [ corollary 2 ] and [ corollary 3 ] , the derived scaling law can accurately reflect the network performance ; while when the conditions are not guaranteed , the results becomes less accurate but can still provide rough suggestions on the performance . in the scaling law analysis in section [ scalinglawres ] , @xmath128 , the average number of cells with pilot contamination is assumed to be a constant . as also mentioned in that section , a decreasing @xmath103 with respect to @xmath14 is possible by pce . to take into consideration the possible variation of @xmath128 with respect to @xmath14 , the following approximate scaling model for @xmath128 can be used : @xmath180 where @xmath181 . the value of the exponent @xmath182 is dependent on the concrete algorithm . for example , @xmath183 corresponds to the imperfect pce case discussed previously , where no pce is available or the pilot contamination is only reduced to a constant value . since the highest achievable scaling of pilot contamination can not surpass the increasing rate of antenna number , we have @xmath184 . with this more general modeling , the corresponding performance scaling law by following similar derivation procedure can be obtained as @xmath185 [ remark 6_170211 ] the expressions in and are approximations since @xmath103 is modeled as an integer in this paper . its variation with respect to @xmath14 can be roughly fitted with the function @xmath186 for a constant @xmath171 . when @xmath103 decreases with @xmath14 ( meaning high pce efficiency and a positive @xmath182 ) , an increasing effective sinr rather than the bounded one in can be obtained if @xmath187 , which will be evaluated by simulations in section [ simulation ] . one important concept in massive mimo analysis is asymptotically deterministic . for example , when the desired signal power and interference power which are random in finite - dimension converge to deterministic values as @xmath14 increases to infinity @xcite , their expected values can be used to replace the random counterparts to simplify the analysis . then a natural question is when the massive mimo system has asymptotically deterministic sinr for the corresponding performance analysis to be reliable . in existing literature , the deterministic equivalence is mainly based on the almost sure convergence@xcite . while many results have been reported based on the aforementioned analysis , the corresponding convergence condition tends to be vague and explicit analysis of the effect of parameter scalings on the convergence is missing . in this section , following the work in @xcite , we adopt the mean square convergence @xcite for the definition of asymptotically deterministic property , and use the scv scaling with respect to @xmath14 to derive sufficient conditions for asymptotically deterministic sinr . finally , typical network scenarios are discussed . [ definition 2 ] the random variable sequence @xmath88 is said to be asymptotically deterministic if its scv decreases linearly with @xmath14 or faster . strictly speaking , there is no constraint on the scaling of the scv for a random sequence to converge to a constant value in the mean square sense other than that the scv converges to 0 as @xmath188 . notice that for any positive number @xmath115 , @xmath189 when @xmath188 . however , the convergence can be very slow for small @xmath115 , in which case , the derived asymptotic results may not be applicable for practical range of large but finite @xmath14 . thus , for practical system applications , the sinr is only considered to be asymptotically deterministic when the scv decreases linearly with @xmath14 or faster . with this definition , we actually derive sufficient conditions for the sinr to be asymptotically deterministic . [ pro1]when @xmath190 , for perfect pce , a sufficient condition for the sinr in to be asymptotically deterministic is @xmath191 for imperfect pce , the sinr is always asymptotically deterministic . please see appendix [ appendix b ] . note that the above sufficient conditions are based on the scaling law in and . they can be utilized to justify whether the cases discussed in d1-d5 ( including the power scaling analysis in existing works @xcite ) satisfy the condition for asymptotically deterministic sinr . for perfect pce , implies @xmath192 , meaning that to make the sinr asymptotically deterministic , the highest possible sinr scaling is @xmath193 . without the constraint for sinr to be asymptotically deterministic , the maximum possible value for @xmath122 is @xmath1 . for imperfect pce , proposition [ pro1 ] implies that no new limit is casted to the sinr scaling for the sinr to be asymptotically deterministic . next , typical scenarios with asymptotically deterministic sinr are investigated in which @xmath119 is allowed to take values from @xmath194 only . the tradeoff between parameters will be revealed . * [ d7 ] for perfect pce , to achieve both @xmath195 ( the sinr increasing linearly with @xmath193 ) and asymptotically deterministic sinr , the sufficient condition reduces to @xmath196 , @xmath197 . it means that the user number should be constant , the product of training energy and the transmission power must scale as @xmath164 where the tradeoff between the energy efficiency in training and transmission period can be found . two typical cases with constant user number may happen : a ) @xmath198 ; and b ) @xmath199 . for case a ) , when the training energy is constant , the transmission power should scales as @xmath164 . for case b ) , when the training energy scales as @xmath164 , the transmission power should be constant . * [ d8 ] for imperfect pce , to achieve both the highest scaling @xmath200 and asymptotically deterministic sinr , the sufficient condition is the same as the condition for non - decreasing sinr in , i.e. , @xmath201 . it means that the product of the training energy and the per - user transmission power should scale as @xmath202 or higher . typical cases are referred to the former discussions in section [ dis_sec_1 ] . * [ d9 ] for perfect pce , to achieve constant and asymptotically deterministic sinr , the sufficient condition reduces to @xmath203 . it means that the product of training energy and the per - user transmission power must scale as @xmath202 . three typical cases may happen : a ) @xmath204 ; b ) @xmath205 and c ) @xmath206 . for case a ) , when the training energy is constant , the per - user transmission power should scales as @xmath202 . for case b ) , when the training energy scales as @xmath164 , the per - user transmission power should also scales as @xmath164 . for case c ) , when the training energy scales as @xmath202 , the per - user transmission power should be constant . in existing works , only constant sinr case ( @xmath207 ) has been considered @xcite . discussion d6-d8 show that for perfect pce , the asymptotically deterministic sinr can scale as @xmath193 . for imperfect pce , at most constant asymptotically deterministic sinr can be achieved without any extra conditions besides the one for non - decreasing sinr in . in this section , main analytical results in this paper will be verified by simulations . we assume that the total number of cells is @xmath208 , the spatial correlation metric is @xmath209 and the inter - cell normalized large scale fading is @xmath210 . the matrix @xmath3 in and ( [ eq5 ] ) is assumed to consist of the first @xmath211 columns of the @xmath14 by @xmath14 discrete fourier transfer ( dft ) matrix . due to the main focus of this paper , we only consider systems with non - decreasing sinr scenario as given in corollary [ corollary 1 ] . first , we consider the case of perfect pce , i.e. , @xmath212 for all simulated antenna numbers . eleven typical parameter settings as specified in table i are considered . [ table 1 ] . parameter settings for perfect pce [ cols="^,^,^,^,^,^,^",options="header " , ] in fig . 4 , the simulated effective sinr with respect to @xmath14 is shown for the network settings given in table ii . for the constant pilot contamination @xmath183 case , the limiting effect of pilot contamination on the performance scaling is shown in cases 4 and 5 . note that these two cases do not match the scaling law as well as that of perfect pce since the pilot contamination term is not dominant enough for the antenna number interval in both cases . however , for the simulated finite antenna numbers , the scaling law for case 4 is more reliable compared with that for case 5 . this is because that the former case has a better compliance with the condition in corollary [ corollary 3 ] . further , as @xmath14 increases , the accuracy of the scaling law for both cases improves . on the other hand , cases 1 - 3 show that for decreasing pilot contamination with respective to @xmath14 , an increasing sinr can be obtained ( case 3 ) , which is in accordance with remark [ remark 6_170211 ] . in fig . 5 , it can been shown that the lower bound given in lemma [ lemma 2 ] also match the simulated achievable ergodic sum - rate well for all parameter settings in table ii . for tdd mutli - cell multi - user massive mimo downlink systems with csi error , channel spatial correlation and mrt precoding , this paper provided a comprehensive general scaling law based performance analysis , which quantitatively shows the effect of important network parameters and the pilot contamination elimination and the tradeoff among them . the application of the scaling law for large but finite antenna number was also studied . moveover , a sufficient condition for the sinr to be asymptotically deterministic in the sense of mean square convergence was derived . our results cover existing ones on such analysis as special cases and show the effect of pilot contamination more explicitly rather than only considering it as the performance limit in existing works . in the future work , more general spatial correlation distribution and more practical modeling on the effect of pilot contamination elimination should be studied . due to the relationship in , the key step for calculating the means and scvs of the sinr components is the calculation of @xmath213 , @xmath214 , and @xmath215 , @xmath216 . the following results are derived for all @xmath217 and @xmath218 . first , with some tedious calculations which are omitted here for space limit , we have @xmath219 @xmath220 @xmath221 and @xmath222 where - are derived from properties of gamma distribution and in , the central limit theorem is used to approximate the distribution of @xmath223 as @xmath224 for all @xmath225 based on . based on ( [ eq38])-([eq41 ] ) and the relationship in ( [ eq11 ] ) , the next step for calculating the means and scvs of the sinr components is to follow their definitions and find the highest order terms with respect to @xmath14 . details are omitted here due to space limit . the sinr expression in ( [ eq17 ] ) can be reformulated as @xmath226 the sinr is asymptotically deterministic when its scv approaches zero as @xmath227 . however , the complex structure of the sinr expression makes it challenging to obtain its scv directly . alternatively , since @xmath96 is asymptotically deterministic as noted in remark [ remark 1 ] , for the sinr to be asymptotically deterministic , the sufficient and necessary condition is that the denominator in ( [ eq42 ] ) is asymptotically deterministic . one sufficient condition is the scv of the denominator in ( [ eq42 ] ) scales no larger than @xmath202 . when @xmath228 is asymptotically deterministic , @xmath228 asymptotically converges to @xmath229 as defined in ( [ eq27 ] ) . thus from the definition in ( [ eq28 ] ) , i.e. , @xmath230 , and @xmath231 ( refer to lemma [ lemma1 ] ) , we have @xmath232 therefore , the sufficient condition can be rewritten as @xmath233 for some constant @xmath234 . by using the fact that a sufficient condition for ( [ eq45 ] ) is that the variance of each term in ( [ eq45 ] ) scales no larger than @xmath202 ( details are omitted for space limit ) and combining the scaling law in ( [ eq30 ] ) and ( [ eq31 ] ) , the sufficient condition in proposition [ pro1 ] can be obtained via some tedious calculations based on ( [ eq20])-([eq22 ] ) . 99 t. l. marzetta , noncooperative cellular wireless with unlimited numbers of base station antennas , " in _ ieee trans . wireless commun . _ , vol . 9 , no . 11 , pp . 3590 - 3600 , nov e. bjrnson , e. g. larsson and t. l. marzetta , massive mimo : ten myths and one critical question , " in _ ieee commun . 114 - 123 , feb . 2016 . f. rusek et al . , scaling up mimo : opportunities and challenges with very large arrays , " in _ ieee signal process . mag . _ , vol . 30 , no . 40 - 60 , jan . 2013 . s. wagner , r. couillet , m. debbah et al . , large system analysis of linear precoding in correlated miso broadcast channels under limited feedback , " in _ ieee trans . inf . theory _ , vol . 58 , no . 7 , pp . 4509 - 4537 , july 2012 . j. hoydis , s. ten brink and m. debbah , massive mimo in the ul / dl of cellular networks : how many antennas do we need ? , " in _ ieee j. sel . areas commun . 160 - 171 , feb . 2013 . h. yang and t. l. marzetta , performance of conjugate and zero - forcing beamforming in large - scale antenna systems , " in _ ieee j. sel . areas commun . 172 - 179 , feb . 2013 . h. q. ngo , e. g. larsson and t. l. marzetta , energy and spectral efficiency of very large multiuser mimo systems , " in _ ieee trans . 1436 - 1449 , apr . 2013 . q. zhang , s. jin , k. k. wong et al . , power scaling of uplink massive mimo systems with arbitrary - rank channel means , " in _ ieee j. sel . topics signal process . _ , vol . 8 , no . 966 - 981 , oct . 2014 . h. q. ngo , e. g. larsson and t. l. marzetta , massive mu - mimo downlink tdd systems with linear precoding and downlink pilots , " _ 2013 51st annual allerton conference on commun . , control , and comput . ( allerton ) , monticello , il , _ 2013 , pp . a. khansefid and h. minn , achievable downlink rates of mrc and zf precoders in massive mimo with uplink and downlink pilot contamination , " in _ ieee trans . 4849 - 4864 , dec . 2015 . j. zuo , j. zhang , c. yuen et al . multicell multiuser massive mimo transmission with downlink training and pilot contamination precoding , " in _ ieee trans . veh . 65 , no . 8 , pp . 6301 - 6314 , aug a. papazafeiropoulos , impact of general channel aging conditions on the downlink performance of massive mimo , " in _ ieee trans 1428 - 1442 , feb . e. bjrnson , m. matthaiou and m. debbah . massive mimo with non - ideal arbitrary arrays : hardware scaling laws and circuit - aware design , " in _ ieee trans . wireless commun . 4353 - 4368 , aug . h. falconet , l. sanguinetti , a. kammoun et al . asymptotic analysis of downlink miso systems over rician fading channels , " _ 2016 ieee international conference on acoust . , speech and signal process . ( icassp ) , shanghai , _ 2016 , pp . 3926 - 3930 . p. j. smith , c. neil , m. shafi et al . , on the convergence of massive mimo systems , " _ 2014 ieee international conference on commun . ( icc ) , sydney _ , nsw , 2014 , pp . 5191 - 5196 . q. wang and y. jing . performance analysis and scaling law of mrc / mrt relaying with csi error in massive mimo systems . " arxiv preprint arxiv:1606.07480 ( 2016 ) . h. yin , d. gesbert , m. filippou et al . , a coordinated approach to channel estimation in large - scale multiple - antenna systems , " in _ ieee j. sel . areas commun . 264 - 273 , feb . 2013 p. zhao , z. wang , c. qian et al . location - aware pilot assignment for massive mimo systems in heterogeneous networks , " in _ ieee trans . 65 , no . 8 , pp . 6815 - 6821 , aug b. tomasi and m. guillaud , pilot length optimization for spatially correlated multi - user mimo channel estimation , " _ 2015 49th asilomar conference on signals , syst . and comput . , pacific grove , ca , _ 2015 , pp . 1237 - 1241 . o. elijah , c. y. leow , t. a. rahman et al . a comprehensive survey of pilot contamination in massive mimo - 5 g system , " in _ commun . surveys tuts . 905 - 923 , second quarter 2016 . j. jose , a. ashikhmin , t. l. marzetta et al . , pilot contamination and precoding in multi - cell tdd systems , " in _ ieee trans . wireless commun . 10 , no . 8 , pp . 2640 - 2651 , aug . 2011 . fundamentals of statistical signal processing , volume i : estimation theory . " ( 1993 ) . y. wang , s. li , c. li et al . ergodic rate analysis for massive mimo relay systems with multi - pair users under imperfect csi , " in _ 2015 ieee global conf . on sig . and info . process . ( globalsip ) , orlando , fl , _ 2015 , pp . t. v. t. le and y. h. kim , power and spectral efficiency of multi - pair massive antenna relaying systems with zero - forcing relay beamforming , " in _ ieee commun . letters _ , vol . 2 , pp . 243 - 246 , feb . z. zhang , z. chen , m. shen et al . spectral and energy efficiency of multipair two - way full - duplex relay systems with massive mimo , " in _ ieee j. sel . areas in commun . 848 - 863 , apr . 2016 . b. everitt , the cambridge dictionary of statistics , cambridge university press , 1998 . s. boyd and l. vandenberghe , convex optimization , cambridge univ . press , 2004 . h. kobayashi , b. l. mark , and w. turin , probability , random processes , and statistical analysis , cambridge univ . press , 2012 .

one of the most remarkable observational results in the last three decades in cosmology is the lack of the gunn peterson trough in the spectra of high redshift quasars and galaxies . this finding implies that the intergalactic medium ( igm ) is highly ionized at least out to @xmath5 , the redshift of the most distant known quasars and galaxies ly@xmath6 emission lines , implying @xmath7 ( see discussion below ) . ] . since the primeval plasma recombined at @xmath8 , its subsequent ionization requires some form of energy injection into the igm , naturally attributable to astrophysical sources . the two most popular examples of such sources are an early generation of stars , residing in sub galactic size clusters ( hereafter `` mini galaxies '' ) , or accreting massive ( @xmath9 ) black holes in small ( @xmath10 ) halos ( hereafter `` mini quasars '' ) . reionization is interesting for a variety of reasons . first , the most fundamental questions are still unanswered : what type of sources caused reionization , and around what redshift did it occur ? how were the reionizing sources distributed relative to the gas ? what was the size , geometry , and topology of the ionized zones , and how did these properties evolve ? second , and more directly relevant to the topic of this review , reionization leaves distinctive signatures on the cosmic microwave background ( cmb ) through the interaction between the cmb photons and free electrons . although it may complicate the extraction of cosmological parameters from cmb data , this `` contamination '' could yield important information on the ionization history of the igm . through reionization , the cmb is a useful probe of nonlinear processes in the high redshift universe . thomson scattering from free electrons affects the cmb in several ways . because the scattering leads to a blending of photons from initially different lines of sight , there is a damping of the _ primary _ temperature anisotropy . on the other hand , a new _ secondary _ anisotropy is generated by what can be thought of as a doppler effect : as photons scatter off free electrons , they pick up some of their peculiar momentum . finally , the polarization dependence of the thomson cross section creates new polarization from the initially anisotropic photon field . the inhomogeneity of reionization affects these processes in two different ways . first , the inhomogeneity of the medium affects how the _ mean _ ionization fraction evolves with time . this is important since the damping , doppler and polarization effects all occur even in the idealized case of spatially homogeneous reionization . secondly , the spatial fluctuations in the ionization fraction can greatly enhance the contribution from the doppler effect at small angular scales . polarization and damping are much less influenced by the inhomogeneity , as we will see below . the aim of the present review is to summarize our current knowledge of reionization , describe the possible connections to the cmb , and assess what we can hope to learn about the high redshift igm from forthcoming observations of the cmb anisotropies . if the igm was neutral , its optical depth to ly@xmath6 absorption would be exceedingly high : + @xmath11^{3/2}$ ] , wiping out the flux from any source at observed wavelengths shorter than @xmath12 . the spectra of high redshift quasars and galaxies reveal ly@xmath6 absorption by numerous discrete ly@xmath6 forest clouds , separated in redshift , rather than the continuous gunn peterson ( gp ) trough expected from a neutral igm . the detection of the continuum flux in between the ly@xmath6 clouds implies that @xmath13 is at most of order unity , or , equivalently , an upper limit on the average neutral fraction of @xmath14 . the intergalactic medium is therefore highly ionized . note that there is no sharp physical distinction between the intergalactic medium and the low column density ( @xmath15 ) ly@xmath6 forest ( reisenegger & miralda - escud 1995 ) . these ly@xmath6 clouds fill most of the volume of the universe , are highly ionized , and account for most of the baryons ( implied by big bang nucleosynthesis ) at high redshift ( weinberg et al . 1997 ; rauch et al . 1997 ) . the energy requirement of reionization can easily be satisfied either by the first mini galaxies or mini quasars . nuclear burning inside stars releases several mev per hydrogen atom , and thin disk accretion onto a schwarzschild black hole releases ten times more energy , while the ionization of a hydrogen atom requires only 13.6 ev . it is therefore sufficient to convert a fraction of @xmath16 of the baryonic mass into either stars or black holes in order to ionize the rest of the baryons in the universe . in a homogeneous universe , the ratio of recombination time to hubble time in the igm is @xmath17^{-3/2}$ ] so that each h atom would only recombine a few times at most until the present time . unless reionization occurred at @xmath18 , recombinations therefore would not change the above conclusion , and the number of ionizing photons required per atom would be of order unity . based on three - dimensional simulations , gnedin & ostriker ( 1997 ) argued that the clumpiness of the gas increases the average global recombination rate by a factor @xmath19 ( the averages are taken over the simulation box ) . this would imply a corresponding increase in the necessary number of ionizing photons per atom . however , in a more detailed picture of reionization in an inhomogeneous universe ( gnedin 1998 , miralda - escud et al . 1999 ) , the high - density regions are ionized at a significantly later time than the low density regions . if this picture is correct , then one ionizing photon per atom would suffice to reionize most of the volume ( filled by the low - density gas ) ; recombinations from the high - density regions would only contribute to the average recombination rate at a later redshift . the reionization redshift can be estimated from first principles . in the simplest picture [ first discussed by arons & wingert ( 1972 ) in the context of quasars ] , a homogeneous neutral igm is percolated by the individual hii regions growing around each isolated uv source . the universe is reionized when the ionized bubbles overlap , i.e. when the filling factor of hii regions reaches unity . several authors studied this problem ( carr et al . 1984 , couchman & rees 1986 , fukugita & kawasaki 1994 , meiksin & madau 1993 ; shapiro et al . 1994 , tegmark et al . 1994 ; aghanim et al . 1996 ; haiman & loeb 1997 ) . specifically , haiman & loeb ( 1998a ) have used the press schechter formalism to describe the formation of halos that potentially host ionizing sources , in order to estimate the redshift at which overlap occurs . in this study , star and quasar black hole formation was allowed only inside halos that can cool efficiently via atomic hydrogen , since molecular hydrogen would be photodissociated earlier , as argued by haiman , abel & rees ( 1999 ) and haiman , rees & loeb ( 1997 ) . this requirement translates to a minimum virial temperature of @xmath20k , or halo masses above @xmath21^{-3/2}$ ] ( see fig . [ fig : bubbles ] ) . for both types of sources , the total amount of light they produce was calibrated using data from redshifts @xmath22 . the efficiency of early star formation can be estimated from the observed metallicity of the intergalactic medium ( songaila & cowie 1996 , tytler et al . 1995 ) , utilizing the fact that these metals and the reionizing photons likely originate from the same stellar population . similarly , the efficiency of black hole formation inside early mini quasars can be constrained to match the subsequent evolution of the quasar luminosity function at redshifts @xmath22 ( pei 1995 ) . these assumptions generically lead to a reionization redshift @xmath23 . the uncertainty in the redshift reflects a range of cosmological parameters ( @xmath24 ) , cdm power spectra ( @xmath25 , @xmath26 ) , and efficiency parameters for the production and escape fraction of ionizing photons . in the stellar reionization scenario , the predicted redshift could be above or below this range , if the initial mass function was strongly biased ( relative to scalo 1986 ) towards massive or low - mass stars . although the naive picture of overlapping hii spheres may give a good estimate for the `` overlap redshift '' , it is likely to be a crude and incomplete description of the real process of reionization . the ionizing sources are expected to be located in the highly overdense regions that form the complex large scale structure of the universe - along filaments or sheets , or at their intersections ( see fig . [ fig:3d ] ) . the ionizing radiation will have to cross the local dense structures before propagating into the igm , which itself has significant density fluctuations . several important consequences of these inhomogeneities can be noted . first , reionization could occur `` outside - in '' ( miralda - escud et al . 1999 ) , if the radiation from a typical source escapes through a relatively narrow solid angle from its host , without causing significant local ionization . in this case , the low density regions could be ionized significantly earlier than the high density regions . an important effect in this case could be the shadowing of ionizing radiation by the dense concentrations within the igm . the same type of shadowing effect was found to be important at low redshift . based on the number of ly@xmath6 absorbers , madau , haardt & rees ( 1998 ) have concluded that as much as @xmath27% of the uv flux is absorbed by these systems . the redshift at which the igm becomes optically thin to the ionizing continuum is therefore significantly delayed relative to the overlap epoch , when the ionized zones first percolate . an alternative possibility is that the ionizing sources are densely spaced along filaments , and their radiation ionizes most of the dense filament before escaping into the igm . in this `` inside - out '' case , the description of ionized zones surrounding their host source would be more applicable . an important feedback during and after reionization is that the newly established uv background can photo evaporate the gas from small ( @xmath28 ) halos ( shapiro , raga & mellema 1997 ; barkana & loeb 1999 ) . a related point , important for the detection of the gp trough , is that at the time of overlap of the hii zones , each individual hii region could still have a non negligible ly@xmath6 optical depth , due to its residual hi fraction . the gp trough therefore disappears only from the spectra of individual sources located below the later redshift when the average ly@xmath6 optical depth of the igm drops below unity ( haiman & loeb 1998b ; miralda - escud et al . 1999 ; shapiro et al . finally , the inhomogeneity of the reionized igm leads to generation of anisotropies in the cmb which are strongly suppressed in the homogeneous case , as we shall see below . reionization is now also being addressed by three - dimensional simulations , which can eventually shed light on some of the above issues . the first such studies approximated radiative transfer by assuming an isotropic radiation field , but were able to take into account the inhomogeneities in the gas distribution ( ostriker & gnedin 1996 ; gnedin & ostriker 1997 ) . the resulting reionization redshift of @xmath29 in a @xmath30cdm model is in good agreement with the semi analytical estimates , given the differences in the assumed cosmology , power spectra , and star formation efficiency . however , an isotropic radiation field is necessarily a crude approximation , at least in the beginning stages of reionization , when each source still has its own isolated ionized zone . in order to understand the progressive overlap of ionized zones , including effects such as self - shielding and shadowing by dense clumps , it is necessary to incorporate radiative transfer into three dimensional simulations ( norman et al . 1998 ) . the numerical algorithms for the first such attempts are described in abel , norman & madau ( 1998 ) and razoumov & scott ( 1998 ) . results from an approximate treatment ( gnedin 1998 ) also support the outside in picture of reionization described above . several authors have emphasized that the known population of quasars or galaxies provide @xmath3110 times fewer ionizing photons than necessary to reionize the universe ( see , e.g. shapiro et al . 1994 , or madau et al . 1999 , and references therein ) . the two most natural candidates for undetected ionizing sources are the mini galaxies and mini quasars expected to be associated with small ( @xmath32 ) dark halos at @xmath33 . as there are no compelling _ ab initio _ theoretical arguments to strongly favor one type of source over the other , the best hope is to distinguish these two possibilities from observations . one expects at least three major differences between mini galaxies and mini quasars : their spectra , their absolute brightness ( or , alternatively , space density ) , and the angular size of their luminous regions . the flux from stellar populations drops rapidly with frequency above the ionization threshold of hydrogen , while the spectra of mini quasars are expected to be harder and extend into the x ray regime . stars could therefore not reionize heii , while quasars with typical spectra ( elvis et al . 1994 ) would reionize heii at approximately the same redshift as h ( haiman & loeb 1998a ) . however , recent high - resolution spectra of @xmath34 quasars have shown a widely fluctuating heii / hi optical depth ( reimers et al . the authors interpreted this observation as evidence for heiii regions embedded in an otherwise heii medium , i.e. a detection of the heii reionization epoch ( see also wadsley et al . if this interpretation is verified by future data , it would constrain the number of mini quasars with hard spectra extending to x rays at high redshifts ( haiman & loeb 1998b ; miralda - escud 1998 ; miralda - escud & rees 1994 ) . at present , a plausible alternative interpretation of the heii observations is that the observed heii optical depth fluctuations are caused by statistical fluctuations in the igm density or the ionizing background flux ( miralda - escud et al . 1999 ) , rather than by the patchy structure of heii / heiii zones . the x ray background ( xrb ; miyaji et al . 1998 ; fabian & barcons 1992 ) might provide another useful constraint on mini quasar models with hard spectra ( haiman & loeb 1998c ) . these models overpredict the _ unresolved flux _ by a factor of @xmath35 in the 0.11 kev range . if an even larger fraction of the xrb will be resolved into low redshift agns in the future , then the xrb could be used to place more stringent constraints on the x - ray spectrum or the abundance of the mini quasars . a second distinction between galaxies and quasars can be made based on their absolute luminosity and space density . on average , based on their @xmath36 counterparts , high redshift mini quasars are expected to be roughly @xmath37 times brighter , but @xmath38 times more rare , than galaxies . at the overlap epoch , hii regions from mini quasars would therefore be larger but fewer than those from mini galaxies . the typical size and abundance of hii regions has at least three important implications . first , it affects the cmb anisotropies ; the larger the hii regions , the larger the effect on the cmb ( gruzinov and hu 1998 , knox et al . 1998 ) , as we will see below . second , large enough hii regions ( @xmath39mpc ) would allow gaps to open up in the gp trough ; the flux transmitted through such gaps around very bright quasars could be detectable ( miralda - escud et al . third , if sources form inside the rare , high-@xmath40 peaks , they would be strongly clustered ( knox et al . this clustering would increase the effective luminosity of each source , and would enhance both of the effects above . finally , a constraint on the number of mini quasars at @xmath41 can be derived from the hubble deep field ( hdf ) . quasars are expected to appear as faint point sources in the hdf , unless their underlying extended host galaxies are resolved . the properties of faint _ extended _ sources found in the hubble deep field ( hdf ) agree with detailed semi analytic models of galaxy formation ( baugh et al . 1998 ) . on the other hand , the hdf has revealed only a handful of faint _ unresolved _ sources , and none with the colors expected for high redshift quasars ( conti et al . 1999 ) . the simplest mini quasar models predict the existence of @xmath42 b band `` dropouts '' in the hdf , inconsistent with the lack of detection of such dropouts up to the @xmath43 completeness limit at @xmath44 . to reconcile the models with the data , a mechanism is needed that suppresses the formation of quasars in halos with circular velocities @xmath45 ( haiman , madau & loeb 1999 ) . this suppression naturally arises due to the photo - ionization heating of the intergalactic gas by the uv background after reionization . forthcoming data on point sources from nicmos observations of the hdf ( see thompson et al . 1998 ) could further improve these constraints . as described above , theoretical expectations place the reionization redshift at @xmath46 . what about observations ? at present , the best _ lower _ limit in the reionization redshift comes from the detection of high redshift ly@xmath6 emitters ( hu et al . 1998 ; weymann et al . 1998 ) . as argued by miralda - escud ( 1998 ) , the damping wing of ly@xmath6 absorption by a neutral igm has a large residual optical depth that would severely damp any ly@xmath6 emission line . currently the highest redshift at which a ly@xmath6 emitter is seen is @xmath47 ; the existence of this object implies that reionization occurred prior to this redshift ( see haiman & spaans 1998 ) . note that hii regions around individual sources , excepting only the brightest quasars , would be too small to allow the escape of ly@xmath6 photons , because the damping wings of the gp troughs around the hii region would still overlap . the best _ upper _ limit on the reionization redshift can be obtained from the cmb anisotropy data . several experiments have revealed a rise in the power on small angular scales , and a drop at even smaller scales ( e.g. , bond 1999 ) , showing evidence for the first doppler peak expected from acoustic oscillations in the baryon - photon fluid prior to recombination ( e.g. , hu & sugiyama 1994 ) . these observations can also be used to set limits on the electron scattering optical depth ( or the corresponding reionization redshift ) that would suppress the doppler peak . from a compilation of all the existing measurements , griffiths et al . ( 1998 ) have derived the stringent ( although model - dependent ) constraint @xmath48 on the reionization redshift if reionization occurred earlier , the electron scattering optical depth would have reduced the amplitude of the anisotropies below the observed level . note that another upper limit results from the spectral distortion of the cmb caused by scattering on the reionized igm , which is constrained by the upper limit on the compton @xmath49parameter measured by cobe . the inferred upper limit on the reionization redshift , however , is much weaker , @xmath50 for typical parameters in a low density universe ( griffiths et al . 1998 , see also stebbins & silk 1986 ) . in summary , present observations have narrowed down the possible redshift of the reionization epoch to @xmath51 , a relatively narrow range that is in good agreement with the theoretical predictions described above . further observational progress in probing the reionization epoch could come either from the _ next generation space telescope _ ( _ ngst _ ) , or from more precise measurements of the cmb anisotropies by _ map _ and _ planck_. _ ngst _ , scheduled for launch in 2007 , is expected to reach the @xmath52njy sensitivity required to detect individual sources to @xmath33 , and to perform medium resolution spectroscopy to @xmath53 . if reionization occurred close to the low end of the allowed redshift range , @xmath54 , then it may be possible to infer the redshift directly from the spectra of bright sources . one specific method relies on the spectrum of a bright source just beyond the reionization redshift , so that the individual ly@xmath6 , ly@xmath55 , and other gp troughs do not overlap in frequency , leaving gaps of transmitted flux ( haiman & loeb 1998b ) . the measurement of this transmitted flux would be possible with _ ngst _ , despite absorption by the high redshift ly@xmath6 forest , as long as the number density of absorbers does not rise much more steeply with redshift than an extrapolation of the current @xmath36 data would imply ( fardal et al . if reionization is gradual , rather than abrupt , than this method would measure the redshift at which the gp optical depth drops to near unity , i.e. the final stages of the reionization epoch . an alternative signature to look for would be the precise shape of the damping wing of ly@xmath6 absorption from the neutral igm along the line of sight to the source ( miralda - escud 1998 ) . the shape , if measured by high resolution spectroscopy , could be used to determine the total optical depth of the igm . the caveat of this method is the inability to distinguish the neutral igm from a damped ly@xmath6 absorber along the line of sight , close to the source . an alternative signature is the background ly@xmath6 emission from the reionization epoch . recombinations are slow both at high redshifts , when the igm is still neutral , and at low redshifts when the igm density is low . as shown by simulations ( gnedin & ostriker 1997 ) , the global recombination rate has a pronounced peak around the reionization epoch . the resulting recombinant ly@xmath6 emission has been computed by baltz , gnedin & silk ( 1998 ) , and a detailed detectability study has shown that this signal could be measured by _ ngst _ , or perhaps even by _ hst _ ( shaver et al . 1999 ) . yet another signature could result from the 21 cm hyperfine transitions in the igm before reionization . prior to reionization , the excitation of the 21 cm line in neutral hi atoms depends on the spin temperature . the coupling of the spin temperature to that of the cmb is determined by the local gas density , temperature , and the radiation background from the first mini galaxies and mini quasars . in general , the 21 cm line could be seen either in absorption or emission against the cmb , and could serve as a tomographic tool to diagnose the density and temperature of the high redshift neutral gas ( see , e.g. madau et al . 1997 ; scott & rees 1990 ) . if reionization occurred at @xmath56 , the redshifted 21 cm signals would be detectable by the giant metrewave radio telescope ; a study of the effect at higher redshifts would be possible with next generation instruments such as the thousand element array , or the square kilometer array ( shaver et al . 1999 ) . the reionization redshift could turn out to be closer to the high end of the allowed range , @xmath57 , rendering direct detection of emission from the reionization epoch implausible ( except perhaps the 21 cm signal ) . there is , however , a fortunate `` complementarity '' , since the electron scattering optical depth increases with reionization redshift . due to the effect on the polarization power spectrum on large angular scales , _ map _ and _ planck _ may be able to discern an electron scattering optical depth as small as a few percent ( zaldarriaga et al . 1997 , eisenstein et al . 1998 , prunet et al . 1998 ) . by temperature anisotropy data alone , they may be able to determine an optical depth as small as @xmath58 due to the damping effect mentioned earlier and to be discussed in more detail later . the reionizing sources may also change the spectral shape of the cmb . the dust that is inevitably produced by the first type ii supernovae , absorbs the uv emission from early stars and mini quasars and re - emits this energy at longer wavelengths . loeb & haiman ( 1997 ) have quantified the resulting spectral distortion in press schechter type models , assuming that each type ii supernova in a scalo imf yields @xmath59 of dust with the wavelength - dependent opacity of galactic dust , uniformly distributed throughout the intergalactic medium . under these assumptions , the dust remains cold ( close to the cmb temperature ) , and its emission peaks near the cmb peak . the resulting spectral distortion can be expressed as a compton @xmath49parameter @xmath60 , near the upper limit derived from measurement of the cmb spectrum by cobe ( fixsen et al . 1996 ) . a substantial fraction ( @xmath61@xmath62% ) of this total @xmath49parameter results simply from the direct far - infrared emission by early mini quasars and could be present even in the absence of any intergalactic dust . inhomogeneities in the dust distribution could change these conclusions . instead of being homogeneously mixed into the igm , the high redshift dust may remain concentrated inside or around the galaxies where it is produced . in this case , the average dust particle would see a higher flux than assumed in loeb & haiman ( 1997 ) , so that the dust temperature would be higher , and dust emission would peak at a shorter wavelength . the magnitude of the spectral distortion would be enhanced , and may account for the recently discovered cosmic infrared background ( puget et al . 1996 , schlegel et al . 1998 , fixsen et al . 1998 , hauser et al . 1998 ) as shown by the semi analytic models of baugh et al . an angular fluctuation in the magnitude of the distortion would also be expected in this case , reflecting the discrete nature of the sources contributing to the effect . as seen above , reionization has several consequences ; in the rest of this review we focus on the connection to the cmb , and the effects of inhomogeneities . in this section , we give an in - depth treatment of the effects on the cmb photons of thomson - scattering off of the free electrons produced by reionization . after defining the optical depth and giving its dependence on the redshift of reionization and cosmological parameters , we discuss the three separate effects : damping , doppler and polarization generation . then we derive the evolution equations from the boltzmann equation , argue that the doppler effect is the most important _ inhomogeneous _ effect , and calculate the resulting power spectrum for several simple models of inhomogeneous reionization ( ihr ) . the probability of a photon scattering in the time interval from some initial time @xmath63 to the present , @xmath64 is given by @xmath65 where ( t_0 ) _ t n_e(t ) dt is the optical depth to thomson scattering and @xmath66 is the thomson cross - section . if one assumes a step - function transition from a neutral to an ionized igm at a redshift of @xmath67 , the mean optical depth is given by ( e.g. , griffiths et al . , 1998 ) [ eqn : depthofzion ] ( z_ion ) = ^*/_0where ^ * = h_0 _ b _ t 4 g m_p ( 1-y/2 ) 0.033 _ b h , @xmath68 and @xmath69 are the density of baryonic matter and of all matter respectively , in units of the critical density today , @xmath70 is the hubble constant , + @xmath71 , @xmath72 is newton s constant , @xmath73 is the proton mass and @xmath74 is the mass fraction of baryons in helium . the final equality assumes @xmath75 . to get equation [ eqn : depthofzion ] , we assumed that @xmath76 , and adopted the line element , @xmath77 , of a spatially flat universe with only non - relativistic matter and a cosmological constant . note that for @xmath78 ( the observational lower bound ) , @xmath79 and @xmath80 , we get @xmath81 for @xmath82 and @xmath83 for @xmath84 . for @xmath67 only slightly larger than this limiting value , it is a good approximation to rewrite equation [ eqn : depthofzion ] as ( z_ion ) = 0.037/ ( 1+z_ion 11)^3/2 ( _ b h^2 0.02 ) ( 0.65 h ) . thus we see that the fraction of cmb photons that have been scattered may be very small , but that this fraction grows fairly rapidly with increasing redshift . the increase in optical depth for fixed @xmath67 with increasing @xmath85 is due to the fact that the proper time from the present to @xmath67 increases as the cosmological constant increases . here we give simple arguments about the nature of the damping , doppler and polarization effects . before doing so , it is useful to define the two - point correlation function , @xmath86 which is perhaps the most important statistical property of the cmb fluctuations : c ( ) ( _ 1,*x*,_0 ) ( _ 2,*x*,_0 ) ; _ 1 _ 2 where @xmath87 denotes ensemble average . the same information is contained in its legendre transform , the angular power spectrum , @xmath88 : c()= _ c_p _ ( ) . roughly speaking , @xmath89 is the power in modes with wavelength @xmath90 so that @xmath91 corresponds to about a degree . if the fluctuations are gaussian - distributed and statistically isotropic , all other statistics can be derived from @xmath89 . for an initially homogeneous and isotropic photon field , there is just as much probability to scatter into a line of sight as there is to scatter out of it . therefore , the optical depth to scattering produces no net effect . if , however , there is initial anisotropy , then this anisotropy is damped . consider a line of sight along which the temperature differs by @xmath92 from the mean @xmath93 in the absence of damping . if damping is present , then the temperature is changed to [ eqn : simpledamping ] |t + t & & ( |t + t ) - ( |t+t)(1-e^- ) + |t(1-e^- ) + & & |t + t e^- this equation expresses the fact that the final temperature is given by the initial temperature , reduced by the photons that have been lost , and increased by the photons that have been scattered in from other lines of sight . since the photons that have been scattered in come from many different lines of sight , their average temperature is very nearly the global average an assumption made in equation [ eqn : simpledamping ] . the net result is that @xmath94 and therefore @xmath95 or @xmath96 . this simple calculation would imply that the damping is independent of scale , which is not exactly true . our assumption that the average @xmath97 along lines of sight that scattered in equals the global average holds only when the distance _ between _ the scattering surfaces is much larger than the length scale of interest . at very large angular scales , the damping does not occur , as one would expect from simple causality considerations . more precisely , the damping factor is @xmath98-dependent and , defined via @xmath99 , is given by the fitting formula of hu & white 1997 : r_l^2 = 1-(-2 ) a+c_1 x + c_2 x^2 + c_3 x^3 + c_4 x^4 + ( -2 ) , with @xmath100 and @xmath101 and @xmath102 . the characteristic angular scale @xmath103 is roughly the angular scale subtended by the horizon at the new last - scattering surface and is given approximately by ( griffiths et al . 1998 , hu & white 1997 ) : l_r = ( 1+z_ion)^1/2(1 + 0.84_0)-1 . if not for the @xmath98-dependence , the damping effect would be completely degenerate with the amplitude of the primary power spectrum ; i.e. , their effects would be indistinguishable . the similar response to amplitude and @xmath104 over a large range of @xmath98 makes them approximately degenerate and is the reason why the optical depth can only be determined to about 10% ( zaldarriaga et al . 1997 ) or possibly worse ( eisenstein et al . 1998 ) based on temperature anisotropy alone . the thomson scattering _ differential _ cross - section is polarization - dependent , and therefore the scattered radiation may be polarized even if the incident radiation is not . however , by symmetry considerations alone one can see that initially isotropic radiation will not become polarized . in fact , it is easy to show that the particular anisotropy required for the creation of polarization is a quadrupole moment . this is essentially due to the @xmath105 dependence of the differential cross - section . the fourier analogue of the angle - distance relation can be written as @xmath106 where @xmath107 is the comoving wave - number that projects from conformal time @xmath108 into multipole moment @xmath109 at conformal time @xmath110 . conformal time is related to proper time by @xmath111 where @xmath112 is the scale factor of the expansion . applying this relationship twice , and using the fact that polarization is generated by the quadrupole moment , @xmath113 , p ~(_0 - _ ion)(2 _ ion- _ lss ) 2 ( -1 ) . for the last equality , we assumed a matter - dominated universe and that @xmath114 . thus we expect a peak in the polarization power spectrum near @xmath115 . determining the location of this peak allows one to determine the epoch of reionization . the amplitude of the peak is proportional to the optical depth , and thus also helps in determining @xmath67 . unfortunately , the signal is very weak and present only at large angular scales , e.g. , @xmath116 for @xmath117 . since there are only @xmath118 independent modes from which to determine @xmath89 , sample variance is worst at low @xmath98 . also , the large - scale features of polarization maps reconstructed from time - ordered data are likely to be those most sensitive to systematic errors . finally , polarized galactic foregrounds ( dust and synchrotron ) are expected to be most troublesome at small @xmath98 ( bouchet et al . 1998 , knox 1998 ) . however , a benefit of the low-@xmath109 location of this signature is that if it can be measured , its interpretation will not be complicated by the inhomogeneity of the reionization , which is a much smaller - scale phenomenon . if the systematic errors are negligible , the reionization feature in the polarization power spectrum can be used to discern very small optical depths . forecasts of parameter determination by eisenstein et al . ( 1998 ) using both the temperature anisotropy and polarization data are for one sigma errors on @xmath104 of 0.022 for _ map _ and @xmath119 for _ planck_. also see zaldarriaga et al . ( 1997 ) for similar results . note that these sensitivities , especially in the case of _ planck _ , are high enough to detect the effects of reionization _ at any epoch _ given that we know it happened prior to @xmath47 . the contribution to @xmath120 at location @xmath121 , in direction @xmath122 , at conformal time @xmath123 ( today ) is given by [ eqn : doppler ] t t(*x*_0,,_0 ) = _ t__ion^_0 dn_e(*x*)_e(*x * ) a where @xmath124 . as one might expect , it is proportional to the line - of - sight integral of the parallel component of electron velocity , @xmath125 times the number density of scatterers , @xmath126 . as sunyaev ( 1978 ) and kaiser ( 1984 ) pointed out , in the homogeneous case this contribution to @xmath127 is suppressed by cancellations due to oscillations in @xmath125 . one way to think of these cancellations is that photons get nearly opposite doppler shifts on different sides of a density peak , which is a consequence of potential flows generated by gravitational instability . the cancellations are less complete at large angular scales ( @xmath128 ) and there is a measurable effect which is exactly taken into account in boltzmann codes such as cmbfast . if speed is critical , the fitting formulae of griffiths et al . ( 1998 ) can be employed for an approximate treatment . the cancellations at small scales can be greatly reduced by the modulation of the number density of free electrons which we can write as @xmath129 where @xmath130 is the ionization fraction and @xmath131 is the number density of all protons , including those in h and he . this modulation occurs even in the case of homogeneous reionization ( spatially constant @xmath130 ) due to spatial variations in @xmath131 . because @xmath132 is zero to zeroth order , and @xmath131 is uniform to zeroth order , this is a second - order effect . that this second order effect could dominate the first - order one ( which is small due to cancellations ) was pointed out by ostriker and vishniac and is known as the ostriker - vishniac effect ( ostriker & vishniac 1986 , vishniac 1987 ) . subsequent systematic study of all 2nd order contributions have shown that it is the dominant second - order contribution as well ( hu et al . 1994 , dodelson & jubas 1995 ) . another source of modulation of the number density of free electrons is spatial variations in @xmath130 , i.e. , ihr . as a photon streams towards us from the last - scattering surface , it may pick up a doppler `` up kick '' on one side of an overdensity , but avoid the canceling `` down kick '' on the other because the igm there is still neutral . how effectively the cancellation is avoided depends on the matching between the typical sizes of the ionized domains and the correlation length of the velocity field , as we will discuss below . we now derive equation [ eqn : doppler ] , as well as a more general expression which includes the contribution from damping as well . more complete treatments of the evolution of a photon distribution function in an expanding , inhomogeneous universe can be found in , e.g. , ma & bertschinger ( 1995 ) . the boltzmann equation governing the evolution of the photon phase - space distribution function , @xmath133 , is simply @xmath134 , where @xmath135 takes into account the effect of collisions in this case with electrons via thomson scattering . the thomson scattering cross - section is independent of frequency and therefore does not induce spectral distortions ; the effect on the distribution function can be fully described by the change in brightness . thus we use the brightness perturbation variable defined by @xmath136 where @xmath137 is the comoving photon momentum , and the phase - space distribution function has been partitioned into a homogeneous and inhomogeneous part , @xmath138 . for a planckian @xmath133 with thermal fluctuations , @xmath92 , about a mean temperature of @xmath97 , @xmath139 . expanding the total derivative on @xmath133 , keeping terms to first order in the perturbation variables and fourier transforming the result , the boltzmann equation becomes : [ eqn : boltz ] ( , , ) + ik(,,)+ 23h ( , ) & & + + 23 ( 3h_33(,)-h(,))p_2 ( ) & = & + d^3x e^i*k * ( * x*,)s(*x * , , ) & & where @xmath140 is the fourier transform of the brightness perturbation , @xmath141 is the direction of the photon momentum , @xmath142 , the dot indicates differentiation with respect to conformal time and @xmath143 and @xmath144 are the 3,3 component and trace of the synchronous gauge metric perturbation , respectively , in the coordinate system defined by @xmath145 . on the right - hand side of equation ( [ eqn : boltz ] ) we have explicitly separated out the dependence of the collision term on the differential optical depth , @xmath146 . the quantity @xmath147 is a function of the brightness perturbation and the electron velocity . to first order in the perturbation variables its fourier transform is ( neglecting the polarization dependence of the thomson cross - section ) [ eqn : s ] s(*k * , , ) = -+ _ 0 + 4 _ e - 12_2 p_2 ( ) where @xmath148 . if we were to allow both @xmath149 and @xmath147 to depend on @xmath150 , the fourier modes would no longer evolve independently . to avoid this complication , we use an expansion in small @xmath149 . let @xmath151 be the solution for @xmath152 and let @xmath153 . then to first order in @xmath149 : ^(1)(*k * ) + ik^(1)(*k * ) = d^3x e^i*k * ( * x*)s^(0)(*x * , ) . which has the solution in real space : [ eqn : delta ] ^(1)(,*x*,_0)= _ _ i^_0 d^(*x^ * ) s^(0)(*x*^ , ) where @xmath154 . this solution can be derived with a green function approach or simply verified by substitution . it can be taken to higher order , if desired , by getting @xmath155 from @xmath156 , etc . note that for the doppler effect equation [ eqn : delta ] is exact because @xmath132 ( unlike , e.g. , @xmath157 ) receives no corrections due to the optical depth . we can now argue that the doppler effect , due to the @xmath125 term in equation [ eqn : delta ] , is the most important inhomogeneous effect . the dominance of this effect is due to the fact that by the time of reionization , velocities have grown substantially as they react to the gravitational field . on the relevant length scales , the rms peculiar velocities are about @xmath158 . in contrast , the damping of the anisotropy is due to the @xmath159 term in @xmath147 , which has not grown from its primordial value of @xmath160 . similarly , polarization is sourced by the quadrupole moment which is also still at the @xmath160 level at the epoch of reionization . therefore we expect these contributions to be down by two orders of magnitude in amplitude , or four in power . if we break up the correlation function into components according to the expansion in optical depth : c ( ) = c^(0)()+2c^(01)()+c^(1 ) ( ) where @xmath161 , @xmath162 and @xmath163 are the correlations between the two zeroth order terms , between the two first order corrections and between the first order and zeroth order terms , respectively . for the doppler effect , the ( exact ) correction to the homogeneous case comes entirely from the @xmath162 term which is equal to [ eqn : cdoppler ] c^(1 ) ( ) = _ _ i^_0d_1 _ _ i^_0 d_2 ( _ 1 ) _ 1 ( _ 2 ) _ 2 where the numeric subscript on the quantities in parentheses means evaluate at , e.g. , @xmath164 . from equation [ eqn : cdoppler ] we see that the relevant quantity for the cmb 2-point autocorrelation function , is the two - point function of the product , @xmath165 , integrated over the lines of sight . as pointed out by knox et al . ( 1998 ) it is this latter 2-point function which must be calculated from either simulations or analytical models of the reionization process . a very simple toy model was used by gruzinov and hu ( 1998 , hereafter gh ) and by knox et al . ( 1998 , hereafter ksd ) for investigating the effect of ihr on the power spectrum . in this model , independent sources turn on randomly and instantaneously ionize a sphere with comoving radius @xmath166 , which then remains ionized . the rate of source creation is such that the mean ionization fraction is zero prior to @xmath167 and then rises linearly with redshift to unity at @xmath67 . gh derived the following approximate expression for the resulting power spectrum : [ eqn : gh ] l^2 c_l 2 & = & a l^2 _ 0 ^ 2 e^-l^2 _ 0 ^ 2/2 , + _ 0 & & r _ 0 - _ ion , + a & = & 36 _ 0 ^ 2 < v^2 > r/_0 z ( 1+z_ion)^3/2 where @xmath168 is the variance of peculiar velocities and @xmath169 and @xmath170 is the number density of free electrons today . the qualitative shape of this power spectrum is simple to understand . on large scales it is the shape of a white noise power spectrum ( @xmath171 constant ) , due to the lack of correlations between the patches . at small scales there is an exponential cutoff corresponding to the angular extent of the patches themselves , since they have no internal structure . the exact calculations for this model by ksd result in a very similar shape but with the power reduced by a factor of two and the location of the peak shifted to larger @xmath98 by 30% . thus we can write down very simple , and accurate , equations for the location of the maximum , which are those of gh with the prefactors modified to fit the results of ksd : [ eqn : dopplermax ] ( l^2 c_l 2)_max & = & 36e _ 0 ^ 2 < v^2 > r/_0 z ( 1+z_ion)^3/2 , and + l_max & = & 1.8 _ 0 r . note that for sufficiently large @xmath166 , velocities are not coherent within a patch and equations [ eqn : gh ] and [ eqn : dopplermax ] break down ; cancellations once again become important . however , this only happens for @xmath172 mpc since the velocity correlation function only becomes negative for separations greater than about @xmath173 mpc ( ksd ) . as long a @xmath166 is less than 10 s of mpc , the power scales linearly with @xmath166 . aghanim et al . ( 1996 , hereafter `` a96 '' ) used a slightly more complicated model in which there was a distribution of patch sizes due to an assumed distribution of mini quasar luminosities . the larger patches dominate at low @xmath98 and the cutoff at higher @xmath98 is much softer than in the single patch case , due to the existence of the smaller patches . for an order - of - magnitude calculation , their largest patches have @xmath174 mpc and reionization proceeds from @xmath175 to @xmath176 . for this simplification of the a96 model we find the curve in figure [ fig : pspec ] which has @xmath177 and @xmath178 . this amplitude is just large enough to possibly be a source of systematic error in the determination of cosmological parameters from _ map _ data , which has sensitivity out to @xmath179 . to be more precise , _ map _ can determine the power in a broad band centered on @xmath180 with width @xmath181 to about 0.5 % . thus , given the primary signal assumed in the figure , we may expect signals on the order of @xmath182 to be significant contaminants . this ihr signal from a 10 mpc patch would be an even more noticeable effect in the _ planck _ data since _ planck _ has sensitivity out to about @xmath183 and can determine the power in a broad band centered on @xmath184 with width @xmath185 to about 0.2 % . the quasar luminosity function ( qlf ) is relatively well measured in the optical below redshift @xmath36 , with a typical average optical luminosity of @xmath186 . in order to estimate the expected typical hii patch size around the observed quasars , one needs to know their lifetime ( or `` duty cycle '' ) . taking the lifetime of quasars to be universal , and near the eddington time ( @xmath187 yr ) , and further ignoring recombinations , one obtains a comoving radius of @xmath311020mpc . as seen above , this would be large enough for its effects on the cmb to be detectable by _ planck _ and maybe _ map_. we emphasize however , that the typical quasar luminosity at @xmath188 can be much lower than at @xmath189 . in hierarchical structure formation models , the nonlinear mass scale is @xmath190 orders of magnitude smaller at @xmath188 than at @xmath34 . with a linear scaling between quasar luminosity and halo mass , this would reduce the average patch size by a factor of @xmath61 . in addition , the duty cycles of quasars are not well established . in theoretical models where the qlf is derived by associating quasars with collapsed dark halos ( in the press schechter formalism ) , a typical lifetime of @xmath187 yr would generically lead to an overestimate of the number of observed quasars ( e.g. small & blandford 1992 ) . to avoid these overpredictions , one needs to assume that either ( 1 ) the quasar shines at a small fraction of the eddington luminosity , or ( 2 ) the efficiency of black hole formation , expressed as @xmath191 , is 2 orders of magnitude lower than suggested by nearby galaxies ( magorrian et al . 1998 ) , or ( 3 ) the black hole masses grow in an optically inactive phase ( e.g. haehnelt , natarajan & rees 1998 ) . alternatively , one can fit the observed qlf by assuming that the quasar lifetime is @xmath31100 times shorter than the eddington time ( haiman & loeb 1997 ) . in the latter case , the hii patch sizes would be reduced to @xmath192 mpc ( i.e. reionization would correspond to a larger number of smaller patches ) . according to equation [ eqn : dopplermax ] , patches smaller than a few mpc result in power spectra with smaller maxima at values of @xmath98 that are too high , and well outside the range of sensitivity of _ map _ and _ planck_. even for patches as large as 1 mpc , one can see that , at least for reionization at the extreme low end of the redshift range , their effect is negligible ( see figure [ fig : pspec ] ) . a detection of an ihr signal in the cmb with a peak could have important implications for quasar evolution theories , e.g. by ruling out models that lead to short duty cycles and small patch sizes . a detection might also prove that reionization was caused by mini quasars , rather than mini galaxies , since for the mini - galaxy models , the patch sizes are expected to be only a few hundred kpc . however , in light of the small expected patch sizes , it seems unlikely that _ map _ or _ planck _ reaches the required angular resolution for such a detection . the above calculations of the ihr power spectrum all assumed that the ionized regions are uncorrelated . however , the pattern of density fluctuations will affect the pattern of ionized regions and therefore correlations in the density field should give rise to correlations in the ionization fraction field . determining what these correlations should be is complicated by the fact that the density field affects the ionization field in two different ways : by the fact that the sources are most likely located in high density regions and the fact that the recombination rates are highest in the densest regions . to take the former effect into account , kds gave the ionization field the same correlation structure as that of the `` high - peaks '' of the density field . these high - peaks are regions where the fractional density perturbation , @xmath193 , smoothed on the appropriate scale , exceeds the critical value for spherical collapse . the smoothing scale was taken to be that corresponding to @xmath194^{-3/2}$ ] , since objects below this size will not be able to cool sufficiently to fragment and form stars ( haiman , rees & loeb 1997 ) . one further assumption was that the sources ionized a region @xmath195 times larger than that from which they collapsed where the efficiency factor is time - dependent and given by ( haiman & loeb 1997 , haiman & loeb 1998a ) @xmath196 . for this rather high choice for the efficiency , reionization is complete at @xmath197 . the resulting spectrum is shown as the dashed line in figure [ fig : pspec ] . the correlations between the patches which are individually only hundreds of kiloparsecs in radius drastically alter the low @xmath98 shape . even given the high redshift of reionization in this scenario , the corresponding uncorrelated model of equation [ eqn : gh ] would predict orders of magnitude less power at @xmath198 to @xmath199 . if this signal were in the data , but not in ones model of the data , it would lead to a bias in the determination of cosmological parameters . ksd estimated the bias that would occur in a nine parameter fit , assuming the primary signal was that of standard cold dark matter . for planck , the systematic in the estimate of the baryon density was equal to the statistical error . for all the other parameters the systematic error was less than the statistical error . if the mini quasar scenario is correct , with bright , long - lived quasars , then the doppler effect will lead to a significant contamination of the _ planck _ data and possibly a marginally important one for _ map _ data . for the more likely mini galaxy scenario , if one ignores the correlations in the ionized patches , the contamination is completely irrelevant for both _ map _ and _ planck_. however , correlations may indeed be important , and in one attempt to take them into account , we have seen that for an extreme mini galaxy scenario , the contamination is marginally significant for _ planck_. work by oh ( 1999 ) supports the idea that the ionized patches should be correlated . oh has modeled the ionizing sources as characterized by a mean separation ( given by press - schechter theory ) , an isotropic attenuation length which specifies how far the ionizing radiation propagates , and a power - law correlation function ( appropriate for these highly - biased objects ) . placing sources in a box and working out the resulting flux densities he can then calculate the correlation function of the flux density . it is indeed correlated , with significant ( 0.2 ) correlation at a comoving separations of 10 mpc . although a clumping factor for the igm from gnedin & ostriker ( 1997 ) is used to estimate the attenuation length , all propagation from the sources is taken to be isotropic , and hence this work does not yet include the influence of shielding and shadowing . as one can see in fig . [ fig : pspec ] , the contribution from ihr may be sub - dominant to that from the ostriker - vishniac ( o - v ) effect , as it is for all three models of ihr shown . the o - v contribution here is calculated using the formalism of jaffe & kamionkowski ( 1998 , hereafter `` jk98 '' ) . one must keep in mind that as a second - order effect , it is highly sensitive to the normalization . the curve here is cobe - normalized ( @xmath200 ) ; a cluster normalization of @xmath201 would reduce the @xmath89 by a factor of 16 . it is also calculated for the very early reionization redshift ( and resulting high optical depth ) of @xmath202 . however , the amplitude is not very sensitive to @xmath104 because much of the effect actually comes from more recent times the dropping number density of electrons must compete with the growth in the density contrasts and peculiar velocities(hu & white 1995 , jk98 ) . of course , the separation into ihr and o - v is artificial , and efforts to understand the contribution from reionization are surely going to benefit from treating them simultaneously . indeed , the 2-point function of @xmath203 , measured from a simulation , would include both effects . gravity is the only other late - time , frequency - independent , influence on the cmb photons , causing lensing and the rees - sciama effect ( rees & sciama 1968 ) . the angular - power spectrum from the rees - sciama effect has been calculated by seljak ( 1995 ) and is substantially sub - dominant to either the primary spectrum or the o - v effect at all angular scales . gravitational lensing results in a smearing of the primary cmb angular power spectrum and has also been calculated by seljak ( 1996 ) . uncertainties in this calculation may indeed be important for any attempt to recover the contribution from the ihr and ostriker - vishniac effects . the principal unanswered questions about reionization are : what type of sources caused it , and at what redshift did it occur ? these questions have been addressed both theoretically and observationally . from a theoretical point of view , the two leading candidate sources are an early generation of stars ( `` mini galaxies '' ) , or massive black holes in small halos ( `` mini quasars '' ) . it is possible to estimate the efficiencies with which objects form in the earliest collapsed halos : from the metals in the ly@xmath6 forest , and from the evolution of the quasar luminosity function at @xmath36 , respectively . however , the uncertainties in these efficiencies are still too large to allow definite predictions . the expected reionization redshift , depending on the type of source , cosmology , power spectrum , and a combination of the efficiency factors , is between @xmath204 in homogeneous models . a further complication is the inhomogeneous nature of reionization : the sources of ionizing radiation are likely clustered , and embedded in complex dense regions , such as the filaments and sheets seen in 3d simulations . at the same time , the gas to be ionized likely has significant density fluctuations . further theoretical progress on the problem will likely come from 3d simulations . such simulations must be able to follow the propagation of a non - spherical ionization front into a medium that has large density fluctuations , as well as opaque clumps of absorbing material . in a homogeneous medium populated by randomly distributed sources , the background radiation would have negligible fluctuations ( e.g. zuo 1992 ) . however , because the density inhomogeneities absorb the uv flux of the ionizing sources , and re - emit diffuse ionizing photons as they recombine , the ionizing background will likely become significantly inhomogeneous and anisotropic ( norman et al . 1998 ) . the first numerical algorithms to deal in full detail with these problems have been proposed by abel , norman & madau ( 1998 ) and razoumov & scott ( 1998 ) ( see also gnedin 1998 ) . the reionization redshift can also be constrained observationally , with a resulting uncertainty that is comparable to the theoretical range : @xmath205 . on both ends of this redshift range , observational progress is likely in the near feature . if new ly@xmath6 emitters are discovered at higher redshifts , this would improve the lower bound . conversely , as the cmb data are collected , the constraint on the electron scattering optical depth will improve , tightening the upper bound . we have seen how the optical depth to thomson - scattering , and hence the epoch of reionization , can be constrained by determining the amount of damping in the cmb temperature anisotropy and by detection of the polarization contribution at very large angular scales . the spatial inhomogeneity of the reionization process gives us a further opportunity to probe in more detail the epoch of reionization . for this probe to become reality , further development of theoretical predictions is required , as well as more sensitive measurements at arc minute and smaller angular scales . fortunately , the inhomogeneity of reionization is unlikely to spoil our ability to interpret the primary cmb anisotropy , to be measured with exquisite precision over the next decade , although this may not be the case if reionization did indeed occur via mini quasars . for the mini galaxy models , even when reionization occurs at the very high end of the allowed redshift range , very little signal is produced at the relevant angular scales . further progress in these predictions is likely to be motivated by the desire to understand the spectrum at @xmath206 as a means of probing the end of the dark ages , rather than at @xmath207 , as a possible contaminant of the primary signal . we are grateful to our collaborators s. dodelson , a. loeb , m. rees , and r. scoccimarro for contributing to our understanding of this subject , to a. jaffe for supplying the `` o - v '' effect curve in fig . [ fig : pspec ] , and to m. norman for supplying fig . zh was supported at fermilab by the doe and the nasa grant nag 5 - 7092 . gnedin , n. y. 1998 , in proc . of 19@xmath208 texas symposium on relativistic astrophysics and cosmology , held in paris , france , dec . 14 - 18 , 1998 , eds . j. paul , t. montmerle , and e. aubourg ( cea saclay ) , in press haiman , z. , & loeb , a. 1998c , invited contribution to the proceedings of 9@xmath208 annual october astrophysics conference , _ after the dark ages : when galaxies were young _ , october 1998 , college park , md , preprint astro - ph/9811395 norman , m. l. , paschos , p. , & abel , t. 1998 , in proc . of _ @xmath209 in the early universe _ , workshop held in florence , italy , eds . e. corbelli , d. galli , and f. palla , memorie della societa astronomica italiana , p. 455 shapiro , p. r. , raga , a. c. , & mellema , g. 1997 , in _ structure and evolution of the igm from qso absorption line systems _ , 13@xmath208 iap colloquium , eds . p. petitjean and s. charlot ( paris : editions frontiere ) , in press , preprint astro - ph/9710210 wadsley , j. w. , hogan , c. j. , anderson , s. f. 1998 , to appear in the proceedings of `` after the dark ages : when galaxies were young ( the universe at @xmath210 ) '' , 9th annual october astrophysics conference in maryland , preprint astro - ph/9812239

systematic investigations of a(@xmath10 ) reactions at saclay , nikhef , mainz and bates have produced an impressive amount of data for various target nuclei @xcite . for the sake of minimizing the uncertainties with respect to the reaction mechanism , a large fraction of these investigations were done under or close to quasi - elastic kinematics ( bjorken @xmath11 1 ) . these data sets highlight at the same time the success and the limits of the independent - particle - model ( ipm ) for atomic nuclei @xcite . indeed , after corrections for final - state interactions and coulomb distortions of the electron probe , the shape of the deduced proton momentum distributions are systematically in line with the predictions of modern formulations of the nuclear ipm . on the other hand , below the fermi momentum the absolute magnitude of the deduced momentum distributions are systematically , i.e. independent of the nucleon momentum , lower than ipm predictions . to cut a long story short , the major conclusion from this world - wide @xmath5 effort seems that an appropriate non - relativistic picture of the nucleus is roughly compatible with _ 70% mean - field behaviour and 30% `` correlations '' _ an observation which is still frequently ignored in various nuclear structure calculations and model developments . the energy of the available electron beams with a large duty factor ( @xmath12 1 gev ) made that most of this aforementioned @xmath5 work was done at four - momentum transfers of the order @xmath13 0.2 ( gev / c)@xmath3 . with the advent of the tjnaf and an upgraded mainz electron facility higher values of @xmath2 come into reach of experimental exploration . amongst the major physics goals motivating exclusive ( @xmath10 ) measurements from finite nuclei at higher momentum transfer ( @xmath14 ( gev / c)@xmath3 ) one can mention the following ones . the higher @xmath2 conditions and unmistakingly smaller distance scales probed should make it feasible to achieve a better understanding of the short - range mechanisms in nuclei . at the same time , one could hope to find experimental evidence for the onset of quark and gluon degrees of freedom . presumably , the most convincing evidence pointing into that direction could come from measurements for processes that are fairly well understood at lower @xmath2 ( like ( @xmath10 ) at x=1 ) and turn out to be completely at odds with meson / baryon models when higher @xmath2 regimes are entered . another challenge for ( @xmath10 ) measurements at higher momentum transfers is the question whether available relativistic models can succeed in producing a better agreement with the data sets than the non - relativistic ones and if so , which dynamical degrees of freedom make them to be substantially different from what is commonly implemented in the non - relativistic nuclear many - body models . another fundamental question that has received a great share of attention for many years , is the question whether the nucleon properties ( like electromagnetic form factors e.g. ) are modified in the nuclear medium . this information is of unvaluable importance for models that embark on the ambitious program of understanding nuclei in terms of quark and gluon degrees of freedom @xcite . a challenging but at the same time rather ambiguity - free way of probing the medium - dependent form factors , are double polarization observables from a(@xmath0 ) measurements @xcite . indeed , double polarization observables are conceived to be rather insensitive to ambiguities with respect to the final state interactions ( fsi ) that affect most of the other a(@xmath10 ) observables . other effects that are recognized to possibly complicate the interpretation of double polarization observables in terms of medium - dependent nucleon form factors are gauge ambiguities , channel - couplings and two - body current effects . whereas the first two sources of possible dilutions were extensively studied by kelly @xcite and found to produce very small corrections , the two - body current effects are not well studied for finite nuclei . earlier efforts to study the role of two - nucleon currents in exclusive a(@xmath10 ) reactions from finite nuclei ( a@xmath154 ) include the pioneering work of suzuki @xcite , the systematic investigations by the pavia group @xcite and the @xmath16he(@xmath10 ) studies as e.g. reported in refs . @xcite . in order to make their calculations computationally more attractive , the results of the pavia group were obtained with a two - body current operator that was formally reduced to an effective one - body one . this approximation adopts a fermi - gas picture for the residual nucleus that allows to integrate out the coordinates of the second nucleon that gets involved in the photoabsorption process . for the work presented here , we treat the two - body currents in their full ( non - relativistic ) complexity and deal with the multi - dimensional integrals that automatically occur when several nucleons get involved in the ( virtual ) photoabsorption process . a similar sort of exact treatment was earlier adopted in our work reported in references @xcite . there it was found that the effect of the two - body currents on the a(@xmath10 ) cross sections , just as the role of the coupled - channel effects , is gradually decreasing with increasing momentum transfer . this conforms to the findings with respect to the momentum - transfer dependence of the two - body current effects in the d(@xmath10)n and @xmath16he(@xmath10 ) @xcite . nevertheless , even at the highest momentum transfer considered in ref . @xcite ( q=600 mev / c ) the cross sections were predicted to be substantially affected by the two - body currents . in this work the role of meson - exchange and isobar degrees of freedom in a(@xmath0 ) and a(@xmath1 ) is investigated in a wide range of four - momentum transfers ( 0.2 @xmath170.8 ( gev / c)@xmath3 ) . special emphasis is placed on the recoil polarization observables as they open perspectives to investigate possible medium modifications of the nucleon properties . the calculational framework will be introduced in section [ sec : model ] . in section [ sec : observ ] the adopted conventions for the ( @xmath0 ) observables will be introduced . the model for the bound and scattering states is described in section [ sec : fsi ] . in section [ sec : current ] the model assumptions with respect to the one- and two - body current operators will be summarized . the results of the calculations are contained in section [ sec : results ] . our conclusions are summarized in section [ sec : conclusions ] . we consider processes in which a longitudinally polarized electron impinges on a nucleus and induces the following reaction @xmath20 to occur . for such a process the cross section reads in the one photon exchange approximation @xmath21 \biggr ] \ ; , \label{eq : eep}\end{aligned}\ ] ] where @xmath22 is the recoil factor @xmath23 and @xmath24 the mott cross section @xmath25 the electron kinematics is contained in the kinematical factors @xmath26 whereas the structure functions are defined in the standard fashion @xmath27 & w_{tt}= 2 { \cal r}e \left [ \left ( j_{-1}^{fi } \right)^ * \left ( j_{+1}^{fi } \right ) \right ] \\ w_{lt}^ { ' } = -2 { \cal r}e \left [ \left ( j_{0}^{fi } \right)^ * \left ( j_{+1}^{fi } \right ) + \left ( j_{0}^{fi } \right)^ * \left ( j_{-1}^{fi } \right ) \right ] & w_{tt}^ { ' } = \left ( j_{+1}^{fi } \right)^ * \left ( j_{+1}^{fi } \right ) - \left ( j_{-1}^{fi } \right)^ * \left ( j_{-1}^{fi } \right ) \ ; . \\ \end{array}\ ] ] the above definitions for the structure functions and the kinematical variables correspond with those of ref . we remind that apart from a negligible parity - violating component , the structure function w@xmath28 vanishes identically and the w@xmath29 in coplanar kinematics if no recoil polarization is determined . a new set of observables comes into reach of experimental exploration when performing polarimetry on the ejected hadron . this results in knowledge about the spin orientation of the ejectile and as e.g. illustrated in neutron form factor studies represents a powerful tool to address fundamental physical quantities . the formal framework for the electroproduction of polarized nucleons from nuclei is outlined in great detail in refs . @xcite and will not be repeated here . here , we only review some basic concepts which mainly serves at introducing the conventions adopted . for the results presented below , the polarization of the escaping nucleon is expressed in the so - called barycentric reference frame that is defined by the following set of unit vectors ( fig . [ fig : reffram ] ) @xmath30 note that for coplanar kinematics @xmath31 determines the y axis of the reference frame . the escaping nucleon polarization observables can be determined through measuring * ratios*. the _ induced polarization _ can be addressed with unpolarized electrons ( i = n , l , t ) @xmath32 whereas the _ polarization transfer _ also requires polarized electron beams ( i = n , l , t ) @xmath33 - \left [ \sigma ( h=1,s_{n}^{i}= \downarrow ) - \sigma ( h=-1,s_{n}^{i}= \downarrow ) \right ] } { \left [ \sigma ( h=1,s_{n}^{i}= \uparrow ) + \sigma ( h=-1,s_{n}^{i}= \uparrow ) \right ] + \left [ \sigma ( h=1,s_{n}^{i}= \downarrow ) + \sigma ( h=-1,s_{n}^{i}= \downarrow ) \right ] } \ ; , \label{eq : trans}\end{aligned}\ ] ] where @xmath34 denotes that the ejected hadron is spin - polarized in the positive i direction ( i=(n , l , t ) ) and @xmath35 is the helicity of the electron impinging on the target nucleus . @xmath36 is a shorthand notation for the differential cross section for an electrodisintegration process initiated by an electron with helicity @xmath35 and for which the ejectile is detected with a spin polarization characterized by @xmath37 . throughout this work we adopt relativistic kinematics . for the model calculations presented here , the bound and scattering states are produced by solving a schrdinger equation with a mean - field potential that is obtained from a hartree - fock calculation . the latter are performed with an effective nucleon - nucleon force of the skyrme type . the mean - field potential includes central , spin - orbit and coulomb terms . this model does not require any empirical input with respect to the inital and final - state potentials . moreover , the orthogonality condition between the initial and final states is obeyed and gauge invariance is preserved at the one - body current level . a drawback of the model is that the inelastic processes , which are commonly accounted for through imaginary parts in the final - state potential , are only partially included . the major impact of the rescattering processes in exclusive ( @xmath10 ) is generally conceived , however , to be a mere reduction of the absolute cross section . in the kinematics regime of interest here , the reduction factor can be estimated from nuclear transparency @xmath5 measurements as they were recently performed at tjnaf for various target nuclei @xcite . the differential cross section , the various structure functions and polarization observables are all calculated starting from the following set of transition matrix elements @xmath38 where @xmath39 and @xmath40 refer to the quantum states of the target and residual a-1 nucleus and @xmath41 denotes the spin projection of the ejectile along the z - axis . the latter is chosen to coincide with the direction of the momentum transfer @xmath42 . when calculating the induced and transverse polarizations as defined in eqs . ( [ eq : indu ] ) and ( [ eq : trans ] ) the transition matrix element for the ejectile s spin pointing in a certain direction determined by the polar and azimuthal angle ( @xmath43 ) are required . these can be easily obtained from the above matrix elements ( [ eq : matr ] ) by remarking that @xmath44 where @xmath45 is the wigner @xmath46-matrix for @xmath47 @xmath48 in determining the matrix elements of eq . ( [ eq : matr ] ) , a standard multipole expansion for the electromagnetic current operators and the `` distorted '' outgoing nucleon wave is made . the multipole expansion of the current operators tends to converge more slowly as more extended systems and higher momentum transfers are addressed . for the highest momentum transfer considered here ( @xmath49 gev ) , convergence in the @xmath50o calculations could only be reached after including all multipolarities up to @xmath51=30 . this number is compatible with the estimate one would obtain by setting @xmath52 , where @xmath53 is the nuclear radius determined by @xmath54 . the large number of multipoles required for calculations at high momentum transfer is a serious numerical complication that prevents models that were developed for lower energy and momentum transfers from being easily extended to higher energy and momentum domains . apart from the fact that a full treatment of the relativistic effects would be in order , the partial - wave expansion technique hampers extending the presented formalism beyond the range of momentum - transfers considered here ( @xmath551 gev / c ) . a quasi - elastic ( @xmath10 ) process will predominantly excite those a-1 states @xmath56 that bear a sizeable hole state component in their overlap with the ground - state of the target nucleus . this component of the final wave function can be constructed starting from the standard multipole expansion in terms of particle - hole @xmath57 excitations out of the ground state of the target nucleus @xcite : @xmath58 where @xmath59 , @xmath60 is the reduced mass of the outgoing nucleon , @xmath61 is the central phase shift and @xmath62 is the coulomb phase shift related to the electromagnetic part of the mean - field potential . the above expression has been derived for the following conventions with respect to the asymptotic behaviour of the continuum eigenstates @xmath63 of the mean - field potential and the total a - body wave function : @xmath64 where @xmath53 is the nuclear radius . the one - body current is derived from the on - shell covariant single - nucleon current @xmath65 u ( \vec { p } _ m , s_m^z ) \ ; , \label{eq : dirac}\ ] ] where @xmath66 is the dirac ( pauli ) form factor . even when making an abstraction from issues related to off - shell corrections , the derivation of an appropriate current operator with relativistic corrections to be used in calculations that adopt a non - relativistic description for the strong interaction part of the reaction process , is not free from ambiguities @xcite . for example , the two standard techniques to derive a non - relativistic reduction of relativistic hamiltonians , i.e. the foldy - wouthuysen ( fw ) and the `` direct pauli reduction '' method , have been shown to make a difference as far as the final expressions for the current operators are concerned @xcite . admittedly , the problems encountered in implementing relativistic corrections in non - relativistic calculations can be avoided by adopting a fully relativistic description for the nuclear dynamics . over the years , a number of fully relativistic models for ( @xmath10 ) have been developed @xcite . basically , all of these models start from a relativistic formulation of the mean - field idea and do not embody current operators that go beyond the ia . unfortunately , for finite nuclei a realistic , empirically well - founded and practicable relativistic nuclear - structure model that _ also accounts for the multi - nucleon degrees of freedom _ is as yet not available . with this in mind we shall resort to a non - relativistic approach for the strong interaction dynamics of the @xmath5 process . we deem this model to be a reliable and valuable testing ground to estimate the relative importance of subnuclear degrees of freedom . the conventional way of deriving a nonrelativistic reduction of the above one - body current operator is the fw technique which relies on a ( @xmath67 ) expansion . recently , an alternative expansion in terms of @xmath68 , with @xmath69 the momentum of the nucleon on which the absorption occurs , has been suggested @xcite . this expansion is based on the `` direct pauli reduction '' method . in order @xmath70 and @xmath68 respectively , both methods give rise to a nonrelativistic four - current density operator that reads in coordinate space @xmath71 \biggr ) \nonumber \\ \vec{j}^{\perp}(\vec{r } ) & = & \sum _ { i=1 } ^{a } c_3(q , q^2 ) \biggl ( \frac { g_e^i(q^2 ) } { 2m_n i } \left\ { \vec{\nabla}_i , \delta ( \vec{r } - \vec{r}_i ) \right\ } + \frac { g_m^i(q^2 ) } { 2m_n } \delta ( \vec{r } - \vec{r}_i ) \left ( \vec{\sigma}_i \times \vec{\nabla } \right ) \biggr ) \ ; , \label{eq : redcurrent}\end{aligned}\ ] ] where @xmath72 denotes the anticommutator and it is implicitly understood that the operator @xmath73 acts only the photon field . the @xmath74 and @xmath75 are the sachs form factors for which we have adopted the standard dipole form . the major difference between the current operators in the two earlier sketched approaches is not the operatorial form but the derived expression for the coefficients ( @xmath76 ) . whereas with the fw method one finds @xcite @xmath77 a @xmath68 expansion with the aid of the pauli reduction technique leads to @xcite @xmath78 unless otherwise specified all results of this paper are obtained with the four - current operator from eq . ( [ eq : redcurrent ] ) using the @xmath79 coefficients as they are obtained with the fw method . the major difference in comparison with a strict non - relativistic approach is the introduction of a spin - orbit term @xmath80 in the charge density operator . this gives rise to a spin - orbit term in the coulomb transition operator that determines the longitudinal strength @xmath81 in coordinate space the one - body transition matrix element corresponding with this operator reads @xmath82^{(j_{2 } ) } \parallel n_b l_b \right > \ ; .\end{aligned}\ ] ] in determining the two - body current operators we start from the observation that of all mesons carying the nucleon - nucleon interaction the pions play a predominant role . in our model calculations , the current operators that explicitly account for subnuclear @xmath83 and @xmath9 degrees of freedom are effecively written in terms of the coordinates of the two nucleons involved . the meson - exchange and @xmath9-isobar currents as they were implemented in our calculations have been discussed in detail in ref . the two - body pion - exchange currents are derived from the one - pion exchange potential in the standard fashion . the derivation of the @xmath9-current operator is somewhat more complicated . indeed , this operator can not be constrained through charge - current conservation and is therefore often referred to as a `` model - dependent '' operator @xcite . moreover , of all pion - related two - body current contributions the @xmath9-current is the only one that could be unambiguously shown to exhibit a quite strong medium dependence . we do not consider explicit @xmath9 admixtures in the wave functions , so that all isobars in our model calculations are attached to a photon line . in our derivations , the @xmath84 and @xmath85 coupling are considered in the standard form : @xmath86 where @xmath87 and @xmath88 denote the @xmath89 spin and isospin transition operators , @xmath90 is the external electromagnetic field and the coupling constants are @xmath91 = 0.37 and @xmath92 = 0.12 . the electromagnetic form factor of the delta @xmath93 is parametrized as @xcite @xmath94 where @xmath95=0.84 gev / c and @xmath96=1.2 gev / c . it is interesting to note that the @xmath97 electromagnetic form factor @xmath98 is decreasing with @xmath2 faster than the nucleon dipole form @xcite . with the above coupling lagrangians we arrive at the following expression for the @xmath99-current in momentum space @xmath100 \times \vec q \biggr ) \nonumber \\ & & + \left ( g_{\delta}^{i } - g_{\delta}^{ii } \right ) \biggl ( - 2i { \vec \tau _ { 2,z } } \left ( \left ( \vec \sigma _ 1 \times \vec k_2 \right ) \times \vec q \right ) { { \vec \sigma _ 2 \cdot \vec k_2 } \over { \vec{k_2}^2+m_\pi ^2 } } -2 i \vec \tau _ { 1,z } \left ( \left ( \vec \sigma _ 2\times \vec k_1 \right ) \times \vec q \right ) { { \vec \sigma _ 1\cdot \vec k_1 } \over { \vec{k_1}^2+m_\pi ^2 } } \nonumber\\ & & -2i ( \vec \tau _ 1\times \vec \tau _ 2)_z \left [ \vec k_2 { { \vec \sigma _ 2\cdot \vec k_2 } \over { \vec { k_2}^2+m_\pi ^2 } } - \vec k_1 { { \vec \sigma _ 1\cdot \vec k_1 } \over { \vec { k_1}^2+m_\pi ^2 } } \right ] \times \vec q \biggr ) \biggr ] \;.\end{aligned}\ ] ] at the @xmath84 and @xmath101 vertices , monopole form factors @xmath102 are introduced . they correct for finite size effects of the interacting baryons and regularize the @xmath101 interaction at extremely short distances . all results presented below are obtained with a cut - off parameter @xmath103 of 1250 mev / c . this value corresponds with those that are typically obtained in the latest parametrizations of the bonn potential . in the above expression , @xmath104 ( @xmath105 ) denotes the propagator for a @xmath9 resonance that is created after ( before ) photoabsorption . in the calculations we adopt the following propagators @xmath106 where @xmath107 is the intrinsically available energy for the resonance , @xmath108=1232 mev and @xmath109 the `` free '' @xmath110 decay width @xcite @xmath111 where @xmath112 is the decay momentum in the center - of - mass ( c.o.m ) frame of the @xmath110 system . as we are dealing with off - shell nucleons the photon energy is not completely available for internal excitation of the @xmath113 resonance . this effect is partially responsible for the observed ( real ) energy shift of the @xmath9 in the medium . a reasonable substitution for @xmath114 is @xcite @xmath115 where @xmath116 is the binding energy of the mean - field orbit on which the pion is reabsorbed . the above expression accounts for the observation that the delta peak in ( @xmath117 ) spectra section shifts to higher @xmath118 s with increasing momentum transfer . various pion - nucleus @xcite , real photoabsorption @xcite and electron scattering studies @xcite experiments have pointed towards to strong medium modifications of the @xmath99 resonance . for that reason , a medium - dependent term @xmath119 that accounts for the interaction of the @xmath9 resonance with the nucleus was added to the propagator written in ( eq . [ eq : delta1 ] ) . various models that implement a dynamical description of isobar propagation in the medium have been developed @xcite . these models are generally reasonably successful in describing the cross sections for inclusive ( @xmath117 ) reactions in the @xmath99 resonance region @xcite . the medium modifications of the @xmath9-resonance appear to be better under control in electromagnetically induced processes than for example in pion - absorption reactions @xcite . for example , chen and lee @xcite have shown that a fairly good description of the @xmath120c(@xmath117 ) cross sections could be achieved with a @xmath9 propagator of the type as written in eq . ( [ eq : delta1 ] ) provided that one introduces a simple medium correction of the type @xmath121 = -30 - 40 \ ; \mathrm{i}$ ] . similar values for the @xmath9-mass shift and broadening have been found in other theoretical approaches @xcite and were deduced from recent total photoabsorption measurements @xcite . for the calculations presented here , we have used the @xmath9-medium potential @xmath119 as it was calculated by oset and salcedo @xcite . in figure [ fig : width ] the imaginary part of @xmath119 is shown . in the resonance region , this calculation does indeed reproduce the earlier quoted value for the broadening of the resonance . the above two - body current operators have been used to calculate @xmath120c(@xmath122 ) and @xmath120c(@xmath123 ) total cross sections in the resonance region . not only could the position and width of the resonance be reasonably reproduced , also the ( @xmath122)/(@xmath123 ) ratio for the various shell - model orbits agreed fairly well with experiment @xcite . these results lend confidence in the model assumptions with respect to the @xmath9 currents and propagators . introducing the earlier discussed two - body current operators in exclusive single - nucleon knockout calculations one ends with the diagrams sketched in figure [ fig : diagram ] . the diagrams of figure [ fig : diagram](b),(c ) and ( d ) are the pion - exchange contributions to single - nucleon knockout and necessarily imply a charge - exchange mechanism carried by a charged pion . as a consequence , for proton knockout the sum over all occupied single - particle states in the target nucleus ( @xmath124 ) produces solely non - zero contributions for the neutron states . the diagrams from figure [ fig : diagram](e)-(h ) involve @xmath99 excitation after interaction of one of the target s nucleons with the ( virtual ) photon field . for this class of diagrams both charged and neutral pion exchange belongs to the possibilities . the contributions drawn in figures [ fig : diagram](e ) and ( h ) involve exclusively neutral pion exchange . finally , the type of processes drawn in figure [ fig : diagram](i)-(l ) involve a pre - formed @xmath9 that is deexcited after interacting with the photon field . technically speaking , the introduction of the two - body current operators implies that for each combination of a multipole component of the electromagnetic transition operator @xmath125 and a partial wave of the ejectile s wave function ( @xmath126 ) , two - body matrix elements of the type @xmath127}(q)\mid\mid 0^{+}(g.s . ) > = \sum_{h ' j_{1}j_{2 } } \sqrt{2j_{1}+1}\sqrt{2j_{2}+1}(-1)^{j_{h}-j_{h ' } -j - j_{2 } } \left\ { \begin{array}{ccc}j_{h } & j_{h ' } & j_{1}\\j_{2 } & j & j \end{array } \right\}\nonumber\\ & & \times \bigl ( < h h ' \ ; ; \ ; j_{1 } \mid\mid t_{j}^{[2]}(q)\mid\mid p(lj \epsilon ) h ' \ ; ; \ ; j_{2 } > - ( -1)^{j'_h+j+j_2 } < h h ' \ ; ; \ ; j_{1 } \mid\mid t_{j}^{[2]}(q)\mid\mid h ' p(lj \epsilon ) \ ; ; \ ; j_{2 } > \bigr ) \ ; , \label{eq : mattwo}\end{aligned}\ ] ] are to be coherently added to the conventional one - body current contribution from the impulse approximation ( ia ) @xmath128}(q ) \parallel 0^{+}(g.s . ) > \ ; .\ ] ] in the above expression , @xmath129}$ ] and @xmath130}$ ] is the one and two - body current contribution to the electromagnetic transition operator . the explicit expressions for the reduced two - body matrix elements with the meson - exchange and isobar currents can be found in refs . the sum over @xmath131 involves all occupied proton and neutron single - particle states . it speaks for itself that for a nucleus like @xmath132pb this summation can only be performed at a large computational cost . for consistency reasons and to avoid orthogonality deficiencies , the wave functions for all occupied states @xmath131 are calculated in exactly the same mean - field potential in which also the distorted outgoing nucleon wave and the overlap wave function @xmath133 is determined . we remark that the very same type of matrix elements that dictate the two - body current contributions to the single - nucleon knockout processes , determine the two - body current contributions to the cross sections for the a(@xmath134 ) and a(@xmath135 ) processes that are presently the subject of investigation at various laboratories . it speaks for itself that two - nucleon knockout processes represent an intrinsically superior way of exploring the two - nucleon effects in nuclei . the model assumptions with respect to the current operators adopted here , are indentical to those that we have adopted in our two - nucleon knockout studies . also the potential in which the bound and scattering states are calculated are indentical for the single- and two - nucleon knockout studies . to illustrate the potential of two - nucleon knockout studies to acquire a precise understanding of two - nucleon mechanisms in finite systems , figure [ fig : gercoeep ] shows a comparison of recently obtained @xmath50o(@xmath136 ) data and our model calculations for the three lowest bins in the excitation energy spectrum in @xmath137c . the comparison between the calculations and the data is done as a function of the pair missing momentum @xmath138 , which is the c.m . momentum of the pair before it undergoes the electromagnetic interaction with the photon field and is established to be the scaling variable in two - nucleon knockout processes , i.e. the counterpart of the variable @xmath139 in the ( @xmath10 ) . for the data shown in figure [ fig : gercoeep ] the experimental resolution in the excitation ( or missing ) energy of @xmath137c was of the order of 4 mev and the individual states could not be resolved in the missing energy spectrum . on the other hand , high - resolution @xmath50o(@xmath136 ) investigations from mainz @xcite and @xmath140n(@xmath141)@xmath137c transfer reactions @xcite allow to infer that the lowest excitation - energy bin ( -4 mev @xmath1424 mev ) is exclusively fed through the @xmath137c ground - state transition , whereas the second ( 4 mev @xmath1439 mev ) and third bin ( 9 mev @xmath14314 mev ) are mainly fed through the @xmath144 doublet and the @xmath145 state respectively . these states can be safely quoted to be the only ones in the low excitation - energy spectrum of @xmath137c that have a dominant `` two - hole '' structure relative to the ground state of @xmath50o . this information allows to constrain the nuclear - structure input that is required for the @xmath146 calculations . the overall agreement between the calculations , that are parameter - free , and the data is reasonable and lends support for the model assumptions with respect to the @xmath9-current operator and propagator . indeed , for the transitions to the 2@xmath147 and @xmath148 states intermediate @xmath9 creation is predicted to be the dominant reaction process in the ( @xmath136 ) reaction under evaluation . the shaded region in figure [ fig : gercoeep ] is the calculated contribution from the central jastrow correlations to the two - proton knockout cross sections . they make a big contribution to the ground - state transition ( upper panel ) in the low missing momentum regime @xmath149 mev / c which can be inferred to arise from diproton knockout from heavily correlated @xmath150(t=0 ) pairs @xcite . for the results presented in figure [ fig : gercoeep ] the contribution from the jastrow or short - range correlations ( src ) was calculated with a correlation function as it was obtained from a g - matrix calculation by gearhart and dickhoff @xcite . recently , this correlation function was shown to produce a favorable agreement with @xmath120c(@xmath136 ) data @xcite . in comparison with other model predictions for the central correlation function , the one obtained by gearhart and dickhoff should be classified in between the categories of `` hard '' ( with a core at short internucleon distances ) and `` soft '' ( characterized by a finite probability to observe nucleon pairs at very short internucleon distances ) . we start our investigations into the role of two - currents for the kinematics of an @xmath50o(@xmath10 ) experiment made at mainz and reported in reference @xcite . in this experiment , the cross sections were measured in a broad missing momentum range between 100 and 700 mev / c . the data were compared to distorted wave impulse approximation ( dwia ) calculations . at lower missing momenta ( @xmath153300 mev / c ) the data were observed to overshoot these calculations by a factor of two , whereas for the highest missing momenta probed the opposite effect was noticed . this qualitative behaviour was recently confirmed in an independent calculation by j.j . kelly @xcite . despite the fact that rather large bound nucleon momenta were probed it was concluded that the deviations from standard ( woods - saxon ) mean - field wave functions are modest . this observation lends support for the picture that nucleon - nucleon correlations are related to the hard - core part of the nucleon - nucleon interaction and are not expected to bring about large deviations from mean - field models in single - knockout processes as long as low excitation energies in the a-1 system are probed . the data of ref . @xcite were obtained for an initial electron energy of 855.1 mev and @xmath154195 mev . as a large fraction of the data were taken in parallel kinematics ( @xmath155 ) , the momentum transfer had to be decreased as higher missing momenta were probed . consequently , to reach higher missing momenta one had to move out of quasi - elastic kinematics . figures [ fig : blom1 ] and [ fig : blom2 ] show the predicted sensitivity of the mainz results to the two - body currents . the figures show results for four different central values in @xmath2 and parallel kinematics . the kinematics is such that one moves out of quasi - elastic conditions as higher missing momenta are probed . as is commonly done , we present the angular cross - sections in terms of the reduced cross section `` @xmath156 '' . hereby , we adopt the standard convention of dividing the calculated cross sections by a kinematical factor times the so - called `` cc1 '' @xcite elementary electron - proton cross section @xmath157 referring to figures [ fig : blom1 ] and [ fig : blom2 ] , striking features are : ( 1 ) the sensitivity of the cross sections to the two - body currents is substantial and ( 2 ) the two - body currents do not dramatically alter the shape of the reduced cross section for none of the four @xmath2 values considered . as a consequence , the impact of the two - body currents would generally not be noticed when comparing results of calculations performed within the ia with data but would simply be `` effectively '' accounted for in the spectroscopic factor that is used to scale the dwia calculations to the data . the net effect of the two - body currents is a reduction of the cross sections . this reduction is of the order 10 - 20% for the lowest missing momenta range and can amount to a factor of two at @xmath158 mev / c . note that for the highest missing momenta considered in figures [ fig : blom1 ] and [ fig : blom2 ] inclusion of the two - body currents tends to increase the cross section . this is generally the case when higher missing momenta are probed @xcite . the observed reduction at low missing momenta that was ascribed to two - body current effects is not sufficient to explain why the data are substantially lower ( factor of two ) than expected on the basis of dwia models . indeed , the curves of figures [ fig : blom1 ] and [ fig : blom2 ] are obtained with a reduction factor of 0.25 for the ground - state transition and 0.30 for the @xmath50o(@xmath10)@xmath140n(3/2@xmath159,6.32 mev ) transition . with these values we obtain reasonable visual fits of the data . the transparency at the considered four - momentum transfers @xmath2 is experimentally determined to be about 0.75 for a light nucleus like @xmath120c @xcite . when correcting the reduction factors we have applied to obtain visual fits of the data , for transparency ( or attenuation ) corrections that fall beyond the scope of our model calculations , we obtain the spectroscopic factors @xmath160e@xmath161=0.0 mev)@xmath162 0.67 and @xmath163e@xmath161=6.3 mev)@xmath1621.6 . in line with the theoretical conclusions drawn from the analyses of references @xcite and @xcite these values are substantially smaller than the values obtained with high - resolution measurements at lower proton kinetic energies where it was found that @xmath164e@xmath161=0.0 mev)=1.20 and @xmath163e@xmath161=6.3 mev)=2.00 @xcite . note , however , that for @xmath165 knockout the spectroscopic factor that is deduced from the calculations that include the two - body currents is about 30% larger than the value that would be deduced in the ia . another striking feature of the results displayed in figures [ fig : blom1 ] and [ fig : blom2 ] is that the influence of the subnuclear degrees - of - freedom is more pronounced for knockout from the `` stretched '' ( @xmath166 ) orbit ( i.e. @xmath165 ) than from its `` jack - knifed '' spin - orbit partner ( @xmath167 ) ( i.e. @xmath168 ) . the major reason for this behaviour is simply that all the two - body current operators have a strong spin dependency . it is worth mentioning that the relativistic effects arising from the lower ( or negative energy ) components in the bound state wave functions were recently shown to have a similar sort of sensitivity . indeed , the ( relativistic ) @xmath5 results for knockout from the `` stretched '' orbits will be closer to their respective non - relativistic limits than those for `` jack - knifed '' orbits @xcite . accordingly , the predicted relativistic effects attributed to the small components in the bound state wave functions are smallest for these transitions with maximal two - body current contributions . the question arises whether the relatively strong sensitivity to two - body currents noticed in figures [ fig : blom1 ] and [ fig : blom2 ] is an intrinsic property of the specific @xmath2 range probed or is rather due to the fact that these results were acquired at smaller values of x and subsequently clearly out of quasi - elastic conditions . results for real quasi - elastic kinematics at about the same value for the proton s kinetic energy and initial electron energy are shown in figures [ fig : mommainz ] . these results are obtained in quasi - perpendicular kinematics . here , the variation in missing momentum is reached by varying the polar angle @xmath169 of the ejected proton . for the remainder of this paper all calculated ( reduced ) cross sections are shown for full sub - shell occupancy , i.e. @xmath170 . it becomes clear that under strict quasi - elastic conditions the effect of the subnuclear degrees - of - freedom is substantially smaller than at smaller values of bjorken x. it is worth mentioning that the results of figure [ fig : mommainz ] are obtained in more transverse kinematics than those of figures [ fig : blom1 ] and [ fig : blom2 ] . accordingly , with the eye on minimizing the effect of two - body currents it is very essential to move in quasi - elastic conditions _ even if this implies shifting to kinematics which is conceived to be transverse . _ another observation is that the extent to which to the two - body currents are important does not depend on the polar angle @xmath171 of the ejected nucleon . figures [ fig : strmainz ] and [ fig : polmainz ] show the structure functions and polarization observables for the same kinematics in which the cross sections of figure [ fig : mommainz ] are obtained . in presenting the structure functions we prefer not to divide out any kinematical variables from the different contributions to the total angular cross section . formally , this means that we write eq . ( [ eq : eep ] ) in the form @xmath172 where the precise definition of all contributing terms is obvious . this way of presenting results has the outspoken advantage that the relative contribution of each term in the cross section can be easily evaluated . furthermore , possible confusions regarding the adopted definitions for the different structure functions are avoided . in line with the observations made for @xmath173he and the deuteron @xcite case , the interference terms that are an order of magnitude smaller than @xmath174 and @xmath175 exhibit the strongest sensitivity to contributions that go beyond the ia . we now turn to the discussion of the recoil polarization observables . for coplanar kinematics the sole non - vanishing recoil polarization observables are @xmath176 and @xmath177 . the @xmath178 does vanish identically in the plane - wave impulse approximation . as such it allows to constrain the model assumptions with respect to the final - state interaction ( fsi ) . inspecting figure [ fig : polmainz ] it is clear that at low missing momenta the qualitative behaviour of the induced polarization @xmath178 turns out to be relatively insensitive to the current contributions beyond the impulse approximation . this result confirms the findings of the pavia calculations reported in ref . the first data set for the induced polarization in finite nuclei has recently become available @xcite . in figure [ fig : woo ] the calculated reduced cross section and recoil polarization observables are shown for the kinematics of this experiment . for knockout from the @xmath179 shell , where the nucleon is mainly located at high nuclear densities , the influence of the subnuclear d.o.f . is more pronounced than for @xmath180-shell knockout . given that we do not rely on any empirical input for the description of the fsi , the agreement with the data is satisfactory but inferior to the quality of agreement that was reached with the calculations by j.j . kelly @xcite . these calculations constrain the description of the final - state interactions with the aid of optical potentials derived from the analysis of hadronic induced reactions . for most observables , the meson - exchange and isobar effects are acting in opposite directions . at low missing momenta the induced polarization exhibits a small sensitivity to the two - body currents . at moderate and high missing momenta the @xmath178 is predicted to exhibit a large sensitivity to two - body currents . a similar observation is made for the polarization transfer observables . recently it was shown that the recoil proton polarizations for p(@xmath0 ) and d(@xmath0)n are almost identical @xcite . from the theory side , rather small contributions from the mec and ic in the double polarization observables were predicted in the deuteron @xcite and @xmath173he @xcite case in quasi - elastic kinematics . despite the fact that the polarization transfer variable @xmath181 and @xmath177 are related to the interference structure functions @xmath182 and @xmath183 their predicted sensitivity to the two - body currents looks relatively small . recently , the polarization transfer observables raised considerable interest in that they would provide a direct handle on the ( possible ) medium dependence of the nucleon form factors . indeed , in the plane - wave impulse approximation ( pwia ) it can be shown that for electroexcitation from a free proton @xmath184 it is of the utmost importance to investigate all possible mechanisms that could bring about changes in the above ratio without being related to medium modifications of the form factors . whereas it was recently shown that final - state interaction and gauge ambiguities effects are only marginally affecting the ratio @xmath185 @xcite the question arises whether meson - exchange and isobaric currents could bring about any change in the ratio of the double polarization observables . whereas the impact of the subnuclear d.o.f . on the value of polarization observables looks small in quasi - elastic kinematics and small missing momenta , the mec and ic are not necessarily unimportant , especially for such delicate issues like nucleon form factor studies where the medium dependency is predicted to be modest @xcite and the projected accuracy of the planned experiments is astounding . with the aim of studying the sensitivity of the ratio @xmath185 to meson - exchange currents and intermediate isobar creation we have calculated the following double ratio @xmath186 where `` @xmath187 '' refers to the calculated ratio in the impulse approximation , this is when retaining solely the one - body currents in the calculations but keeping all other ingredients fixed . as could be expected , the observed deviation from the ia result enhances as the missing momentum increases . the most favorable regime to perform the recoil polarization measurements for knockout from p - shell orbits , is probably in the peak of the cross section ( @xmath188100 mev / c ) . for the @xmath50o(@xmath10)@xmath140n(1/2@xmath159 ) transition , the deviation in the ratio @xmath185 that is attributed to the mec and ic is about 5 - 10% in the peak of the cross section . for knockout from its spin - orbit partner ( @xmath165 ) the calculated effect is 10 - 15% . in both cases , at 200 mev / c the modifications in the ratio due to two - body currents has grown to 20% and with increasing @xmath69 the effect appears to be rather out of control . we continue our investigations into the role of two - body currents by considering the kinematics of the tjnaf experiments e89 - 003 and e8 - 9033 @xcite . the e89 - 003 experiment has measured the separated structure functions and momentum distributions for @xmath50o(@xmath10 ) under quasi - elastic conditions at @xmath189=2.445 gev , @xmath118=445 mev and q=1 gev / c . in e89 - 033 on the other hand , the @xmath50o(@xmath190 ) polarization observables were measured for approximately the same kinematics . figures [ fig : mome89003]-[fig : sstate ] summarize the calculated results for the reduced cross sections , structure functions and recoil polarization observables for knockout from the two @xmath180-shell and the @xmath191-shell state . in comparing these results with those obtained in the previous subsection for quasi - elastic conditions at lower four - momentum transfer , we can study the @xmath2 dependency of the meson and isobar degrees of freedom . in analogy with the obvervations made at lower @xmath2 , the pion - exchange currents ( `` mec '' ) increase the cross section and this effect is completely counterbalanced by the @xmath9 current that produces the largest contributions . the @xmath2 dependency of the two - body current effects can maybe be better estimated from the individual structure functions . comparing figures [ fig : strmainz ] and [ fig : str3e89003 ] one does indeed observe a decreasing trend in the relative importance of the two - body currents . a similar tendency is observed for the recoil polarization observables ( figure [ fig : pol2a3 ] . the deviation in the ratio @xmath192 in the peak of the cross section is of the order of a few percent for the ground - state transition ( knockout from the @xmath168 orbit ) , whereas for the excitation of the @xmath193 hole state at 6.32 mev excitation energy in @xmath140n ( knockout from the stretched @xmath165 orbit ) the effect is close to 10% . with the aim of studying the medium - dependent effects one could be tempted to probe the region of highest nuclear density which for a nucleus like @xmath50o implies studying the region of @xmath194 knockout . in light of the mass - independence of the two - body current effects that is frequently alluded to , our findings for @xmath194-knockout in @xmath50o can to a certain extent serve as a guideline for the effects than could be expected in @xmath16he . from all orbits studied here , knockout from the @xmath194 orbit exhibits the strongest sensitivity to the two - body currents . after all , this observation is not that surprising given its stretched status and the fact that the region of highest nuclear density is probed . for example , the @xmath195 and @xmath196 interference responses in figure [ fig : str1e89003 ] grow substantially after including the two - body currents . indications for the strong sensitivity of the @xmath197 structure function to two - body currents was reported for the @xmath173he case in ref . @xcite . also for the recoil polarization observables in the missing - energy range of @xmath194 knockout ( figure [ fig : sstate ] ) , sizeable contributions from the meson - exchange and isobar currents are predicted . for the ratio @xmath185 the two - body currents bring about a reduction which is slightly bigger than 10% . this is qualitatively very similar with the result obtained for the `` stretched '' p - shell state ( @xmath165 ) ( figure [ fig : pol2a3 ] ) so that some general qualitative behaviour for the sensitivity of the @xmath185 ratio to multi - nucleon currents seems to emerge . indeed , at low missing momenta and quasi - elastic conditions the two - body currents reduce the ratio @xmath185 , an effect that slowly decreases as one goes to higher @xmath2 and is larger for knockout from `` stretched '' than from the `` jack - knifed '' single - particle orbits . finally we turn our attention to the role of the relativistic corrections ( section [ sec : current ] ) in the one - body current operator . the dotted lines in figures [ fig : str2e89003 ] , [ fig : str3e89003 ] and [ fig : str1e89003 ] show the calculated contributions to the cross section when neglecting the relativistic corrections in the one - body current operator , i.e. when adopting @xmath198 and @xmath199 in eq . ( [ eq : redcurrent ] ) . when comparing these curves with the solid ones one can estimate the effect of the relativistic corrections in the one - body current . for the transverse responses @xmath175 and @xmath200 the foldy - wouthuysen prescription does not lead to any relativistic correction into lowest order . whereas the spin - orbit term in the charge - density operator produces rather small corrections in the longitudinal response @xmath174 in the considered kinematics , the longitudinal - transverse response @xmath201 for the `` stretched '' single - particle states @xmath165 and @xmath194 is doubled after including the relativistic corrections in the charge - density operator . a strong sensitivity of the @xmath195 response to relativistic corrections was earlier found for @xmath202 @xcite . to our knowledge , beyond the giant resonance region no ( @xmath1 ) measurements have been made . nevertheless , neutron knockout investigations with moderate energy resolution could be done at mami , bates or tjnaf . results for neutron knockout in kinematic conditions that can be reached with electron beam energies below 1 gev are shown in figure [ fig : een ] . these results are obtained with a neutron formfactor @xmath203=0 . the @xmath6 calculations were done in the same kinematical conditions for the ( @xmath10 ) curves of figure [ fig : mommainz ] . the reduced cross sections for neutron knockout are subject to larger two - body current corrections than corresponding proton results . as a matter of fact , the absolute magnitude of the two - body current contributions is comparable for proton and neutron knockout . the fact that @xmath2042 and the absence of a substantial neutron charge , however , makes them to be relatively more important in the neutron knockout channel . as could be expected on the basis of the isospin structure of the underlying current operators , the qualitative behaviour is similar for proton and neutron knockout . the net reduction of the cross sections displayed in figure [ fig : een ] is of the order 10 - 20% . note that coupled - channel - effects between the proton knockout channels and the considerably weaker ( @xmath1 ) channels are expected to compensate largely for the loss of strength brought about by the meson and isobar degrees - of - freedom @xcite . our ( @xmath1 ) results are qualitatively similar to those reported in @xcite . the mec are observed to increase the cross section , an effect that is largely overcompensated by the isobaric current that works in the opposite direction . this reduction can be attributed to the fact that the one - body magnetization current and the isobaric current interfere destructively . in this section we report on calculations for an @xmath132pb(@xmath10 ) experiment that was done at nikhef . this experiment was performed in very transverse kinematics ( @xmath20596@xmath206 ) at an incident electron energy of 462.14 mev and near quasi - elastic conditions ( @xmath207=161 . mev , x=0.94 ) . as this corresponds with very `` transverse '' kinematics we consider this as a worst - case - scenario benchmark for investigating the sensitivity to two - currents that could be expected in quasi - elastic ( @xmath10 ) from heavy nuclei . this is particularly important in view of the fact that the @xmath132pb target is sometimes advocated as an appropriate surrogate for nuclear matter studies . in comparison with studies on light nuclei , for heavy nuclei there is a price to pay in that coulomb distortion effects and reduced nucleon transparencies make both the electromagnetic and the fsi part of the reaction process more difficult to handle and subject to likely enhanced theoretical uncertainties . here , we only want to study the _ relative contribution _ from the two - body currents that could be expected for a heavy nucleus like @xmath132pb . for computational reasons we have neglected the coulomb distortion effects . indeed , an exact treatment of these would imply that the two - body matrix elements of eq . ( [ eq : mattwo ] ) that involve a sum over all nucleons in the target nucleus , are to be calculated for a whole range of q values . whereas calculations of this type could be performed at lower momentum transfer @xcite , the amount of multipoles required at higher momentum transfer makes it an enormous computational task . figure [ fig : pb208 ] displays the reduced @xmath132pb(@xmath10 ) cross section for knockout from the @xmath208 and @xmath209 orbits . it becomes clear that the predicted two - nucleon effects are modest and of the same relative size in comparison with the findings for light nuclei . whereas the transparency for nucleon knockout has been established to decrease as the mass number increases @xcite the impact of the two - body currents on single - nucleon knockout appears to be rather mass independent . within the context of ongoing research activities in two - nucleon and pion production from nuclei a fair understanding of the two - nucleon currents in finite nuclei has been reached . with this knowledge the contribution of two - body currents to exclusive ( @xmath10 ) and ( @xmath1 ) reactions can be calculated with some confidence . reasonable estimates of their contribution is essential in view of the fact that subnuclear d.o.f . frequently represent unwanted background that could be at the origin of ambiguities in the interpretation of ( @xmath0 ) studies . most of our investigations were carried out in quasi - elastic kinematics and momentum transfers @xmath210 1 gev / c . the effect of two - body currents dramatically increases as one moves out of quasi - elastic conditions and/or higher missing momenta are probed . in quasi - elastic kinematics and the whole range of @xmath2 values studied here , the two - body currents decrease the @xmath5 differential cross sections when lower missing momenta are probed . as the shape does not seem to be substantially affected by the two - body currents it appears virtually impossible to acquire experimental evidence for the role of two - body current effects from plain @xmath5 cross - section measurements in quasi - elastic kinematics . at the same time , our findings imply that the spectroscopic factors as they are usually derived from ia calculations are subject to corrections that would make them bigger . these corrections are not unimportant and exhibit a @xmath2 dependence . even at four - momentum transfers of @xmath211 0.8 ( gev / c)@xmath3 the effect of the two - body currents on the spectroscopic factor can be as large as 20% . the role of the two - body currents can be made more explicit by measuring the interference response functions that , admittedly , represent a rather small part in the total angular cross section . the induced polarization @xmath178 exhibits a moderate sensitivity to the two - body currents at low missing momenta @xmath212 mev / c . therefore , it retains its status as a proper variable to constrain the final - state interaction effects . similar sort of sensitivities are found for the polarization transfer variables . in the light of exploiting these variables to increase our understanding of the ( possible ) medium dependency in the nucleon form factors , the effect of the meson and isobar degrees of freedom on the ratio @xmath185 can be boiled down to a few percent at higher values of @xmath2 . under typical mami kinematics , which implies four - momentum transfers of the order @xmath2130.50 ( gev / c)@xmath3 , the impact of subnuclear degrees of freedom on the @xmath185 ratio is predicted to be of the order of 10% . of all pion - related two - body currents studied here the @xmath9-isobar current has by far the strongest impact on the ( @xmath0 ) and ( @xmath1 ) observables . multi - coincidence experiments of the type ( @xmath214 ) and ( @xmath215 ) at higher four - momentum transfers @xmath2 would greatly help in further constraining the model assumptions with respect to the @xmath9-isobar current operator when reaching higher energy and momentum transfers . kelly , adv . * 23 * ( 1996 ) 75 . v. phandaripande , i. sick and p.k.a . de witt huberts , rev . mod * 69 * ( 1997 ) 981 . lu _ et al . _ , nucl . phys . * a634 * ( 1998 ) 443 . lu _ et al . _ , phys . lett . * b417 * ( 1998 ) 217 . laget , nucl . phys . * a579 * ( 1994 ) 333 . tjnaf experiment e89 - 003 `` measurement of recoil polarization in the @xmath50o(@xmath10 ) reaction with 2.4 gev electrons '' ( spokesperson c. glashausser ) . tjnaf experiment e93 - 043 `` polarization transfer in the reaction @xmath173he(@xmath10)@xmath216h in the quasi - elastic scattering region '' ( spokesperson j. van den brand ) tjnaf experiment e91 - 006 `` study of nuclear medium effects by recoil polarization up to high momentum transfers '' ( spokesperson a. saha ) j.j . kelly , phys . c * 56 * ( 1997 ) 2672 . kelly , nucl - th/980990 toshio suzuki , nucl . phys . * a495 * ( 1989 ) 581 . s. boffi , c. giusti , f.d . pacati and m. radici , nucl . phys . * a518 * ( 1990 ) 639 . s. boffi and m. radici , phys . b242 * ( 1990 ) 151 . s. boffi , m. radici , j.j . kelly and t.m . payerle , nucl . phys . * a539 * ( 1992 ) 597 . * 70 * ( 1993 ) 2868 . v. van der sluys , j. ryckebusch and m. waroquier , phys . c * 54 * ( 1996 ) 1322 . v. van der sluys , j. ryckebusch and m. waroquier , phys . c * 49 * ( 1994 ) 2695 . a.s . raskin and t.w donnelly , ann . of phys . * 191 * ( 1989 ) 78 . a. picklesimer and j.w . van orden , phys . c * 35 * ( 1987 ) 266 . s. boffi , c. giusti , f.d . pacati and m. radici , electromagnetic response of atomic nuclei , oxford studies in nuclear physics ( clarendon press , oxford , 1996 ) . d. potterfeld _ et al . _ , in _ proceedings of the conference on perspectives in hadronic physics _ , edited by s. boffi , c. ciofi degli atti and m. giannini , ( world scientific , singapore , 1998 ) , 164 . j. ryckebusch , m. waroquier , k. heyde , j. moreau and d. ryckbosch , nucl . phys . * a476 * ( 1988 ) 237 . c. mahaux and h. weidenmller , _ in _ shell model approach to nuclear reactions ( north - holland , amsterdam , 1969 ) . amaro , m.b . barbaro , j.a . caballero , t.w . donnelly and m. molinari , nucl . * a643 * ( 1998 ) 349 . fearing , g.i . poulis and s. scherer , nucl . phys . * a570 * ( 1994 ) 657 . s. boffi , f. capuzzi , p. demetriou and m. radici , nucl . phys . * a637 * ( 1998 ) 585 . s. jeschonnek and t.w . donnelly , phys . c * 57 * ( 1998 ) 2483 . yanhe jin and d.s . onley , phys . c * 50 * ( 1994 ) 377 . udas , p. sarriguren , e. moya de guerra , e. garrido and j.a . caballero , phys . c * 48 * ( 1993 ) 2731 . m. hedayati - poor , j.i . johansson and h.s . sheriff , phys . c * 51 * ( 1995 ) 2044 . mcdermott , phys . * 65 * ( 1990 ) 1991 . j. carlson and r. schiavilla , rev . * 70 * ( 1998 ) 743 . j. ryckebusch , v. van der sluys , k. heyde , h. holvoet , w. van nespen , m. waroquier and m. vanderhaeghen , nucl . phys . * a624 * ( 1997 ) 581 . marcucci , d.o . riska and r. schiavilla , phys . c * 58 * ( 1998 ) 3069 . stoler , phys . * 226 * ( 1993 ) 103 . frolov _ et al . * 82 * ( 1999 ) 45 . e. oset , h. toki and w. weise , phys . rep . * 83 * ( 1982 ) 281 . dekker , p.j . brussaard and j.a . tjon , phys . c * 49 * ( 1994 ) 2650 . t. wilbois , p. wilhelm , and h. arenhvel , phys . c * 54 * ( 1996 ) 3311 . j. ryckebusch , l. machenil , m. vanderhaeghen , v. van der sluys , and m. waroquier , phys . c * 54 * ( 1996 ) 3313 . b. krfgen , p. oltmanns , f. osterfeld and t. udagawa , phys . c * 55 * ( 1997 ) 1819 . c * 50 * ( 1994 ) 1637 . n. bianchi _ c * 54 * ( 1996 ) 1688 . j.k . koch and n. ohtsuka , * a435 * ( 1985 ) 765 and j. koch _ in _ moderns topics in electron scattering , eds . b. frois and i. sick , ( world scientific , singapore ) ( 1991 ) 28 . oconnell _ et al . c * 35 * ( 1987 ) 1063 . r. cenni , p. christillin and g. dillon , phys . lett . * b139 * ( 1984 ) 341 . b. krfgen , f. osterfeld and t. udagawa , phys . c * 50 * ( 1994 ) 1637 . c.r . chen and t .- s.h . lee , phys . c * 38 * ( 1988 ) 2187 . kondratyuk , m.i . krivoruchenko , n. bianchi , e. de sanctis and v. muccifora , nucl . phys . * a579 * ( 1994 ) 453 . e. oset and l.l . salcedo , nucl . phys . * a468 * ( 1987 ) 631 . et al . _ , * 80 * ( 1998 ) 245 . j. ryckebusch , m. vanderhaeghen , l. machenil and m. waroquier , nucl . phys . * a568 * ( 1994 ) 828 - 854 . g. rosner , _ in _ proc . conf . on perspectives in hadronic physics , ictp trieste , italy , may 12 - 16 , 1997 , eds . s. boffi , c. ciofi degli atti and m.m . gianninni ( world scientific , singapore ) 185 . p. bartsch _ _ , _ `` investigation of short - range nucleon - nucleon correlations using the reaction @xmath50o(e , e@xmath217pp)@xmath137c in super - parallel kinematics '' _ mami proposal a1/1 - 97 ( spokesperson g. rosner ) , 1997 . g. kaschl , g. mairle , h. mackh , d. hartwig and u. schwinn , nucl . phys . * a178 * ( 1971 ) 275 . * 81 * ( 1998 ) 2213 . c.c.gearhart , phd thesis , washington university ( st . louis , 1994 ) , unpublished and w. dickhoff , private communication . b421 * ( 1998 ) 71 . et al . _ , b344 * ( 1995 ) 85 . t. de forest , nucl . * a392 * ( 1983 ) 232 . * b366 * ( 1996 ) 44 . g. garino _ et al . c * 45 * ( 1992 ) 780 . m. leuschner _ et al . _ , c * 49 * ( 1994 ) 955 . caballero , t.w . donnelly , e. moya de guerra and j.m . udas , nucl . phys . * a643 * ( 1998 ) 189 . s. gilad , w. bertozzi and z .- l . zhou , nucl . a631 * ( 1998 ) 276c . j. mandeville _ et al . _ , * 72 * ( 1994 ) 3325 . et al . _ , * 80 * ( 1998 ) 456 . * 80 * ( 1998 ) 452 . kelly , adv . nucl . * 23 * ( 1996 ) 75 . e. hummel and j.a . tjon , phys . c * 49 * ( 1994 ) 21 . c * 49 * ( 1994 ) 1783 . j. ryckebusch , k. heyde , d. van neck and m. waroquier , nucl . phys . * a503 * ( 1989 ) 694 .

justification logic is a family of logics of justified beliefs . where epistemic modal logic treats formulas of the form @xmath1 with the intended meaning that an agent knows / believes @xmath2 , in justification logic we consider formulas of the form @xmath3 with the intended meaning that @xmath4 is a justification for @xmath2 - or that the agent has justification @xmath4 for @xmath2 . the first justification logic was lp , the logic of proofs , and appeared in @xcite by artemov , but it has since developed in a wide system of explicit epistemic logics with notable complexity properties that significantly differ from the corresponding modal logics : while every single - agent justification logic whose complexity has been studied has its derivability problem in @xmath5 ( the second level of the polynomial hierarchy ) , the corresponding modal logics have -complete derivability problems . furthermore certain significant fragments of these justification logics have an even lower complexity- , or even in some cases . for an overview of justification logic see @xcite . for an overview of complexity results of ( single - agent ) justification logic , see @xcite . in epistemic situations we often have multiple agents , so as it is the case with modal logic , there is a need for a multi - agent justification logic . in @xcite , yavorskaya presents two - agent variations of lp . these logics feature interactions between the two agents justifications : for @xmath6 , for instance , every justification for agent 1 can be converted to a justification for agent 2 for the same fact and we have the axiom @xmath7 , while @xmath8 comes with the extra axiom @xmath9 , so agent 2 is aware of agent 1 s justifications . in @xcite , we extended yavorskaya s logics to two - agent variations of other justification logics , as well as to combinations of two different justification logics . we then gave tableau procedures to prove that most of these logics were in the second step of the polynomial hierarchy , an expected result which mimics the ones for single - agent justification logics from @xcite . for some cases , however , we were able to prove -completeness , which was a new phenomenon for justification logic . in this paper we continue our work from @xcite . we provide a general family of multi - agent logics . each member of this family we call @xmath10 , where @xmath11 are parameters of the logic . for @xmath10 we consider @xmath12 agents and the interactions between the agents justifications are described by binary relations on agents , @xmath13 and @xmath14 . furthermore , not all agents are equally reliable : @xmath15 and @xmath16 are sets of agents , all agents in @xmath15 have consistent beliefs and all agents in @xmath16 have true beliefs . these concepts are made precise in section [ multidefinitions ] . it is our goal to provide a flexible system capable of modelling situations of many diverse agents , or diverse types of justifications , allowing for reasonably general interactions among their justifications . for this family of logics we provide semantics and a general tableau procedure and then we make observations on the complexity of the derivation problem for its members . in particular , we demonstrate that all logics in this family have their satisfiability problem in - under reasonable assumptions . this family demonstrates significant variety , as it also includes - and -complete members , while of course some of its members have their satisfiability problem in @xmath17 . this is a somewhat surprising result , as all single - agent justification logics whose complexity is known have their satisfiability problem in @xmath17 . this paper is organised as follows . in section [ multidefinitions ] , we give the base definitions of the syntax , axioms and semantics for each logic in the family . then we reintroduce the star calculus , an invaluable tool , and our first complexity results , mirroring the ones for single - agent justification logic ( see @xcite ) . the version of the star calculus we provide is somewhat different than the usual ones in that it is based on a given frame . if the frame includes a single world , we get the usual , more familiar version . in section [ tableaux ] we give general tableau rules for each of our logics . naturally , the rules are parameterized by the logic s parameters , including the interactions between the agents , so special attention is given in that section to these interactions . we then go on and further optimize the tableau procedure with respect to the number of world - prefixes it produces ; this results in a @xmath17 upper bound for the satisfiability of a general class of logics . in this section we present the system we study in this paper , its semantics and the basic tools we will need later on . most of the proofs for the claims here can be adjusted from the one- or two- agent versions of justification logic . the reader can see @xcite or @xcite for an overview of single - agent justification logic and @xcite for a two - agent version of this system . in this paper , if @xmath18 , @xmath19 $ ] will be the set @xmath20 . for every @xmath21 , the justification terms of the language @xmath22 will include all constants @xmath23 and variables @xmath24 and if @xmath25 and @xmath26 are terms , then the following are also terms : @xmath27 , [ t_1\cdot t_2 ] , ! t_1 $ ] . the set of terms will be referred to as @xmath28 . we also use a set @xmath29 of propositional variables , or sentence letters . these will usually be @xmath30 . formulas of the language @xmath22 include all propositional variables and if @xmath31 are formulas , @xmath32 $ ] and @xmath4 is a term , then the following are also formulas of @xmath22 : @xmath33 . the remaining propositional connectives , whenever needed , are treated as constructed from @xmath34 and @xmath35 in the usual way . the operators @xmath36 and @xmath37 are explained by the following axioms . intuitively , @xmath38 applies a justification for a statement @xmath39 to a justification for @xmath40 and gives a justification for @xmath41 . using @xmath42 we can combine two justifications and have a justification for anything that can be justified by any of the two initial terms - much like the concatenation of two proofs . finally , @xmath37 is a unary operator called the proof checker . given a justification @xmath4 for @xmath2 , @xmath43 justifies the fact that @xmath4 is a justification for @xmath2 . let @xmath18 , @xmath44 $ ] and @xmath45^{2}$ ] . the logic @xmath46 is the logic with modus ponens as a derivation rule and the following axioms : propositional axioms : : : finitely many schemes of classical propositional logic ; application : : : @xmath47 \colon_{i } \ ! } \psi)$ ] ; concatenation : : : @xmath48 \colon_{i } \ ! } \phi$ ] , @xmath49 \colon_{i } \ ! } \phi$ ] ; @xmath16-factivity : : : for every @xmath50 , @xmath51 ; @xmath15-consistency : : : for every @xmath52 , @xmath53 ; @xmath13-verification : : : for every @xmath54 , @xmath55 ; @xmath14-conversion : : : for every @xmath56 , @xmath57 , where in the above , @xmath2 and @xmath58 are formulas in @xmath22 , @xmath59 are terms and @xmath60 $ ] . @xmath16-factivity and @xmath15-consistency are the usual factivity and consistency axioms for every agent in @xmath16 and @xmath15 respectively . positive introspection is seen as a special case of @xmath13-verification - in this context , if agent @xmath61 has positive introspection , then @xmath62 . a constant specification for @xmath63 is any set @xmath64\}.\ ] ] we say that axiom @xmath40 is justified by a constant @xmath65 for agent @xmath61 , when @xmath66 . a constant specification is : _ axiomatically appropriate with respect to @xmath67 $ ] _ if for every @xmath68 , each axiom is justified by at least one constant , _ schematic _ if every constant justifies only a certain number of schemes from the ones above ( as a result , if @xmath65 justifies @xmath40 for @xmath61 and @xmath41 results from @xmath40 and substitution , then @xmath65 justifies @xmath41 for @xmath61 ) and _ schematically injective _ if it is schematic and every constant justifies at most one scheme . let @xmath69 be the smallest set such that @xmath70 and for every @xmath71 , it is the case that for every @xmath72 $ ] , @xmath73 . @xmath74 is @xmath75 , where @xmath76 just outputs all elements of @xmath69 . @xmath77 is consistent : just map each formula to the propositional formula that is the result of removing all terms from the original one ; then , all axioms are mapped to propositional tautologies and modus ponens preserves this mapping . we now introduce models for our logic . in the single - agent cases , m - models ( introduced in @xcite ) and f - models ( introduced in @xcite ) are used ( also in @xcite for two - agent logics ) and they are both important in the study of complexity issues . in this paper we are mostly interested in f - models , which we will usually just call models . these are essentially kripke models with an additional machinery to accommodate justification terms . let @xmath78 . then , an f - model @xmath79 for @xmath80 is a quadruple @xmath81 } , ( { { \mathcal{a}}}_i)_{i \in [ n]},{\mathcal{v}})$ ] , where @xmath82 is a set , for every @xmath83 $ ] , @xmath84 is a binary relation on @xmath85 , @xmath86 and for every @xmath83 $ ] , @xmath87 . @xmath85 is called the _ universe _ of @xmath79 and its elements are the worlds or states of the model . @xmath88 assigns a subset of @xmath85 to each propositional variable , @xmath89 , and @xmath90 assigns a subset of @xmath85 to each pair of a justification term and a formula . furthermore , @xmath91}$ ] will often be seen and referred to as @xmath92\times tm \times l_n { \longrightarrow}2^{w}$ ] and @xmath93 is called an admissible evidence function . additionally , @xmath93 must satisfy the following conditions : application closure : for any @xmath83 $ ] , formulas @xmath31 , and justification terms @xmath94 , + @xmath95 sum closure : for any @xmath83 $ ] , formula @xmath2 , and justification terms @xmath94 , + @xmath96 @xmath97-closure : for any formula @xmath71 , @xmath98 . @xmath13-verification closure : if @xmath54 , then @xmath99 @xmath14-conversion closure : if @xmath56 , then @xmath100 @xmath13-distribution : for any formula @xmath2 , justification term @xmath4 , @xmath54 and @xmath101 , if @xmath102 and @xmath103 , then @xmath104 . the accessibility relations , @xmath105 , must satisfy the following conditions : * if @xmath106 , then @xmath105 must be reflexive . * if @xmath107 , then @xmath105 must be serial ( @xmath108 ) . * if @xmath109 , then for any @xmath110 , if @xmath111 , we also have @xmath112 . has positive introspection ( i.e. @xmath113 ) , then @xmath105 is transitive . ] * for any @xmath56 , @xmath114 . truth in the model is defined in the following way , given a state @xmath115 : * @xmath116 and if @xmath89 is a propositional variable , then @xmath117 iff @xmath118 . * if @xmath31 are formulas , then @xmath119 if and only if @xmath120 , or @xmath121 . * if @xmath2 is a formula and @xmath4 a term , then @xmath122 if and only if @xmath103 and @xmath123 for all @xmath124 such that @xmath125 . a formula @xmath126 is called satisfiable if there are some @xmath127 ; we then say that @xmath79 satisfies @xmath2 in @xmath115 . if @xmath97 is axiomatically appropriate with respect to @xmath15 , then @xmath77 is sound and complete with respect to its models ; it is also sound and complete with respect to its models that have the _ strong evidence property : _ @xmath122 iff @xmath128 _ _ ; furthermore , if @xmath2 is satisfiable , then it is satisfied by a model @xmath79 of at most @xmath129 states - and in fact , it is satisfied by a model @xmath79 of at most @xmath129 states that has the strong evidence property ( see @xcite for proofs of all the above that can be easily adjusted for this general case ) . a pair @xmath130})$ ] as above is called a frame for @xmath77 . we present the @xmath131-calculi for @xmath77 . the @xmath131-calculi for the single - agent justification logics have proven to be an invaluable tool in the study of the complexity of these logics . this concept and results were adapted to the two - agent setting in @xcite and here we extend them to the general multi - agent setting . although the calculi have significant similarities to the ones of the single - agent justification logics , there are differences , notably that each calculus depends upon a frame and operates upon @xmath131-expressions ( defined below ) prefixed by states of the frame . a @xmath131-calculus was first introduced in @xcite , but its origins can be found in @xcite . if @xmath4 is a term , @xmath2 is a formula , and @xmath83 $ ] , then @xmath132 is a star - expression ( @xmath131-expression ) . given a frame @xmath133})$ ] and @xmath134 ^ 2 $ ] and constant specification @xmath97 , the @xmath135 calculus on the frame @xmath136 is a calculus on @xmath131-expressions prefixed by worlds from @xmath85 with the axioms and rules that are shown in figure [ fig : starcalc ] . + [ cols="^,^",options="header " , ] notice that the calculus rules correspond to the closure conditions of the admissible evidence functions . in fact and because of this , given a frame @xmath137 and a set @xmath138 of @xmath131-expressions prefixed by states of the frame , the function @xmath93 such that @xmath139 is an admissible evidence function and in fact it is the minimal admissible evidence function such that for every @xmath140 , @xmath141 in the sense that always @xmath142 for any other admissible evidence function @xmath143 such that for every @xmath140 , @xmath144 . therefore , given a frame @xmath137 and two set @xmath145 of @xmath131-expressions prefixed by states of the frame there is an admissible evidence function @xmath93 on @xmath146 such that for every @xmath147 , @xmath141 and for every @xmath148 , @xmath149 , if and only if there is no @xmath150 such that @xmath151 . this observation yields the following . [ prp : proofbystarcalc ] for any holds , but to prove it we need to introduce m - models , but we do not use m - models anywhere else . thus , the reader can see @xcite , or @xcite for a proof of proposition [ prp : proofbystarcalc ] for all constant specifications . ] constant specification @xmath97 , frame @xmath146 with universe @xmath85 and @xmath152 , @xmath153 . [ thm : calccompnew ] let @xmath154 be a schematic constant specification in @xmath155 and @xmath134 ^ 2 $ ] . then , the following problem is in @xmath156 : given a finite frame @xmath157})$ ] and a finite set @xmath138 of @xmath131-expressions prefixed by worlds from @xmath85 , a formula @xmath2 , a term @xmath4 , a @xmath158 and @xmath32 $ ] , is it the case that @xmath159 the proof of this proposition is very similar to the one that can be found in @xcite . what is different here is the additional assignment of a state set to each node of the derivation tree , which does not change things a lot . for this proof and every @xmath160 $ ] , let @xmath161 , s.t . for every @xmath162 , @xmath163 . * nondeterministically construct a rooted tree with pairs of the form @xmath164 , where @xmath160 $ ] and @xmath165 is a subterm of @xmath4 , as nodes , such that @xmath166 is the root and the following conditions are met . node @xmath164 can be the parent of @xmath167 or of both @xmath167 and @xmath168 as long as there is a rule @xmath169 or @xmath170 , respectively , of the @xmath131-calculus and as long as @xmath171 . to keep this structure a tree , we can ensure at this step that there are no cycles , which would correspond to consecutive applications of @xmath131c@xmath172 and which would be redundant . * nondeterministically assign to each leaf , @xmath173 , either * * some formula @xmath58 and the closure under @xmath174 of some set @xmath175 , s.t . for every @xmath176 , @xmath177 or , * * as long as @xmath178 is of the form @xmath179 , where @xmath65 a constant , @xmath180 , then we can also assign some @xmath181 and @xmath182 , where @xmath40 an axiom scheme , s.t . @xmath183 . * if for some node @xmath184 all its children , say @xmath185 have been assigned some scheme or formula @xmath186 and world sets @xmath187 , assign to @xmath188 some scheme or formula @xmath189 , such that @xmath190 can be unified to @xmath191 such that @xmath192 is a rule in the @xmath193-calculus and world set @xmath13 , where @xmath13 is the closure of @xmath194 under @xmath174 . apply this step until the root of the tree has been assigned some scheme or formula and a @xmath195 subset of @xmath85 . * unify @xmath2 with the formula assigned to @xmath166 and verify that @xmath196 if some step is impossible , the algorithm rejects . otherwise , it accepts . using efficient representations of schemes using dags and robinson s unification algorithm , the algorithm runs in polynomial time . we can see that as the tree is constructed , if @xmath197 is assigned scheme @xmath189 and set @xmath13 , then the construction effectively describes a valid derivation of any expression of the form @xmath198 , where @xmath199 and @xmath58 an instance of @xmath189 . therefore , if the algorithm accepts , there exists a valid @xmath200-calculus derivation of @xmath201 . on the other hand if there is some @xmath200-calculus derivation for @xmath202 from @xmath138 , then the algorithm in the first two steps can essentially describe this derivation by producing the derivation tree and the formulas / schemes by which the derivation starts . therefore , the algorithm accepts if and only if there is a @xmath200-calculus derivation for @xmath202 from @xmath138 . see @xcite and @xcite for a more detailed analysis . the number of nondeterministic choices made by the algorithm in the proof of proposition [ thm : calccompnew ] is bounded by @xmath203 , where @xmath204 . therefore , if there is some formula @xmath58 such that @xmath205 is a subformula of @xmath58 and for every @xmath206 , @xmath207 is a subformula of @xmath58 , then @xmath208 and therefore we can simulate all nondeterministic choices in time @xmath209 . thus the algorithm can be turned into a deterministic one running in time @xmath210 . this observation , the fact that a satisfiable @xmath2 can be satisfied by a model of at most @xmath129 states ( see the previous subsection ) and the previous two propositions give the following results : let @xmath211 , where @xmath212 is schematic . then , 1 . deciding for @xmath205 that @xmath213 is in . 2 . if @xmath97 is axiomatically appropriate with respect to @xmath15 , then the satisfiability problem for @xmath80 is in . additionally notice that if the term @xmath4 has no @xmath42 , @xmath97 is schematically injective and @xmath214 , we have essentially eliminated nondeterministic choices from the procedure above . thus , we conclude ( for the original result , see @xcite ) : let @xmath211 , where @xmath212 is schematically injective . then , deciding for @xmath205 , where @xmath4 has no ` @xmath42 ' , that @xmath213 is in . in this section we give a general tableau procedure for every logic which varies according to each logic s parameters . we can then use the tableau for a particular logic and make observations on its complexity , as we do in the following section . a version of the tableau which is more efficient for some cases follows after . to develop the tableau procedure we need to examine the relations on the agents more carefully than we have so far . for this section and the following one fix some @xmath215 and we assume @xmath97 is axiomatically appropriate with respect to @xmath15 and schematic . if @xmath216 are binary relations on the same set , then @xmath217 is the binary relation on the same set , such that @xmath218 if and only if there are @xmath219 in the set , such that @xmath220 . if @xmath40 is a binary relation , then @xmath221 is the reflexive , transitive closure of @xmath40 ; if @xmath40 is a set ( but not a set of pairs ) , then @xmath221 is the set of strings from @xmath40 . we also use the following relation on strings : @xmath222 iff there is some string @xmath65 such that @xmath223 . we define the following subsets of and relations on @xmath19 $ ] . @xmath224 : : @xmath225 | ( \exists j \in d \cup f ) \ i c^ * j \}$ ] attempts to capture exactly those agents that require a serial accessibility relation ; from now and on these agents monopolize our attention ; @xmath226 : : @xmath225 | ( \exists j \in f ) \ i c^ * j \}$ ] attempts to capture exactly those agents that require a reflexive accessibility relation ; @xmath227 : : @xmath228 notice that if @xmath229 , and @xmath230 , then @xmath231 ; in @xmath232 is to indicate that @xmath232 is a variation of @xmath14 influenced by the agents in @xmath16 , not that it is available with other subscripts ] @xmath233 : : @xmath234 : if @xmath235 , then @xmath61 s justifications somehow affect @xmath236 s justifications ; conversely , @xmath236 s accesibility relations somehow affect @xmath61 s accessibility relations ; @xmath237 : : @xmath238 : @xmath239 if and only if @xmath240 and @xmath241 ; we can easily see that @xmath242 is an equivalence relation and that if @xmath243 , then @xmath61 and @xmath236 have the same accessibility relations ; similarly , for the above equivalence relations we can define equivalence classes on @xmath138 , @xmath244 . @xmath245 is the equivalence class @xmath246 s.t . we can define relations @xmath248 and @xmath249 on @xmath250 in the following way : @xmath251 iff @xmath252 s.t . @xmath253 and @xmath254 iff @xmath252 s.t . @xmath255 . also , @xmath256 iff @xmath257 and @xmath258 and similarly for @xmath259 . then , define @xmath260 iff there are @xmath261 and @xmath262 s.t . @xmath263 , that is , there are @xmath264 , where @xmath265 . @xmath266 iff @xmath260 and @xmath267 . let @xmath32 $ ] . then , @xmath268 and @xmath269 s.t . @xmath270 . the formulas used in the tableau will have the form @xmath271 , where @xmath272 or is a @xmath131-expression , @xmath273 ( the world prefixes are strings of equivalence classes of agents ) , @xmath274 is ( either the empty string or ) of the form @xmath275 , @xmath276 $ ] , and @xmath277 . furthermore , @xmath278 will be called a world - prefix or state - prefix , @xmath165 a truth - prefix and world prefixes will be denoted as @xmath279 , instead of @xmath280 , where for all @xmath281 $ ] , @xmath282 . a tableau branch is a set of formulas of the form @xmath283 , as above . a branch is complete if it is closed under the tableau rules ( they follow ) . it is propositionally closed if @xmath284 and @xmath285 are both in the branch . we say that a tableau branch is constructed by the tableau rules from @xmath2 , if it is a closure of @xmath286 under the rules . for every @xmath287 , @xmath288 if there are some @xmath289 such that @xmath290 , @xmath291 and @xmath292 . the tableau rules will include certain classical rules to cover propositional cases of formulas , as well as the ones that follow : @xmath293{\sigma\ t\ { t \colon_{i } \!}\psi}{\sigma\ t\ * _ i(t,\psi ) \\\\ \sigma\ t\ \box_i\psi } \ ] ] if @xmath294 ; @xmath295{\sigma\ t\ { t \colon_{i } \!}\psi } { \sigma.\chi(j)\ f\ \bot } \ ] ] if @xmath294 , @xmath296 , and @xmath297 ; @xmath298{\sigma\ t\ { t \colon_{i } \!}\psi}{\sigma\ t\ * _ i(t,\psi)}\ ] ] if @xmath299 ; @xmath300{\sigma\ f\ { t \colon_{i } \!}\psi}{\sigma\ f\ * _ i(t,\psi)}\ ] ] @xmath301{\sigma.\chi(j ) \ f \ \bot } { \sigma.\chi(j).\chi(i ) \ f \ \bot}\ ] ] if @xmath288 ; @xmath302{\sigma\ t\ \box_{i}\psi}{\sigma.\chi(i)\ t\ \psi}\ ] ] if @xmath303 has already appeared ; @xmath304{\sigma\ t\ \box_{i}\psi}{\sigma\ t\ \psi}\ ] ] if @xmath106 ; @xmath305{\sigma\ t\ \box_{i}\psi}{\sigma\ t\ \box_{j}\psi}\ ] ] if @xmath306 ; @xmath307{\sigma\ t\ \box_{i}\psi}{\sigma\ t\ \box_{j}\box_i\psi}\ ] ] if @xmath308 . we do not explicitly mention it anywhere else , but of course , we need a set of rules to cover propositional cases as well . in particular we can use + @xmath309 and @xmath310 the separator @xmath311 indicates a nondeterministic choice between the two options it separates . if @xmath312 is a tableau branch , then here and wherever else it may appear is the set @xmath313 . ] @xmath314 let @xmath315}$ ] be such that for every @xmath32 $ ] , @xmath316 then @xmath317})$ ] , where @xmath318}$ ] is the closure of @xmath319}$ ] under the conditions of frames for the accessibility relations , except for seriality . @xmath318}$ ] is constructed in the following way : for every @xmath32 $ ] , let @xmath320 and for every @xmath321 , @xmath322 and then , @xmath323})$ ] . finally , let @xmath324 and @xmath325 . a branch @xmath312 of the tableau is rejecting when it is propositionally closed or there is some @xmath326 such that @xmath327 . otherwise it is an accepting branch . by induction on the construction of @xmath328 , it is not hard to see that for every @xmath329 , it must be the case that @xmath240 or that @xmath330 and @xmath331 . by induction on the frame construction we can see that if @xmath332 appears in @xmath312 and @xmath333 , then @xmath334 appears in @xmath312 . [ prp : tableaucomplete ] if there is a complete accepting tableau branch @xmath335 , then the formula @xmath2 is satisfiable by a model for @xmath336 . let @xmath337 } , { { \mathcal{a } } } , { \mathcal{v}})$ ] , where @xmath130 } ) = { \mathcal{f}}(b)$ ] , @xmath338 , and @xmath339 . let @xmath340},({{\mathcal{a}}}_i)_{i\in[n]},{\mathcal{v}})$ ] , where for every @xmath32 $ ] , if @xmath294 , then @xmath341 s.t . @xmath342 and @xmath343 , otherwise . @xmath344 is an f - model for @xmath336 : @xmath345}$ ] easily satisfy the appropriate conditions , as the extra pairs of the accessibility relations do not affect the @xmath131-calculus derivation , and we can prove the same for @xmath346}$ ] . if @xmath347 and @xmath348 , if @xmath349 , then @xmath350 and thus @xmath351 . if @xmath352 , then , from rule s , there must be some @xmath353 , so @xmath354 and thus , @xmath355 . if @xmath356 and @xmath357 , then , trivially , whether @xmath352 or not , @xmath358 . by induction on @xmath359 , we prove that for every formula @xmath359 and @xmath360 , if @xmath361 , then @xmath362 and if @xmath363 , then @xmath364 . propositional cases are easy . if @xmath365 and @xmath363 , then @xmath366 , so @xmath367 . on the other hand , if @xmath368 , then @xmath369 and by rule trd , for every @xmath370 such that @xmath240 , there is some @xmath371 . therefore , for every @xmath372 , it is the case that @xmath373 , so by rule trb , a previous observation about formulas of the form @xmath374 , and the inductive hypothesis , for every @xmath375 , @xmath376 and therefore , @xmath377 . now we can prove the following proposition . [ prp : tableaux ] let @xmath126 . @xmath2 is @xmath77-satisfiable if and only if there is a complete tableau branch @xmath312 that is produced from @xmath378 , such that * for all @xmath379 , not both @xmath380 and @xmath381 appear in @xmath312 and * for any @xmath382 , @xmath383 . the `` if '' direction was handled by proposition [ prp : tableaucomplete ] . we will prove the `` only if '' in the following . let @xmath79 be an f - model , @xmath384},({{\mathcal{a}}}_i)_{i\in [ n]},{\mathcal{v}})$ ] that has the strong evidence property and a state @xmath385 such that @xmath386 . furthermore , fix some @xmath387 , such that @xmath388 and the following conditions are met . for any @xmath389 , @xmath390 is some element of @xmath85 s.t . @xmath391 . let @xmath392 \}$ ] . given a state @xmath115 of the model , and @xmath393 , @xmath394 has the usual , modal interpretation , @xmath394 iff for every @xmath352 , @xmath395 we can see in a straightforward way and by induction on the tableau derivation that there is a branch , such that if @xmath396 appears in the branch and @xmath397 , then @xmath398 , if @xmath399 appears in the branch and @xmath397 , then @xmath400 , if @xmath401 appears in the branch , then @xmath402 and if @xmath403 appears in the branch , then @xmath404 . the proposition follows . by taking a closer look at the interactions between the agents we can further improve the efficiency of our tableau procedure and we do that in this section . we use this improvement to prove an upper bound on the complexity of a general class of logics . we need the following definitions and lemma [ lem : clusters ] , which is a generalization of a result from @xcite and has appeared in simpler forms in @xcite . first we define an additional equivalence relation : @xmath405 . as an equivalence relation , this one too gives equivalence classes on @xmath138 and they are @xmath406 . notice that as @xmath407 , @xmath408 and therefore @xmath250 is a refinement of @xmath409 . we ll call @xmath410 . furthermore , notice that for any @xmath411 , either @xmath412 s.t . @xmath413 , or @xmath246 . in the first case , @xmath414 will be called a _ v - class _ of agents and in the second case it will be called a _ c - class _ of agents . for each agent @xmath32 $ ] , @xmath415 will be the equivalence class @xmath411 s.t . @xmath247 and to help keep the relationship between the sets of equivalence classes in mind , notice that @xmath416 $ ] . furthermore , we can extend relations @xmath248 and @xmath249 on @xmath189 in the same way they were defined on @xmath250 . [ lem : clusters ] let @xmath337 } , ( { { \mathcal{a}}}_i)_{i\in [ n ] } , { \mathcal{v}})$ ] be a @xmath77 f - model of at most @xmath129 states , @xmath417 a v - class of agents , and @xmath418 . then , there are states of @xmath85 , @xmath419 , such that 1 . for any @xmath420 , @xmath421 . 2 . for any @xmath422 , @xmath423 , if @xmath424 , then @xmath425 . @xmath419 will be called _ a @xmath426-cluster for @xmath427_. for this proof we need to define the following . let @xmath32 $ ] , @xmath428 . _ an @xmath429-path ending at @xmath61 _ ( and starting at @xmath430 ) from @xmath431 to @xmath432 is a finite sequence @xmath433 , such that for some @xmath434 $ ] , @xmath435 , where for some @xmath436 $ ] @xmath437 and @xmath438 ( and @xmath439 ) , for every @xmath440 $ ] , @xmath441 and if @xmath442 , then @xmath443 , while if @xmath444 , then @xmath445 and @xmath446 , @xmath447 , @xmath448 . the @xmath429-path _ covers _ a set @xmath449 $ ] if @xmath450 . for this path and @xmath451 $ ] , @xmath452 is a @xmath453-state . notice that if there is an @xmath429 path ending at @xmath61 from @xmath431 to @xmath432 and some @xmath454 and @xmath455 such that the path covers @xmath165 and @xmath456 , it must also be the case that @xmath457 . let @xmath458 { \longrightarrow}p_a$ ] be such that @xmath459 , @xmath460 = p_a$ ] and for every @xmath461 $ ] , either @xmath462 or @xmath463 and there is some @xmath461 $ ] such that @xmath463 . for any @xmath385 , @xmath464 let @xmath465 be the following : @xmath466 , for all @xmath467 $ ] , @xmath468 will be such that there is an @xmath429 path ending at @xmath469 from @xmath165 to @xmath468 and covering @xmath426 and if @xmath470 , @xmath471 is such that @xmath472 is an @xmath429 path ending at @xmath473 . let @xmath474 , x\in { \ensuremath{\mathbb{n}}}},(a^x_i)_{i \in [ m ] , x\in { \ensuremath{\mathbb{n}}}}$ ] be defined in the following way . for every @xmath475 $ ] , @xmath476 and for every @xmath464 , @xmath477 . finally , for @xmath478 , @xmath474}$ ] is defined in the following way . if there are some @xmath479 , @xmath422 , such that @xmath480 , @xmath481 and not @xmath482 , then for all @xmath420 , @xmath483 . otherwise , @xmath484 . by induction on @xmath485 , we can see that for every @xmath486 , @xmath420 , if @xmath487 , then @xmath488 . since the model has a finite number of states , there is some @xmath464 such that for every @xmath489 , @xmath490 . therefore , we can pick appropriate @xmath419 among @xmath491 that satisfy conditions 1 , 2 . we recursively define relation @xmath34 on @xmath492 : * if @xmath493 then @xmath494 ; * if @xmath348 , then @xmath495 ; * if @xmath496 , then @xmath497 . @xmath498 is the reflexive , transitive closure of @xmath34 . @xmath498 tries to capture the closure of the conditions on the accessibility relations of a frame . this is made explicit by observing that if for some frame @xmath130})$ ] , @xmath499 and @xmath500 , then @xmath501 . furthermore , if , in addition , @xmath502 , then for every @xmath503 $ ] , @xmath504 . for every agent @xmath294 , we introduce a new agent , @xmath505 and we extend @xmath498 , so that when @xmath415 a v - class , then @xmath506 for every @xmath507 such that @xmath508 . for each @xmath246 , we fix some @xmath509 and @xmath510 . furthermore , if @xmath511 , then @xmath512 @xmath513 . this extended definition of @xmath498 tries to capture the closure of the conditions on the accessibility relations of a frame like the ones that will result from a tableau procedure as defined in the following . let @xmath411 and @xmath514 be a finite string of elements from @xmath250 . then , @xmath414 is _ visible _ from @xmath278 if and only if there is some @xmath515 , some @xmath516 and some @xmath517 such that @xmath518 and @xmath519 ; @xmath520 is then called the @xmath414-view from @xmath514 . notice that there is a similarity between this definition and the statement of lemma [ lem : clusters ] - this will be made explicit later on . then we adjust the tableau by altering rules trd and s and introduce rule svb : @xmath295{\sigma\ t\ { t \colon_{i } \!}\psi } { \sigma.\chi(j)\ f\ \bot } \ ] ] if @xmath294 , @xmath296 , @xmath297 , and @xmath521 is not a @xmath13-class visible from @xmath514 ; @xmath301{\sigma.\chi(j ) \ f \ \bot } { \sigma.\chi(j).\chi(i ) \ f \ \bot}\ ] ] if @xmath288 and it is not the case that @xmath415 is a @xmath13-class visible from @xmath522 ; @xmath523{\sigma\ t\ \box_{i}\psi}{\tau.\chi(i)\ t\ \psi}\ ] ] if @xmath524 a @xmath13-class , visible from @xmath514 , @xmath525 is the @xmath415-view from @xmath514 and @xmath520 has already appeared in the tableau . then we have to redefine the frame @xmath328 . let @xmath315}$ ] be such that for every @xmath32 $ ] , @xmath316 @xmath526 @xmath527 and @xmath317})$ ] , where @xmath318}$ ] is the closure of @xmath319}$ ] as it was defined before . proposition [ prp : tableaucomplete ] and its proof remain the same . to prove proposition [ prp : tableaux ] for this version of the rules , follow the same proof , but for every @xmath360 and v - class @xmath414 fix some @xmath414-cluster for @xmath115 and if @xmath415 is a @xmath13-class , then for @xmath528 , the fixed @xmath414-cluster for @xmath529 , @xmath530 - and observe that if @xmath415 a @xmath13-class , visible from @xmath514 and @xmath525 is the @xmath415-view from @xmath514 , then in model @xmath79 there is some @xmath432 such that @xmath531 , which by the definition of clusters in turn means that @xmath532 . the remaining proof is the same . notice that if for every appearing world - prefix @xmath303 , @xmath61 is always in the same v - class @xmath414 , then all prefixes are of the form @xmath533 , where @xmath534 . in that case we can simplify the box rules and in particular just ignore rule v and end up with the following result . [ cor : easycasesinpi2 ] when there is some v - class @xmath414 such that for every @xmath535 there is some @xmath536 such that @xmath537 , then @xmath77-satisfiability is in @xmath17 . in this section we look into some more specific cases of multi - agent justification logics and demonstrate certain jumps in the complexity of the satisfiability problem for these logics . we first revisit the two - agent logics from @xcite . like in the previous sections , we assume our constant specifications are schematic and axiomatically appropriate ( and in for upper bounds ) . our definition here of @xmath77 allows for more two - agent logics than the ones that were studied in @xcite . it is not hard , though , to extend those results to all two - agent cases of @xmath77 : when there are @xmath538 $ ] , @xmath539 , @xmath540 , @xmath541 , and @xmath542 , then @xmath543-satisfiability is -complete ; otherwise it is in @xmath17 ( see @xcite ) . we will further examine the following two cases . @xmath544 are defined in the following way . @xmath545 ; @xmath546 ; @xmath547 ; @xmath548 , @xmath549 ; @xmath550 ; finally , for @xmath551 $ ] , @xmath552 , where @xmath553 is some axiomatically appropriate and schematic constant specification . by an adjustment of the reductions in @xcite , as it was done in @xcite , it is not hard to prove that @xmath554 is -hard and @xmath555 is -hard . would correspond to what is defined in @xcite as @xmath556 and @xmath557 to @xmath558 . then we can pick a justification variable @xmath485 and we can either use the same reductions and substitute @xmath559 by @xmath560 , or we can just translate each diamond - free fragment to the corresponding justification logic in the same way . it is not hard to see then that the original modal formula behaves exactly the same way as the result of its translation with respect to satisfiability - just consider f - models where always @xmath98 . ] notice that the way we prove @xmath561-hardness for @xmath562 is different in character from the way we prove the same result for the two - agent logics in @xcite . for @xmath562 we use the way the tableau prefixes for it branch , while for @xmath563 the prefixes do not branch , but they increase to exponential size . in fact , we can see that @xmath562 is -complete , while @xmath557 is -complete . the respective tableau rules as they turn out for each logic are - notice that neither logic has any @xmath564-maximal v - classes : notice that the maximum length of a world prefix is at most @xmath566 , since the depth ( nesting of terms ) of the formulas decrease whenever we move from @xmath514 to @xmath565 . also notice that when we run the @xmath131-calculus , there is no use for rule @xmath131v - dis , so we can simply run the calculus on one world - prefix at the time , without needing the whole frame . therefore , we can turn the tableau into an alternating polynomial time procedure , which uses a nondeterministic choice when the tableau would make a nondeterministic choice ( when we apply the propositional rules ) and uses a universal choice to choose whether to increase prefix @xmath514 to @xmath567 or to @xmath568 . this means that @xmath562-satisfiability is -complete . for the tableau procedure of @xmath557 we have no such bound on the size of the largest world - prefix , so we can not have an alternating polynomial time procedure . as before , though , the @xmath131-calculus does not use rule @xmath131v - dis , so again we can run the calculus on one world - prefix at the time . furthermore , for every prefix @xmath431 , @xmath569 is polynomially bounded ( observe that we do not need more than two boxes in front of any formula ) , so in turn we have an alternating polynomial space procedure . therefore , @xmath557-satisfiability is -complete . antonis achilleos . a complexity question in justification logic . in _ proceedings of the 18th international conference on logic , language , information and computation _ , wollic11 , pages 819 , berlin , heidelberg , 2011 . springer - verlag . roman kuznets . on the complexity of explicit modal logics . in peter clote and helmut schwichtenberg , editors , _ csl _ , volume 1862 of _ lecture notes in computer science _ , pages 371383 . springer , 2000 . errata concerning the explicit counterparts of @xmath570 and @xmath571 are published as @xcite . roman kuznets . complexity through tableaux in justification logic . in _ plenary talks , tutorials , special sessions , contributed talks of logic colloquium ( lc 2008 ) , bern , switzerland _ , pages 3839 , 2008 . roman kuznets . self - referentiality of justified knowledge . in edward a. hirsch , alexander a. razborov , alexei l. semenov , and anatol slissenko , editors , _ csr _ , volume 5010 of _ lecture notes in computer science _ , pages 228239 . springer , 2008 . alexey mkrtychev . models for the logic of proofs . in sergei adian and anil nerode , editors , _ logical foundations of computer science , 4th international symposium , lfcs97 , yaroslavl , russia , july 612 , 1997 , proceedings _ , volume 1234 of _ lecture notes in computer science _ , pages 266275 . springer , 1997 .

single molecule biomechanical experiments allow to measure the forces generated by biomolecules and their response to applied forces @xcite . for `` passive '' biopolymers , that do not produce active movement against load , the resulting force - extension curves provide a probe of the nonlinear elasticity of the chains and its relationship to configurational changes . such studies were carried out for single stranded and double stranded dna @xcite , for the muscle protein titin @xcite , the extracellular matrix protein tenascin @xcite , the polysaccharides dextran @xcite and xanthan @xcite , as well as the synthetic polymer poly(ethylene - glycol ) @xcite . theoretical modeling of the elastic behavior of these polymers is often hampered by lack of detailed knowledge concerning the configurations involved and the relevant interactions . a direct and detailed confrontation between experiment and theory is however possible for the case of homopolypeptides capable of forming an @xmath4-helix . in this case the molecular configurations are well understood . furthermore , one can relate the elastic force law to the known zimm - bragg parameters that characterize the helix - coil transition of the free polypeptides . as we shall discuss , the extension force law of a long polypeptide just above the melting temperature of the helix , @xmath0 , exhibits two plateaus . both are associated with a coexistence of helical and coil domains . the low - tension plateau is due to helix formation induced by the chain extension and the accompanying loss of configurational entropy . the second , high - tension plateau is due to the extension induced melting of the helices when the end - to - end length exceeds the length of the fully helical chain . for @xmath5 only the second plateau survives but the force law for weak extensions exhibits a `` quasi plateau '' _ i.e. , _ a regime with a weak slope that is not associated with a coexistence . this last regime is due to the facile alignment of the long persistent regions in the helix . in the following we analyze the force laws , their dependence on temperature and the corresponding population distribution of helical and coil segments . while we are unaware of single molecule experiments on this system , similar behavior was observed experimentally on fibers of collagen during the 1950s @xcite . two different treatments of this problem were recently advanced @xcite . in the present article we present a unified analysis of the problem , tracing the physical origins of the disagreements between the force laws obtained in the two earlier publications and analyzing their range of validity . our treatment is distinctive in that it is based on free energy argument instead of the customary transfer matrix approach @xcite . this free energy argument allows to recover the results of the transfer matrix method but is physically transparent and relatively simple mathematically . within this approach we consider the extension of a very long homopolypeptide capable of forming @xmath4-helices . the discussion is limited to quasi - static extension assuming that the rate of equilibration of the configurations of the chain is much faster than the rate of extension . two models are explored . in one , the helical domains are viewed as rigid rods while in the second , they are modeled as semiflexible chains . for each model we present a rigorous derivation of the corresponding force law allowing for the polydispersity of the helical and coil domains and the associated mixing entropy ( see fig . [ multi ] ) . in addition , we explore two useful approximations allowing for simplified calculations of the force law . in one approximation , the mixing entropy of the helices and domains is neglected and the plateau is associated with coexistence of one helical domain with one coil domain ( see fig . [ phases ] ) . within this `` diblock '' or `` @xmath6 '' approximation the coexistence regimes involve a first order phase transition . in spite of this erroneous conclusion , this approximation correctly identifies the important length and force scales in the problem . the second , mean field approximation , neglects the direct coupling of the distribution of domains sizes and the applied tension . as we shall see , this approximation is actually exact when the helical domains are modeled as long semiflexible chains . the problem and the analysis are of interest from a number of perspectives . homopolypeptides are the simplest representatives of helicogenic molecules . in this case it is not necessary to allow for the sequence heterogeneity which affects the elasticity of heteropolypeptides . nor is it necessary to consider the long range interaction that occur in multiple helices . as a result , the number of parameters invoked in the theory is smaller and the mathematical analysis is simpler . the force laws obtained are determined by known zimm - bragg parameters with no adjustable parameters . accordingly this system enables a direct comparison between experiment and theory . in this juncture it is important to note the availability of synthetic methods for producing long polypeptides with a well defined architecture @xcite . furthermore , the formalism described provides a convenient basis for the analysis of more complex helicogenic polymers such as dna and collagen . from a polymer science point of view the elasticity of homopolypeptides is of interest because it reflects the effects of internal degrees of freedom arising from intrachain self - assembly . the study of these systems enables the exploration of the distinctive non - linear force laws that are the signatures of such intrachain self - assembly . finally , this elasticity plays a role in the stabilization of the helical state of grafted polypeptides @xcite . in turn this is of importance for the function of fusogenic polypeptides and the design of surfaces that favor spreading and growth of cells @xcite . the paper is organized as follows . in section ii we introduce the free energy argument _ via _ an analysis of the helix - coil transition in free , undeformed homopolypeptides . a semiquantitative analysis of the effect of extension is presented in section iii . two approaches are used . one focuses on the intersections of the free energy curves of a pure coil and a pure helix as a function of the end - to - end distance , @xmath2 . as we shall see , each intersection is a rough diagnostic for a transition between the two configurations and for the occurrence of a plateau in the force law . the second , more quantitative approach is the diblock or @xmath6 approximation . within this approximation the homopolypeptide is assumed to consist of two domains , a helix and a coil , and the mixing entropy associated with the polydispersity of the helical and coil domains is set to zero , @xmath6 . finally , in section iv we present a rigorous analysis of the force laws allowing for the polydispersity of the coexisting helices and coil domains . as stated earlier , we implement this analysis for two models . in one the helical domains are viewed as rigid rods and in the second as semiflexible chains described by the worm like chain ( wlc ) model . in this section we also examine the validity of the mean field approximation introduced in @xcite . within this approximation , the explicit coupling between the tension and the distribution of domain sizes is ignored . as we shall see , it recovers the exact result when the helices are semiflexible _ i.e. , _ when the persistence length of the helices is finite and the helical domains are large . the analysis of the rigid helices model recovers the results obtained by tamashiro and pincus @xcite using a transfer matrix formalism . a comparison between the two models and an outline of some open problems are presented in the discussion . it is helpful to first summarize the main features of the helix - coil transition of free polypeptides in the absence of imposed extension . we will consider only homopolypeptides , consisting of a single type of amino acid residues , in order to avoid complications introduced by sequence of different residues found in heteropolypeptides . our discussion focuses on the case of long chains where the degree of polymerization of the polymer @xmath7 is large , @xmath8 . to this end we first discuss the zimm - brag parameters and then introduce the free energy analysis of the transition . as we shall see , this is equivalent to the transfer matrix formalism @xcite that is traditionally used to analyze the transition @xcite . our discussion focuses on helicogenic polypeptides that are capable of assuming two configurations : a random coil and an @xmath4-helix . in the coil state the monomers are free to rotate thus leading to a small kuhn length with typical value of @xmath9 corresponding to the length of few monomers . for simplicity we will identify the kuhn length with the span of a single monomer , @xmath10 . the helical configuration involves hydrogen bonds between residues @xmath11 and @xmath12 . this constrains the orientation of the three intermediate monomers . as a result the persistence length @xmath13 of the helical configuration is large , of order @xmath14 . at the same time the projected length of a monomer along the axis of the helix decreases to @xmath15 . the length of the polypeptide in a helical configuration is thus smaller than the span of a fully extended coil by a factor of @xmath16 @xcite . the homopolypeptides are modeled as a two state system . each monomer can exist either in a helical or a coil state . choosing the coil state as a reference state , the excess free energy @xmath17 of a monomer within a helical domain reflects two contributions . first is the change in enthalpy upon formation of an intrachain @xmath18-bond , @xmath19 . the second is associated with the loss of configurational entropy , @xmath20 , in the helical state . altogether @xmath21 the enthalpy term favors the helical state while the entropy term favors the coil state . thus , at low temperatures the free energy @xmath17 is negative and the helical configuration is preferred while at high temperatures @xmath17 is positive and the coil state is dominant . the two terms are comparable at @xmath22 when @xmath23 . @xmath0 provides an approximate value for the melting temperature of the @xmath4-helix . an additional parameter is needed to describe the helix - coil transition . the number of @xmath18-bonds in a helical domain consisting of @xmath24 monomers is @xmath25 . it is thus necessary to allow for the special state of the terminal monomers of the helical domain . these two monomers lose their configurational entropy with no gain due to the formation of @xmath18-bonds . each of the terminal bonds is thus assigned an excess free energy of @xmath26 with respect to the coil state . in the following we will use @xmath27 measured with respect to the helical state , that is @xmath28 as we shall discuss later , this definition is consistent with the known experimental results . clearly , this definition is consistent with the traditional one when the coil state is used as a reference . @xmath27 plays the role of an interfacial free energy associated with the boundary between helix and coil domains . this term makes the helix - coil transition cooperative since it favors the formation of large domains . it is customary to describe the helix - coil transition in terms of the zimm - bragg parameters @xcite : @xmath29 where @xmath30 is the boltzmann constant . the @xmath31 dependent @xmath32 represents the boltzmann factor associated with adding one monomer to a helical domain . @xmath33 at the transition temperature @xmath0 while @xmath34 for @xmath5 and @xmath35 for @xmath36 . @xmath37 corresponds to the boltzmann factor associated with the creation of a helical domain . the definition of @xmath27 as @xmath38 suggests that @xmath37 is independent of @xmath31 . the definition of @xmath27 as @xmath39 does not predict that @xmath37 is independent of @xmath31 but it yields an identical estimate for the numerical value of @xmath37 @xcite . the experimentally measured @xmath37 , as obtained from plots of the helical content _ vs. _ @xmath31 , are in the range of @xmath40 , depending on the residue , and exhibit a weak @xmath31 dependence . as stated earlier , in the following we will use @xmath41 . as we shall discuss , a plot of the helical content _ vs. _ @xmath32 exhibits a sigmoid behavior and the width of the transition scales with @xmath42 . thus , the transition becomes sharper , more cooperative , as @xmath37 decreases . having defined the relevant molecular parameters , @xmath17 and @xmath43 or equivalently @xmath32 and @xmath44 , we are in a position to analyze the helix - coil transition as it occurs in free chains upon changing the temperature @xmath31 . the minimal description of the transition requires the specification of two properties , the total number of monomers in a helical configuration , @xmath45 and the number of helical domains , @xmath46 . while @xmath47 and @xmath48 are sufficient for a discussion of the helix - coil transition of free chains , the analysis of the effect of extension requires additional information . in particular , the probability distribution of helical domains , @xmath49 , and of coil domains , @xmath50 , comprising of @xmath24 monomers . in order to determine these parameters we minimize the appropriate phenomenological free energy . this method recovers the results obtained by the transfer matrix approach and is easily generalized to allow for the effect of applied tension . the free energy allows for two principal contributions : ( i ) the free energy of the monomers . all @xmath51 helical monomers , including the terminal monomers of the domains are assigned an excess free energy of @xmath17 . the @xmath52 terminal monomers are assigned an additional free energy of @xmath53 . as a result their excess free energy with respect to the coil state is @xmath54 . ( ii ) a mixing entropy term associated with the different possible placements of @xmath51 helical monomers and @xmath52 terminal monomers constituting the domain boundaries . since the helical and coil domains alternate , there is no entropy due to their mixing . rather , the mixing entropy arises because of the polydispersity of the domains _ i.e. _ , helical and coil domains of different size @xmath24 are considered distinguishable . since there are @xmath46 helical and @xmath46 coil domains the mixing entropy is @xmath55 the free energy per monomer in the unperturbed chain is thus @xmath56 the lagrange multipliers , @xmath57 and @xmath58 , ensure the normalization @xmath59 and @xmath60 . the two remaining lagrange multipliers , @xmath61 and @xmath62 , impose the average sizes @xmath63 and @xmath64 of the helical and coil domains . this description corresponds to a simplified version of the transfer matrix method involving a two by two matrix . in both cases there is no constraint ensuring that helical domains incorporate at least three monomers @xcite . since we focus on the case of @xmath65 and @xmath8 the weight of small domains is negligible and this description yields the correct results . the probabilities @xmath49 and @xmath50 are determined by minimization of @xmath66 with respect to @xmath49 and @xmath50 ( appendix a ) leading to @xmath67 note that @xmath49 and @xmath50 decay exponentially with decay constants @xmath68 and @xmath69 $ ] . thus , @xmath49 and @xmath50 are not sharply peaked at the mean values @xmath63 and @xmath70 . in particular , there is a significant population of small helices even when the average size of the helical domain @xmath63 is large . in this limit , @xmath71 . similarly , when @xmath70 is large @xmath72 . upon substituting @xmath49 and @xmath50 in @xmath73 as given by ( [ eq_free2 ] ) , we obtain @xmath66 as a function of @xmath47 and @xmath48 only @xcite : @xmath74 the equilibrium conditions @xmath75 and @xmath76 then lead to @xmath77 in turn , these yield ( appendix b ) the expression for the equilibrium values of @xmath78 and @xmath48 @xmath79^{-1/2}}{s+1 + \sqrt{(s-1)^{2}+4\,\sigma s}}.\end{aligned}\ ] ] these expressions are identical to the ones obtained _ via _ the transfer matrix method @xcite . the behavior of @xmath47 and @xmath48 as a function of @xmath32 and @xmath37 is plotted in fig . [ theta0 ] . when @xmath80 we recover the boltzmann distribution for independent two - level systems : @xmath81 . in marked contrast when @xmath82 , the fraction of helical monomers is steplike : @xmath83 for @xmath34 and @xmath84 for @xmath35 . the chain is either a pure helix or a pure coil . for intermediate values of @xmath37 , the plot of @xmath47 _ vs. _ @xmath32 yields a sigmoid curve with a transition region of width @xmath85 centered around @xmath86 . the number of domains is maximal in the vicinity of @xmath33 and it decreases with @xmath37 . the parameter @xmath87 serves as a measure of the temperature , @xmath31 . for low temperatures , @xmath88 and the polymer is mainly in a helical state . at high temperatures , @xmath89 and the polymer is mostly in a coil state . the crossover occurs at @xmath90 for which @xmath86 . near this crossover , @xmath91 and the width of the crossover regime is @xmath92 . the most notable feature of the force laws of homopolypeptides are the plateaus . in this section we present qualitative and semiquantitative analysis of the plateaus . each plateau in the force law corresponds to a coexistence between two states , a helix and a coil . the situation is somewhat analogous to a liquid - gas transition which leads to a plateau in the pressure - volume diagram . in our case , the tension @xmath1 plays the role of the pressure and the end - to - end distance @xmath2 is analogous to the volume . this analogy is however of limited validity since the homopolypeptides are one - dimensional systems with short range interactions and thus incapable of undergoing a first order phase transition . in such systems the boundary free energy is independent of the size of the domain . as a result , a first order phase transition involving a coexistence of two domains is prevented by the mixing entropy @xcite . accordingly , the average domain sizes @xmath64 and @xmath93 , as obtained in the previous section , are independent of the size of the chain , @xmath7 . nevertheless , the main features of the system can be recovered if one ignores the role of the mixing entropy . as expected , within this rough approximation the plateaus are erroneously associated with first order phase transitions . in spite of this deficiency , this approximation recovers the correct length and force scales but the force laws obtained are wrong in detail . before we present this approximation we investigate the crossing of the free energy curves associated with the `` pure '' helix and coil states . this is a rough diagnostic for first order phase transitions . in our case it provides an insight concerning the physical origin of the plateaus as well as a reasonable estimate for some of the length and force scales involved . the occurrence of the plateaus is signaled by the crossing of the free energy curves of the pure helix and the pure coil as a function of @xmath2 . each crossing point signals a coexistence between the two `` phases '' and thus a plateau in the force law . with the unperturbed coil as a reference state , the free energy of the coil is simply its elastic free energy . in this section , we will utilize the fixed @xmath2 ensemble and we thus denote the elastic free energy by @xmath94 . to allow for the finite extensibility of the chain we use the freely jointed chain ( fjc ) model ( appendix c ) for @xmath94 and the free energy per monomer is @xmath95 where @xmath96 $ ] , @xmath97 is the reduced end - to - end distance of the polymer and @xmath98 is the reduced force . @xmath99 and @xmath100 are related _ via _ @xmath101 where @xmath102 is the langevin function . the free energy of the helix reflects two contributions . first is the excess free energy of a helix , @xmath103 . in the limit of an infinite chain , when the contribution of the terminal monomers is negligible , the free energy per monomer is @xmath104 . at this stage we assume that the persistence length of the helix is infinite and that it may be viewed as a perfectly rigid and unextendable rod . within this view , extending the helix beyond its natural length is associated with an infinite free energy penalty . this penalty is important because the helix is shorter than the fully extended coil . as noted earlier , the contribution of each helical monomer to the overall length of the helix is @xmath105 . altogether , @xmath104 so long as @xmath106 while in the range @xmath107 it is @xmath108 . three scenarios are now possible ( fig . [ crossing ] ) . at low temperatures , @xmath109 and @xmath34 . accordingly , @xmath110 while @xmath111 . in this regime the helix is more stable than the coil . when @xmath107 the free energy of the helix diverges and the coil state becomes preferable . this curve crossing corresponds to a plateau involving the breakdown of the helix due to the applied tension . above the melting temperature , for @xmath112 and @xmath113 the unstretched polypeptide is in a coil state . upon stretching @xmath114 increases and eventually intersects with @xmath115 this intersection signals a plateau in the force law due to the formation of helical domains . at @xmath116 the free energies @xmath114 and @xmath115 cross again because of the infinite free energy of a helix with @xmath107 . this second intersection signals a second plateau involving the melting of the helix and the formation of coil domains . finally , for sufficiently high temperatures , @xmath117 , no intersections occur and the polypeptide remains in a coil state for all @xmath99 values . the threshold temperature for this regime , @xmath118 , is defined by the equality @xmath119 where @xmath120 is the free energy of a stretched coil with @xmath116 . the intersection of the free energy curves provides a rough indicator for the occurrence of plateaus in the force law . however , it does not allow to calculate an explicit force - extension diagram . a simple approach yielding the main features of the force law is the @xmath6 or diblock approximation @xcite . within this approximation the force law is calculated for a diblock copolymer with one helical domain and one coil domain . the size of the annealed domains varies with the extension and is determined by the equilibrium conditions . as is implied by its name , in this approximation @xmath121 is neglected . as expected , setting @xmath6 leads to a first order phase transition associated with a coexistence of two `` phases '' . consequently , some features of the force law are wrong . in particular , this approximation yields a perfect plateau with @xmath122 while in reality @xmath1 increases weakly with @xmath2 . these features notwithstanding , this simple approximation does recover the correct length and force scales . within this discussion we will also briefly examine the effect of the persistence length of the helical domains . in particular we compare the behavior of a helix endowed with an infinite persistence length to that of a helix with a finite one . within the @xmath6 or diblock approximation the free energy @xmath123 of the chain is @xmath124 here , the first term allows for the excess free energy of the helical monomers . since @xmath6 the helical and coil phases separate into two domains . the `` interfacial '' free energy of the single helical domain gives rise to the second , constant term . these two terms are supplemented by an elastic free energy @xmath125 that depends on @xmath47 , @xmath126 and the end - to - end distance , @xmath2 . different versions of the @xmath6 are possible , depending on the choice of @xmath125 . in the following we focus on the simplest version where @xmath125 is given by the gaussian elasticity of an ideal random coil in the fixed @xmath2 ensemble . two additional assumptions are invoked : ( i ) the helical domain is perfectly rigid and unextendable and ( ii ) the helical domain is perfectly aligned with the applied force . these assumptions specify the elastic free energy per monomer @xmath127 this simple version of the diblock approximation permits an explicit analytical solution for the force law . the corresponding equilibrium condition is @xmath128 within this approximation the equilibrium condition @xmath129 is an identity yielding no additional information . the solution of this quadratic equation in @xmath130 is @xmath131 . this defines a critical @xmath17 @xmath132 corresponding to a critical @xmath133 or @xmath118 . while @xmath134 , the equilibrium condition can not be satisfied , that is , the polypeptide remains in a coil state irrespective of the imposed tension . a critical point occurs for @xmath135 . when @xmath136 two solutions exist , each defining a plateau in the force _ vs. _ extension curve . the solution @xmath137 corresponds to the lower plateau , obtained for @xmath138 , where @xmath139 this first regime disappears for @xmath140 or @xmath5 . the condition @xmath84 defines a lower boundary of the plateau , @xmath141 @xmath142 increases as @xmath17 increases until @xmath143 for @xmath135 . for @xmath144 the fraction of helices is @xmath84 and the chain is in a purely coil form . the upper boundary of this plateau occurs at @xmath145 when the chain is fully helical , @xmath83 . at intermediate extensions , @xmath146 the fraction of helical bonds increases as @xmath147 and @xmath148 occurs at the midpoint of the plateau @xmath149 . the equilibrium free energy of the chain for @xmath144 is @xmath150 while for @xmath151 it is @xmath152 the second plateau , is described by the solution , @xmath153 , leading to @xmath154 the lower boundary of this plateau occurs at @xmath155 when @xmath83 . the upper boundary , when @xmath84 , occurs at @xmath156 for @xmath157 the fraction of helical monomers decreases as @xmath158 and @xmath148 occurs at the midpoint of the plateau @xmath159 . notice that the width of the two plateaus shrinks as @xmath160 or @xmath161 . for @xmath162 the fraction of monomers in a helical state is again @xmath84 and the corresponding free energy is @xmath150 . the equilibrium free energy of the chain in the range @xmath163 is @xmath164 note that only the lower plateau is obtained if @xmath165 is assumed@xcite . the @xmath6 approximation allowed us to obtain rough force laws corresponding to the three scenarios identified by the curve crossing argument . thus , for @xmath134 or @xmath117 , the polypeptide remains in a coil configuration for the whole range of extension . the `` two plateau scenario '' occurs for @xmath136 corresponding to @xmath166 or @xmath167 . in this case we can distinguish five regimes ( fig . [ phases ] ) : ( i ) for low extensions the chain is in a coil state , @xmath84 , and the elastic behavior is gaussian leading to @xmath168 . ( ii ) in the range @xmath146 helix and coil domains coexist according to the lever rule @xmath169 . this coexistence is associated with a plateau with a constant tension @xmath170 ( iii ) at @xmath171 the force curve exhibits a discontinuity associated with a step like increase to a second plateau . the magnitude of the step , @xmath172 is @xmath173 ( iv ) in the range @xmath157 helix and coil domains coexist according to the lever rule @xmath174 . this coexistence is associated with a second plateau with a constant tension @xmath175 ( v ) finally , for @xmath176 the chain is in a strongly stretched coil state obeying @xmath168 . when @xmath177 or @xmath178 the unperturbed chain is in a helical state . consequently , the lower plateau disappears . it is replaced by a `` quasi plateau '' which is not associated with a helix - coil coexistence . rather , it is due to the facile orientation of the helical chain which is modeled as a rigid rod . at @xmath116 a jump occurs to the upper plateau . the analysis presented above utilized the gaussian elastic free energy in the fixed @xmath2 ensemble to describe the coil . while this allows us to obtain explicit expressions for all the characteristics of the force curve , it also introduces errors . clearly , this choice of @xmath125 does not capture the behavior of the coil in the high extension regime , where finite extensibility begins to play a role . as a result , the values of @xmath179 are overestimated . this has two consequences . first , within this approximation the span of the two plateaus are equal @xmath180 while for more realistic choice of @xmath125 the higher plateau is narrower . a second more important pathology is the disappearance of regime ( v ) in the @xmath181 case . to allow for the finite extensibility it is necessary to use the freely jointed chain model ( fjc ) for the coil segment @xcite . it is important to note that so long as @xmath182 and @xmath183 the discrepancies between the two schemes are rather small ( fig . [ force_db ] ) . another ingredient of our simplified analysis is the modeling of the helices as rigid and unextendable rods . an alternative description of the elastic properties of helices models them as worm like chains _ i.e. , _ semiflexible chains capable of undergoing undulations ( see appendix d ) @xcite . as we shall see later , this semiflexible approximation of the helices has a qualitative effect on the results of the rigorous analysis of the system . however , in the context of the @xmath6 approximation for @xmath183 , it mainly affects the transition between the two plateaus ( fig . [ force_db ] ) . when the helix is modeled as a rigid rod the transition occurs as a jump that takes place at @xmath116 . on the other hand , when the wlc model is used the jump , while abrupt , takes place over a finite interval _ @xmath184 is replaced by @xmath185 . for the @xmath140 ( fig . [ force_db2 ] ) the transition at the upper boundary of the `` quasi plateau '' is more gradual when the elasticity of the semiflexible helix is allowed for . finally , note that the diblock approximation is recovered from the more rigorous treatments presented below when the @xmath186 limit is taken before the @xmath187 limit . when @xmath188 limit is approached for a finite chain it leads to a single domain . the analysis presented in the previous section fails to allow for a number of important ingredients . the most important is the mixing free energy , @xmath189 , associated with the polydispersity of the domains . in this section we present an analysis that allows for the role of this mixing entropy . because of @xmath121 the helix - coil transitions no longer take place as first order phase transitions . in turn this has two important consequences : ( i ) the `` perfect '' plateaus , with constant forces @xmath190 and @xmath191 are replaced by `` imperfect '' plateaus with a finite slope . ( ii ) the crossovers between the plateaus and the neighboring regimes become smoother . the analysis is based on a generalization of the approach described in section ii that is , all chain characteristics , @xmath49 , @xmath50 , @xmath78 and @xmath48 are obtained from the minimization of the appropriate free energy _ i.e. _ , the equilibrium conditions . as in section iii , the free energy is of the form @xmath192 where , @xmath66 is the free energy of the unperturbed free chain and @xmath125 , accounts for the elasticity of the chain . however , in the following we utilize different expressions for these two terms . since we aim to obtain @xmath49 and @xmath50 it is necessary to utilize @xmath66 as given by equation ( [ eq_free2 ] ) . the handling of the elastic free energy is significantly different . in section iii we used the gaussian elasticity to describe the deformation behavior of the coils and assumed that the helices contribute to the elastic behavior as a geometric constraint . while this rough description is sufficient at the level of the @xmath6 approximation , a more refined treatment is needed for the detailed analysis to be undertaken below . in particular , it is necessary to allow for two additional features : ( i ) the finite extensibility of the chain and ( ii ) the gradual alignment of the helical domains with the applied tension . the freely jointed chain ( fjc ) model allows us to incorporate the finite extensibility of the coil domains . with regard to the helices we will explore two different descriptions that impose finite extensibility on the helices . first we will model the helical domains as rigid and unextendable rods and allow for their length dependent alignment by using the fjc model _ i.e. , _ the helical domains are considered as freely jointed rigid rods forming a chain . this amounts to the assumption of an infinite persistence length for each helical domain . this picture is reasonable so long as the size of the helical domains is small compared to the persistence length , @xmath13 . in the second approach , each of the helical domains is viewed as worm like chain ( wlc ) that is , semiflexible chain characterized by a finite persistence length and capable of undergoing undulations . it is assumed that the domains are large in comparison to @xmath13 . an important difference between the fjc and the wlc models is that in the later the persistence length depends on the applied tension . our approach utilizes the analysis of marko and siggia of the extension elasticity of the wlc @xcite . the utilization of these models makes the constant @xmath1 ensemble mathematically convenient . this choice of ensemble is also preferable because it corresponds better to the experimental situation encountered in single molecule measurements . in each case , we first obtain the equilibrium values of @xmath49 , @xmath50 , @xmath47 and @xmath48 of a chain subject to a given tension , @xmath1 , by minimizing @xmath193 with respect to @xmath49 , @xmath50 , @xmath47 and @xmath48 . this allows us to determine the equilibrium @xmath193 thus enabling us to calculate the @xmath3 diagram by using @xmath194 when the helical domains are viewed as rigid rods and the fjc is invoked , the reduced end - to - end distance @xmath97 of the polypeptide is @xmath195 where @xmath98 is the reduced tension . the first part of this expression corresponds to the projection of the coil monomers along the direction of the tension . the second part reflects to the contribution of the helical domains to @xmath99 . because the individual monomers within the coil domains are aligned independently , the domain size does not play a role . the extension of each coil domain is extensive with respect to the number of monomers , @xmath24 . consequently , this contribution may be expressed in terms of the first moment of the probability distribution @xmath196 . as a result the full probability @xmath50 is irrelevant . in marked contrast , the monomers within the helical domains are not aligned individually but as part of the rod - like domains . the non - linearity introduced by the @xmath197 term in the expression for @xmath99 couples @xmath1 and @xmath49 . as a consequence , @xmath49 will be explicitly modified by the applied tension while @xmath50 depends on @xmath1 only implicitly , _ via _ @xmath47 and @xmath48 . eliminating @xmath50 from ( [ r_rigid ] ) , we obtain @xmath198 this expression for @xmath199 enables us to obtain the corresponding elastic free energy @xmath200 ( appendix c ) @xmath201 where @xmath96 $ ] . altogether the free energy per monomer is thus @xmath202 since the @xmath1 dependence of @xmath123 is due only to @xmath125 , ( [ rdef ] ) yields @xmath2 as given by ( [ rdef2 ] ) . as before , @xmath49 and @xmath50 are determined by minimizing @xmath193 with respect to these probabilities subject to two constraints : ( i ) the normalization of the probabilities @xmath49 and @xmath50 and ( ii ) the average sizes @xmath64 and @xmath93 of the coil and helical domains . upon substituting the resulting @xmath49 and @xmath50 in @xmath123 we obtain @xmath193 as a function of @xmath47 and @xmath48 . @xmath78 and @xmath48 are then determined by the equilibrium conditions @xmath203 . the details of the calculation involved are described in appendix e. @xmath47 and @xmath48 for a given tension @xmath1 are @xmath204 where @xmath205 the probabilities @xmath50 and @xmath49 are @xmath206 @xmath49 , as @xmath50 , is essentially an exponentially decaying function . however , the characteristic decay constants are now dependent on @xmath1 . the dependence is explicit for @xmath49 while for @xmath50 the @xmath1 dependence is implicit , _ via _ @xmath47 and @xmath48 . in fig . [ theta2 ] , the equilibrium @xmath47 and @xmath48 , are plotted as functions of @xmath98 for different values of @xmath37 . clearly , the results of the @xmath6 or diblock approximation are approached in the limit of strong cooperativity , @xmath188 . in particular , @xmath48 approaches zero while @xmath47 exhibits a step like behavior . thus , @xmath207 outside the plateaus range , @xmath208 and @xmath209 , while within the plateaus range , @xmath210 we find @xmath211 and @xmath212 . consequently , the mean size of the helical domains , @xmath93 , becomes very large compared to @xmath13 . in turn , this suggests that the validity of the model of helical domains as rigid rods is questionable . clearly , the rigid helices model is meaningful only while the persistence length is much larger than @xmath213 . once the equilibrium characteristics of the chain for a given @xmath1 are found , it is possible to obtain the force - extension diagram . the reduced end - to - end distance , @xmath99 , as a function of the reduced tension , @xmath100 , is ( appendix e ) @xmath214 .\ ] ] the corresponding force - extension diagram is plotted in fig . [ force ] . again , the results of the @xmath6 or diblock approximation are approached in the @xmath188 limit . note the `` jump '' in the force at @xmath215 @xcite . at this point , between the two plateaus , the polypeptide is in a `` purely '' helical state , @xmath211 and @xmath212 . this jump is another indication of difficulties due to the approximation of the helices as rigid rods . as we shall see in the next section , this unphysical feature disappears when the finite persistence length of the helices is allowed for . it is of interest to compare the exact solution of the model to the mean field approximation ( mf ) @xcite . the mf approximation involves the removal of the explicit coupling between @xmath50 , @xmath49 and @xmath1 . as noted earlier , this decoupling is exact for @xmath50 . for @xmath49 the decoupling is achieved by replacing @xmath216 in ( [ rdef2 ] ) by @xmath217 . in effect , all helical domains are assigned an equal size of @xmath93 and the effect of their polydispersity on @xmath125 is ignored . the corresponding free energy is @xmath218 the equilibrium conditions @xmath219 @xmath220 lead to @xmath221 , \\ & & \frac{y^{2}}{(\theta -y)(1-\theta -y ) } = \\ \nonumber & & \hskip 0.4 cm = \frac{\sigma y}{\theta \gamma x } \sinh \left ( \frac{\theta \gamma x}{y}\right ) \exp \left [ -\frac{\theta \gamma x}{y}{\cal l}\left ( \frac{\theta \gamma x}{y}\right ) \right ] .\end{aligned}\ ] ] the overall features of the mf solution are rather similar to the exact solution ( fig . [ force ] ) . the two primary differences are : ( i ) the mf solution indicates that the transition between the plateaus involves a second order phase transition @xcite in marked distinction to the exact solution . ( ii ) because of ( i ) the plateaus do not exhibit an inflection point and the characteristic value of the slope , at @xmath116 , scales as @xmath222@xcite . the assumption of perfectly rigid helical sequences implies an infinite persistence length . this assumption is reasonable so long as the size of the helical domains is small compared to the known values of their persistence length , @xmath13 : a typical value is @xmath223 corresponding to @xmath224 monomers . as we have seen , the rigid rod approximation for an infinite chain leads to helical domains that are larger . in turn , this leads to an unphysical jump , located at @xmath225 , in the force - extension curve . two approaches allow to circumvent this difficulty . one approach treats each helical monomer as an extendable spring _ i.e. _ , it introduces an elastic penalty associated with the deviation of the length of the helical monomer from its equilibrium length @xmath226 . this approach was utilized in the analysis of the extension behavior of dna @xcite and of poly-(ethylene glycol ) @xcite . in the following we adopt a different approach that proved successful in fitting the force curves of dna . within this approach the helical domains are considered as worm like chains ( wlc ) instead of rigid rods ( appendix d ) . in the wlc model the helical domains are viewed as semiflexible chains undergoing undulations . the angular correlations along the chain decay because of the thermal undulations . within this model the decay length of the angular correlations , @xmath227 depends on the tension , @xmath228 for weakly stretched chains @xmath229 is identical to the persistence length , @xmath13 , of the unperturbed chain , @xmath230 while for stronger tensions @xmath231 and the correlations decay faster . a long helical domain of length @xmath232 may thus be considered as a fjc of @xmath233 `` rigid '' segments of length @xmath234 and the reduced end - to - end distance is @xmath235 where @xmath102 is the langevin function and @xmath236 @xcite . as before , the expression for @xmath99 reflects two contributions . the first term corresponds to the alignment of the coil monomers with the applied tension using the fjc model . the second term describes the alignment of the semiflexible helical domains . in the following we assume that the helical domains are larger than @xmath237 that is , @xmath238 . the error introduced by this assumption diminishes as @xmath1 increases and @xmath237 decreases . within this approximation each @xmath234 segment in a helical domain is aligned independently . this is in marked contrast to the rigid helices model where all the monomers within a helical domain align as a unit . as a result we may utilize @xmath239 in order to obtain @xmath240 . consequently , @xmath99 is independent of the probabilities @xmath50 and @xmath49 . expression ( [ dist_flex ] ) for @xmath199 enables us to obtain ( appendix d ) the corresponding elastic free energy @xmath200 @xmath241 in marked distinction to the `` rigid helices '' scenario , this elastic free energy is independent of the probabilities @xmath49 and @xmath50 . accordingly , @xmath49 and @xmath50 are not modified explicitly by the extension and obey ( [ proba_h ] ) and ( [ proba_c ] ) . in other words , the mean field approximation provides an exact solution of the problem . altogether , the free energy of a polypeptide , using the wlc model for the helical domains , is @xmath242{\cal l}(\alpha ) . \nonumber\end{aligned}\ ] ] this expression is similar to @xmath123 for a free , undeformed chain ( [ eq_free1 ] ) . the two free energies differ in two respects : ( i ) ( [ eq_free_flex ] ) contains an additive term @xmath243 which is independent of @xmath47 and @xmath48 . ( ii ) the force independent @xmath32 in ( [ eq_free1 ] ) is replaced by a force dependent @xmath244 @xmath245 . \label{stilde}\ ] ] consequently , the equilibrium conditions @xmath203 lead to expressions of the same form as those found for the free chains but with @xmath246 replacing @xmath32 @xmath247 ^{-1/2}}{\tilde{s}+1+\sqrt{(\tilde{s}-1)^{2}+4\,\sigma \tilde{s}}}.\end{aligned}\ ] ] substitution of the equilibrium values of @xmath47 and @xmath48 for a given @xmath100 in ( [ dist_flex ] ) yields the corresponding equilibrium end - to - end distance . plots of @xmath47 _ vs. _ @xmath248 and @xmath248 _ vs. _ @xmath249 , as obtained for the `` semiflexible helices '' scenario for different @xmath37 , are depicted in fig . [ thetawlc ] . the replacement of the `` rigid helices '' by semiflexible ones leads to the disappearance of the `` jump '' in the force law . another difference with respect to fig . [ theta2]a is that @xmath83 is never attained . again , we recover the appropriate diblock approximation in the limit of @xmath188 . all the force - extension curves in fig . [ thetawlc]b cross at three different points . the external points correspond to a reduced force @xmath250 satisfying @xmath251 . in turn , this condition is equivalent to the equilibrium condition @xmath252 for @xmath123 corresponding to the appropriate diblock approximation in which the coil is modeled as fjc and the helix as a wlc @xcite . in particular , @xmath250 correspond to points in the plateaus where @xmath148 , @xmath253 /2 $ ] and @xmath254 . the inner crossing point corresponds to @xmath255 for which @xmath244 is maximal . @xmath255 satisfies the condition @xmath256 $ ] _ i.e. _ , to an equality between the projected length of the coil and the helical monomers along the force . from ( [ dist_flex ] ) and ( [ stilde ] ) , it is apparent that this condition is independent of @xmath37 . the plateaus depicted in fig . [ thetawlc]b exhibit a small but finite slope . its characteristic value , at the inflection points @xmath250 , scales with @xmath257 @xcite . our analysis of the extension behavior of polypeptides utilized two models . the two differ in the description of the helical domains . in one model , the helical domains are viewed as perfectly rigid and unextendable rods . the `` rigid helices '' model is reasonable so long as the length of the helical domains is small compared to the persistence length . in the second approach the helical domains are treated as long semiflexible polymers within the worm like chain ( wlc ) model . this approach is preferable when the length of the helical domains is larger than the force dependent decay length @xmath258 . for each of the models , we presented a rigorous solution and two approximations . within the diblock approximation one neglects the mixing entropy associated with the polydispersity of the domains . in the mean field approximation there is no direct coupling between the distribution of sizes of the domains and the applied tension . in all cases considered , our analysis is based on the minimization of a phenomenological free energy . this approach is equivalent to the transfer matrix method but it is physically transparent and mathematically simple . the general features of the extension behavior of the polypeptides are independent of the model and the approximation scheme . three different scenarios are possible for the extension of long polypeptides : ( a ) for @xmath117 the polypeptide remains in a coil configuration for all extensions . ( b ) in the range @xmath167 the chain is initially in a coil state but it undergoes a helix - coil transition upon extension . the helical domains eventually melt upon stronger extension . consequently , the force law exhibits two plateaus associated with a helix - coil coexistence . ( c ) for @xmath5 the unstretched polypeptide is a helix which melts upon stretching thus leading to a force law with a single plateau . in the second scenario , @xmath167 , it is possible to distinguish between five regimes : ( i ) for weak extensions the polypeptide behaves as a random coil exhibiting a linear response . ( ii ) stronger extensions , and the associated loss of configurational entropy , favor the formation of helical domains . the associated coexistence of helical and coil domains gives rise to a lower plateau . ( iii ) a steep crossover corresponding to the stretching of a stiff , mainly helical , polymer leads to ( iv ) a second , higher plateau due to the break - up of the helices when the imposed end - to - end distance is larger then the length of a fully helical chain . this plateau is also associated with helix - coil coexistence . ( v ) when the helical content is reduced to zero the chain exhibits a strong extension behavior of a random coil , with deviations from the linear response behavior due to finite extensibility effects . in the third scenario , @xmath5 , the regimes ( i ) and ( ii ) disappear and regime ( iii ) extends down to @xmath259 . the small slope of the force law in this regime ( fig . [ f_s1.1 ] ) is reminiscent of a plateau . however , this quasi - plateau is not associated with a helix - coil coexistence . rather , it is due to the easy alignment of the persistent segments in the helix . the two models , `` rigid '' and `` semiflexible '' helices , differ with regard to the precise details of the force diagram . thus , region ( iii ) occurs as a `` jump '' within the `` rigid helices '' model but is a steep yet gradual increase in the `` semi - flexible helices '' model . the slope of the plateaus within the rigid helices model scales as @xmath260 ^{2}$ ] @xcite while in the `` semi - flexible helices '' model it scales as @xmath257 . overall , the crossovers between the various regimes are sharper in the `` rigid helices '' model . the diblock approximation , with the appropriate elastic term , allows to recover the main features , length and force scales , of each model . however , within this @xmath6 approximation the transition takes place as a first order phase transition . as a result it yields perfect plateaus , @xmath261 or @xmath262 , and sharp crossovers . the mean field approximation , where the polydispersity is decoupled from the tension is exact for the `` semiflexible helices '' model . in the case of the `` rigid helices '' model this approximation , while it captures the overall features of the force law also gives rise to a number of artifacts @xcite . neither of the models considered captures the physics completely . the `` rigid helices '' model fails to describe the behavior of helical domain that are long compared to the persistence length . within the `` semiflexible helices '' model all helical domains are assumed to be long in comparison to @xmath237 . accordingly , this model is likely to mishandle short helical domains that do behave as rigid rods . the importance of this deficiency diminishes however as @xmath1 increases and @xmath237 decreases . a full description of the system requires a model combining the positive features of the two models described above . while our analysis focused on the case of homopolypeptides forming @xmath263 -helices , our approach can be extended to more complicated systems . in the case of heteropolypeptides it is necessary to allow for the residue dependence of @xmath17 . for double stranded dna it is necessary to allow for long range interactions as given by the stockmayer formula for @xmath44 @xcite . we would like to thank m. n. tamashiro and p. pincus as well as i. rouzina and v. a. bloomfield for providing us with their results prior to publication . the work of ab was supported by the marie curie fellowship hpmf - ct-1999 - 00328 . the minimization of the free energy ( [ eq_free2 ] ) with respect to the probabilities @xmath49 and @xmath50 yields @xmath264 these are solved by @xmath265 and @xmath266 where @xmath267 the parameters @xmath268 , @xmath269 , @xmath270 and @xmath271 are determined using the constraints @xmath272 yielding @xmath273 , @xmath274 , @xmath275 and @xmath276 thus leading to the expressions ( [ proba_h ] ) for @xmath49 and ( [ proba_c ] ) for @xmath50 . the equilibrium conditions @xmath277 with @xmath66 given by ( [ eq_free1 ] ) lead to @xmath278 from ( [ appendixa2 ] ) , we obtain @xmath279\ ] ] or @xmath280 where @xmath281 . the second condition ( [ appendixa1 ] ) yields @xmath282 combining these two equations allows us to eliminate @xmath48 and obtain an equation for @xmath78 as a function of @xmath32 and @xmath44 , @xmath283 leading to @xmath284 the freely jointed chain model may be considered as the macromolecular analog of the langevin theory of paramagnetism @xcite . a flexible chain comprising @xmath7 monomers is the counterpart of @xmath7 non interacting magnetic dipoles . the monomer length @xmath285 is the analog of the magnetic moment @xmath286 , the applied tension @xmath287 plays the role of the magnetic field @xmath288 and the induced magnetic moment , @xmath289 is the analog of the end to end distance @xmath290 each monomer is assigned an orientational energy @xmath291 where @xmath292 is the angle between the force and the segment . the reduced end - to - end distance @xmath249 of the polymer , at a temperature @xmath31 , is the mean value of @xmath293 : @xmath294 where @xmath98 is the reduced force and @xmath295 is the langevin function . the corresponding elastic free energy in the fixed @xmath1 ensemble , as given by @xmath296 , leads to the following free energy per monomer : @xmath297 where @xmath298 $ ] , is the integrated langevin function . the elastic free energy in the fixed @xmath2 ensemble @xmath299 is the legendre transform of @xmath300 where @xmath97 and @xmath100 are related by @xmath301 . for weak extensions @xmath94 reduces to the familiar gaussian form @xmath302 while for strong extensions it diverges as @xmath303 . within the wormlike chain model ( wlc ) the polymer is viewed as a bendable rod @xcite . in the following we consider the case of a chain of constant contour length @xmath304 and a single bending modulus @xmath305 where @xmath13 is the persistence length of the unperturbed chain . the chain trajectory is described by the position of a point on the chain , @xmath306 at the contour length @xmath307 the unit tangent vector @xmath308 specifies the local orientation of the chain . a straight rod corresponds to @xmath309 or @xmath310 accordingly , the bending energy is a quadratic function of @xmath311 . since @xmath312 , the only quadratic form is @xmath313 and the conformational distribution of the chain , its partition function , up to a contour length @xmath32 , is @xmath314\ ] ] where @xmath315 is a function of @xmath316 and @xmath317 _ i.e , _ this describes a gaussian process with the constraint @xmath319 . as such it is analogous to rotational diffusion and obeys @xmath320 where @xmath321 is the rotational operator that is related to the angular momentum operator in quantum mechanics , @xmath322 , as @xmath323 . the correlations along the unperturbed chain decay as @xmath324 when the chain is subjected to an external field @xmath325 the expression for @xmath326 becomes @xmath327\ ] ] the effect of tension @xmath287 is specified by @xmath328 thus leading to @xmath329 it is possible to find eigenstate satisfying @xmath330 . for a long chain the free energy is extensive and the eigenfunction expansion of @xmath331 involves terms of the form @xmath332 where @xmath333 is the free energy per unit length due to the bending fluctuations . an analytical solution of equation ( [ eq_psi ] ) is not available . marko and siggia @xcite argued that the wlc free energy is determined by the lowest @xmath334 in the spectrum and utilized a variational approach in order to determine @xmath334 . the force law is then recovered from @xmath335 . to this end they chose the trial function @xmath336 $ ] where @xmath4 is a variational parameter and @xmath337 is the angle between @xmath338 and @xmath287 . as we shall see later , @xmath339 where @xmath340 is a force dependent decay length that replaces @xmath13 in characterizing the decay of @xmath341 . assuming @xmath342 @xmath343 this leads to @xcite @xmath344 where @xmath295 is the langevin function and @xmath98 is the reduced force . the argument @xmath345 that minimizes @xmath334 is specified by @xmath346 which leads to : @xmath347{\cal l}^{\prime } ( \alpha ) .\ ] ] where @xmath348 . here the end - to - end distance is simply deduced from : @xmath350 this end - to - end distance is identical to the one obtained from the fjc model if the reduced force @xmath351 @xcite is replaced by @xmath352 . for small forces , when @xmath353 and @xmath354 ( [ eq_alpha ] ) leads to @xmath355 or @xmath356 . for high tension @xmath4 increases more slowly . in this limit @xmath357 while @xmath358 and ( [ eq_alpha ] ) yields @xmath359 or @xmath360 . the difference between the variational approximation and the exact solution of the wlc model is less than @xmath361 over the whole range of forces . the associated elastic free energy , @xmath362 is @xmath363 since @xmath364 and where @xmath334 corresponds to the explicit expression in ( [ eq_g ] ) with @xmath365 according to ( [ eq_alpha ] ) . in eq . ( [ eq_fel ] ) this elastic free energy is used for a helical domain of length @xmath366 with @xmath24 monomers . in this appendix we determine the probabilities @xmath367 and @xmath368 of a coil and a helical segment with @xmath24 monomers as well as the fractions of helical monomers @xmath78 and of helices @xmath48 . the probability @xmath49 is modified by the stretching and is more difficult to obtain . to do so we consider the two remaining equilibrium conditions @xmath374 and @xmath375 after substitution of the equilibrium form of @xmath50 as given by ( [ prob_pc ] ) to obtain the end - to - end distance @xmath2 as a function of the tension @xmath1 we use @xmath393 which leads to @xmath394 the second term may be tackled exactly using the definition @xmath395 : @xmath396 to be precise , this approach is equivalent to a transfer matrix description utilizing a second order transfer matrix . in turn , this description is sufficient for the analysis of the helix - coil transition in an infinite chain . the final expression for the mixing entropy for the free chain can be calculated directly @xcite . @xmath406 where @xmath407 is the product of two factors : @xmath408 the number of possible placements of @xmath46 boundaries to separate @xmath409 helical bonds into @xmath46 helical domains and @xmath410 the number of ways to place @xmath46 boundaries to separate @xmath411 coil bonds into @xmath46 coil domains . using the sterling formula @xmath412 we obtain @xmath413 $ ] where @xmath414 stands for @xmath415 . in the fixed @xmath1 ensemble , the free energy is @xmath416 where @xmath96 $ ] ( see appendix c ) and the boundaries of the plateau are specified by the equilibrium condition @xmath417 while the end - to - end distance is : @xmath418 $ ] . in this case the @xmath6 approximation leads to ( see appendix d and section ivb ) @xmath419 { \cal l}(\alpha)$ ] . the domain boundaries are specified by @xmath420{\cal l } ( \alpha _ { \pm } ) /4n_{h}$ ] with @xmath421 and the end - to - end distance is : @xmath422 $ ] . to characterize the slope we expand @xmath244 around @xmath250 where @xmath251 . in the immediate vicinity of @xmath250 we have @xmath426 where @xmath427 is a constant . accordingly , @xmath428 $ ] and @xmath429\right\}$ ] leading to @xmath430 . the variational eigenstate @xmath431 $ ] where @xmath289 is the normalization factor . with @xmath432 , the normalization condition @xmath433 yields @xmath434 . since there is no @xmath435 dependence @xmath436 where @xmath437 is the appropriate unit vector . consequently ( [ eq_psi ] ) leads to @xmath438

we choose as the starting point of the present work the fundamental problem of channel simulation . roughly speaking , this problem asks when a communication channel @xmath1 from alice ( a ) to bob ( b ) can be used to simulate another channel @xmath2 , also from a to b ? @xcite this problem has many variants according to the resources available to a and b. in particular , the case when a and b can access unlimited amount of shared entanglement has been completely solved . let @xmath3 denote the entanglement - assisted classical capacity of @xmath1 @xcite . it was shown that , in the asymptotic setting , to optimally simulate @xmath2 , we need to apply @xmath1 at rate @xmath4 @xcite . in other words , the entanglement - assisted classical capacity uniquely determines the properties of the channel in the simulation process . furthermore , even with stronger resources such as no - signalling correlations or feedback , this rate can not be improved otherwise we would violate causality , see @xcite for a discussion . here we are interested in the zero - error setting @xcite . it is well known that the zero - error communication problem is extremely difficult , already for classical channels . indeed , the single - shot zero - error classical communication capability of a classical noisy channel equals the independence number of the ( classical ) confusability graph induced by the channel , and the latter problem is well - known to be np - complete . the behaviour of quantum channels in zero - error communication is even more complex as striking effects such as super - activation are possible @xcite . the most general zero - error simulation problem remains wide open . to overcome this difficulty , many variants of this problem have been proposed . the most natural way is to introduce some additional resources and see how this changes the capacity . indeed , extra resources such as classical feedback @xcite , entanglement @xcite , and even a small ( constant ) amount of forward communication @xcite , have been introduced . it has been shown these extra resources can increase the capacity , and generally simplify the problem . in particular , it was shown that even for classical communication channel , shared entanglement can strictly increase the asymptotic zero - error classical capacity @xcite . however , determining the entanglement - assisted zero - error classical capacity remains an open problem even for classical channels . more powerful resources are actually required in order to simplify the problem . @xcite introduced classical no - signalling correlations into the zero - error communication for classical channels , and showed that the well - known fractional packing number of the bipartite graph induced by the channel , gives precisely the zero - error classical capacity of the channel . previously , it was known by shannon that this fractional packing number corresponds to the zero - error classical capacity of the channel when assisted with a feedback link from the receiver to the sender and when the unassisted zero - error classical capacity is not vanishing @xcite . for general background on graph theory see @xcite , and for `` fractional graph theory '' the delightful book @xcite . another major motivation for this work is to further explore the connection between quantum information theory and the so - called `` non - commutative graph theory '' suggested in @xcite . such a connection has been well - known in classical information theory . in @xcite , shannon realized that the zero - error capacity of a classical noisy channel only depends on the confusability graph induced by the channel . he further pointed out that in the presence of classical feedback , the zero - error capacity is completely determined by the bipartite graph of possible input - output transitions associated with the channel . thus it makes sense to talk about the zero - error capacity of a ( bipartite ) graph . the notion of non - commutative graph naturally occurs when we use quantum channels for zero - error communication . for any quantum channel , the non - commutative graph associated with the channel captures the zero - error communication properties , thus playing a similar role to confusability graph . most notably , this notion also makes it possible to introduce a quantum lovsz @xmath0 function to upper bound the entanglement - assisted zero - error capacity that has properties quite similar to its classical analogue @xcite . very recently , it was shown that the zero - error classical capacity of a quantum channel in the presence of quantum feedback only depends on the kraus operator space of the channel @xcite . in other words , the kraus operator space plays a role that is quite similar to the bipartite graph . now it becomes clear that any classical channel induces a bipartite graph as well as a confusability graph , while a quantum channel induces a non - commutative bipartite graph and a non - commutative graph . the new insight is that we can simply regard a non - commutative ( bipartite ) graph as a high - level abstraction of all underlying quantum channels , and study its information - theoretic properties , not limited to zero - error setting . this leads us to a very general viewpoint : graphs as communication channels . for instance , we can define the entanglement - assisted classical capacity of a non - commutative bipartite graph as the minimum of the entanglement - assisted classical capacity of quantum channels that induce the given kraus operator space . it was shown that this quantity enjoys a number of interesting properties including additivity under tensor product and an operational interpretation as a sort of entanglement - assisted conclusive capacity of the bipartite graph @xcite . it remains a great challenge to find tractable forms of various capacities for non - commutative ( bipartite ) graphs . in this paper we consider a more general class of quantum no - signalling correlations described by two - input and two - output quantum channels with the no - signalling constraints . this kind of correlations naturally arises in the study of the relativistic causality of quantum operations @xcite ; see also the more recent @xcite . distinguishability of these correlations from an information theoretic viewpoint has also been studied @xcite . we provide a number of new properties of these correlations , and establish several structural theorems of these correlations . then we generalize the approach of @xcite to study the zero - error classical capacity of a noisy quantum channel assisted by quantum no - signalling correlations , and the reverse problem of perfect simulation . we show that both problems can be completely solved in the one - shot scenario , revealing some nice structure : 1 . the answers are given by semidefinite programmes ( sdps , cf . @xcite ) ; 2 . at the same time they generalize the results of cubitt _ et al . _ @xcite ; 3 . for the simulation , the question is really how to form a constant channel by a convex combination of the one we want to simulate and an arbitrary other quantum channel , and the number of bits needed is just @xmath5 , where @xmath6 is the probability weight of the target channel in the convex combination ( throughout this paper , @xmath7 denotes the binary logarithm ) ; 4 . for assisted communication , there is an analogous problem of convex - combing a certain channel from b to a which has some kind of orthogonality relation with the given channel from a to b , with another one to form a constant channel . if the target channel has weight @xmath6 , then the number of bits sent is again @xmath5 . most interestingly , the solution to the communication problem only depends on the kraus operator space of the channel , not directly on the channel itself . for the simulation problem , the solution is given by the conditional min - entropy @xcite of the channel s choi - jamiokowski matrix , and is actually additive , thus also gives the asymptotic cost of simulating the channel . if we are interested in simulating the cheapest channel contained in the kraus operator space , we obtain an sdp in terms of the projection of the choi - jamiokowski matrix . both the capacity and the simulation sdps are in general not known to be multiplicative under the tensor product of channels , thus we do not know the optimal asymptotic simulation cost . we then focus on the asymptotic zero - error classical capacity and simulation cost assisted with quantum no - signalling correlations . this requires determining the asymptotic behaviour of a sequence of sdps . in general the one - shot solution does not give the asymptotic result , since the corresponding sdp is not multiplicative with respect to the tensor product of channels . a simple formula for the asymptotic channel capacity remains unknown . however , for the special cases of classical - quantum ( cq ) channels , we find that the zero - error capacity is given by the solution of a rather simple sdp suggested earlier by harrow as a natural generalization of the classical fractional packing number @xcite , which we call _ semidefinite packing number_. this result has two interesting corollaries . first , it implies that the zero - error classical capacity of cq - channels assisted by quantum no - signalling correlations is additive . second , and more importantly , we show that for a classical graph @xmath8 , the celebrated lovsz number @xmath9 @xcite , is actually the minimum zero - error classical capacity of any cq - channel that has the given graph as its confusability graph . in other words , lovsz @xmath0 function is the zero - error classical capacity of a graph assisted by quantum no - signalling correlations . to the best of our knowledge , this is the first information theoretic operational interpretation of the lovsz number since its introduction in 1979 . previously , it was known that it is an upper bound on the entanglement - assisted zero - error classical capacity of a graph @xcite . it remains unknown whether the use of quantum no - signalling correlations could be replaced by shared entanglement . the asymptotic simulation cost for kraus operator spaces associated with cq - channels is rather simpler , and is actually given by the one - shot simulation cost . before we proceed to the technical details , it may be helpful to present an overview of our main results . let @xmath1 be a quantum channel from @xmath10 to @xmath11 , with a kraus operator sum representation @xmath12 where @xmath13 . let @xmath14 denote the kraus operator space of @xmath1 . the choi - jamiokowski matrix of @xmath1 is given by @xmath15 , where @xmath16 and @xmath17 are isomorphic hilbert spaces , @xmath18 ( @xmath19 ) is orthonormal basis over @xmath16 ( @xmath17 , resp . ) , and @xmath20 is the unnormalized maximally entangled state over @xmath21 . recall that @xmath22 . let @xmath23 denote the projection onto the support of @xmath24 , which is the subspace @xmath25 , sometimes called the choi - jamiokowski support of @xmath26 ( or of the channel ) . it is worth noting that many results below can be defined on any matrix subspace @xmath26 , not just those corresponding to a quantum channel @xmath1 . however , we have to make sure that @xmath26 is actually corresponding to some quantum channel @xmath1 . this puts an additional constraint on @xmath26 . more precisely , suppose @xmath27 for some orthonormal basis @xmath28 such that @xmath29 . then we should be able to find a quantum channel @xmath30 such that @xmath31 , for some invertible matrix @xmath32 $ ] and @xmath33 . this is equivalent to @xmath34 if such a positive definite matrix @xmath35 $ ] can not be found , @xmath26 will not correspond to a quantum channel ( nor any @xmath36 for @xmath37 ) . in this case one might still be able to find @xmath38 that is a kraus operator space for some quantum channel @xmath39 . for instance , @xmath40 , and @xmath41 . ( note however that @xmath26 is the limit of the kraus subspaces of genuine quantum channels , namely amplitude damping channels with damping parameter going to @xmath42 ; hence it might still be considered as admissible kraus space of an infinitesimally amplitude damping channel . ) we will , therefore , always assume that @xmath26 corresponds to some quantum channel @xmath1 such that @xmath43 . from now on , any such kraus operator space @xmath26 will be alternatively called `` non - commutative bipartite graph '' in fact , below we shall argue why it is a natural generalization of bipartite graphs . [ capacity-1-shot ] the one - shot zero - error classical capability , ( quantified as the largest number of messages ) , of @xmath1 assisted by quantum no - signalling correlations depends only on the non - commutative graph @xmath26 , and is given by the integer part of the following sdp : @xmath44 where @xmath23 denotes the projection onto the the subspace @xmath45 . + hence we are motivated to call @xmath46 the _ no - signalling assisted independence number of @xmath26 . _ the proof of this theorem will be given in section [ subsec:1-shot - capacity ] , where we also explore other properties of the above sdp . for another direction of investigation , looking at the specific channel @xmath1 with choi - jamiokowski matrix @xmath47 and trying to minimize the error probability for given number of messages , we refer the reader to the recent and highly relevant work of leung and matthews @xcite , where exactly this is done in an environment with free no - signalling resources subject to other semidefinite constraints . it is evident from this theorem that the one - shot zero - error classical capacity of @xmath1 only depends on the kraus operator space @xmath26 . that is , any two quantum channels @xmath1 and @xmath2 will have the same capacity if they have the same kraus operator space . for this reason , we usually use @xmath46 to denote @xmath48 . furthermore , we can talk about the capacity of the kraus operator space @xmath26 directly without referring to the underling quantum channel . notice that a classical channel @xmath49 naturally induces a bipartite graph @xmath50 , where the input and output alphabets @xmath51 and @xmath52 are the two sets of vertices , and @xmath53 is the set of edges such that @xmath54 if and only if @xmath55 . we shall also use the notation @xmath56 if @xmath54 , and @xmath57 otherwise . in this case , we have @xmath58 and our notion @xmath59 generalizes this to arbitrary quantum channels . similarly , the simulation cost is given as follows . [ review - simulation-1-shot ] the one - shot zero - error classical simulation cost ( quantified as the minimum number of messages ) of a quantum channel @xmath1 under quantum no - signalling assistance is given by @xmath60 . here , @xmath61 is the choi - jamiokowski matrix of @xmath1 , and @xmath62 is the conditional min - entropy defined as follows @xcite : @xmath63 for example , the asymptotic zero - error classical simulation cost of the cq - channel @xmath64 and @xmath65 , is given by @xmath66 where @xmath67 is the trace distance between @xmath68 and @xmath69 . this gives a new operational interpretation of the trace distance between @xmath68 and @xmath69 as the asymptotic exact simulation cost for the above cq - channel . since there might be more than one channel with kraus operator space included in @xmath26 , we are interested in the exact simulation cost of the `` cheapest '' among these channels . more precisely , the one - shot zero - error classical simulation cost of a kraus operator space @xmath26 is defined as @xmath70 where @xmath71 means that @xmath59 is a subspace of @xmath26 . then it follows immediately from theorem [ review - simulation-1-shot ] that [ thm : l-1-shot - cost ] the one - shot zero - error classical simulation cost of a kraus operator space @xmath26 under quantum no - signalling assistance is given by the integer ceiling of @xmath72 where @xmath23 is the projection onto the choi - jamiokowski support of @xmath26 . we will prove these two theorems in section [ subsec:1-shot - cost ] . we introduce the asymptotic zero - error channel capacity of @xmath26 by considering the number of bits that can be communicated over @xmath73 copies of the channel @xmath1 , i.e. @xmath74 , having kraus operator space @xmath36 , per channel use as @xmath75 ; we denote it as @xmath76 . likewise , the asymptotic number of bits needed per channel use to simulate @xmath74 as @xmath75 , denoted @xmath77 , and the same minimized over all channels with kraus operator space @xmath36 ( not necessarily product channels ! ) , which we denote @xmath78 . from these definitions , it is clear that they are given by the regularizations of the respective one - shot quantities : @xmath79 the first one because @xmath80 is additive under tensor products . so far , we are unable to determine closed formulas for the latter two in general . interestingly , the special case of @xmath26 coming from a cq - channel can be solved completely . note that if @xmath26 corresponds to a cq - channel @xmath81 , with @xmath82 the projection over the support of @xmath83 , then @xmath26 can be uniquely identified by a set of projections @xmath84 ( up to a permutation over inputs ) . in this case , @xmath85 any such kraus operator space will be called `` non - commutative bipartite cq - graph '' or simply `` cq - graph '' . the case of assisted communication seems very complicated , and most interesting . we show that the zero - error classical capacity of cq - graphs is given by the solution of the following sdp : @xmath86 this number was suggested by harrow as a natural generalization of the shannon s classical fractional packing number @xcite , and we will refer to it as _ semidefinite ( fractional ) packing number _ associated with a set of projections @xmath84 . our result can be summarized as [ thm : cq - capacity ] the zero - error classical capacity of a cq - channel @xmath87 assisted by quantum no - signalling correlations is given by the logarithm of the semidefinite packing number @xmath88 , i.e. , @xmath89 to be precise , @xmath90 the proof of this theorem is given in section [ subsec : asymptotic - capacity - cq ] . the asymptotic zero - error classical simulation cost for cq - graphs @xmath26 is relatively easy and straightforward . indeed , we show that the one - shot zero - error classical simulation cost for cq - channels is multiplicative , _ i.e. _ @xmath91 for cq - graphs @xmath92 and @xmath93 . the equality is proved by simply combining the sub - multiplicativity of the primal sdp and the super - multiplicativity of the dual problem , and then applying strong duality of sdps . it readily follows that the asymptotic simulation cost for any kraus operator @xmath26 corresponding to a cq - channel is given by the one - shot simulation cost , namely @xmath94 it is worth noting that the above equality is valid for many other cases . in particular , it holds when @xmath26 corresponds to a quantum channel that is an extreme point in the set of all quantum channels . it remains unknown whether this is true for more general @xmath26 . as an unexpected byproduct of our general analysis we obtain the following . to understand it , note that any cq - channel @xmath95 naturally induces a confusability graph @xmath8 on the vertices @xmath96 , by letting @xmath97 if and only if @xmath98 , i.e. inputs @xmath96 and @xmath99 are confusable . [ thm : theta ] for any classical graph @xmath8 , the lovsz number @xmath9 @xcite is the minimum zero - error classical capacity assisted by quantum no - signalling correlations of any cq - channel that induces @xmath8 , _ i.e. _ @xmath100 where the minimization is over cq - graphs @xmath26 and @xmath101 is the non - commutative graph corresponding to @xmath8 ( see eq . ( [ sg - graph ] ) in section v ) . in particular , equality holds for any cq - channel @xmath102 such that @xmath103 is an optimal orthogonal representation for @xmath8 in the sense of lovsz original definition @xcite . the above result gives the first clear information theoretic interpretion of the lovsz @xmath0 function of a graph @xmath8 , as the zero - error classical capacity of @xmath8 assisted by quantum no - signalling correlations , in the sense of taking the worst cq - channel with confusability graph @xmath8 . its proof will be given in section [ sec : theta ] . in section [ sec : feasibility ] , we provide results which completely characterize the feasibility ( i.e. , positivity ) of zero - error communication with a non - commutative bipartite graph , or a quantum channel , assisted by no - signalling correlations . finally , in section [ sec : conclusion ] , we conclude and propose several open problems for further study . we will provide an intuitive explanation for how the quantum no - signalling correlations naturally arise from the information - theoretic viewpoint . let @xmath2 and @xmath1 be two quantum channels both from a to b , and assume that a and b can access any quantum resources that can not be used for communicating between them directly . an interesting question is to ask when @xmath1 can exactly simulate @xmath2 ? this class of resources clearly includes shared entanglement , and actually has some other more important members that are the central interest of this paper . we will derive some general constraints that all these resources should satisfy . let us start with the one - shot case , that is , a and b can establish quantum channel @xmath2 by using all `` allowable resources '' and one use of channel @xmath1 . we can abstractly represent the whole procedure as in fig . [ fig : simulation ] ; note that it falls within the formalism of `` quantum combs '' @xcite , but as it is a rather special case we can understand it without explicitly invoking that theory . using another channel @xmath1 just once , and the correlations between a and b ; b ) . this is just an equivalent way to redraw a ) , and we have highlighted all correlations between a an b , and their pre- and/or post- processing as @xmath104.,width=415 ] here is how this simulation works . first , a performs some pre - processing on the input quantum system @xmath105 together with all possible resources at her hand , and outputs quantum system @xmath106 as the input of channel @xmath1 . the channel @xmath1 then outputs a quantum system @xmath107 . b will do some post - processing on this system together with all possible resources he has , and finally generates an output @xmath108 . if we remove @xmath1 , we are left with a network with two inputs @xmath105 and @xmath107 , and two outputs @xmath106 and @xmath108 . clearly this represents all possible pre- and/or post - processing that a and b have done and all resources that are available to a and b. in the framework of quantum mechanics , this network can be formulated as a quantum channel @xmath104 with two inputs and two outputs . thus we can redraw the simulation procedure as fig . [ fig : simulation].b . however , as @xmath1 is the only communicating device from a to b , we must have that @xmath104 can not be used to communicate from a to b. furthermore , the output @xmath106 represents the input of a to the channel @xmath1 , and thus can be uniquely determined by @xmath16 , but not @xmath107 , which is the output of @xmath1 . we will see this constraint is equivalent to b can not communicate to a using @xmath104 . these constraints have led us to a fruitful class of resources that a and b could use in communication . as before , we do nt attempt to solve the most general channel simulation problem . instead , we will focus on two simpler but most interesting cases : i ) . @xmath2 is a noiseless classical channel and @xmath1 is the given noisy channel . the optimal solution to this problem will lead us to the notion of zero - error classical capacity of @xmath1 ; ii ) . @xmath1 is a noiseless classical channel and @xmath2 is the given noisy channel . the optimal solution will lead us to the notion of zero - error classical simulation cost of @xmath2 . in the communication problem , we want to maximize the number of messages we can send exactly by the given channel ; while in the simulation problem , we want to minimize the amount of the noiseless classical communication to simulate the given channel . in the rest of this section , we will study the mathematical structures of quantum no - signalling correlations in detail . as discussed before , quantum no - signalling correlations are linear maps @xmath109 with additional constraints . first , @xmath104 is required to be completely positive ( cp ) and trace - preserving ( tp ) . this makes @xmath104 a physically realizable quantum operation . furthermore , @xmath104 is a to b no - signalling ( a@xmath110b ) . that is , a can not send classical information to b by using @xmath104 . more precisely , for any density operators @xmath111 and @xmath112 , we have @xmath113 or equivalently , @xmath114 likewise , @xmath104 is required to be b to a no - signalling ( b@xmath110a ) . that is , b can not send classical information to a by using @xmath104 . this constraint can be formulated as the following @xmath115 let the choi - jamiokowski matrix of @xmath104 be @xmath116 where @xmath117 is the identity operator over @xmath118 , @xmath119 , and @xmath120 the un - normalized maximally entangled state . we now show that all above constraints on @xmath104 can be easily reformulated into the semidefinite programming constraints in terms of the choi - jamiokowski matrix @xmath121 . for convenience , we often use unprimed letters such as @xmath105 and @xmath107 to denote the quantum systems inputting to quantum channels , and the primed letters @xmath122 and @xmath123 for the reference systems which are isomorphic to @xmath105 and @xmath107 , respectively . the constraints on @xmath104 can be equivalently formulated in terms of @xmath121 as follows : @xmath124 where @xmath51 and @xmath52 are arbitrary hermitian operators , so the transpose is not really necessary . the first two constraints guarantee that @xmath125 corresponds to a cptp map @xmath104 , while the latter two make sure that @xmath125 can not be used for communicating from a to b and b to a , respectively . both constraints need only be verified on a hermitian matrix basis of @xmath122 , @xmath123 , respectively . the key to deriving the above constraints is the following useful fact : @xmath126 where @xmath17 is an isomorphic copy of @xmath16 . it is worth noting that the class of quantum no - signalling correlations is closed under convex combinations . that is , if @xmath127 and @xmath128 are quantum no - signalling correlations and @xmath129 , then @xmath130 are also no - signalling correlations . furthermore , this class is also stable under the pre- or post - processing by a or b. that is , if @xmath104 is a no - signalling correlation from @xmath131 to @xmath132 . then @xmath133 is also no - signalling , where @xmath134 are cptp maps on suitable hilbert spaces . it is instructive to compare the quantum no - signalling correlations and the classical no - signalling correlations . recall that any classical no - signalling correlations can be described as a classical channel @xmath135 with two classical inputs @xmath136 and @xmath137 , and two classical outputs @xmath138 and @xmath139 , where @xmath140 evidently , @xmath141 can also be represented as a quantum channel in the following way @xmath142 one can easily verify the above constraints are exactly the same as treating @xmath141 a quantum no - signalling correlation . from this viewpoint , quantum no - signalling correlations are natural generalizations of their classical correspondings . finally , we would like to mention another interesting fact . that is , any two - input and two - output quantum channel @xmath104 can be reduced to a two - input and two - output classical channel @xmath141 that has the same signalling property by simply doing pre- or post - processing , and all inputs and outputs of @xmath141 are binary , i.e. , if @xmath104 is a to b ( and/or b to a ) signalling then @xmath141 is also a to b ( resp . b to a ) signalling . due to its significance , we formulate it as [ binary - signalling ] for any cptp map @xmath104 from @xmath143 to @xmath132 such that @xmath104 is b to a ( and/or a to b ) signalling , one can obtain a classical channel @xmath135 with all @xmath144 , by doing suitable local pre- and post - processing on @xmath104 , such that @xmath141 is also b to a signalling ( resp . a to b ) . assume @xmath104 is both way signalling ( the case that it is one - way signalling is similar , and in fact simpler ) . then we can find a pair of states @xmath145 and @xmath146 , such that the following two maps @xmath147 and @xmath148 are _ non - constant _ cptp maps : @xmath149 i.e. , we can find another two states @xmath150 and @xmath151 such that @xmath152 by helstrom s theorem , for any two different quantum states @xmath153 , we can find a projective measurement @xmath154 to distinguish the given states with equal prior probably with an average success probability @xmath155 as a direct consequence , the cptp map @xmath156 will produce two different classical binary probability distributions @xmath157 and @xmath158 . we will need this fact below . let @xmath159 and @xmath160 be the measurements to optimally distinguish @xmath161 and @xmath162 , respectively . then we can define four cptp maps as follows : @xmath163 using these as pre- and post - processing on @xmath104 , we obtain the desired channel @xmath164 in @xmath141 , if b inputs @xmath42 or @xmath165 , and a inputs @xmath42 , then by the above construction , a must output two binary probability distributions that are different , hence can be used for signalling from b to a. similarly , if a inputs @xmath42 or @xmath165 , and b inputs @xmath42 , then b must output another two different binary probability distributions that can be used for signalling from a to b. we will establish several structure theorems regarding quantum no - signalling correlations . note that @xmath104 is a two - input and two - output quantum channel . so there are two natural ways to think of @xmath104 according to the relation between the inputs and outputs . the first way is to partition @xmath104 as @xmath166 , so the output of @xmath105 and @xmath107 would be @xmath106 and @xmath108 , respectively ; this is perhaps the most standard way . the second way is to focus on the communication between @xmath16 and @xmath167 and partition @xmath104 as @xmath168 . in this case the output of @xmath105 and @xmath107 will be regarded as @xmath108 and @xmath106 , respectively . this kind of partition is quite useful when we are interested in the communication between @xmath16 and @xmath167 . let us start with the bipartition @xmath166 of @xmath104 . in this case , we can have a full characterization of no - signalling maps . ( bear in mind that we have assumed @xmath104 to be a cptp map in the following discussion ) . [ ns - structure1 ] let @xmath169 be a bipartite cptp map with a decomposition @xmath170 according to bipartition @xmath166 , where @xmath171 , @xmath172 , and @xmath173 are complex numbers . then we have the following : \i ) @xmath104 is b to a no - signalling iff @xmath174 can be chosen as cptp for any @xmath175 . in this case @xmath176 will be also cptp . \ii ) @xmath104 is a to b no - signalling iff @xmath177 can be chosen as cptp for any @xmath175 . in this case @xmath178 will be also cptp . \iii ) @xmath104 is no - signalling between a and b iff both @xmath177 and @xmath174 can be chosen as cptp maps , @xmath179 , and @xmath173 are real for all @xmath175 . the most interesting part is item iii ) . intuitively , any no - signalling correlation @xmath104 between a and b can be written as a real affine combination of product cptp maps @xmath180 . the proofs are relatively straightforward and we simply leave them to interested readers as exercises . now we turn to the bipartition @xmath168 of @xmath104 . we do nt have a full characterization of no - signalling correlations in this setting . nevertheless , a simple but extremely useful class of no - signalling correlations can be constructed using the following facts . [ ns - structure2 ] let @xmath169 be a bipartite cptp map with a decomposition @xmath181 according to bipartition @xmath168 , where @xmath182 , @xmath183 , and @xmath184 are complex numbers . then we have the following : \i ) if @xmath185 is a constant map and @xmath186 is tp for any @xmath175 , then @xmath104 is a to b no - signalling . \ii ) if @xmath187 is a constant map and @xmath188 is tp for any @xmath175 , then @xmath104 is b to a no - signalling . \iii ) if all @xmath188 and @xmath186 are tp and both @xmath185 and @xmath187 are constant maps , then @xmath104 is no - signalling between a and b. for example , we can choose @xmath189 , where @xmath188 and @xmath186 are cptp maps from @xmath190 to @xmath191 and @xmath192 to @xmath193 , respectively , and @xmath194 is a probability distribution . if we further have that @xmath185 and @xmath187 are constant maps , then we obtain a no - signalling correlation @xmath104 . clearly , for any such @xmath104 , neither of @xmath105 and @xmath107 can send classical information to @xmath106 or @xmath108 . to emphasize this special feature , this class of no - signalling correlations are said to be _ totally no - signalling_. we shall see later even this class of correlations could be very useful in assisting zero - error communication and simulation . in particular , we need the following technical result : [ rank2-ns ] let @xmath195 and @xmath196 be two cptp maps from @xmath190 to @xmath191 , and let @xmath197 be two cp maps from @xmath192 to @xmath193 . furthermore , assume there is unique @xmath129 such that @xmath198 is a constant channel . then @xmath199 is a no - signalling correlation if and only if both @xmath200 and @xmath201 are cptp maps , and @xmath202 is a constant map . this can be proven by directly applying the definition of quantum no - signalling correlations . two key points are : first , the uniqueness of @xmath6 such that @xmath198 is constant ; secondly , if @xmath203 for any traceless hermitian operator @xmath51 , @xmath204 is a constant map . a third result about the structure of no - signalling correlations is the following @xcite ( see also @xcite for an alternate proof , and @xcite for more discussions ) : [ ns - structure3 ] @xmath104 is b@xmath110a iff @xmath205 for two cptp maps @xmath206 and @xmath207 , where @xmath208 is an internal memory . this result is quite intuitive . it says that in the case that b can not communicate to a , we can actually have the output of a before we input to b. suppose now we are given a no - signalling map @xmath209 and a cptp map @xmath210 . when @xmath211 and @xmath212 , we can construct a new and unique map @xmath213 by feeding the output @xmath106 of @xmath104 into @xmath1 , and use the output @xmath167 of @xmath1 as the input @xmath107 to @xmath104 ( refer to fig . next we will show how to derive more explicit forms of this map . , width=415 ] let us start by considering the partition according to @xmath166 to look at @xmath104 . if @xmath104 is of product form @xmath214 where @xmath215 and @xmath216 , then clearly the composition must be given by @xmath217 . we can simply extend this construction to the superpositions of these product maps by linearity . i.e. , for a general @xmath218 , we define the new map by composing @xmath1 and @xmath104 as the following @xmath219 and this is clearly well - defined . we can also formulate @xmath2 in another useful way by considering the bipartition @xmath168 . suppose @xmath104 is given in the form of @xmath220 where @xmath182 , @xmath183 . then to compose @xmath104 and @xmath1 , the only thing we need to do is to compose @xmath1 and @xmath186 directly and take trace . that is , we have @xmath221 where the trace of super - operator @xmath222 on @xmath193 is given by @xmath223 for any orthonormal basis @xmath224 on @xmath193 in the sense of @xmath225 . this is just the natural generalization of the trace function of the square matrices to super - operators . it is easy to verify that @xmath226 holds for any @xmath227 and @xmath1 , and the trace is independent of the choice of the orthonormal basis @xmath224 . for instance , for any bipartite classical channel @xmath135 , and another classical channel @xmath228 , we can easily show that the map @xmath229 constructed by composing @xmath141 and @xmath230 is given by @xmath231 which coincides with the discussions in @xcite . note that the above two constructions of the map @xmath2 from @xmath104 and @xmath1 are quite general , and purely mathematical . indeed , we can form a map @xmath2 from any @xmath1 and @xmath104 . an interesting and important question is to ask whether @xmath2 is a cptp map when both @xmath1 and @xmath104 are . it turns out that whenever @xmath1 and @xmath104 are cp maps , @xmath2 should also be cp . this fact is actually a simple corollary of the teleportation protocol . indeed , we can easily see from the fig . 3 that @xmath232(\1_{b_o}\ox\phi_{bb_i'})\},\ ] ] where @xmath211 , and @xmath167 , @xmath107 , and @xmath123 are all isomorphic . the way to get the above equality is to first apply the teleportation protocol to the special case that @xmath233 , and then extend the result to the general case by linearity . the above expression not only provides a proof for the complete positivity of @xmath2 , but also the uniqueness of such @xmath2 . and @xmath104 from the viewpoint of teleportation.,width=340 ] so we have obtained that for any two maps @xmath234 and @xmath235 , we can uniquely construct a map @xmath236 by composing @xmath104 and @xmath1 , and trace out @xmath106 and @xmath107 . an explicit construction can be equivalently given by any of eqs . ( [ connection-1 ] ) , ( [ connection-2 ] ) , and ( [ connection-3 ] ) . furthermore , if @xmath104 and @xmath1 are cp , so is @xmath2 . to guarantee that @xmath2 is tp , we have to put further constraints on @xmath104 , but not on @xmath1 except that it is an arbitrary cptp map . we will show that such constraints are precisely the same as the constraints that b to a no - signalling ( b@xmath110 a ) . let @xmath234 and @xmath235 be two cptp maps , and let @xmath236 be the unique cp constructed by composing @xmath104 and @xmath1 . if @xmath104 is cptp and b to a no - signalling , then @xmath2 is also cptp for any cptp map @xmath1 . conversely , if @xmath2 is cptp for any @xmath1 , then @xmath104 has to be b to a no - signalling . the sufficiency is just a direct corollary of item i ) of proposition [ ns - structure1 ] or proposition [ ns - structure3 ] . the later actually provides a physical realization of this map @xmath2 . this constraint is logically reasonable as b to a no - signalling simply means that the input of the channel @xmath1 can not depend on its output ( otherwise we will have a closed loop , against the causality principle ) . now we will prove that the constraint of no - signalling from b to a is also necessary for @xmath2 to be cptp for any cptp @xmath1 . our strategy is to first prove this fact for two - input and two - output binary classical channel @xmath141 . then we can apply proposition [ binary - signalling ] for the general case . we can actually show that if @xmath141 is signalling from b to a , and the composition of @xmath141 and one - bit noiseless classical channel is not a legal classical channel anymore ( the output for some input is not a legal probability distribution ) . by contradiction , assume that the composition of @xmath104 and any cptp map @xmath1 is always cptp , but that @xmath104 is b to a signalling . applying proposition [ binary - signalling ] , we can find a two - input and two - output binary classical channel @xmath135 such that @xmath141 is constructed from @xmath104 by simply performing pre- and/or post - processing such that i ) @xmath237 ; ii ) @xmath141 is also b to a signalling ; iii ) the composition of @xmath141 and any classical channel @xmath16 to @xmath52 is also a legal classical channel . we will show such @xmath141 does not exist , hence complete the proof . actually , ii ) implies there are @xmath238 , such that @xmath239 w.o.l.g , we can assume @xmath240 , @xmath241 , @xmath242 , and @xmath243 . then the above inequality can be rewritten explicitly as @xmath244 on the other hand , the composition of @xmath141 and a one - bit noiseless classical channel @xmath245 is given by @xmath246 in particular , taking @xmath240 and applying the above inequality , we have @xmath247 this contradicts the fact that @xmath248 is a classical channel . thus , to guarantee that the composition @xmath2 of a two - input and two - output channel @xmath104 and any quantum channel from a to b is always a cptp map , @xmath104 is only required to be b to a no - signalling . the constraint of no - signalling from a to b is another natural requirement as we have assumed that a can not communicate to b directly , and the given channel @xmath1 is the only directed resource from a to b we can use . these considerations naturally lead us to quantum no - signalling correlations . for any integer @xmath249 , we denote by @xmath250 and @xmath251 classical registers with size @xmath249 . we use @xmath252 to denote the noiseless classical channel that can send @xmath249 messages from @xmath16 to @xmath167 , _ i.e. _ @xmath253 , or equivalenyly , @xmath254 if a channel @xmath1 can simulate @xmath255 , we say that @xmath249 messages can be perfectly transmitted by one use of @xmath1 . the problem we are interested in is to determine the largest possible @xmath249 , which will be called one - shot zero - error classical transmission capability of @xmath1 assisted with no - signalling correlations . we restate here the result that we will prove in this subsection . * theorem [ capacity-1-shot ] * _ the one - shot zero - error classical capability ( quantified as the largest number of messages ) of @xmath1 assisted by quantum no - signalling correlations depends only on the non - commutative graph @xmath26 , and is given by the integer part of the following sdp : @xmath44 where @xmath23 denotes the projection onto the the subspace @xmath45 . + hence we are motivated to call @xmath46 the _ no - signalling assisted independence number of @xmath26 . _ _ to make the indices of the choi - jamiokowski matrices more readable , we reserve the unprimed letters such as @xmath105 and @xmath107 for the reference systems ( thus as the indices of choi - jamiokowski matrices ) , and use the primed versions @xmath122 and @xmath123 as the inputs to the quantum channels . we will use a general no - signalling correlation @xmath104 with choi - jamiokowski matrix @xmath125 and the channel @xmath1 , to exactly simulate a noiseless classical channel @xmath255 . our goal is to determine the maximum integer @xmath249 when such simulation is possible . in this case both @xmath256 and @xmath257 are classical , and @xmath258 , @xmath259 . we will show that @xmath125 can be chosen as the following form : @xmath260 where @xmath261 is the choi - jamiokowski matrix of the noiseless classical channel @xmath255 , and @xmath262 and @xmath263 are positive semidefinite operators . we will show that the no - signalling conditions are translated into @xmath264 with some state @xmath265 . in other words , both @xmath266 and @xmath267 are choi - jamiokowski matrices of channels from @xmath11 to @xmath268 whose weighted sum is equal to a constant channel mapping quantum states in @xmath167 to a fixed state @xmath269 . let us first have a closer look at the the mathematical structure of @xmath125 . rewrite @xmath125 into the following form , @xmath270 where @xmath271 is the choi - jamiokowski matrix of the classical channel @xmath272 that sends each @xmath273 into a uniform distribution of @xmath274 . in other words , the no - signalling correlation @xmath104 should have the following form : @xmath275 and @xmath204 and @xmath227 are cp maps from @xmath167 to @xmath16 corresponding to choi - jamiokowski matrices @xmath266 and @xmath267 , respectively . now we can directly apply lemma [ rank2-ns ] to @xmath104 in eq . ( [ simplified - ns ] ) to obtain the no - signalling constraints in eqs . ( [ ns - condition1 ] ) and ( [ ns - condition2 ] ) . first , the constraints for @xmath276 are automatically satisfied due to the special forms of @xmath255 and @xmath272 , and @xmath277 is the unique number such that @xmath278 is constant . the no - signalling constraint @xmath279 is equivalent to that @xmath204 and @xmath227 are cptp maps , and @xmath280 is some constant channel . when composing @xmath1 with @xmath104 in eq . ( [ simplified - ns ] ) , we have the following channel @xmath281 the zero - error constraint requires @xmath282 , which is equivalent to @xmath283 , or @xmath284 where @xmath285 is the choi - jamiokowski matrix of @xmath1 ( strictly speaking , this is the complex conjugate of the choi - jamiokowski matrix of @xmath1 , but this does not make any difference to the problem we are studying ) . since @xmath266 and @xmath267 depend on each other , we can eliminate one of them . for instance , we only keep @xmath266 . then the existence of @xmath267 will be equivalent to @xmath286 by absorbing @xmath249 into @xmath269 and introducing @xmath287 , we get @xmath249 as the integer part of @xmath288 which ends the proof . now we provide a detailed derivation of the form ( [ omega - cns0 ] ) of the no - signalling correlation . assume the channel is @xmath289 , and the no - signalling correlation we will use is @xmath290 . suppose we can send @xmath249 messages exactly by one use of the channel @xmath1 when assisted by @xmath104 . then we should have ( refer to fig . 3 and note @xmath291 , @xmath292 , @xmath293 , @xmath294 , and @xmath295 ) @xmath296=\ketbra{m}{m}_{m_b}\ \forall m\in\{1,\ldots , m\}.\ ] ] the choi - jamiokowski matrix of @xmath104 , say @xmath297 , should satisfy the following constraints @xmath298 furthermore , noticing that @xmath299 we have @xmath300 thus the left hand side ( l.h.s ) of eq . ( [ teleportation ] ) gives , for all @xmath301 , @xmath302=\ketbra{m}{m}_{m_b}.\ ] ] in other words , for any @xmath303 , we have @xmath304 the next crucial step is to simplify the form of @xmath125 . we will study all the possible forms of @xmath125 satisfying the above equation . ( any such operator is said to be feasible ) . since both @xmath250 and @xmath251 are classical registers , we can perform the dephasing operation on them and assume @xmath125 has the following form @xmath305 where @xmath306 might not be identical for different pairs @xmath307 . to further simplify the form of @xmath125 , we next exploit the permutation invariance of @xmath255 . more precisely , for any @xmath308 permutation @xmath309 , if @xmath297 is feasible , then @xmath310 is also feasible . furthermore , if @xmath311 and @xmath312 are feasible , so is any convex combination @xmath313 for @xmath314 . with these two observations , from any feasible @xmath125 that has been dephased , we can always construct a new feasible @xmath315 by performing the following twirling operation @xmath316 by applying schur s lemma to the symmetry group @xmath317 , we can see that @xmath315 can be chosen as the following form @xmath318 which has exactly the same form as eq . ( [ omega - cns0 ] ) . the sdp characterization of theorem [ capacity-1-shot ] , @xmath319 has a dual . it is given by @xmath320 which can be derived by the usual means ; by strong duality , which applies here , the values of both the primal and the dual sdp coincide . for an easy example of the evaluation of these sdps , @xmath321 for the cq - graph @xmath322 of the noiseless classical channel @xmath323 of @xmath324 symbols . the kraus operator space of the noiseless @xmath324-level quantum channel @xmath325 is @xmath326 , and has @xmath327 . by inspection of the primal sdp , for any @xmath92 and @xmath93 , @xmath328 , because the tensor product of feasible solutions of the sdp ( [ eq : upsilon ] ) is feasible for @xmath329 . in particular , we do not know whether equality holds here ( sometimes it does ) ; the most natural way for proving this would be to use the dual sdp ( [ eq : upsilon - dual ] ) , but to use it to show `` @xmath331 '' by tensoring together dual feasible solutions would require that @xmath332 . leaving this aside , this last observation is why @xmath46 is rightfully called the no - signalling assisted independence number , and not its integer part . indeed , the number of messages we can send via @xmath333 is @xmath334 , for non - integer @xmath46 and sufficiently large @xmath324 . for convenience we restate here the two theorems already announced in the introduction . * theorem [ review - simulation-1-shot ] * _ the one - shot zero - error classical simulation cost ( quantified as the minimum number of messages ) of a quantum channel @xmath1 under quantum no - signalling assistance is given by @xmath60 . here , here , @xmath61 is the choi - jamiokowski matrix of @xmath1 , and @xmath62 is the conditional min - entropy defined as follows @xcite : @xmath63 _ * theorem [ thm : l-1-shot - cost ] * _ the one - shot zero - error classical simulation cost of a kraus operator space @xmath26 under quantum no - signalling assistance is given by the integer ceiling of @xmath72 where @xmath23 denotes the projection onto the subspace @xmath45 . _ in this case we have @xmath335 and @xmath336 are classical , @xmath337 , @xmath338 . we will show that w.l.o.g . @xmath339 with positive semidefinite @xmath266 and @xmath267 . then according to lemma [ rank2-ns ] , the no - signalling conditions are equivalent to @xmath340 for a state @xmath341 . now , there are two variants of the problem . first , to simulate the precise channel @xmath1 , @xmath266 has to be equal to @xmath24 . by identifying @xmath342 and eliminating @xmath267 , we get the solution for the minimal @xmath249 as smallest integer @xmath343 proving theorem [ review - simulation-1-shot ] . the latter is then also asymptotic cost of simulating many copies of @xmath230 because the conditional min - entropy is additive , eq . ( [ eq : channel - sim ] ) . furthermore , to simulate the `` cheapest '' @xmath1 with choi - jamiokowski matrix supporting on @xmath23 , we get @xmath344 which is the claim of theorem [ thm : l-1-shot - cost ] . now we provide a more detailed derivation of the form of no - signalling correlations . suppose we can use a noiseless classical channel @xmath255 to simulate a quantum channel @xmath345 . the no - signalling correlation we will use is @xmath346 . then we have @xmath347.\ ] ] denote the choi - jamiokowski matrix of @xmath104 as @xmath348 thus the choi - jamiokowski matrix of @xmath1 is given by @xmath349 \\ & = \sum_m \tr_{m_am_b}[\ketbra{mm}{mm}_{m_am_b}\o_{m_am_bab } ] \\ & = \tr_{m_am_b } d_{m_am_b}\o_{m_am_bab},\end{aligned}\ ] ] where @xmath350 is the choi - jamiokowski matrix of the noiseless classical channel @xmath351 , as before . in summary , to simulate @xmath1 exactly , we have @xmath352 and the no - signalling constraints on @xmath125 . by dephasing and twirling classical registers @xmath250 and @xmath251 ( refer to the case of zero - error communication ) , we can choose w.l.o.g @xmath297 to have the form in eq . ( [ simulation : ns ] ) . we end this subsection , like the previous one , recording the primal and dual sdp form of @xmath353 , again equal by strong duality : @xmath354 its dual sdp is @xmath355 just as for the no - signalling assisted independence number @xmath46 , @xmath356 and hence @xmath357 . for cq - graphs @xmath26 we can say a bit more : since @xmath322 is a cq - graph , too , we find @xmath358 , by proposition [ prop : sigma - cq - multi ] in section [ subsec : asymptotic - cost - cq ] below . for a classical channel with bipartite graph @xmath359 , such that @xmath360 is a special type of cq - graph , it was shown in @xcite that @xmath361 where @xmath362 is the _ fractional packing number _ @xcite ( which is equal to its _ fractional covering number _ ) : @xmath363 in fact , there it was shown ( and one can check immediately from theorems [ capacity-1-shot ] and [ thm : l-1-shot - cost ] ) that in this case @xmath364 furthermore , to attain the simulation cost @xmath77 , as well as @xmath78 , asymptotically no non - local resources beyond shared randomness are necessary @xcite . we will see in the following that quantum channels exhibit more complexity . indeed , while @xmath365 is clearly multiplicative in the channel ( or equivalently in @xmath47 ) , cf . @xcite , and @xmath362 is well - known to be multiplicative under direct products of graphs , cf . @xcite , by contrast @xmath366 and @xmath367 are only super- and sub - multiplicative , respectively : @xmath368 we know that the first inequality can be strict ( see subsection [ 2-state - example ] below ) , and suspect that the second can be strict , too . thus we are facing a regularization issue to compute the zero - error capacity and the zero - error simulation cost , assisted by no - signalling correlations : @xmath369 a standard causality argument , together with entanglement - assisted coding for the simulation of quantum channels @xcite , implies also @xmath370 where @xmath371 is the minimum of the entanglement - assisted classical capacity of quantum channels @xmath1 such that @xmath71 , i.e. , @xmath59 is a subspace of @xmath26 . with the quantum mutual information @xmath372 of the state @xmath373 , where @xmath374 is a purification of @xmath375 : @xmath376 where the last equality follows from the sion s minimax theorem @xcite . for more properties , including the above minimax formulas , and an operational interpretation of @xmath377 as the entanglement - assisted capacity of @xmath26 ( i.e. the maximum rate of block codings adapted simultaneously to all channels @xmath378 such that @xmath379 ) , we refer the reader to @xcite . to put better and easier to use bounds on @xmath380 and @xmath381 , we introduce the semidefinite packing number : @xmath382 which we have given in primal and dual form ; this generalizes the form given in eq . ( [ eq : aram - cq ] ) for cq - graphs . it was suggested to us in the past by aram harrow @xcite for its nice mathematical properties . a slightly modified and more symmetric form is given by @xmath383 again both in primal and dual form . from these it is straightforward to see that @xmath88 and @xmath384 are both sub- and super - multiplicative , and so @xmath385 both definitions reduce to the familiar notion of fractional packing number @xmath362 when @xmath26 is associated to a bipartite graph @xmath359 , coming from a classical channel @xmath1 . furthermore , @xmath88 and @xmath384 are equal for cq - graphs . however , we will see later that in general they are different . for any non - commutative bipartite graph @xmath386 , @xmath387 consequently , @xmath388 . from the sdp ( [ eq : sigma ] ) for @xmath353 , we get operators @xmath389 and @xmath267 such that @xmath390 , with @xmath391 and @xmath392 . hence @xmath393 and so @xmath394 i.e. @xmath395 is feasible for the dual formulation of @xmath384 , hence the result follows . below we will see that this bound ( both in its one - shot and regularized form ) is in general strict , indeed already for cq - channels it is not a strict equality . we do not know whether @xmath46 is in general related to @xmath88 , but we will show bounds in either direction for cq - channels . suppose that the cq - channel @xmath1 acts as @xmath396 , with support @xmath397 and support projection @xmath82 of @xmath83 . then , the non - commutative bipartite graph @xmath26 associated with @xmath1 is given by @xmath398 , and the projection for the choi - jamiokowski matrix is @xmath399 . the sdp ( [ eq : upsilon ] ) easily simplifies to @xmath400 the semidefinite packing number ( [ eq : aram ] ) , on the other hand , simplifies to @xmath401 as this is an sdp relaxation of the problem ( [ eq : upsilon : cq - channel ] ) , we obtain : [ lemma : u - aram - bound ] for a non - commutative bipartite cq - graph @xmath386 , @xmath402 consequently , @xmath403 . in general , @xmath46 can be strictly smaller than @xmath88 , see subsection [ 2-state - example ] below , but we shall prove equality for the regularization , by exhibiting a lower bound @xmath404 the way to do this is to take a feasible solution @xmath405 of the sdp ( [ eq : aram : cq - channel ] ) ; w.l.o.g . its value @xmath406 , otherwise the above statement is trivial . on strings @xmath407 of length @xmath73 this gives a feasible solution @xmath408 for @xmath36 , with value @xmath409 now for @xmath410 different symbols @xmath411 there are at most @xmath412 many types of strings , hence there is one type @xmath413 such that @xmath414 restricting the input of the channel to @xmath415 ( while the output is still @xmath416 ) , we thus loose only a polynomial factor of the semidefinite packing number . what we gain is that the inputs are all of the same type , which means that the output projectors @xmath417 are related to each other by unitaries @xmath418 permuting the @xmath73 @xmath167-systems . note that then also all of the @xmath419 on the left hand side above are the same , say @xmath420 , and hence the left hand side is @xmath421 . abstractly , we are thus in the following situation : assume that there exists a transitive group action by unitary conjugation on the @xmath82 , i.e. we have a finite group @xmath8 acting transitively on the labels @xmath96 ( running over a set of size @xmath230 ) , and a unitary representation @xmath422 , such that @xmath423 for @xmath424 . in other words , the entire set of @xmath84 is the orbit @xmath425 of a fiducial element @xmath426 under the group action . then we can twirl the sdp ( [ eq : aram : cq - channel ] ) and w.l.o.g . assume that all @xmath405 are identical to @xmath427 , so the constraint reduces to @xmath428 , meaning that the largest admissible @xmath427 is @xmath429 , and the semidefinite packing number @xmath430 . from this we see that the representation theory of @xmath422 has a bearing on the semidefinite packing number @xmath88 ; cf . @xcite for some basic facts that we shall invoke in the following . indeed , it also governs @xmath46 since we can do the same twirling operation and find that in the sdp ( [ eq : upsilon : cq - channel ] ) we have w.l.o.g . that all @xmath405 are equal to the same @xmath427 , but also that for @xmath431 , w.l.o.g . @xmath432 . in particular , in this case @xmath433 let us be a little more explicit in the reduction of the sdp ( [ eq : upsilon : cq - channel ] ) to the above sdp : let @xmath427 and @xmath434 be feasible as above , and denote by @xmath435 the subgroup of @xmath8 leaving @xmath42 invariant , @xmath436 . by lagrange s theorem , @xmath437 . then , @xmath438 is also feasible with the same @xmath427 , using @xmath439 for all @xmath440 . letting @xmath441 for any @xmath442 such that @xmath443 , and @xmath444 , then yields a feasible solution for ( [ eq : upsilon : cq - channel ] ) and this is well - defined . thus @xmath46 is not smaller than the above sdp . in the other direction , let @xmath405 and @xmath445 be feasible for ( [ eq : upsilon : cq - channel ] ) , i.e. @xmath446 . letting @xmath447 and @xmath448 yields a feasible solution for ( [ eq : upsilon : covariant - cq - channel ] ) . if the representation @xmath422 happens to be irreducible , we are lucky because then the group average in the second line in eq . ( [ eq : upsilon : covariant - cq - channel ] ) is automatically proportional to the identity , by schur s lemma . hence the optimal choice is @xmath449 and we find @xmath450 . in general this wo nt be the case , but if the representation @xmath422 is `` not too far '' from being irreducible , in a sense made precise in the following proposition , then @xmath46 is not too much smaller than @xmath88 : [ key - lemma ] for a set of projections @xmath82 on @xmath167 with a transitive group action by conjugation under @xmath422 , let @xmath451 be the isotypical decomposition of @xmath167 into irreps @xmath452 of @xmath422 , with multiplicity spaces @xmath453 . denote the number of terms @xmath454 by @xmath455 , and the largest occurring multiplicity by @xmath456 . then , for the corresponding cq - graph @xmath26 , @xmath457 if @xmath458 . assume that we have a feasible @xmath459 for @xmath88 such that @xmath460 . our point of departure is the sdp ( [ eq : upsilon : covariant - cq - channel ] ) : for given @xmath461 and @xmath462 , schur s lemma tells us @xmath463 where @xmath464 is the projection onto the irrep @xmath452 , @xmath465 is a semidefinite operator on @xmath453 . feasibility of @xmath427 and @xmath434 ( to be precise : the equality constraints ) is equivalent to @xmath466 , the projection onto @xmath453 , for all @xmath454 . now , for each @xmath454 choose an orthogonal basis @xmath467 of hermitians over @xmath453 , with @xmath468 and @xmath469 for @xmath470 . then the @xmath471 form a basis of the @xmath422-invariant operators , hence our sdp can be rephrased as @xmath472 notice that here , the semidefinite constraints on @xmath434 leave quite some room , whereas we have `` only '' @xmath473 linear conditions to satisfy . given @xmath459 satisfying the constraint of @xmath88 , our strategy now will be to show that we can construct a @xmath474 such that the above equations are true with @xmath475 on the left hand side , and with a factor @xmath476 on the right hand side . we will choose @xmath477 and thus there is a feasible solution with @xmath478 to ( [ eq : upsilon : covariant - cq - channel-2 ] ) , hence @xmath479 as claimed . in detail , introduce a new variable @xmath480 , with @xmath481 which makes sure that @xmath434 is automatically supported on the orthogonal complement of @xmath426 . rewrite the equations @xmath482 in terms of @xmath51 , introducing the notation @xmath483 this gives , noting @xmath484 for all @xmath96 because of the @xmath422 invariance of @xmath485 , @xmath486 what we need of the coefficients @xmath487 is that they can not be too large : from @xmath488 we get @xmath489 our goal will be to find a `` nice '' dual set @xmath490 to the @xmath491 , i.e. @xmath492 , with which we can write a solution @xmath493 . to this end , we construct first the dual set @xmath494 of the @xmath495 , which is easy : @xmath496 so that indeed @xmath497 . now , consider the @xmath498-matrix @xmath395 , @xmath499 with the deviation @xmath500 here , @xmath501 using @xmath502 , the unitary invariance of the trace norm , and lemma [ lemma : tracenorm - trace ] below . since @xmath503 is invariant under the action of @xmath422 , we have @xmath504 for all @xmath96 , and using @xmath505 we get @xmath506 with this we get that @xmath507 if @xmath508 . assuming @xmath509 ( which will be the case with our later choice ) , we thus know that @xmath395 is invertible ; in fact , we have @xmath510 with @xmath511 , hence @xmath512 and so @xmath513 i.e. , writing @xmath514 we get @xmath515 the invertibility of @xmath395 implies that there is a dual set to @xmath491 in @xmath516 . indeed , from the definition of @xmath517 and the dual sets , @xmath518 now we can finally write down our candidate solution to eqs . ( [ eq : x - and - t ] ) : @xmath519 the remainder term `` rest '' can be bounded as follows : @xmath520 using eqs . ( [ eq : t - bound ] ) and ( [ eq : tilde - delta - bound ] ) . thus we find @xmath521 if @xmath522 and @xmath523 . in this case , @xmath524 and we can wrap things up : @xmath525 satisfies @xmath526 as well as @xmath527 i.e. , we get a feasible solution @xmath528 for @xmath46 . [ lemma : tracenorm - trace ] let @xmath375 be a state and @xmath529 a projection in a hilbert space @xmath530 . then , @xmath531 more generally , for @xmath480 and a povm element @xmath532 , @xmath533 we start with the first chain of inequalities . the left hand one follows directly from the definition of the trace norm . for the right hand one , choose a purification of @xmath534 on @xmath535 . now , @xmath536 and @xmath537 . thus , by the monotonicity of the trace norm under partial trace , @xmath538 the second chain is homogenous in @xmath51 , so we may w.l.o.g . assume that @xmath539 , i.e. @xmath540 is a state . for a general povm element @xmath262 there is an embedding @xmath541 of the hilbert space @xmath530 into a larger hilbert space @xmath542 and a projection @xmath529 in @xmath542 such that @xmath543 . then , @xmath544 as before by the definition of the trace norm , and using the invariance of the trace number under unitaries and the first part , @xmath545 which concludes the proof . for the permutation action of @xmath546 on @xmath416 , the irreps @xmath454 are labelled by young diagrams with at most @xmath547 rows , hence @xmath548 , and it is well - known that @xmath549 ( * ? ? ? 6.2 ) , @xcite . thus the previous proposition yields directly the following result , observing that @xmath455 and @xmath249 are polynomially bounded in @xmath73 , whilst @xmath550 grows exponentially . [ c - q - capacity - aram ] let @xmath26 be a non - commutative bipartite cq - graph with @xmath551 inputs and output dimension @xmath547 . then for sufficiently large @xmath73 , @xmath552 consequently , @xmath553 . lemma [ lemma : u - aram - bound ] and proposition [ c - q - capacity - aram ] together prove theorem [ thm : cq - capacity ] . [ c - q - capacity - additivity ] for any two non - commutative bipartite cq - graphs @xmath92 and @xmath93 , @xmath554 . here we study the asymptotic zero - error simulation cost of a non - commutative bipartite cq - graph @xmath398 , where the subspace @xmath397 is the support of the projection @xmath82 . thus , @xmath399 and the sdp ( [ eq : sigma ] ) easily simplifies to @xmath555 similarly , the dual sdp ( [ eq : sigma - dual ] ) simplifies to @xmath556 [ prop : sigma - cq - multi ] for non - commutative bipartite cq - graphs @xmath26 , @xmath353 is multiplicative under tensor products , i.e. @xmath557 where @xmath92 and @xmath93 are arbitrary non - commutative bipartite cq - graphs . the sub - multiplicativity of @xmath353 is evident from ( [ eq : sigma : cq - channel ] ) . we will show that the super - multiplicativity follows from the dual sdp ( [ eq : dual : cq - channel ] ) . indeed , let @xmath92 and @xmath93 correspond to @xmath84 and @xmath558 , respectively , and assume that @xmath559 and @xmath560 are optimal solutions to @xmath561 and @xmath562 in dual sdps , respectively . then we have @xmath563 where @xmath564 denotes the minimal eigenvalue of the linear operator @xmath565 in the support of @xmath82 . similarly , we have @xmath566 clearly , we have @xmath567 so @xmath568 is a feasible solution to the dual sdp for @xmath569 . since the dual sdp takes maximization , we have @xmath570 from the above result we can read off directly for any non - commutative bipartite cq - graph @xmath386 , @xmath571 in fact , @xmath572 for any two non - commutative bipartite cq - graphs @xmath92 and @xmath93 . this theorem motivates us to call @xmath573 the _ semidefinite covering number _ , at least for cq - graphs @xmath26 , in analogy to a result from @xcite which states that the zero - error simulation rate of a bipartite graph is given by its fractional packing number . note however that while fractional packing and fractional covering number are dual linear programmes and yield the same value , the semidefinite versions @xmath574 are in general distinct ; already in the following subsection [ 2-state - example ] we will see a simple example for strict inequality . for general non - commutative bipartite graph @xmath26 , we do not know whether the one - shot simulation cost also gives the asymptotic simulation cost . however , this is true when @xmath575 corresponds to an extremal channel @xmath576 , which is well - known to be equivalent to the set of linear operators @xmath577 being linearly independent @xcite . actually , in this case , there can only be a unique channel @xmath1 such that @xmath43 , and furthermore , a unique channel @xmath578 such that @xmath579 . hence @xmath580 thus we have the following result : [ extremal - channel - cost ] let @xmath575 be an extremal non - commutative bipartite graph in the sense that @xmath577 is linearly independent . then @xmath581 for the choi - jamiokowski state of the unique channel @xmath1 with @xmath71 . furthermore , @xmath582 if both @xmath92 and @xmath93 are extremal non - commutative bipartite graphs . the fact that the set of extremal non - commutative bipartite graphs has a one - to - one correspondence to the set of extremal quantum channels has greatly simplified the simulation problem . how to use this property to simplify the assisted - communication problem is still unclear . here we will examine our above findings of one - shot and asymptotic capacities and simulation costs for the simplest possible cq - channel , which has only two inputs and two pure output states @xmath583 , w.l.o.g . @xmath584 with @xmath585 . in fact , we shall assume @xmath586 since the two equality cases are trivial ( noiseless classical channel and completely noisy channel , respectively ) . note @xmath587 ; the non - commutative bipartite cq - graph @xmath588 . we can work out all the optimization problems introduced before : @xmath589 eq . ( [ twostate - cq - n - shot ] ) directly gives us @xmath590 since the signal ensemble is symmetric under the pauli @xmath591 unitary ( which exchanges two output states ) , it is easy to evaluate @xmath592 , eq . ( [ twostate - cq - aram ] ) . also @xmath377 is easy to compute , yielding eq . ( [ twostate - cq - cmine ] ) . for pure state cq - channels @xmath1 , by dephasing the input of any channel with kraus operators in @xmath593 , we get a simulation of @xmath1 itself , hence @xmath594 proving eq . ( [ twostate - cq - cost ] ) . noticing that @xmath1 is the unique extremal channel in @xmath26 , we can also apply theorem [ extremal - channel - cost ] to obtain eqs . ( [ twostate - cq - cmine ] ) and ( [ twostate - cq - cost ] ) directly . so we have @xmath595 and both inequalities become strict when @xmath596 ( refer to fig . ( green ) , @xmath597 ( red ) and @xmath381 ( blue ) for the cq - channel of two pure states , as a function of @xmath598.,width=415 ] the largest effort goes into calculating or bounding the numbers @xmath599 in eq . ( [ twostate - cq - n - shot ] ) . in this case we can obtain much better lower bounds ( only at most one bit less than optimal value by @xmath73 uses ) compared to theorem [ thm : cq - capacity ] . _ one copy @xmath600 _ : let @xmath601 be the projection orthogonal to @xmath82 ( which is rank - one ) , so that @xmath602 with a number @xmath603 . because of the @xmath591-symmetry , @xmath604 we can symmetrize any solution and assume @xmath605 and @xmath606 . then the normalization condition reads @xmath607 which implies @xmath608 . hence the maximum value of @xmath609 . _ many copies @xmath610 _ : in this case it is difficult to find the optimum , but it is enough that we show the achievability of @xmath611 for sufficiently large @xmath73 . note that this already implies that the sdp for the zero - error number of messages is not multiplicative ! somehow , what s happening is that the normalization condition of @xmath612 is a non - trivial constraint because @xmath445 has to be supported on the orthogonal complement of @xmath82 ; this hurts us in the case @xmath600 . now in the case of many copies , @xmath82 is a tensor product of single - system projections , hence the orthogonal complement is asymptotically dominating . we have seen how these considerations help in understanding the general case . we have @xmath613 states @xmath614 , indexed by @xmath73-bit strings @xmath615 , which are related by qubit - wise @xmath591-symmetry : @xmath616 which motivates that we find @xmath617 ( orthogonal to @xmath618 ) and define @xmath619 let @xmath620 where @xmath621 is the weight ( the number of @xmath165 s ) of the @xmath73-bit string @xmath615 . we also denote @xmath622 which is always a projection for @xmath623 and vanishing for @xmath624 . for @xmath625 , we have @xmath626 clearly @xmath627 for @xmath628 . now we set all @xmath629 equal to @xmath427 , and propose the following ansatz @xmath630 where @xmath427 and @xmath631 are non - negative eigenvalues to be determined . note that @xmath617 is automatically supported on the orthogonal complement of @xmath618 . we need @xmath632 or equivalently , @xmath633 using the @xmath634 symmetry , we can calculate @xmath635 similarly @xmath636 and @xmath637 applying the property that @xmath638 , we can see that ( * ) is equivalent to @xmath639 and @xmath640 then @xmath641 we need @xmath642 , which is equivalent to @xmath643 , or @xmath644 the first term of the left hand side ( lhs ) of the above inequality achieves the maximum when @xmath645 , and the second term of the lhs reaches the maximum when @xmath646 . so we only need @xmath647 or @xmath648 which is always satisfied when @xmath73 is large enough . note that when @xmath649 , we have @xmath650 the above constraint is tight in the sense @xmath651 if @xmath652 ; otherwise @xmath653 . more precisely , we have @xmath654 when @xmath652 , and @xmath655 when @xmath656 . let us consider now the more general case with two output states @xmath68 and @xmath69 with projections @xmath426 and @xmath657 , respectively . denote @xmath658 @xmath659 is known as the maximal fidelity between @xmath68 and @xmath69 , but depends only on their supports @xmath660 and @xmath92 . a key property of the maximal fidelity is the following @xcite : there exists a cptp map @xmath661 such that @xmath662 and @xmath663 if and only if @xmath664 . applying this result , we can show that the case of two general output states @xmath68 and @xmath69 is simply equivalent to the case of two pure output states @xmath665 and @xmath666 such that @xmath667 . thus we have @xmath668 as we have seen , different non - commutative bipartite graphs @xmath26 , even cq - graphs , having the same confusability graph @xmath8 , can have different assisted zero - error capacity @xmath76 and simulation cost @xmath78 ; c.f . the last subsection [ 2-state - example ] in the previous section . a classical undirected graph is given by @xmath669 , where @xmath670 is the set of vertices , and @xmath671 is the set of edges . as shown in previous work @xcite , @xmath8 is naturally associated with a non - commutative graph , denoted @xmath101 , via the following way : @xmath672 where @xmath673 means confusability : @xmath674 or @xmath675 is an edge of the graph @xcite . hence the questions we are facing are the maximum and minimum @xmath76 and @xmath78 over all cq - graphs @xmath26 with @xmath676 . while the maximum is clearly @xmath677 for both quantities , the minima turn out to be much more interesting . we restate here the main result we will go on to prove in this section . * theorem [ thm : theta ] * _ for any classical graph @xmath8 , the lovsz number @xmath9 is the minimum zero - error classical capacity assisted by quantum no - signalling correlations of any cq - channels that have @xmath8 as non - commutative graph , _ i.e. _ @xmath100 where the minimization is over cq - graphs @xmath26 . + in particular , equality holds for any cq - channel @xmath102 such that @xmath103 is an optimal orthogonal representation for @xmath8 in the sense of lovsz original definition @xcite . _ the proof of this result is achieved by combining two facts about the semidefinite packing number for cq - channels : 1 ) @xmath88 gives the zero - error classical capacity assisted with no - signaling correlations for a cq - channel ; 2 ) the lovsz number @xmath9 of a graph is given by the minimization of @xmath88 for all non - commutative bipartite graph @xmath26 that generate the same confusability graph @xmath8 . the first fact has been proven in theorem [ c - q - capacity - aram ] , so we focus on the second for the rest of the section . let @xmath678 be a kraus operator space with @xmath679 , and let @xmath680 be the non - normalized maximally entangled state . then @xmath529 , the projection on the support of the choi - jamiokowski state , can be written as @xmath681 we can rewrite the semidefinite packing number @xmath88 using these kraus operators @xmath682 : @xmath683 note that @xmath682 spans a valid kraus operator space of a quantum channel . so @xmath684 ( positive definite ) . the dual sdp is @xmath685 and we can easily verify that both the primal and the dual are strictly feasible by choosing @xmath686 and @xmath687 ( here @xmath688 is sufficiently large ) , respectively . hence strong duality holds . we will start by deriving some minimax representations of @xmath88 . let us introduce @xmath689 where @xmath690 and @xmath691 represent the maximal and the minimal eigenvalues of a hermitian operator @xmath395 , respectively , and @xmath692 , @xmath693 are cp maps ( but not necessarily trace or unit preserving ) ; @xmath375 and @xmath265 range over all density operators , i.e. @xmath694 and @xmath695 . in the second and the fourth equalities above , we have employed the following well - known characterizations : @xmath696 in the third equality we have employed the obvious equality @xmath697 , and von neumann s minimax theorem @xcite , since @xmath698 is a linear function with respect to @xmath375 and @xmath265 , and @xmath375 and @xmath265 range over convex compact sets . [ lemma : aram - dual ] under the above definitions , @xmath699 for any non - commutative bipartite graph @xmath26 . suppose that @xmath700 for some @xmath462 . let us construct a density operator @xmath701 . by the assumption @xmath702 , we have @xmath703 or equivalently @xmath704 by the definition of @xmath705 , we have @xmath706 conversely , suppose that @xmath707 for some density operator @xmath68 . then we have @xmath708 , and thus @xmath709 that means @xmath710 is a feasible solution of the sdp defining the semidefinite packing number . by the definition of @xmath88 , we know that @xmath711 concluding the proof . focussing on the special class of cq - channels , which are in some sense the direct quantum generalizations of classical channels , we are now ready for of theorem [ thm : theta ] let @xmath26 correspond to a cq - channel @xmath712 , and let @xmath82 be the projection on the support of quantum states @xmath83 . it will be more convenient to study @xmath705 instead of @xmath88 . actually , applying the above minimax representation to this special case , we have @xmath713 where @xmath714 ranges over probability distributions , and @xmath265 ranges over density operators . the right - most expression motivates us to introduce some notations . by definition , @xmath84 is an orthogonal representation ( or ) of the confusability graph @xmath8 induced by the cq - channel . the value of an or @xmath84 is defined as follows : @xmath715 we introduce the following function of a graph @xmath8 , @xmath716 where the maximization ranges over all possible ors of @xmath8 . clearly , if we require that an or consists of only rank - one projections and @xmath265 takes only rank - one projection , then @xmath717 , the reciprocal of the lovsz number of @xmath8 @xcite . it has been shown in @xcite that even allowing @xmath82 to be general projection but @xmath265 to be rank - one projection , there is no difference between @xmath718 and @xmath719 . however , if @xmath265 is a mixed state , we can only have @xmath720 . interestingly , we can show that equality does hold . in fact , it is evident that if @xmath84 is an or for a graph @xmath8 , then @xmath721 remains an or for the same graph , where @xmath167 is any auxiliary system . now the value of @xmath84 with respect to general mixed states , is the same as the value of @xmath721 with respect to pure states . that is , @xmath722 where @xmath723 . the above equality follows directly from the fact that @xmath724 where @xmath725 is any purification of @xmath265 . summarizing , we have @xmath726 and we are done . as a final comment on the above proof , note that for a fixed or @xmath84 , we can not always choose the optimal handle @xmath265 as a pure state the restriction to rank - one projectors and pure state handle only emerges as we optimize over both elements . we would like to interpret theorem [ thm : theta ] to say intuitively that the zero - error capacity of a graph @xmath8 assisted by no - signalling correlations is @xmath9 . the problematic part of such a manner of speaking is that there are many cq - graphs @xmath26 with the same confusability graph @xmath8 , but the no - signalling assisted capacity may vary with these @xmath26 . however , note that for any ( finite ) family of cq - graphs @xmath727 such that @xmath728 , @xmath729 , we can construct a cq - graph @xmath26 that `` dominates '' all of the @xmath727 in the sense that any no - signalling assisted code for @xmath26 can be used directly for @xmath727 because actually @xmath730 : @xmath731 by going from direct sums to direct integrals , we can thus construct a universal cq - graph @xmath732 which dominates _ all _ @xmath727 with confusability graph @xmath8 , in fact contains them up to isomorphism . any no - signalling assisted code for this object will deal in particular with every eligible channel simultaneously . the only caveat is that the @xmath733 are subspaces in an a priori infinite dimensional hilbert space , and all of our proofs ( in particular that of theorem [ thm : cq - capacity ] ) require finite dimension as a technical condition . we conjecture however that the capacity result of theorem [ thm : cq - capacity ] still holds in that setting . just as we were looking at the smallest zero - error capacity over all cq - channels with a given confusability graph @xmath8 in this section , we can study the minimum simulation cost over all cq - graphs @xmath26 with @xmath734 . to be precise , we are interested in @xmath735 and the asymptotic simulation cost ( regularization ) @xmath736 where @xmath737 denotes the @xmath73-fold strong graph product . the latter limit exists and equals the infimum because evidently @xmath738 . for a cq - graph @xmath26 , we have @xmath739 in fact for every eligible cq - channel @xmath1 with non - commutative bipartite graph @xmath26 , @xmath740 the reason is that @xmath741 is the zero - error capacity assisted by no - signalling correlations ( theorem [ thm : cq - capacity ] ) ; while @xmath742 is the holevo ( small - error ) capacity , which is the same as the entanglement - assisted capacity for cq - channels and which is not increased by any other available no - signalling resources because of the quantum reverse shannon theorem @xcite ; and @xmath743 is the perfect simulation cost of the channel when assisted by no - signalling resources , with the choi - jamiokowski matrix @xmath744 , eq . ( [ eq : g0-equals - sigma ] ) . in eq . ( [ eq : k - chain ] ) , @xmath745 because @xmath26 is a cq - graph , so we only need to consider cq - channels @xmath746 in the minimization , for which @xmath747 letting now @xmath748 we then have the following additivity result . [ lemma : c - min - add : graph ] for any two graphs @xmath8 and @xmath749 , @xmath750 the subadditivity , @xmath751 , is evident from the definition , because if @xmath752 and @xmath753 , then @xmath754 . it remains to show the opposite inequality `` @xmath755 '' . this relies crucially on the minimax identity @xmath756 where the maximum is over probability distributions @xmath757 and the infimum is over cq - channels @xmath1 with confusability graph contained in @xmath8 . this is a special case of sion s minimax theorem @xcite , since the holevo mutual information is well - known to be concave in @xmath757 and convex in @xmath1 , while the domain of @xmath757 is the convex compact simplex of finite probability distributions and the domain of @xmath1 is an infinite - dimensional convex set . now , for a cq - channel @xmath758 with confusability graph contained in @xmath759 , and an arbitrary distribution @xmath760 of the two input variables @xmath51 and @xmath52 , we have @xmath761 here , the first term refers to the cq - channel @xmath762 while the @xmath96-th summand in the second term sum refers to the cq - channel @xmath763 note that @xmath764 is eligible for @xmath8 ( since @xmath765 implies @xmath766 for all @xmath767 , hence @xmath768 ) , while similarly for all @xmath96 , @xmath769 is eligible for @xmath749 . thus , in eq . ( [ eq : chain - rule ] ) , we can take the infimum over eligible cq - channels , to obtain @xmath770 where the minimizations are over cq - channels @xmath1 eligible for @xmath759 , @xmath771 eligible for @xmath8 and @xmath772 eligible for @xmath749 , whereas @xmath773 and @xmath774 are independent copies of @xmath51 and @xmath52 , i.e. they are jointly distributed according to @xmath775 . now , taking the maximum over distributions @xmath760 completes the proof because of the minimax formula ( [ eq : c_min : minimax ] ) . as a corollary , we get the following chain of inequalities : @xmath776 note that it may be true that @xmath777 , but to prove this we would need to show the additivity relation @xmath778 , which remains unknown . another observation is that if in the respective minimizations , the channels are restricted to classical channels , then the results of @xcite show that @xmath779 we now demonstrate by example that the rightmost inequality in ( [ eq : min - over - graphs ] ) can be strict , when quantum channels are considered . namely , for even @xmath73 let @xmath780 the complement of the hadamard graph @xmath781 , whose vertices are the vectors @xmath782 and two vectors @xmath783 are adjacent in @xmath781 if and only if they are orthogonal in the euclidean sense , @xmath784 . in other words , the cq - channel @xmath785 has confusability graph @xmath8 . since the output dimension is @xmath73 , we see from this that @xmath786 , which happens to coincide with the lovsz number , @xmath787 , hence @xmath788 . on the other hand , the clique number of @xmath8 is known to be upper bounded @xmath789 @xcite , hence @xmath790 , meaning @xmath791 . we think that also the leftmost inequality in ( [ eq : min - over - graphs ] ) can be strict . but although the pentagon @xmath792 seems to be a candidate , for which we conjecture ( based on some ad hoc calculations ) that @xmath793 , whereas @xmath794 , a rigorous proof of this has so far eluded us . whether the other two inequalities can be strict remains an open question . given a non - commutative bipartite graph @xmath26 , it is important to know when @xmath26 is able to send classical information exactly in the presence of quantum no - signalling correlations . it turns out that these channels can be precisely characterized ; we will start with cq - graphs . [ feasibility - cq ] let @xmath26 be a non - commutative bipartite cq - graph specified by a set of projections @xmath795 with supports @xmath796 . then the following are equivalent : 1 . @xmath797 ; 2 . @xmath798 ; 3 . @xmath799 . the equivalence of i ) and ii ) follows directly from proposition [ c - q - capacity - aram ] . the equivalence of ii ) and iii ) is only a simple application of the sdp of @xmath88 for bipartite cq - graphs . first , we show that iii ) implies ii ) . by contradiction , assume that the intersection of supports of @xmath82 is empty while @xmath800 . recall that the dual sdp of @xmath88 is given by @xmath801 then @xmath800 implies that we can find @xmath802 such that @xmath803 and @xmath804 for any @xmath96 . clearly @xmath805 is a density operator , and we should have @xmath806 for any @xmath96 . the only possibility is that @xmath805 is in the intersection of the supports of @xmath82 , which is a contradiction . now we turn to show that ii ) implies iii ) . again by contradiction , assume that @xmath798 while the intersection of supports is nonempty . then we can find a pure state @xmath807 from the intersection such that @xmath808 . so @xmath809 thus any feasible solution should have @xmath810 , which indeed implies that @xmath800 . [ feasibility - k ] let @xmath26 be a non - commutative bipartite graph with choi - jamiokowski projection @xmath23 , and let @xmath811 be the orthogonal complement of @xmath23 . then the following are equivalent : 1 . @xmath797 ; 2 . @xmath798 ; 3 . @xmath812 ; 4 . @xmath813 is positive definite . as a matter of fact , we have @xmath814 the meaning of i ) and ii ) are very clear , while iii ) and iv ) need some explanation . essentially , iv ) means we can find a cp map from @xmath167 to @xmath16 with choi - jamiokowski matrix @xmath815 supporting on some subset of @xmath816 . note that in the one - shot sdp formulation of @xmath46 , we need @xmath815 to be a cptp map . however , the trace - preserving condition is not necessary for asymptotic case but only @xmath817 is positive definite , which is the most nontrivial part of this theorem . \iii ) is directly equivalent to iv ) as we have @xmath818 , hence @xmath819 in the following we only focus on i ) , ii ) , and iii ) . the equivalence of ii ) and iii ) is straightforward , simply noticing that @xmath820 is a feasible solution to the primal sdp for @xmath88 , where @xmath821 . the equivalence of i ) and iii ) is much more difficult and non - trivial . in the following we want to explain a little bit more about this equivalence as there are some tricky points . first , let s see how to use iii ) to derive i ) . we can apply the standard super - dense coding protocol , and obtain a cq - channel with @xmath822 outputs @xmath823 , and the projections are given by @xmath824 , where @xmath825 are generalized pauli matrices acting on @xmath16 . so we can compute the semidefinite packing number as @xmath826 this is also the zero - error no - signalling assisted classical capacity of this cq - channel . noticing that when @xmath827 strictly holds , the right - hand side of the above equation is strictly larger than @xmath165 . the fact that i ) implies iii ) can be proven by contradiction together with the one - shot sdp formulation for @xmath46 . assume i ) holds but @xmath828 does not have full rank . then we can find a non - zero vector @xmath829 such that @xmath830 . or equivalently , @xmath831 . \i ) means for some @xmath610 we have @xmath832 . by the one - shot sdp formulation of @xmath599 , we can find positive semidefinite operators @xmath833 and @xmath834 , such that @xmath835 so we can find @xmath836 with @xmath837 , with full rank . on the other hand , we also have @xmath838 supported on @xmath839 , and the later is the summation of product terms such as @xmath840 , containing at least one factor @xmath816 each . so we have @xmath841 which immediately implies that @xmath842 is vanishing on the product vector @xmath843 , contradicting the fact that @xmath844 has full rank . the simple bound @xmath845 is very interesting , and in some important cases it is tight , such as the cq - channels with symmetric outputs , and the class of pauli channels . it would be interesting to know whether this kind of `` entanglement - assisted coding '' could provide a possible way to resolve our puzzle between @xmath76 and @xmath88 , eventually . we have shown that there is a meaningful theory of zero - error communication via quantum channels when assisted by quantum no - signalling correlations . in the terminology of non - commutative graph theory , both the one - shot zero - error classical capacity and simulation cost for non - commutative bipartite graphs assisted by quantum no - signalling correlations have been formulated into feasible sdps . the asymptotic problems for non - commutative bipartite cq - graphs have also been successfully solved , where the capacity turns out to involve a nontrivial regularization of super - multiplicative sdps , which nevertheless leads to another sdp , the semidefinite packing number . we found analogously that the zero - error simulation cost of a cq - graph is given by a semidefinite covering number , which in contrast to the classical case is in general larger than the packing number . the zero - error classical capacity of a classical graph assisted by quantum no - signalling correlations is given precisely by the celebrated lovsz number . for the most general non - commutative bipartite graphs , we are able to provide a necessary and sufficient condition for when these graphs have positive zero - error classical capacity assisted with quantum no - signalling correlations we know rather little about the asymptotic capacity and simulation cost for non - commutative bipartite graphs that are not classical - quantum , however . a very interesting candidate is the non - commutative bipartite graph @xmath846 associated with the amplitude damping channel @xmath847 , with @xmath848 , @xmath849 and @xmath850 . the extremes @xmath851 and @xmath852 correspond to the noiseless qubit channel and a constant channel , respectively , and are trivial ; we will hence assume @xmath853 . this channel is quite interesting because it gives a separation between the semidefinite packing number and the modified version , and other quantities . it is also interesting because there is only one unique channel @xmath854 which has @xmath855 as the kraus operator space . actually it is a non - unital extreme point of the convex set of cptp maps . this channel is also able to communicate classical information without error when assisted with no - signalling correlations . by applying the above super - dense coding bound , we have @xmath856 by some routine calculation , we can show that the semidefinite packing number is @xmath857 , the revised version @xmath858 , and @xmath859 where @xmath860 is the binary entropy . clearly , @xmath861 , while @xmath862 . this means that the revised semidefinite packing number can not be equal to the zero - error classical capacity assisted by quantum no - signalling correlations . whether the original form has the same problem remains unknown . there are many other interesting open problems , of which we highlight a few here . first , it would be very interesting to explore the mathematical structures of quantum no - signalling correlations in greater detail , and to characterize quantitatively how much non - locality is contained in a quantum no - signalling correlation . second , it is of great importance to solve the general asymptotic capacity and cheapest simulation problems in the zero - error setting when assisted with quantum no - signalling correlations . in particular , are there examples of strict sub - multiplicativity of @xmath573 ? the third problem of great interest is to explore the relationship between the quantum lovsz @xmath0 function introduced in ref . @xcite for a non - commutative graph @xmath863 , and the zero - error classical capacity of this non - commutative graph assisted with quantum no - signalling correlations , _ i.e. _ of kraus operator spaces @xmath26 with @xmath864 . likewise , what is the minimum simulation cost @xmath573 or @xmath865 over all such @xmath26 ? this latter is open already for a classical graph @xmath8 and cq - graphs @xmath26 with confusability graph @xmath8 , see the end of section [ sec : theta ] . in any case , it seems that @xmath866 and @xmath867 are two new interesting , multiplicative graph parameters , different from both @xmath9 and @xmath868 . fourth : are no - signalling correlations really necessary to achieve the asymptotic simulation cost @xmath78 , or is perhaps entanglement enough ? the motivation for this question comes from @xcite where it was shown that for classical channels , optimal asymptotic simulation is possible using only shared randomness , no other non - local resources are needed . a nice test case for this question is provided by cq - graphs , for which it is easy to see that @xmath869 . on the other hand , the entanglement - assisted simulation of cq - channels , also known as _ remote state preparation _ @xcite is far harder to understand . indeed , for generic channels the best known protocol requires @xmath870 bits of communication , which is the communication cost of teleportation @xcite . fifth , which no - signalling resources are actually required to achieve the asymptotic zero - error capacity @xmath76 ? this seems an innocuous question after all , for each channel the sdp @xmath871 tells us precisely which no - signalling correlation achieves the maximum . but looking at the case of classical channels and bipartite graphs @xcite , in the light of a classical result of elias @xcite , shows that there only a very specific resource is required , which is universal for all channels . namely , according to ( * ? ? ? 4 ) , a list size-@xmath455 list code for a bipartite graph @xmath359 can achieve a rate @xmath872 , which is arbitrarily close to @xmath873 for sufficiently large @xmath455 . using such a code , alice can send a message @xmath96 out of @xmath874 over @xmath73 uses of the channel , and bob will end up with a list , i.e. a subset @xmath875}$ ] of @xmath876 possible messages such that @xmath877 . to resolve the remaining ambiguity , alice and bob now require a no - signalling resource @xmath878 that we call _ subset correlation _ , defined for @xmath879 $ ] , @xmath880\choose l}$ ] , @xmath881 $ ] , with @xmath882 alice will input @xmath96 into the box , bob the set @xmath883 , and then alice will send her output @xmath884 to bob ; since @xmath877 , bob knows that he can recover the index @xmath427 of @xmath885 as @xmath886 . do these or other universal quantum no - signalling correlations allow to achieve the @xmath76 ( or at least @xmath887 for cq - graphs ) ? finally , in @xcite , the observation was made that the optimal orthogonal representation of a graph @xmath8 in the sense of lovsz ( see theorem [ thm : theta ] ) seems to satisfy @xmath888 , which was confirmed by direct calculation for several graph families . if this held in general , it would mean that @xmath9 can in a certain sense be achieved by a single channel use , rather than requiring a many - copy limit . it is a pleasure to thank many people for their interests and feedback on the present work , including salman beigi , mick bremner , eric chitambar , min - hsiu hsieh , aram harrow , masahito hayashi , richard jozsa , ching - yi lai , debbie leung , will matthews , jonathan oppenheim , xin wang , and stephanie wehner . further thanks are due to giulio chiribella , marco piani , and john watrous for delightful discussions on quantum no - signalling correlations . rd was supported in part by the australian research council ( arc ) under grant dp120103776 ( with aw ) and by the national natural science foundation of china under grants no . he was also supported in part by an arc future fellowship under grant ft120100449 . aw was supported by the european commission ( streps `` qcs '' and `` raquel '' ) , the european research council ( advanced grant `` irquat '' ) and the philip leverhulme trust . furthermore , by the spanish mineco , projects fis2008 - 01236 and fis2013 - 40627-p , with the support of feder funds , as well as the generalitat de catalunya cirit , project no . 2014 sgr 966 . c. h. bennett , i. devetak , a. w. harrow , p. w. shor , a. winter , `` the quantum reverse shannon theorem and resource tradeoffs for simulating quantum channels '' , _ ieee trans . inf . theory _ * 60*(3):2926 - 2959 ( 2014 ) ; arxiv[quant - ph]:0912.5537 . t. s. cubitt , j. chen , a. w. harrow , `` superactivation of the asymptotic zero - error classical capacity of a quantum channel '' , _ ieee trans . inf . theory _ * 57*(12):8114 - 8126 ( 2011 ) . arxiv[quant - ph]:0906.2547 . r. duan , s. severini , a. winter , `` zero - error communication via quantum channels , non - commutative graphs and a quantum lovsz number '' , _ ieee trans . inf . theory _ * 59*(2):1164 - 1174 ( 2013 ) . arxiv[quant - ph]:1002.2514 . d. leung , l. maninska , w. matthews , m. ozols , a. roy , `` entanglement can increase asymptotic rates of zero - error classical communication over classical channels '' , _ commun . phys . _ * 311*:97 - 111 ( 2012 ) . t. eggeling , d. schlingemann , r. f. werner , `` semicausal operations are semilocalizable '' , _ europhys . lett . _ * 57*(6):782 - 788 ( 2002 ) . m. piani , m. horodecki , p. horodecki , r. horodecki , `` properties of quantum nonsignaling boxes '' , _ phys . rev . a _ * 74*:012305 ( 2006 ) . m. hayashi , `` exponents of quantum fixed - length pure state source coding '' , arxiv : quant - ph/0202002 ( 2002 ) . choi , `` completely positive linear maps on complex matrices '' , _ lin . alg . appl . _ * 10*(3):285 - 290 ( 1975 ) . c. h. bennett , g. brassard , c. crpeau , r. jozsa , a. peres , w. k. wootters , `` teleporting an unknown quantum state via dual classical and einstein - podolsky - rosen channels '' , _ phys . * 70*(13):1895 - 1899 ( 1993 ) .

it is well known that , studying the spatial structures and dynamics of globular clusters ( gcs ) is of great importance for understanding both their formation condition and dynamical evolution within the environment of their galaxies @xcite . for example , these clusters are ideal laboratories for detailed studies on two - body relaxation , mass segregation , stellar collisions and mergers , and core collapse @xcite . the correlations of structures with galactocentric distance can provide information on the role of the galaxy tides towards the clusters , while the distribution of ellipticity can shed light on the primary factor for cluster elongation . in addition , comparisons of structures of gcs located in different environment of galaxies offer clues to differences in the early formation and evolution of the galaxies or in their subsequent accretion histories @xcite . the `` fundamental plane '' for clusters in parameter space reflects universal cluster formation conditions , regardless of their host environments @xcite . the structural and dynamical parameters of clusters are often determined by fitting the surface brightness profiles to structure models , combined with mass - to - light ratios estimated from velocity dispersions or population - synthesis models . an accurate and well resolved density profile can be obtained by studying the distribution of integrated light coupling with star counts @xcite . several models are often used in the fits : the empirical , single - mass , modified isothermal spheres @xcite ; the isotropic multi - mass models @xcite ; the anisotropic multi - mass models @xcite ; the power - law surface - density profiles @xcite . the nearest large gc system outside the milky way ( mw ) is that of m31 , with a distance of @xmath1 kpc from us @xcite . it is so close to us that most gcs in it can be well resolved with _ hubble space telescope _ ( _ hst _ ) . @xcite first estimated core radii for several clusters in m31 , and subsequently , a number of studies @xcite focused on the internal structures of m31 gcs , including the core radius , half - light radius , tidal radius , and ellipticity , using the images from large ground - based telescopes and _ @xcite derived structural and dynamical parameters for 34 m31 gcs , and construct a comprehensive catalog of these parameters for 93 m31 gcs with corrected versions of those in a previous study @xcite . combined with the structures and dynamics for clusters from the mw , magellanic clouds , the fornax dwarf spheroidal , and ngc 5128 , these authors found the gcs have near - universal structural properties , regardless of their host environments . @xcite found that bright young clusters in m31 are larger and more concentrated than old ones , and are expected to dissolve within a few gyr and will not survive to become old gcs . with measurements of structural parameters for 13 extended clusters ( ecs ) in the halo regions of m31 , @xcite presented that the faintest ecs have magnitudes and sizes similar to palomar - type gcs in the mw halo . @xcite measured structures and kinematics for 10 newly discovered gcs in the outer halo of m31 , and found that they have larger ellipticities than most of gcs in m31 and the mw , which may be due to galaxy tides from satellite galaxies of m31 or may be related to the merger and accretion history that m31 has experienced . using the same sample clusters in @xcite , @xcite found that some gcs in m31 exhibit cuspy cores which are well described by @xcite models . these authors also confirmed the exist of luminous and compact globulars at large galactocentric radii of m31 , with no counterparts found in the mw . the last three studies extended the structural analysis of clusters in m31 out to @xmath2 kpc , providing important information on the accretion history of m31 outer regions . in this paper , we determined structures and kinematics for 79 clusters in m31 by fitting several structural models to their surface brightness profiles . this paper is organized as follows . in section 2 , we present the _ hst _ observations for the sample clusters , and the data - processing steps to derive the surface brightness profiles . in section 3 , we determine structures and kinematics of the clusters and make some comparisons with previous studies . a discussion on the correlations of the derived parameters is given in section 4 . finally , we summarize our results in section 5 . _ hst _ has imaged a large fraction of globular clusters ( gcs ) in m31 . @xcite used _ hst_/wide field planetary camera 2 ( wfpc2 ) images to measure ellipticities , position angles , and best - fit @xcite model ( hereafter `` king model '' ) parameters for a large sample of m31 gcs . @xcite determined structures and kinematics for m31 gcs by fitting surface brightness profiles to different models , however , for all the gcs but g001 studied in @xcite , only king model was used . in addition , there were some clusters located at the edges of the images or observed with only one filter . with updated observations by _ hst _ , new data can be derived for them now . so , we decided to re - estimate the structure parameters for these gcs in @xcite . however , @xcite determined structures for g001 using three models and new observation , while @xcite determined structure parameters for five clusters ( b315 , b318 , b319 , b368 , and b374 ) using updated _ hst _ data . in addition , there is no new observation for b077 , which is at the edges of the _ hst _ images . we also noticed that bh20 and bh21 had been classified as stars , while bh24 a galaxy @xcite , which would not be discussed again in this paper . the remaining 69 clusters from @xcite would be included in the sample . @xcite estimated metallicities , distance moduli and reddening values for 10 newly - discovered halo gcs in m31 using the acs / wide field camera ( wfc ) images . although @xcite and @xcite determined structures for the 10 gcs , @xcite only used king model , while @xcite presented few structure parameters . so , we re - estimated the structure parameters for these gcs . finally , there are 79 clusters in our sample . we obtained the combined drizzled images from the hubble legacy archive . the images in the bandpass close to @xmath3 band ( f555w or f606w ) and @xmath4 band ( f814w ) were preferred , otherwise the images of f300w , f435w or f475w were selected . the images with high resolution were preferentially adopted ( wfpc2/pc or acs / wfc ) , followed by those obtained with wfpc2/wfc . figure 1 shows the spatial distribution for the sample gcs . the data analysis procedures to measure surface brightness profiles of clusters have been described in @xcite . when the center positions of these clusters were determined using the imcentroid task in iraf , the ellipse task was run in two passes to derive the surface brightness profiles . the ellipse showed inability to converge for several individual clusters . in these cases we first smoothed the images with a boxcar filter @xcite , and then ran the ellipse to derive the density profiles . the overall ellipticity and position angle ( pa ) were determined by averaging the ellipse output in the first pass , with the errors estimated as the standard deviation of the mean . several clusters ( b330 , b468 , bh04 , bh11 , bh29 , and nb39 ) showed errors of p.a.s larger than 15 degrees . we checked the images for these clusters , and found that the random fluctuations due to individual stars @xcite may account for the high errors , leading to a more difficult business for accurate measurements of pas table 1 lists the average ellipticity , p.a . and some additional integrated data for the sample clusters . considering that the metallicities for young ( @xmath5 gyr ) and old clusters are quite different , first , we averaged the ages for the sample 79 clusters from several previous studies @xcite to distinguish young from old clusters . of which , there are 13 clusters ( bh23 , bh29 , nb39 , and the 10 gcs from * ? ? ? * ) with no available age values in the literatures , and we assumed them as old ones . finally , there are six young clusters ( b097d , b324 , bh05 , bh12 , dao38 , and m091 ) in our sample , all of which are from @xcite . for the 69 gcs from @xcite , the metallicities with uncertainties from @xcite were preferentially adopted as our reference , followed by those of @xcite for old clusters , while the solar metallicity was assumed for young clusters @xcite . the reddening values were from @xcite , while the other integrated data were from rbc v.5 @xcite . for the 10 gcs from @xcite , we used the integrated data presented in @xcite , including the @xmath6 magnitudes , galactocentric distances , reddening values , and metallicities ( see * ? ? ? * and references therein ) . in addition , @xmath7 magnitudes for six of the 10 gcs are from rbc v.5 . old clusters with no metallicity measurements are assigned with a mean m31 gc metallicity of [ fe / h]@xmath8 @xcite , while the uncertainties of [ fe / h ] are assumed to be 0.6 as for the standard deviation of the metallicity distribution of the m31 gc system @xcite . clusters with no reddening values are assigned with the galactic reddening in the direction of m31 of @xmath9 @xcite . raw output from package ellipse is in terms of counts s@xmath10 pixel@xmath10 , which needs to multiply by a number ( 400 for acs / wfc , 100 for wfpc2/wfc , and 494 for wfpc2/pc ) to convert to counts s@xmath10 arcsec@xmath10 . a formula was used to transform counts to surface brightness in magnitude calibrated on the vegamag system , @xmath11 as noted by @xcite , occasional oversubtraction of background during the multidrizzling in the automatic reduction pipeline led to `` negative '' counts in some pixels , so we worked in terms of linear intensity instead of surface brightness in magnitude . with updated absolute magnitudes of the sun @xmath12 ( c. willmer , private communication ) listed in table 2 , the equation for transforming counts to surface brightness in intensity was derived , @xmath13 converting from luminosity density in @xmath14 to surface brightness in magnitude was done according to @xmath15 table 2 presents the zeropoints , conversion factors , and coefficients used in these transformations for each filter . table 3 gives the final , calibrated intensity profiles for the 79 clusters but with no extinction corrected . the reported intensities are calibrated on the vegamag scale . column 7 gives a flag for each point , which has the same meaning as @xcite and @xcite defined . the points flagged with``ok '' are used to constrain the model fit , while the points flagged with `` dep '' are those that may lead to excessive weighting of the central regions of clusters ( see * ? ? ? * ; * ? ? ? * for details ) . in addition , points marked with `` bad '' are those individual isophotes that deviated strongly from their neighbors or showed irregular features , which were deleted by hand . the point - spread function ( psf ) models are critical to accurately measure the shapes of objects in images taken with _ hst _ @xcite . @xcite found that by fitting models without psf convolution , the scale radii were systematically larger , and the concentrations smaller than those estimated from the convolved models . compared to ground - based telescopes , the psf of _ hst _ is very stable , although it is also known to vary with time @xcite . tiny tim has been the standard modeling software for _ hst _ psf simulation for 20 years , with a variety of uses ranging from deconvolution , model convolution , psf fitting photometry and astrometry , and psf subtraction @xcite . in this paper , we derived the acs / wfc and wfpc2 psf models using tiny tim , and then the models were fitted using a function of the form @xmath16^{-{\beta}/{\alpha } } , \label{eq : psf}\ ] ] where @xmath17 , @xmath18 , and @xmath19 for each filter are given in table 4 . it can be seen that the parameters in table 4 are slightly different with those from @xcite for acs / wfc in the f606w and f814w filters . @xcite selected a number of isolated stars on a number of images , and combined them to produce a single , average psf for each filter . here we derived a few model psfs at different positions of the camera , and averaged them to produce the final psf for each filter . the discrepancies of these parameters from the two studies are due to the different methods , but are negligible . the psf variation over the cluster extent was ignored since the clusters are small compared to the camera field of view @xcite . we used three structural models to fit star cluster surface profiles , including king model , @xcite , and @xcite model ( hereafter `` wilson model '' and `` srsic model '' ) . @xcite have described the three structural models in detail , here we briefly summarized some basic characteristics for them . king model is most commonly used in studies of star clusters , which is the simple model of single - mass , isotropic , modified isothermal sphere . @xcite found that m31 clusters are better fitted by king models . the phase - space distribution function for king model is defined as @xmath20 - 1 , & & e < 0 , \\ 0 , & & e \geq 0 , \end{array}\right.\ ] ] where @xmath21 is the stellar energy , @xmath22 is a velocity scale . wilson model is an alternate modified isothermal sphere based on the ad hoc stellar distribution function of @xcite . these models have more extended envelope structures than the standard king isothermal spheres @xcite . several studies presented that wilson models fit the majority of gcs in the milky way ( mw ) and some of its satellites and ngc 5128 as well as or better than king models @xcite . the phase - space distribution function for wilson model is defined as @xmath20 - 1+e/{\sigma}_0 ^ 2 , & & e < 0 , \\ 0 , & & e \geq 0 . \end{array}\right.\ ] ] srsic model has a @xmath23 surface - density profile , and has been the standard model for parameterizing the surface brightness profiles of early - type galaxies and bulges of spiral galaxies @xcite . @xcite found that some classical gcs in m31 which exhibit cuspy core profiles are well described by srsic models of index @xmath24 . the clusters with cuspy cores have usually been called post - core collapse ( see * ? ? ? * and references therein ) . the srsic model is defined as @xmath25.\ ] ] before we fitted models to the brightness profiles of the sample clusters , the intensity profiles were corrected for extinction . table 2 lists the effective wavelengths of the acs and wfpc2 filters from the instrument handbook . with the extinction curve taken from @xcite with @xmath26 , we derived the @xmath27 values for each filter . we first convolved the three models with psfs for the filters used . given a value for the scale radius @xmath17 , a dimensionless model profile @xmath28 was computed , and then the convolution was carried out , @xmath29dx^\prime dy^\prime , \label{eq : convol}\ ] ] where @xmath30 , and @xmath31 . @xmath32 was approximated using the equation ( [ eq : psf ] ) ( see * ? ? ? * for details ) . the best fitting model was derived by calculating and minimizing @xmath33 as the sum of squared differences between model and observed intensities , @xmath34 ^ 2}{\sigma_{i}^{2}}},\ ] ] in which a background @xmath35 was also fitted . the uncertainties of observed intensities listed in table 3 were used as weights . as an example , we plotted the fitting for one sample cluster in figure 2 . the observed intensity profile with extinction corrected is plotted as a function of logarithmic projected radius . the open squares are the data points included in the model fitting , while the crosses are points flagged as `` dep '' or `` bad '' , which are not used to constrain the fit @xcite . the best - fitting models , including the king model , wilson model , and srsic model are shown with a solid line from the left to the right panel , with a fitted @xmath35 added . the dashed lines represent the shapes of the psfs for the filters used . there are some clusters showing individual isophotes with ellipse intensities that showed irregular features or deviated strongly from their neighbors . as @xcite reported , such bumps and dips may skew the following model fits . in these cases , we first derived the ellipse output through a boxcar filter to make a smoothed cluster profile @xcite , and then fitted these surface profiles using structural models . if some individual isophotes still can not be well fitted , these points were deleted by hand , which were masked with `` bad '' in table 3 . most profiles of the sample clusters were well fitted by the models , except for several clusters with different reasons . there are one or several bright objects at the intermediate radii of b097d , gc9 , and m091 ; the shape is very loose for dao38 ; the signal - to - noise ratio ( snr ) is low for b205 ; there are one or several bright objects near the outer region of b328 , b331 , and bh11 ; the images are not well resolved for b092 , b101 , b145 , bh05 , and nb39 ; several clusters ( b324 , b330 , b333 , b468 , gc3 , gc7 , bh04 , and bh29 ) are lack of a central brightness concentration , which may be candidates of `` ring clusters '' . the `` ring clusters '' have been reported in the magellanic clouds ( mcs ) and m33 @xcite , with irregular profiles such as bumps and dips which may not be attributed to the random fluctuations due to a few luminous stars . the images of these `` ring cluster '' candidates in our sample are displayed in figure 3 . there are three clusters ( b018 , b114 , and b268 ) showing bumps only in the luminosity profiles of the f814w , with the intrinsic color of @xmath36 0.95 , 1.14 , and 1.2 , respectively . some bright redder stars may locate at the intermediate radii of these clusters . there are 15 clusters ( b124 , b127 , b128 , b148 , b153 , b167 , b268 , b331 , bh05 , bh12 , m091 , nb39 , b064 , b092 , and b205 ) with high errors of @xmath35 . we checked the images carefully , and found that several reasons may account for those high errors : most of these clusters are not well resolved in the images ; there are one or several bright objects in the intermediate and outer part of clusters b331 and m091 ; clusters b064 and b205 only have images in one band ( f300w ) , of which the snrs are low . we should notice that if the fitted background is too high , the tidal radius may be estimated smaller artificially . only one band of observations can be derived for six clusters ( b009 , b020d , b064 , b092 , b101 , b205 ) , and none of the bands is close to @xmath3 band . so , we would not include the six objects in the following discussions . we used the same method mentioned in @xcite and @xcite to transform the magnitudes from acs and wfpc2 to @xmath3 on the vegamag scale @xcite . for clusters with available data of f555w or f606w band , we briefly used the extinction - corrected color @xmath37 or @xmath38 to do the transformations , while for clusters with no data of f555w or f606w band , we first transformed the acs or wfpc2 magnitudes to @xmath4 magnitude using the color @xmath39 , and then computed the @xmath3 magnitude using the color @xmath37 . the @xmath40 magnitudes and the reddening values are listed in table 1 . we estimated a precision of @xmath41 mag in the transformation , which was propagated through the parameter estimates @xcite . the mass - to - light ratios ( @xmath42 values ) , which were used to derive the dynamical parameters , were determined from the population - synthesis models of @xcite , assuming a @xcite initial mass function . the ages and metallicities used to computed @xmath42 values in @xmath3 band were derived as follows . an age of 13 gyr with an uncertainty of @xmath43 gyr was adopted for old clusters , while the derived average ages ( in section [ brightness.sec ] ) were adopted for the young clusters . the metallicities with uncertainties are listed in table 1 . the errors for @xmath42 of old clusters include the uncertainties in age and metallicity , while for young clusters , we simply adopted @xmath44 in @xmath42 as the errors as @xcite did . the basic structures and various derived dynamical parameters of the best - fitting models for each cluster are listed in table 5 to table 7 , with a description of each parameter / column at the end of each table ( see for details of their calculation in * ? ? ? the uncertainties of these parameters were estimated by calculating their variations in each model that yields @xmath33 within 1 of the global minimum for a cluster , while combined in quadrature with the uncertainties in @xmath42 for the parameters related to it ( see * ? ? ? * for details ) . in this paper , we determined structural parameters for 79 clusters in m31 by comparing their surface brightness profiles with three structural models . in figure 4 , some estimated structural parameters for clusters in this paper were compared with those from previous studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite presented structural and dynamical parameters in @xmath3 band for 51 clusters in our sample , while @xcite determined structures for 10 gcs using king model . so we used the results on the bandpass close to @xmath3 band ( e.g. , f475w , f555w , and f606w ) and fitted by king model for comparison . it is not unexpected to see that most of our parameters are larger than the results from @xcite , since the isophotes flagged as `` dep '' in this paper may not be excluded from the fitting process in @xcite , resulting in excessive weighting of the inner regions in the fits . in addition , the ellipticities presented by @xcite were averaged over different filters for each cluster , while our results in figure 4 were on the bandpass close to @xmath3 band . the cluster with the largest discrepancy of ellipticity is bh05 , with an estimate of 0.19 by @xcite , and 0.55 in this paper . as discussed above , the image of bh05 is not well resolved , so it is difficult to derive accurate ellipticities and structural parameters . @xcite also concluded that the shapes of outer parts of faint clusters are strongly affected by the galaxy background or low snrs , leading to a difficult business to accurately measure the ellipticities . the largest scatter of the @xmath45 is b018 , which is @xmath46 pc derived by @xcite , while @xmath47 pc in this paper . although the psfs and some calibration factors adopted here are slightly different from those of @xcite , the parameters derived here are in good agreement with their results . @xcite presented observed velocity dispersions for a number of gcs in m31 using new high - resolution spectra from mmt / hectochelle . these authors estimated the @xmath42 values in @xmath3 band for clusters using the virial masses and luminosities , since the virial masses are the nominal estimates of the `` global '' mass of the system and are less sensitive to the accuracy of the king model fit ( see * ? ? ? * ; * ? ? ? * for details ) . here we estimated the @xmath42 values in @xmath3 band from population - synthesis models by giving their metallicities and various ages @xcite . in figure 5 , we presented the comparison of some parameters derived by @xcite and this paper with those from @xcite . there is a large discrepancy between the @xmath42 values derived from the two methods , and most of the @xmath42 values from population - synthesis models are larger than those from observed velocity dispersions . the model masses from @xcite and this paper are slightly larger than the virial masses from @xcite , which were estimated using the half - mass radius and global velocity dispersion @xmath48 . the @xmath49 values from @xcite were the central velocity dispersions estimated using the observed velocity dispersions by integrating a known king model over the fiber aperture , and are consistent with the predicted line - of - sight velocity dispersions at the cluster center from @xcite and this paper . however , there are few young clusters with velocity dispersion measured by @xcite , so more observation data and analysis are needed to check the conclusion of the comparisons . figure 6 plots the correlations of velocity dispersion and mass with @xmath42 for m31 clusters . the left two panels show these parameters of clusters from @xcite , while the right two panels show those derived by king model from @xcite and this paper . we can see that the @xmath50 and @xmath51 from king - model fits show large dependence on the @xmath42 . the old and young clusters show a distinct boundary of the @xmath42 values . there is no clear correlation for @xmath42 and @xmath49 from @xcite , while the decrease in @xmath42 toward lower masses is expected from dynamical evolution like evaporation , which is the steady loss of low - mass ( high @xmath42 ) stars from the cluster driven by relaxation @xcite . considering the sensitive dependence on @xmath42 of @xmath50 , in the following discussion of the correlations of velocity dispersion with ellipticity and [ fe / h ] , we decided to use the @xmath49 values from @xcite for clusters in @xcite and this paper . in order to determine which model can describe the structure of clusters best , @xcite and @xcite defined a relative @xmath33 index , which compares the @xmath33 of the best fit of an `` alternate '' model with the @xmath33 of the best fit of king model , @xmath52 it is evident that the `` alternate '' model is a better fit than king model if @xmath53 is negative , while king model is a better fit if @xmath53 is positive . figure 7 shows the relative quality of fit , @xmath53 for wilson- and srsic - model fits versus king - model fits for the sample clusters in this paper . the @xmath53 values are plotted as a function of age and some structures , including the half - light radius @xmath45 , the total model luminosity @xmath54 , and the surface brightness over the half - light radius in the @xmath3 band @xmath55 . the circles refer to clusters with @xmath56 , while triangles present those with @xmath57 , where @xmath58 is the most large radius for the available observation data . it can be seen that most clusters with @xmath57 are those having fainter luminosity @xmath54 or surface brightness @xmath55 . @xcite presented that the fitting data out to @xmath56 can effectively constrain the model fitting to the outer regions , which are essential to determine which structural model best describes the clusters . however , when @xmath58 is small ( @xmath59 ) , few data are available at large cluster radii , and the fitting are mostly dependent on the inner part . so , these clusters can not be used to determine a preference of one model or the other . similarly , we can not conclude that these clusters are well fitted by both two models even @xmath53 for them are small . @xcite determined a catalogue of structural and dynamical parameters for gcs in ngc 5128 using the three models , and showed that the bright clusters ( @xmath60 ) are better fitted by wilson model and srsic model , indicating that the halos of clusters in ngc 5128 are more extended than what king model describes . there is no evident correlation of @xmath53 with @xmath54 and @xmath55 for clusters in this paper . several studies @xcite showed that young clusters in the large magellanic cloud ( lmc ) and m33 do not appear to be tidally truncated and seem to be better fitted by power - law profiles than king models , while old clusters show no clear differences between the qualities of the fittings . however , there is no correlation of @xmath53 with age in figure 7 . we should notice that there are only six young clusters in our sample , and 13 clusters with no ages estimated in previous studies are assumed to be 13 gyr . a large sample of young star clusters with precise age estimates are needed for the study on correlation of @xmath53 with age . however , we do see that the king- and wilson - model fits are better than the srsic - model fits . we concluded that clusters in m31 can be well fitted by both king model and wilson model ( also reported in * ? ? ? * ; * ? ? ? figure 8 compares the relative quality of fit , @xmath53 values with a number of structure parameters ( @xmath61 , @xmath45 , @xmath62 , @xmath54 , @xmath50 , and @xmath63 ) for the sample clusters in this paper . the grey and open circles show the physical properties of clusters with @xmath56 derived from wilson- and srsic - model fits comparing to king - model fits , respectively . the triangles refer to clusters with @xmath57 . there are some clusters with comparable @xmath33 , but large discrepancy of @xmath45 and @xmath54 values for king- and wilson - model fits . we can see that most of these clusters have @xmath57 . as discussed above , the few constrain of the fitting to outer regions results in much different extrapolations of models , and it is hard to determine which model does the correct fitting in those cases @xcite . most parameters derived from wilson model are slightly larger than those from king model , while parameters derived by srsic model are smaller than those from king model , especially for @xmath61 , @xmath50 , and @xmath63 . @xcite concluded that rather than an intrinsic difference between clusters in m31 and other galaxies , the preference for king models over wilson models for m31 clusters is due to some subtle features of the observations . we combined the newly derived parameters here with those derived by king - model fits for m31 young massive clusters ( ymcs ) @xcite , m31 globulars @xcite , m31 extended clusters ( ecs ) @xcite , and mw globulars @xcite to construct a large sample to look into the correlations between the parameters . the ellipticities and galactocentric distances for the mw gcs are from @xcite ( 2010 edition ) . the parameters used in the following discussion for m31 gcs @xcite are those derived on the bandpass close to @xmath3 band ( e.g. , f555w and f606w ) , while the data from bluer filters are preferred for the ymcs in @xcite . the metallicities for most young clusters in @xcite were assumed to be solar metallicity , while metallicities from @xcite were adopted for some older clusters ( b083 , b222 , b347 , b374 , and nb16 ) . @xcite derived structure parameters of 13 ecs , including the core radius , half - light radius , tidal radius , and the central surface brightness in magnitude . the metallicities of 4 ecs ( hec4 , hec5 , hec7 and hec12 ) were determined by @xcite , and the integrated cluster masses for them were derived here using the absolute magnitudes and @xmath42 values in @xmath3 band from population - synthesis models ( see * ? ? ? * for the details ) . figure 9 shows structural parameters as a function of galactocentric distance @xmath64 for the new large sample clusters in m31 and the mw . some global trends can be seen . both @xmath45 and @xmath65 increase with the @xmath64 as expected , although the trend is largely driven by the ecs from @xcite . the star clusters with large @xmath64 can keep more stars with weaker tides relatively to those located at bulge and disk . @xcite explained that the @xmath45 and @xmath65 of the clusters in the halo of galaxies , which may be accreted into the galaxies from dwarf galaxies , can expand due to the change from strong to a weaker tidal field . @xcite concluded that the correlation of @xmath45 with @xmath64 reflects physical formation conditions as suggested by @xcite for gcs in the mw . however , @xcite found that no clear correlation between @xmath45 and @xmath64 exists beyond @xmath66 kpc for gcs in ngc 4649 , and they suggested that the sizes of gcs are not generically set by tidal limitation . it can be seen that there are no compact clusters at large radii ( @xmath67 kpc ) in the mw , while in m31 , there is a bimodality in the size distribution of gcs at large radii ( also reported in * ? ? ? * ; * ? ? ? it is interesting that there are few gcs having @xmath45 in the range from 8 to 15 pc at large radii ( @xmath67 kpc ) in m31 . there are three clusters ( b124 , b127 , and nb39 ) with large @xmath45 and @xmath65 at small @xmath64 , leading to more diffuse trends . however , the total model masses for them are @xmath68 , @xmath69 , and @xmath70 , respectively , meaning that they are massive clusters . it is not unexpected that these massive clusters can contain more stars than other clusters , although they are at small @xmath64 . the luminosities of clusters decrease with increasing @xmath64 , implying that either strong dynamical friction drives predominantly more massive gcs inwards , or massive gcs may favor to form in the nuclear regions of galaxies with the higher pressure and density @xcite . however , the lack of faint clusters with small galactocentric distance may be due to selection effects , since these clusters are difficult to detect against the bright background near m31 center , which is also reported by @xcite using the correlation of central surface brightness with galactocentric distance . the metallicities of clusters decrease with increasing @xmath64 , indicating that metal - rich clusters are typically located at smaller galactocentric radii than metal - poor ones , although with large scatter . @xcite found that metal - rich clusters ( [ fe / h]@xmath71 ) in m31 seem to form in a rotating disk extending to @xmath72 kpc . the metal - rich gcs may have undergone accelerated internal evolution due to strong tidal shocks @xcite from both bulge and disk passages . there are two matal - poor clusters ( b114 , b324 ) at small @xmath64 , which are at the bottom - left in the [ fe / h]-@xmath64 panel . the diffuse distribution of these parameters vs. @xmath64 may be caused by the different galactocentric distance used , which is true three - dimensional distances for galactic gcs while projected radii for m31 clusters . figure 10 shows the distribution of ellipticity with galactocentric distance , metallicity , and some structure parameters for clusters in the mw and m31 , which may show clues to the primary factor for the elongation of clusters : rotation and velocity anisotropy , cluster mergers , `` remnant elongation '' , and galaxy tides @xcite . @xcite presented that the tidal forces can only distort a cluster s outer region , while the elongation of the inner parts are due to its own gravitational potential and the total orbital angular momentum . \1 ) rotation and velocity anisotropy . cluster rotation is the generally accepted explanation for cluster flattening @xcite . dynamical models show that internal relaxation coupled to the external tides will drive a cluster toward a rounder shape over several relaxation times ( see * ? ? ? * and references therein ) . @xcite presented that more compact clusters which experience more relaxation processes , and clusters with larger velocity dispersions which rotate more slowly , should be rounder . @xcite presented that dynamical evolution could reduce the initial flattening caused by rotation or velocity anisotropy , indicating that more evolved clusters would be rounder . in order to check these predictions , we show the correlations of @xmath17 , @xmath73 , the observed velocity dispersion @xmath49 , and @xmath74 with the ellipticity . the observed velocity dispersion @xmath49 for m31 clusters are from @xcite , while @xmath49 for clusters in the mw are from @xcite . all the @xmath49 have been extrapolated to their central values with an aperture of 0 . there is no clear correlation of ellipticity with these parameters . however , we do see that more compact clusters smaller @xmath17 and larger @xmath73are more rounder , which is consistent with previous studies @xcite . \2 ) cluster mergers and `` remnant elongation '' . @xcite presented that ellipticity correlates strongly with luminosity for clusters in lmc : more luminous clusters are more flattened , while @xcite presented that the most flattened gcs in both the mw and m31 are also brightest . this may be explained by the cluster mergers and `` remnant elongation '' , which is from some clusters former lives as dwarf galaxy nuclei . so , the larger ellipticities for clusters located at large galactocentric radii of m31 may be related to the merger or accretion history that m31 has experienced ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? . however , no clear correlation of ellipticity with luminosity can be seen . \3 ) galaxy tides . a strong tidal field might rapidly destroy the velocity anisotropies , and force an initially triaxial , rapidly rotating elliptical gc to a more isotropic distribution and spherical shape , while weak tidal fields are unable to change the initial shapes of gcs @xcite . it seems plausible that clusters located at different galactocentric radii or different galaxies have various distributions of ellipticities , due to the diverse galaxy tides . @xcite and @xcite found the distributions of ellipticities for m31 and ngc 5128 are very similar , but differ from the mw distribution , which has few very round clusters . no clear correlation of ellipticity with @xmath64 can be seen for clusters in these galaxies ( also see * ? ? ? * ) . however , the innermost clusters are slightly more spherical @xcite , which may due to the strong tidal field near galaxy center that reduces ellipticities of them . some clusters located at large projected radii of m31 do have larger ellipticities than most gcs in m31 and the mw , which may be caused by galaxy tides coming from satellite galaxies @xcite . @xcite summarized the orientation of a large sample of clusters in m33 and found that the distribution of pas shows a strong peak at @xmath75 , which is close to the direction towards m31 . these authors suggested that the elongation of clusters in m33 may be attributed to the tidal forces of m31 , considering that a recent encounter between m33 and m31 @xcite may have led to significant effects on the properties of m33 disk . figure 11 depicts the distribution of pas of star clusters in this paper and 34 clusters from @xcite , which were not included in @xcite . no clear trend is present in the orientation vectors towards m33 , although a small peak at @xmath76 does exist in the distribution of the pas . here we concluded that the elongation of clusters seems to be due to various factors @xcite . figure 12 plots structural parameters as a function of [ fe / h ] for the new large sample clusters in m31 and the mw . the trends of the parameters with [ fe / h ] nearly disappear when we add the data for the m31 ymcs , which were fitted with solar metallicities @xcite . however , we noticed that the metallicities for most of these ymcs obtained by @xcite are poorer than [ fe / h]@xmath77 . if we do not consider the ymcs in @xcite , it is evident that the metal - rich clusters have smaller average values of @xmath45 than those of metal - poor ones @xcite . @xcite presented that the sizes of metal - rich gcs are smaller than the metal - poor ones in ngc 4649 , which is a massive elliptical galaxy in the virgo galaxy cluster . these authors explained that as an intrinsic size difference rather than projection effects . @xcite carried out n - body simulations of metallicity effects on cluster evolution , and found that there is no evident difference for the half - mass radii of metal - rich and metal - poor cluster models . so , they explained that metal - rich and metal - poor clusters have similar structures , while metallicity effects combined with dynamical effects such as mass segregation produce a larger difference of the half - light radii . @xcite also found that @xmath45 decreases with increased metallicity for gcs in the mw , the mcs , fornax dwarf spheroidal , and m31 , except for gcs in ngc 5128 . an evident feature is that these four ecs are all metal - poor and have large @xmath45 . no clear correlation of model luminosity @xmath54 with [ fe / h ] is present , but we do see that the metal - rich clusters tend to be more luminous . both the observed velocity dispersion @xmath49 and central `` escape '' velocity @xmath78 increase with metallicity . @xcite presented that the higher @xmath79 of more metal - rich clusters in the observation may reflect the metallicity dependence of the terminal velocities of the stellar winds , since the @xmath79 of a metal - rich cluster should be higher to retain the fast stellar winds . core - collapsed clusters in general show a power - law slope in the central surface brightness profiles , which can be better fitted by a power - law model than king model @xcite . @xcite presented that the process of core collapse can be divided into two stages . in the first stage , stars are driven to the halo of the cluster due to close encounters . stellar evaporation occurs and the core contracts due to energy conservation . in the second stage , the low - mass stars are scattered to high velocities and escape to the halo , while the high - mass stars sink to the core due to energy equipartition . the increasing core density also increases the interaction rate of the binaries and single stars , which can generate energy in the core , reverse the contraction process , and produce an expansion . after a long time , the core shrinks again , and the process repeats . m15 , which has a central cusp , may be at an intermediate state between the extremes of collapse and expansion @xcite . @xcite presented that the core - collapsed clusters are those that have reached or are about to reach the `` binary burning '' stage , while the non core - collapsed clusters are still contracting under two - body relaxation . @xcite presented a catalogue of surface brightness profiles of 125 galactic gcs , and classified 16% of their sample as core - collapsed clusters and 6% as core - collapsed candidates . @xcite noticed that a number of clusters ( more than half of their sample ) in ngc 5128 have the index @xmath0 derived from srsic model , indicating that these clusters have strongly peaked central density profiles . these authors explained that the psf may flatten the models , and then the cuspy central density profiles are shown relatively . @xcite presented that galactic clusters with steep cusps are all close to the center of the galaxy , indicating that an increased incidence of tidal shocking @xcite may accelerate the core collapse process . @xcite also concluded that most of the m31 core - collapsed candidates are within 2 kpc of the center of m31 . figure 13 shows the distributions of galactocentric distance , and some structure parameters derived from srsic - model fitting for clusters in this paper and m31 gcs in @xcite . there are 25 clusters bested fitted by srsic model ( @xmath80 and @xmath81 ) , and 11 of them have @xmath0 , indicating that they have cuspy central density profiles . we assumed these 11 clusters to be possible core - collapsed clusters . in fact , the clusters with @xmath0 do have smaller @xmath61 and larger @xmath82 than their counterparts with @xmath83 . @xcite presented several clusters ( b011 , b064 , b092 , b123 , b145 , b231 , b268 , b343 , and bh18 ) as core - collapsed candidates in m31 . however , most of these clusters are best fitted by king model or wilson model in this paper except b145 and b343 . we checked the images of these clusters and found that the images of b123 , b231 , and b268 are not fully resolved , while the snrs of images for b064 and b092 are low . actually , the discrimination between the core - collapsed clusters and the `` king model clusters '' @xcite may often become unclear for several reasons : ( 1 ) statistical noise due to some unresolved bright stars in the cores of clusters ; ( 2 ) the similarity between the high - concentration king - model and power - law profiles ( see * ? ? ? * for details ) . b145 is best fitted by srsic model in this paper , but @xmath84 . the cluster b343 shows a cuspy density profile in the center , and has been classified as a core - collapsed cluster @xcite . we also find a better fitting by srsic model for it with @xmath85 . the distribution of core - collapsed cluster candidates is not constrained to the center of m31 . there are only two out of the 11 clusters with @xmath86 kpc , while four ones with @xmath87 kpc . it has been widely noticed that gcs do not occupy the full four - dimensional parameter space ( concentration @xmath73 , scale radius @xmath17 , central surface brightness @xmath62 , and central @xmath42 or velocity dispersion @xmath88 ) instead locate in a remarkably narrow `` fundamental plane '' ( fp ) . it is interesting to learn which of the structural and dynamical properties are either universal or dependent on characteristics of their parent galaxies @xcite . @xcite explored some tight correlations between various properties using a large sample of galactic clusters , and @xcite defined an fp for gcs in terms of velocity dispersion , radius , and surface brightness . @xcite investigated the mass - binding energy relationship for a few gcs and ellipticals , while @xcite defined an fp using the binding energy and luminosity , which is formally different from that of @xcite . @xcite found that the ngc 5128 gcs describe a relation between binding energy and luminosity tighter than in the mw . here we show the two forms of the fp for the new large sample clusters in the mw and m31 . figure 14 shows the mass - density - based fp relations . the left two panels show the correlations of the properties derived with the @xmath42 values from the velocity dispersions @xcite , using king model and the same fitting process in this paper ( section [ fit.sec ] ) , while the right two panels show the correlations of the properties derived with the @xmath42 values from population - synthesis models . @xcite presented the surface - brightness - based fp relations and found a large offset between the young m31 clusters and old clusters . they explained that as a result from lower @xmath42 values for the young clusters in m31 . it is obvious that the velocity dispersion , characteristic radii , and surface density for these clusters show tight relations , both on the core and half - light scales . the exist of fp for clusters strongly reflects some universal physical conditions and processes of cluster formation . figure 15 shows the correlation of binding energy with the total model mass . the left panel shows the @xmath51 and @xmath63 derived with the @xmath42 from the directly observed velocity dispersions @xcite , using king model and the same fitting process in this paper ( section [ fit.sec ] ) , while the right panel shows those properties derived with the @xmath42 from population - synthesis models . all the clusters locate in a remarkably tight region although in the widely different galaxy environments , which is consistent with the previous studies @xcite . @xcite concluded that the scatter around this relation is so small that the structures of star clusters may be far simpler than those scenarios derived from theoretical arguments . high - resolution imaging can be derived from _ hst _ observations for m31 star clusters . in this paper , we presented surface brightness profiles for 79 clusters , which were selected from @xcite and @xcite . structural and dynamical parameters were derived by fitting the profiles to three different models , including king model , wilson model , and srsic model . we found that in the majority of cases , king models fit the m31 clusters as well as wilson models , and better than srsic models . we discussed the properties of the sample gcs here combined with gcs in the mw @xcite and clusters in m31 @xcite . in general , the properties of the m31 and the galactic clusters fall in the same regions of parameter spaces . there is a bimodality in the size distribution of m31 clusters at large radii , which is different from their galactic counterparts . there are 11 clusters in m31 best fitted by srsic models with index @xmath0 , meaning that they have cuspy central density profiles , which are classified as core - collapsed cluster candidates . we investigated two forms of the fp , including the correlation of velocity dispersion , radius , and surface density , and the correlation of binding energy with the total model mass . the tight correlations of cluster properties indicate a tight fp for clusters , regardless of their host environments , which is consistent with previous studies @xcite . in addition , the tightness of the relations for the internal properties indicates some physical conditions and processes of cluster formation in different galaxies . we thank the anonymous referee for providing a rapid and thoughtful report that helped improve the original manuscript greatly . we would like to thank dr . mclaughlin for his help in deriving the parameters of the three structure models , and richard hook in understanding the tinytim package , and pauline barmby in understanding the psf of _ hst _ , and christopher willmer in providing the updated table of absolute solar magnitudes . this work was supported by the chinese national natural science foundation grands no . 10873016 , and 10633020 , and by national basic research program of china ( 973 program ) , no . 2007cb815403 . cccccccccccc name & @xmath89 & @xmath90 & @xmath91 & @xmath92 & @xmath7 & @xmath3 & @xmath4 & @xmath93 & @xmath94 & @xmath95}$ ] & @xmath96 + & & ( deg e of n ) & & ( deg e of n ) & ( vegamag ) & ( vegamag ) & ( vegamag ) & ( kpc ) & & & ( gyr ) + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11 ) & ( 12 ) + b006 & @xmath97 & @xmath98 & @xmath99 & @xmath100 & 16.49 & 15.50 & 14.33 & 6.43 & 0.11 & @xmath101 & 13.25 + b011 & @xmath102 & @xmath103 & @xmath104 & @xmath105 & 17.46 & 16.58 & 15.56 & 7.73 & 0.09 & @xmath106 & 15.85 + b012 & @xmath107 & @xmath108 & @xmath97 & @xmath109 & 15.86 & 15.09 & 14.03 & 5.78 & 0.11 & @xmath110 & 8.00 + b018 & @xmath111 & @xmath112 & @xmath113 & @xmath114 & 18.37 & 17.53 & 16.38 & 9.31 & 0.20 & @xmath115 & 1.20 + b027 & @xmath116 & @xmath117 & @xmath118 & @xmath119 & 16.49 & 15.56 & 14.41 & 6.02 & 0.19 & @xmath120 & 15.75 + b030 & @xmath121 & @xmath122 & @xmath123 & @xmath124 & 18.75 & 17.39 & 15.68 & 5.66 & 0.57 & @xmath125 & 9.62 + b045 & @xmath102 & @xmath126 & @xmath127 & @xmath128 & 16.72 & 15.78 & 14.51 & 4.90 & 0.18 & @xmath129 & 11.40 + b058 & @xmath130 & @xmath131 & @xmath132 & @xmath133 & 15.81 & 14.97 & 13.87 & 6.96 & 0.12 & @xmath134 & 8.01 + b068 & @xmath135 & @xmath136 & @xmath116 & @xmath137 & 17.60 & 16.39 & 14.84 & 4.32 & 0.42 & @xmath138 & 7.90 + b070 & @xmath102 & @xmath139 & @xmath127 & @xmath140 & 17.61 & 16.76 & 15.72 & 2.46 & 0.12 & @xmath141 & 8.75 + ccccccc filter & pivot @xmath142 & @xmath143 & zeropoint@xmath144 & @xmath145 & conversion factor@xmath146 & coefficient@xmath147 + & ( ) & & & & & + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) + + f435w & 4318.9 & 4.20 & 25.779 & 5.459 & 3.1693 & 27.031 + f475w & 4746.9 & 3.72 & 26.168 & 5.167 & 1.6926 & 26.739 + f555w & 5361.0 & 3.19 & 25.724 & 4.820 & 1.8508 & 26.392 + f606w & 5921.1 & 2.85 & 26.398 & 4.611 & 0.8207 & 26.183 + f814w & 8057.0 & 1.83 & 25.501 & 4.066 & 1.1349 & 25.638 + + f300w & 2986.8 & 5.66 & 21.448 & 6.061 & 297.9614 & 27.633 + f450w & 4555.4 & 3.93 & 24.046 & 5.263 & 13.0545 & 26.835 + f555w & 5439.0 & 3.14 & 24.596 & 4.804 & 5.1542 & 26.376 + f606w & 5996.8 & 2.81 & 24.957 & 4.581 & 3.0100 & 26.153 + f814w & 8012.2 & 1.85 & 23.677 & 4.074 & 6.1342 & 25.646 + cccccccc name & detector & filter & @xmath148 & @xmath4 & uncertainty & flag + & & & ( arcsec ) & ( @xmath149 pc@xmath150 ) & ( @xmath149 pc@xmath150 ) & + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) + b006 & wfpc2/pc & f555w & 0.0234 & 44001.773 & 155.603 & ok + & wfpc2/pc & f555w & 0.0258 & 43776.562 & 171.971 & dep + & wfpc2/pc & f555w & 0.0284 & 43528.629 & 188.743 & dep + & wfpc2/pc & f555w & 0.0312 & 43258.270 & 206.935 & dep + & wfpc2/pc & f555w & 0.0343 & 42964.875 & 227.137 & dep + & wfpc2/pc & f555w & 0.0378 & 42647.070 & 249.659 & dep + & wfpc2/pc & f555w & 0.0415 & 42303.609 & 274.847 & dep + & wfpc2/pc & f555w & 0.0457 & 41919.195 & 299.011 & dep + & wfpc2/pc & f555w & 0.0503 & 41381.914 & 308.833 & ok + & wfpc2/pc & f555w & 0.0553 & 40666.176 & 309.009 & dep + & wfpc2/pc & f555w & 0.0608 & 39884.902 & 322.641 & dep + & wfpc2/pc & f555w & 0.0669 & 38966.234 & 328.825 & dep + & wfpc2/pc & f555w & 0.0736 & 37925.949 & 316.808 & ok + lccccc detector & filter & @xmath17 & @xmath18 & @xmath19 + & & ( arcsec ) & & + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) + acs / wfc & f435w & 0.068 & 3 & 3.80 + & f475w & 0.064 & 3 & 3.60 + & f555w & 0.057 & 3 & 3.39 + & f606w & 0.053 & 3 & 3.14 + & f814w & 0.056 & 3 & 3.05 + wfpc2/wfc & f300w & 0.076 & 2 & 5.05 + & f450w & 0.073 & 2 & 4.89 + & f555w & 0.064 & 2 & 4.35 + & f606w & 0.059 & 2 & 4.11 + & f814w & 0.051 & 2 & 3.71 + wfpc2/pc & f300w & 0.051 & 2 & 3.76 + & f555w & 0.045 & 2 & 3.10 + & f606w & 0.045 & 2 & 2.96 + & f814w & 0.059 & 2 & 3.18 + cccccccccccc name & detector & band & @xmath151@xmath152 & model & @xmath153@xmath144 & @xmath35@xmath154 & @xmath155@xmath146 & @xmath156@xmath147 & @xmath157@xmath158 & @xmath159@xmath160 & @xmath159@xmath161 + & & & & & & ( @xmath149 pc@xmath150 ) & & & @xmath162 & ( arcsec ) & ( pc ) + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11 ) & ( 12 ) + b006 & wfpc2/pc & 555 & 49 & k66 & 35.44 & @xmath163 & @xmath164 & @xmath165 & @xmath166 & @xmath167 & @xmath168 + & & & 49 & w & 5.53 & @xmath169 & @xmath170 & @xmath171 & @xmath172 & @xmath173 & @xmath174 + & & & 49 & s & 130.08 & @xmath175 & @xmath176 & @xmath177 & @xmath178 & @xmath179 & @xmath180 + b006 & wfpc2/pc & 814 & 49 & k66 & 14.97 & @xmath181 & @xmath182 & @xmath183 & @xmath184 & @xmath185 & @xmath186 + & & & 49 & w & 1.88 & @xmath187 & @xmath188 & @xmath189 & @xmath190 & @xmath191 & @xmath192 + & & & 49 & s & 32.24 & @xmath193 & @xmath176 & @xmath194 & @xmath195 & @xmath196 & @xmath197 + cccccccccccccc name & detector & band & model & @xmath198@xmath152 & @xmath199@xmath144 & @xmath200@xmath154 & @xmath201@xmath146 & @xmath202@xmath147 & @xmath203@xmath158 & @xmath204@xmath160 & @xmath205@xmath161 & @xmath206@xmath207 & @xmath55@xmath208 + & & & & ( pc ) & ( pc ) & ( pc ) & & ( @xmath209 pc@xmath150 ) & ( @xmath209 pc@xmath210 ) & ( @xmath209 ) & ( mag ) & ( @xmath209 pc@xmath150 ) & @xmath162 + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11 ) & ( 12 ) & ( 13 ) & ( 14 ) + b006 & wfpc2/pc & 555 & k66 & @xmath211 & @xmath212 & @xmath213 & @xmath214 & @xmath215 & @xmath216 & @xmath217 & @xmath218 & @xmath219 & @xmath220 + & & & w & @xmath221 & @xmath222 & @xmath223 & @xmath224 & @xmath225 & @xmath226 & @xmath227 & @xmath228 & @xmath229 & @xmath230 + & & & s & @xmath231 & @xmath180 & @xmath232 & @xmath233 & @xmath234 & @xmath235 & @xmath236 & @xmath237 & @xmath238 & @xmath239 + b006 & wfpc2/pc & 814 & k66 & @xmath240 & @xmath241 & @xmath242 & @xmath243 & @xmath215 & @xmath244 & @xmath217 & @xmath218 & @xmath245 & @xmath246 + & & & w & @xmath247 & @xmath248 & @xmath249 & @xmath250 & @xmath225 & @xmath226 & @xmath227 & @xmath228 & @xmath251 & @xmath252 + & & & s & @xmath231 & @xmath197 & @xmath253 & @xmath254 & @xmath234 & @xmath255 & @xmath236 & @xmath237 & @xmath256 & @xmath257 + cccccccccccccc name & detector & band & @xmath258@xmath152 & model & @xmath259@xmath144 & @xmath260@xmath154 & @xmath261@xmath146 & @xmath262@xmath147 & @xmath263@xmath158 & @xmath264@xmath160 & @xmath265@xmath161 & @xmath266@xmath207 & @xmath267@xmath208 + & & & ( @xmath268 ) & & ( @xmath12 ) & ( erg ) & ( @xmath12 pc@xmath150 ) & ( @xmath12 pc@xmath210 ) & ( @xmath12 pc@xmath150 ) & ( km s@xmath10 ) & ( km s@xmath10 ) & ( yr ) & ( @xmath12 ( pc km s@xmath269 + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11 ) & ( 12 ) & ( 13 ) & ( 14 ) + b006 & wfpc2/pc & 555 & @xmath270 & k66 & @xmath271 & @xmath272 & @xmath273 & @xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 & @xmath279 + & & & @xmath280 & w & @xmath281 & @xmath282 & @xmath273 & @xmath283 & @xmath284 & @xmath285 & @xmath286 & @xmath287 & @xmath288 + & & & @xmath280 & s & @xmath289 & @xmath290 & @xmath291 & @xmath292 & @xmath293 & @xmath294 & @xmath295 & @xmath296 & @xmath297 + b006 & wfpc2/pc & 814 & @xmath270 & k66 & @xmath271 & @xmath298 & @xmath273 & @xmath299 & @xmath300 & @xmath301 & @xmath302 & @xmath303 & @xmath304 + & & & @xmath280 & w & @xmath281 & @xmath305 & @xmath273 & @xmath283 & @xmath306 & @xmath285 & @xmath286 & @xmath307 & @xmath288 + & & & @xmath280 & s & @xmath289 & @xmath308 & @xmath291 & @xmath309 & @xmath310 & @xmath311 & @xmath312 & @xmath313 & @xmath314 +

the physics of many 1 + 1 dimensional lattice models at criticality is captured in the continuum limit by two - dimensional conformal field theory ( 2d cft ) . as a result , one can numerically extract the conformal data , for instance scaling dimensions of the primary operators from the scaling of the two - point correlation functions in the critical lattice model , see @xcite for the standard reference . on the other hand , measures of entanglement proved to be useful quantities for exploring the critical points of many - body systems . for example , in the ground state of a critical chain , the entanglement entropy of a block of spins has a universal logarithmic scaling with the size of the block that is proportional to the central charge @xcite . this provides an efficient numerical way to obtain the central charge of the cft that governs the critical point , see refs . @xcite for review . it is then natural to ask if entanglement measures in excited states can be used to extract more cft data , such as for instance the modular s or t matrices or quantum dimensions , numerically . in rational cfts local elementary excitations are catalogued into finite number of conformal families containing primary operators and their descendants . the simplest excited states can then be obtained by inserting local cft operators at some spatial points . in such states , one can study how the local operator changes the structure of entanglement in the ground state . more precisely , it is possible to compute the time evolution of the change in rnyi entropies for a reduced density matrix of a single interval due to the operator insertion @xcite see ref . @xcite for some results in various cft setups and ref . @xcite for entanglement in a related class of globally excited states . in 2d cft this analysis can be preformed analytically and rnyi entropies detect an increase in entanglement equal to the logarithm of the quantum dimension of the conformal family @xcite . having such a clear and elegant prediction from the cft , it is then natural to wonder if and how the logarithms of quantum dimensions are reproduced on the lattice . in this article we initiate such program for the simplest case of the critical ising chain . the advantage of the ising model is that it is exactly solvable and the action of a family of local operators can be efficiently simulated for large system sizes . the main subtle point of this analysis is the identification between the cft operators and their lattice counterpart . in fact , there are only few models where such map is well established see for instance the discussion in @xcite and a general belief is that a given lattice operator corresponds in the continuum to a primary operator plus its descendants . in the case of the two - point functions these extra contributions from descendants lead to corrections that are suppressed as higher powers with the distance . in this work , given a well established identification for the ising model operators , we will be able to check the contribution form this non - unique identification to the physics of entanglement propagation . we will see that for truly local operators on the lattice like the ising spin we recover the cft answer , but for operators with non - local support like ising energy the subleading contributions modify the leading answer and lead to a mismatch . the computations of rnyi entropies in cft are done using the replica method that , for excited states , boils down to calculation of correlation functions on complicated riemann surfaces . even for states locally excited by more than a single operator such objects are notoriously difficult to compute analytically and features of entanglement measures in this class of states remain unexplored . similarly , measures of distance between quantum states like , e.g. , relative entropy for locally excited states require the access to higher - point correlators @xcite . in this work , we will further explore the ising model to shed a new light in these directions by numerically performing the time evolution of the relative entropy , as well as rnyi entropies in more general states excited by multiple local operators . this paper is organized as follows . in section [ sec : cft ] we summarize the relevant results from the two dimensional cft . in section [ sec : num ] we present our main numerical results for the evolution of entanglement in the critical ising model in states excited by single local operators . in section [ sec : gen ] we consider the evolution away from the critical point , as well as more general operator excitations . in section [ sec : rel ] we present the evolution of the relative entropy after local operator excitation in this model . finally , we conclude and present the details of our numerical approach in the appendix [ numer ] . in this section we briefly review the existing results for the evolution of rnyi entropies in locally excited states in 2d cfts and then show some details of the computation for the ising cft . finally , we discuss the minor modifications that appear for the cfts on the cylinder that , in the following sections , we will be comparing to numerics from the periodic chain . let us start with a 2d cft on the real line and a state excited by a local operator @xmath0 at distance @xmath1 from the entangling interval @xmath2 $ ] , as presented pictorially on fig . [ fig : schematic ] . the density matrix is given by ( t)&=&ne^-ite^-(0,-l)|00|^(0,-l)e^-e^it & & n(w_2,|w_2)|00| ^(w_1,|w_1 ) , [ denmat ] where the insertion points of the local operators are defined as & & w_1=i(-it)-l , w_2 = -i(+it)-l , + & & |w_1=-i(-it)-l , |w_2=i(+it)-l . [ points ] the factor of @xmath3 is the uv regulator for the local operators and we take @xmath4 at the end of the computation . the normalization @xmath5 ensures that the trace of the density matrix is equal to 1 . + using the replica trick , we can compute how a family of rnyi entropies @xmath6 , indexed by integer @xmath7 , changes due to the local operator insertion . the answer to this question is expressed in terms of the logarithm of the ratio @xcite s_a^(n ) , [ eq : delta ] where the correlator in the numerator is computed on the @xmath7-sheeted surface @xmath8 with cuts on each copy corresponding to interval @xmath9 , and the two - point function in the denominator is on a single sheet @xmath10 with an interval cut @xmath9 . for a detailed derivation and further illustrative explanations see @xcite . in 2d cft , one can apply a conformal map from @xmath8 to a complex plane and evaluate the correlators explicitly . it turns out that the answer is universal and the increase in the rnyi entanglement entropies is equal to a constant that is the same for all the members of a conformal family , i.e. primary operators and their descendants @xcite . in rational cfts this constant is equal to the logarithm of the quantum dimension of the local operator @xcite which is defined as d_a=,[qdim ] where @xmath11 denotes the elements of the modular s - matrix of the cft , see e.g. ref . @xcite . in this work we focus on the 2d ising model which is the @xmath12 minimal model with three primary operators : the identity @xmath13 , the energy @xmath14 with conformal dimensions @xmath15 @xmath16 and the spin @xmath17 with @xmath18 @xmath19 , where @xmath20 marks the total scaling dimensions . + the modular s - matrix of the ising model is given by s= ( ccc 1 & 1 & + 1 & 1 & - + & - & 0 + ) , so the three quantum dimensions are d_=d_=1,d_=. this way , at criticality , only excitations with primary @xmath17 can non - trivially change the entanglement in the vacuum state , and for all the rnyi entropies we have @xcite s^(n)_a=. the standard , chiral ( anti - chiral ) descendants are obtained by either acting on the primary operators with chiral ( anti - chiral ) derivatives @xmath21 ( @xmath22 ) or taking the operator product expansion ( ope ) of the primary operators with the energy momentum tensor . such descendants increase the entropies by the same amount as the primaries . if we however act with the linear combination of the two derivatives , there is an additional contribution to the entropy equal to @xmath23 @xcite . we will see this in case of the spatial derivative @xmath24 acting on @xmath25 , which we consider in sec . [ sec : num ] c. .,width=264 ] in order to compare the cft results with numerics we have to take into account finite size of the system , @xmath26 . in the cft computation it enters through the invariant cross - ratios . let us , for simplicity , consider the change in the second renyi entropy @xmath27 that requires the correlator on two cylinders . the correlators entering are computed using a composition of the conformal map @xmath28 from each cylinder to the plane with a cut and the uniformization map @xmath29 . after some standard cft manipulations , the change in the entropy can be written as s^(2)_a=-(|z(1-z)|^4h_^_(z,|z)),[secren ] where @xmath30 is the canonical 4-point function on the complex plane with operators @xmath31 inserted at @xmath32 and the cross - ratios are defined as z=,|z= , with @xmath33 , @xmath34 and similarly for @xmath35 . from the conformal map , we can also show that @xmath36 and @xmath37 ( similarly for @xmath35 ) . once we plug the insertion points of the operators , in the @xmath4 limit , the cross - ratios become periodic functions of time , as shown on fig . [ fig : cr1 ] . the difference with the cft on the infinite line is that in one cycle of time equal to @xmath26 , both @xmath38 and @xmath35 reach their maximal value of @xmath39 . more precisely , @xmath40 in the time @xmath41=[l+\alpha n , l+l+\alpha n]$ ] and zero outside , whereas @xmath42 inside @xmath43=[n-(l+l)+\alpha n , n - l+\alpha n]$ ] and zero outside , for integer period @xmath44 . in order to extract the increase in the second rnyi entropy analytically in the @xmath4 limit , we simply take the limit of @xmath45 or @xmath46 in in the appropriate time intervals . let us focus on the explicit example of the ising model . for the @xmath17 operator , from the fusion rule @xmath47 , the correlator can be decomposed as @xcite ^_(z,|z)=(c^1_)^2|^_(1|z)|^2+(c^_)^2|^_(|z)|^2 , with the three - point coefficients @xmath48 and @xmath49 , as well as the conformal blocks ^_(1|z)=,^_(|z)=. we can then check that that the non - zero contribution in the two time intervals comes from the identity block and is equal to @xmath50 , in accordance with the modular s - matrix elements . this behavior is naturally explained from the quasi - particle picture where left and right moving sets of quasi - particles propagate from the insertion point of the operator and , on the circle , there are two time intervals where either left or right particles are inside the entangling interval @xmath9 . on the other hand , for the @xmath14 excitation , using the fusion @xmath51 , we can write the correlator as ^_(z,|z)=(c^1_)^2|^_(1|z)|^2 , with @xmath52 and conformal block ^_(1|z)=. clearly , inserting the cross - ratios and taking @xmath4 yields @xmath53 for all times . this suggests that the quasiparticles produced by @xmath14 are in a product state . after this short review , below we perform the numerical analysis and compare how the cft predictions are reproduced on the discrete chain . we consider quantum ising model in a transverse magnetic field on a 1d chain of @xmath26 spins-@xmath54 described by the hamiltonian h=-_n=1^n , [ eq : spin_ising ] where @xmath55 are the standard pauli matrices acting on @xmath7-th spin and we assume periodic boundary conditions @xmath56 . the model is critical for @xmath57 and unless stated otherwise , in this article we set @xmath58 . the model can be mapped onto the free fermion system using the jordan - wigner transformation and our numerical results are performed in such a setup . we refer to the appendix [ numer ] for details . after the mapping , the hamiltonian can be diagonalized as @xcite h=_k _ k ( _ k^_k - 12 ) , [ eq : h_ising ] where @xmath59 are the fermionic annihilation operators . the ground state of @xmath60 is the vacuum state annihilated by all @xmath59 and the dispersion relation reads _ k = 2 , [ eq : epsilon_k ] where the quasi - momenta @xmath61 take discrete values @xmath62 $ ] . even at the critical point , for @xmath58 , the dispersion relation is not strictly linear for all values of @xmath63 $ ] due to the discrete nature of the system , approaching linear behaviour only in the limit of long - wavelength @xmath64 . as a result , the velocity of the quasiparticles depends on the momentum @xmath61 , especially for larger values of @xmath65 in contrast with what is the case for cft . in order to account for that , as well as for comparisons with cft , we simultaneously consider the model with linearized dispersion relation fixing the velocity of quasiparticles , h_lin=_k ^lin_k ( _ k^_k - 12 ) , where @xmath66 and @xmath59 are the annihilation operators diagonalizing ising hamiltonian in eq . . in our conventions , @xmath67 is the velocity of quasiparticles at the critical point for that hamiltonian in the long - wavelength limit of @xmath68 , but we will present our plots appropriately rescaled for comparisons with cft . next , we study the evolution of rnyi entanglement entropies in states locally excited by operators on the lattice . more precisely , we excite the ground state of the critical ising model with local operators @xmath69 , which can have support on more then one lattice site , and perform a unitary time evolution . we then numerically calculate the entropy of a block @xmath9 of @xmath70 consecutive spins at a distance @xmath1 from the excitation for different times and subtract from it the entropy of the block with no excitation . this setup is schematically illustrated in fig . [ fig : schematic ] . the lattice operators that have @xmath17 and @xmath3 fields as their leading contributions in the continuum are @xcite ( n)&=&^x_n , + ( n)&=&^x_n ^x_n+1- ^z_n . moreover , to confront the cft predictions for other members of a given conformal family , we also consider the simplest descendant , namely the spatial discrete derivative of the @xmath17 field d(n)=^x_n+1 - ^x_n . that corresponds to descendant @xmath71 . + in the following sections , we discuss the results for classes of states locally excited by these three different operators . the results for @xmath72 excitation are collected in fig . [ fig : x ] . column ( a ) shows the change in entropy for fixed @xmath70 and @xmath1 . we observe the expected plateau appearing at a time when @xmath73 and subsequent plateaus when the signal enters the block @xmath9 from the opposite side or after making some number of circle around the chain . the plateau is however oscillating slightly and then vanishing in a long tail for @xmath74 . the mean values in the first plateau are @xmath75 and @xmath76 , close to the expected value of @xmath77 . both the oscillations and the tail visible for the first plateau are mostly independent of the system size . when we use linearized ising hamiltonian @xmath78 ( dashed lines ) in place of ising hamiltonian @xmath60 ( solid lines ) both the oscillations and the tails disappear and fully periodic structure consistent with the cft prediction is recovered . this supports the natural interpretation that the tail is related to different velocities of excited quasiparticles which naturally appear for local hamiltonian on a chain and in this case are smaller then @xmath79 . notice also that the dashed line has smooth edges , what can be interpreted as manifestation of the nonzero @xmath3 on the lattice , compare with fig . [ fig : cr1 ] . finally , for @xmath60 , the subsequent plateaus appearing for longer times are visibly shifted up , as the signal is slowly dissipating . column ( b ) shows the change of entropy when both @xmath1 and @xmath70 are proportional to the system size @xmath26 and the time is properly rescaled . this validates that the obtained value of @xmath80 is independent of the block size in contrast to the subtracted background entropy of the block without excitation which in this case grows logarithmically with @xmath81 . additionally , we observe that after the rescaling the tails for different @xmath81 are collapsing at least up to corrections which are not visible in this scale and for short enough times which suggests that the characteristic time scale at which the tails are disappearing is proportional to the block size . finally , in column ( c ) we place the excitation symmetrically with respect to the block , so that the quasiparticle with the same absolute momentum traveling left and right should be entering the block at the same time . even in this setup @xmath80 acquires large non - zero value when the fastest quasiparticles traveling with @xmath79 reach block @xmath9 from both sides . this shows that the simplest interpretation valid for global ( translationally invariant ) quench that only the pairs of quasiparticles with opposite momenta @xmath82 contribute to the entanglement is no longer valid for local excitation which breaks translational invariance . this signal almost disappears when @xmath78 is used and all the quasiparticles enter or leave the block in the same instance of time . we refer for a consistent discussion in the context of local quenches to @xcite and evolution of the negativity to @xcite . the results for @xmath83 excitation are presented in fig . [ fig : e ] . for fixed @xmath1 and @xmath70 in column ( a ) we obtain non - zero signal with sharp peaks when the signal is first entering and leaving the block , which then disappears in a long tail . when the evolution is governed by @xmath78 the tails are not present and we recover the periodic structure of plateaus , which however acquire non - zero values of @xmath84 , and @xmath85 . interestingly , in this case , there are still additional sharp peaks when the signal is entering or leaving the block , which are similar to the structure obtained from cft with small but finite @xmath3 . we are however not able to explain the value of @xmath80 in the plateaus as the logarithm of the quantum dimension of @xmath14 that should be zero . we conclude that the sub - leading contributions to @xmath83 in the continuum turn out to be more relevant than for @xmath72 . at the same time the observed results suggest that the quasiparticles with large @xmath61 significantly contribute to the observed value of entanglement , which might be related here with the fact that @xmath83 is supported on two lattice sites . the stability of the obtained signal is validated in column ( b ) where we set @xmath70 and @xmath1 proportional to system size @xmath26 , and after properly rescaling of the time we observe the collapse of @xmath80 for different values of @xmath26 . finally , in ( c ) we show that the non - zero value of @xmath80 obtained in the plateaus for @xmath78 can be also recovered from the evolution governed by the original ising hamiltonian @xmath60 when the block @xmath9 is placed just next to excitation . this way difference in velocity of excited quasiparticles turn out to be unimportant for the first plateau as ( almost ) all right - moving quasiparticles are able to enter the block and the situation resembles that for @xmath78 . this suggests that similar strategy can be used in general spin chains , which can not be mapped onto system of free fermions which we use to construct @xmath78 . such systems can be conveniently simulated using the toolbox of matrix product states ( mps ) @xcite , where , for instance , specific algorithms to study the dynamics of localized excitations in infinite systems have been put forward @xcite . the results for the discrete spatial derivative of @xmath72 are collected in fig . [ fig : dx ] . we note that , on the contrary to the two previous cases , the operator @xmath87 is not unitary , so even at @xmath88 it changes the entropy of a block supported on sites other then @xmath7 and @xmath89 . we however see that this effect is relatively small and vanishing with increasing @xmath1 . in column ( a ) , for fixed @xmath90 and @xmath70 , we observe that @xmath80 does not form plateaus and then disappears in a long tails , which suggests that the slower quasiparticles with large @xmath61 significantly contribute to the observed signal . this is further corroborated by using @xmath78 which allows to recover the periodic structure of plateaus with the values at the peaks close to @xmath91 predicted by cft . similar observation also hold in column ( b ) for @xmath1 and @xmath70 proportional to the system size @xmath26 . finally , in column ( c ) , similarly to the situation in the previous section , we observe that for local ising hamiltonian @xmath60 we are able to recover the structure of the first plateau with the peak value @xmath92 if the block @xmath9 is placed just next to the excitation and @xmath93 . in this section we present numerical results that , in principle , could be reproduced from cft , but in practice the analytical computations become very difficult . in such cases numerics is a great tool for understanding the phenomenology of entanglement evolution and we explore it below . since we successfully recovered the cft predictions for the spin @xmath17 operator , we will mostly consider states excited by applying @xmath94 to the lattice sites , but we believe that the universal features of our analysis remain valid for other conformal families . + we begin with evolution of entanglement of a block of spins after exciting the ground state by local operator but in the non - critical model . having the diagonalized ising hamiltonian for any value of the parameters , we can study the evolution of the rnyi entropies away from the critical point . in fig . [ non - crit ] we plot the evolution of the second rnyi entropy after acting with @xmath72 for a few values of @xmath95 around the critical one . clearly , in each case the entropy only changes after time of order @xmath1 , however the clear plateau with the logarithm of the quantum dimension only appears at criticality . this , in principle , might serve as a proxy for seeing the critical point with local excitations but in general the value of the plateau might be a more complicated expression in terms of the quantum dimensions of the model . figure [ non - crit ] also indicates that the critical hamiltonian leads to the smallest increase of entanglement of the block after we excite a vacuum by a local operator . based on our limited analysis , it is not obvious if this lower bound is universal and satisfied by general families of hamiltonians that have a critical point in their parameter space . it would be very interesting to provide further checks using other available methods like e.g. mps in order to support this observation or possibly find counter - examples . + excitation for different values of the magnetic field @xmath95 . the distance of excitation from the block is ( a ) @xmath93 and ( b ) @xmath96 . @xmath97 , @xmath98.,width=529 ] next , by using the map to free fermions we can also study more general excited states by acting with multiple local operators on the critical chain . on the other hand , the computations in cft using the replica trick become very cumbersome and were only done for two excitations in @xcite . this part will then serve as a collection of new predictions for the evolution of the rnyi entanglement entropies in ( rational ) cfts . we begin with states where few operators were inserted on sufficiently separated sites and in some distance to the entangling block . we chose the block to be large enough so that there is a time where all the excited quasiparticles are inside the entangling region . from numerical results it is clear that in such excited states the entanglement entropy is a sum of the quantum dimensions of the operators , see fig . [ steps ] for a case of 3 and 5 @xmath17-excitations . the time evolution of entanglement in such states is very similar to periodic quench studied , e.g. , in ref . @xcite . on ( a ) @xmath99 sites , @xmath100 and ( b ) @xmath101 sites , @xmath102 . results for ising hamiltonian ( solid lines ) and linearized ising hamiltonian ( dashed lines ) . block @xmath9 supported on sites @xmath103 ( @xmath104 ) . , width=529 ] based on these observations , we conjecture that in rational cfts , and more generally cfts for which the quasi - particle picture remains a good effective description of the dynamics of entanglement , for @xmath106 such excitations the change in maximal contribution to the entanglement rnyi entropies is simply s^(n)_l=^m_i=1d_i with quantum dimension of the @xmath107th operator @xmath108 ( can be different or the same ) . we verify numerically , using @xmath78 , that if we use some combination of operators @xmath72 , @xmath109 and @xmath110 placed in the setup considered in fig . [ steps ] where the insertion points are sufficiently separated , then the total increase of the entropy of the block @xmath9 indeed is a sum of contributions from single excitations discussed in section [ sec : num ] . a similar result was reported in free quantum field theories in @xcite but it would be very interesting to formulate at least a necessary conditions for the validity of this formula in arbitrary interacting 2d cft and we leave this as an open future problem . + another class of excited states that we consider is defined by acting with local operators on all the sites of the chain | _ g > = ^n_i=1 _ i| 0 > . they could be thought of as a version of a global quench @xcite and have recently been employed in large @xmath111 holographic cfts as states dual to the matter collapsing to a black hole @xcite . . block size @xmath112 and the system size @xmath105 . , width=302 ] as we can see , the evolution looks qualitatively similar to the global quench in the finite size system @xcite , but the final value in that case is given by the entropy density times the length of the interval . here , the evolution of entanglement entropies can still be interpreted in terms of the quasiparticles propagating from the lattice sites to the left and right . this picture suggests that the maximal value of the entropies is equal to @xmath113 , where the factor of two comes from the fact that at each site we can have two ( left and right ) quasiparticles . on the other hand , the maximal value of the entropy of the block of @xmath70 spins is attained by the density matrix with all @xmath114 eigenvalues equal and is also equal to @xmath115 . apparently these two numbers coincide for @xmath116 and the value of the plateaux for @xmath117 is equal to this maximal value minus the ground state entropy , i.e. @xmath118 attains its maximal possible value , which can be seen in fig . [ allsx ] for @xmath78 . if ising hamiltonian @xmath60 is used instead , the maximal value of the plateaux is not reached and the clean periodic structure of revivals is obscured , showing the relevance of coherent dynamics of quasiparticles for obtaining those effects . in other words , the memory effects for the quasiparticles in the critical model seem to be suppressed by entanglement between quasiparticles with different momenta and for late times entanglement measures saturate . i would be interesting to compare how this behaviour changes for local operators with different quantum dimensions and for different densities of excitations leading possibly to a saturation , so we leave it as another open problem . moreover , according to @xcite , in the large @xmath111 cfts , entanglement entropy saturates at the thermal value with an effective temperature and the time for returning into the initial value ( poincare recurrence ) is expected to be exponential in the central charge . it would be also interesting to further explore what happens in between these two regimes and how our memory effects are corrected once we consider states in chaotic or non - local toy models for black holes as for instance in ref . [ sec : rel_ent ] we finish our numerical explorations with evolution of the relative entropy at criticality . as far as we are aware , numerical analysis of the relative entropy in critical systems has been much less explored than the rnyi entropies . nevertheless , it is an important tool for understanding the notion of the distance between quantum states in field theories and plays an interesting role in uncovering the features of holographic cfts @xcite . the relative entropy is defined for two reduced density matrices @xmath119 and @xmath120 as s(|)=()- ( ) . [ eq : rel_ent ] in general excited state of a 2d cft , the second term makes it very hard to compute analytically . even for locally excited states the replica method requires the knowledge of the correlation function of @xmath121 operators in order to continue to @xmath122 , see , e.g. , @xcite . on the other hand , if we compare excited states @xmath119 with @xmath120 obtained in the vacuum , given that the reduced density matrix @xmath120 can be written as the exponent of a known modular hamiltonian , @xmath123 , we can express the entropy as s(|)=h_m - s^(1 ) . [ rre ] the expectation value of the vacuum modular hamiltonian @xmath124 is computed in the excited state and @xmath125 denotes the difference of the von - neumann entropies of the two density matrices , see @xcite for more details . now , in 2d cfts , the expectation value of the vacuum modular hamiltonian for an interval of length @xmath70 in a state locally excited by a primary operator is universal . namely , it follows from the ope of the stress tensor with the primary operator , and can be computed from ( see e.g. @xcite ) h_m=_ldx ( l - x)xt_00(x ) . moreover , as we argued in section [ sec : cft ] , the change of the entanglement entropy of the block is equal to the logarithm of the quantum dimension of the primary operator . from these two results we evaluate the relative entropy in our cft setup and we can possibly compare it with numerics . . colors indicate different block sizes @xmath126 ( blue ) , @xmath127 ( red ) , @xmath128 ( green ) . results for ising hamiltonian ( solid lines ) and linearized ising hamiltonian ( dashed lines ) . distance of the entangling block from the excitation ( a ) @xmath93 , ( b ) @xmath96 . @xmath105 . inset ( c ) shows the maximum of relative entropy from panel ( b ) for linearized ising hamiltonian , where the linear fit @xmath129 . , width=529 ] in the following , we simply take @xmath120 to be the reduced density matrix of a block of @xmath70 spins in the ground state and @xmath119 as the density matrix of the block for the state locally excited by operator @xmath72 in a distance @xmath1 from the block . we find an elegant expression for the relative entropy in terms of fermionic covariance matrices for gaussian @xmath119 and @xmath120 @xmath130 and we refer to the appendix [ numer ] for details . we only notice here that as @xmath131 and @xmath132 , or equivalently @xmath119 and @xmath120 , can not be simultaneously diagonalized , finite numerical precision limits the calculation of the second term in the above expression only to relatively small block sizes . we present the results in fig . [ fig : rel_entropy ] . we observe that the signal strongly changes depending if @xmath60 or @xmath78 is governing the time evolution . for the ising hamiltonian @xmath60 the maximum of the relative entropy for given block size @xmath70 quickly disappears with the distance between the block and the excitation suggesting that modes with high momenta contribute significantly to the observed value . if linearized ising hamiltonian @xmath78 is used instead , the signal does not dissipate , with both the width and the value at the peak being proportional to @xmath70 . one can verify that , for non - zero @xmath3 , the numerical evolution of the relative entropy is consistent with the cft computation with the expectation value of the stress tensor in our locally excited state . moreover , it is possible to compute in cft the value of the peak of the relative entropy . the maximum of the relative entropy is universal in the small @xmath3 limit ( s(| ) ) e_l+o(),[fl ] where the energy due to the operator insertion is @xmath133 . numerically , this is shown on the inset in fig . [ fig : rel_entropy ] where we see the linear growth of the peak of the relative entropy with the length of the interval . our numerical results in the ising chain are consistent with the energy @xmath134 close to the slope @xmath135 fitted in fig . [ fig : rel_entropy](c ) . let us finally compare our formula with the first - law for entanglement entropy for a family of the nearby equilibrium states @xcite that reads @xmath136 with @xmath137 . this relation is a consequence of the vanishing relative entropy . interestingly , we observe an analogous first - law like relation for the maximal value of the distance between our two quantum states as measured by the relative entropy . moreover , the maximal value is universal and contained in @xmath138 . the details of the cft analysis for this class of locally excited states will appear elsewhere @xcite . we have shown that , by using local operators on the critical lattice , one is able to extract further non - trivial cft data like quantum dimensions numerically . we have successfully done so for the spin operator @xmath17 and its derivative descendant . however , we also saw that , even after linearization of the dispersion relation , the lattice energy operator gave a non - zero contribution to the entropy of a block . this fact might be explained by the ambiguity in the identification between the lattice and the cft operators or a non - trivial contribution from the tail " of operators in the continuum and deserves further investigation . our analysis could be naturally extended to other cfts with operators that enjoy known lattice counterparts . a natural setup might be the three - state potts model in which @xcite recently analyzed the lattice realization of the local operators in the parafermionic cft . clearly , it would be much more interesting to explore analytically what is the contribution from a general lattice operator and what kind of information can be extracted from the increase in rnyi entropies . in fact there has been a lot of work on extracting local cft operators from the multiscale entanglement renormalization ansatz ( mera ) @xcite . it would be interesting to apply those developments in the studies of entanglement evolution after local operator excitations . the evolution of entanglement measures after local excitations away from the critical point appears to be relatively unexplored . as we saw , the critical behavior appears to be very special characterized by formation of a clear plateaux and might be used as a smoking gun of a critical point . moreover , the contribution to the rnyi entropies appears to be the smallest for the ising hamiltonian with critical parameters what might be a sign of a general bound . we hope that our analysis will serve as a starting point for more complicated systems and will help to uncover other unknown universal phenomena in the propagation of entanglement . last but not least , we performed the evolution of the relative entropy between the vacuum and a locally excited state . interestingly , this distance measure shows universal features analogous to the first law of entanglement and its maximum maximal quantum distance is proportional to the change in the energy with an effective temperature . exploring this relation in cfts or free field theories where explicit computations are under control opens a new interesting path for investigation . + * acknowledgements * we would like to thank john cardy , paul fendley , masahiro nozaki , tokiro numasawa , tadashi takayanagi , kento watanabe , luca tagliacozzo and alvaro veliz - osorio for correspondence and discussions . we would also like to thank xueda wen for comments sharing some of his unpublished results . we acknowledge support by the swedish research council ( vr ) grant 2013 - 4329 ( p.c . ) and narodowe centrum nauki ( ncn , national science center ) under project no . 2013/09/b / st3/01603 ( m.m.r ) . p.c . would like to thank yukawa institute for theoretical physics for hospitality and support during the `` quantum information in string theory and many - body systems '' workshop where some of this work was performed . we would like to thank the organizers of the tensor network summer school in ghent where this work was initiated . _ hamiltonian. _ the ising hamiltonian is diagonalized in a standard way @xcite by mapping it onto a system of free fermions using the jordan - wigner transformation @xmath139 where @xmath140 are fermionic annihilation operators . for convenience , we introduce majorana fermions @xmath141 , @xmath142 which are hermitian , unitary and satisfy canonical anticommutation relations @xmath143_+ = 2 \delta_{m , n}$ ] . the ising hamiltonian with periodic boundary conditions can be rewritten as @xmath144 above , @xmath145 are projectors on the subspaces with respectively even and odd number of fermions , where the parity operator @xmath146 commutes with @xmath147 . the hamiltonians in both subspaces can be expressed in terms of fermionic operators as @xmath148 differing only at the boundary term . this is accounted for by enforcing the boundary conditions : antiperiodic @xmath149 for @xmath150 , and periodic @xmath151 for @xmath152 . for @xmath153 , which we use in this work , the ground state of @xmath147 belongs to the subspace with even parity for any value of @xmath26 , see e.g. ref . @xcite . it is convenient to introduce @xmath154 , where @xmath155 is a column vector composed of operators @xmath156 and @xmath157 are @xmath158 hermitian matrices . in the following , we employ the matrix notation whenever is serves to simplify the notation . in each parity subspace the system is solved independently by a canonical transformation to a new base of majorana fermions , @xmath159 , where @xmath160 are real and orthogonal matrices . the hamiltonian then reads @xmath161 expressed above in terms of standard annihilation operators @xmath162 . this can be done analytically by a subsequent fourier and bogoliubov transformations @xcite , resulting in @xmath163 given by eq . . the fermionic modes are indexed by quasi - momenta consistent with the respective boundary conditions , i.e. @xmath164 $ ] for @xmath165 , and @xmath166 $ ] for @xmath167 . this procedure is equivalent to bringing @xmath168 into the canonical form @xmath169 . we can now formally introduce the linearized ising hamiltonian by @xmath170 defined as @xmath171 , where the linearized dispersion relation @xmath172 . the ground state of the system , @xmath173 , which is the starting point for our further considerations , is the even parity state annihilated by all annihilation operators diagonalizing @xmath174 , namely @xmath175 . all the information about this state is encoded in the matrix @xmath176 . _ local operators and time evolution. _ we consider local operators which in the fermionic language read @xmath177 up to the irrelevant phase factors . we simulate the action of those operators on the ground state by employing the heisenberg picture . as all operators in eq . are expressed as a product of linear combinations of operators @xmath156 , this can be done iteratively . starting with the simplest case of operator @xmath178 , which is unitary by construction , we define the operator in the heisenberg picture as ^o_1_n = _ 1^_n _ 1 = - ( -1)^_nm_1 _ n = - _ j ( -1)^_nm_1 [ u^+]_nj ^+_j , where a new matrix @xmath179_{nj } = - ( -1)^{\delta_{nm_1 } } [ { \rm u}^+]_{nj}$ ] describes the transformation to the fermionic base in which the initial state is the vacuum note that operator @xmath180 is applied to the ground state of the ising hamiltonian @xmath181 in eq . . subsequent application of unitary operators @xmath182 for @xmath183 is obtained similarly as @xmath184 , and @xmath185_{nj } = - ( -1)^{\delta_{nm_l } } [ { \rm u}^{o_{l-1}}]_{nj}$ ] for @xmath186 , corresponding to the action of the operator @xmath187 on the initial ground state . in a general case when @xmath188 , with @xmath189 being a row vector of coefficients , we have to generalize the above procedure . notice that @xmath180 , in principle , does not have to be unitary . as we are only interested in the action of @xmath180 on the ground state , @xmath173 , we introduce unitary operator @xmath190 , such that @xmath191 up to normalization . for @xmath190 to be unitary , it is enough to require that @xmath192 , where @xmath193 is a row vector of real coefficients . we can now define the operators in the heisenberg picture ^o_1 = _ 1^ _ 1 = u^+ ( _ 1^^+ _ 1 ) = u^+ ( 2 - ) ^+ , defining new transformation matrix @xmath194 . the vector @xmath193 is found as @xmath195 _ { 2n-1 } - i [ \vec v \mathrm{u}^+ ] _ { 2n})$ ] and @xmath196 _ { 2n-1 } - i [ \vec v \mathrm{u}^+ ] _ { 2n})$ ] . this follows from the condition @xmath197 , where the two operators may differ by operators @xmath198 annihilating the vacuum state @xmath173 . finally the time evolution is simulated in the heisenberg picture as @xmath199 $ ] , which leads to @xmath200 , where @xmath201 , and the matrix @xmath202 for the even ( odd ) parity of the excited state @xmath203 . notice that operator @xmath109 is preserving the parity , and @xmath72 , @xmath110 are changing it . finally , the time evolution with the linearized hamiltonian is obtained by using matrices @xmath204 instead of @xmath205 above . _ calculating entropies. _ the entropy of a block of consecutive spins is calculated in a standard way @xcite , as all the information about the reduced density matrix of a block is encoded in the @xmath206 covariance matrix @xmath207_{m , n = 2l_0 + 1,2l_0 + 2 , \dotsc,2l_0 + 2l}\ ] ] supported on @xmath70 consecutive lattice sites @xmath208 . it is found as a submatrix of @xmath209 , with @xmath210 being the correlation matrix in the canonical ( vacuum ) base . rnyi entropy of the block is then found as , s^()_l = _ j=1^l , where @xmath211 are the eigenvalues of @xmath212 . for @xmath213 , the von - neumann entropy reads s^(1)_l = - _ j=1^l = - ( c_l c_l ) , where the second equation provides the direct expression in term of covariance matrix @xmath212 . relative entropy can be similarly found similarly as @xmath214 as the matrices @xmath215 and @xmath132 ( calculated as the correlation matrix in the ground state ) can not be diagonalized simultaneously , it becomes difficult to numerically calculate the second term in the above equation for large blocks . in that case @xmath132 has many eigenvalues approaching @xmath216 , falling below numerical precision , which prevents the calculation of the logarithm . for this reason , in sec . [ sec : rel_ent ] , we calculate residual entropy only for small enough block where this problem does not yet occur . finally , in order to derive the non - standard second term on the r.h.s . of eq . we employ the fact that @xmath120 is gaussian and can be expressed as @xmath217 , where the diagonal form in terms of fermionic annihilation ( creation ) operators @xmath218 ( @xmath219 ) can be directly obtained from the correlation matrix @xmath220 @xcite . we introduce here @xmath221 as a column vector consisting of majorana fermions @xmath222 in the entangling block . then @xmath223 , where @xmath224 is a column vector consisting of @xmath225 , and @xmath226 is a unitary matrix diagonalising @xmath227 \mathrm{u}^{\vartheta \dagger}$ ] . this allows to compute @xmath228 . using matrix notation this equals @xmath229 \left [ \bigoplus_{j = 1}^l \begin{pmatrix } \log q_j & 0 \\ 0 & \log ( 1-q_j ) \\ \end{pmatrix } \right ] \right ) = \tr \left ( \mathrm{u}^{\vartheta \dagger } c_l^\rho \mathrm{u}^{\vartheta } \left [ \bigoplus_{j = 1}^l \begin{pmatrix } \log q_j & 0 \\ 0 & \log ( 1-q_j ) \\ \end{pmatrix } \right ] \right ) = \tr \left ( c^\rho_l \log c^\vartheta_l \right)$ ] , which gives eq . . calabrese p and cardy j 2016 j. stat . mech . 064003 nozaki m , numasawa t and takayanagi t 2014 phys . 112 * 111602 nozaki m 2014 jhep * 1410 * 147 caputa p , nozaki m and takayanagi t 2014 ptep * 2014 * 093b06 + caputa p , simon j , stikonas a and takayanagi t 2015 jhep * 1501 * 102 + asplund c t , bernamonti a , galli f and hartman t 2015 jhep * 1502 * 171 + caputa p , simon j , stikonas a , takayanagi t and watanabe k 2015 jhep * 1508 * 011 + shiba n 2014 jhep * 1412 * 152 + guo w z and he s 2015 jhep * 1504 * 099 + guo w z 2015 arxiv:1510.07142 + nozaki m , numasawa t and matsuura s 2016 jhep * 1602 * 150 + caputa p , numasawa t and veliz - osorio a 2016 arxiv:1602.06542 + sivaramakrishnan a 2016 arxiv:1604.00965 + david j r , khetrapal s and kumar s p 2016 arxiv:1605.05987 + caputa p , nozaki m and numasawa t 2016 phys . rev . d * 93 * 105032 + nozaki m , and watamura n , arxiv:1606.07076 + zhou t , arxiv:1607.08631 alcaraz f c , berganza m i and sierra g 2011 phys . lett . * 106 * , 201601 + berganza m i , alcaraz f c and sierra g 2012 j. stat . * 1201 * p01016 + v. alba , m. fagotti and p. calabrese , 2009 j. stat . mech . * 0910 * , p10020 + palmai t 2014 phys . b * 90 * 161404 + sheikh - jabbari m m and yavartanoo h 2016 arxiv:1605.00341 + chen b and wu j q 2016 arxiv:1605.06753 + taddia l , ortolani f and palmai t 2016 arxiv:1606.02667 + giusto s , and russo r , 2014 phys . d * 90 * , no . 6 , 066004 he s , numasawa t , takayanagi t and watanabe k 2014 phys . d * 90 * 041701 caputa p and veliz - osorio a 2015 phys . d * 92 * 065010 chen b , guo w z , he s and wu j q 2015 jhep * 1510 * 173 mong r s k , clarke d j , alicea j , lindner n h and fendley p 2014 j. phys . a * 47 * 452001 lashkari n 2016 phys . * 117 * 041601 zauner v , ganahl m , evertz h g , nishino t 2015 j. phys . : condens . matter * 27 * 425602 + phien h n , vidal g and mcculloch i p 2012 phys . * 245107 + milsted a , haegeman j , osborne t j and verstraete f 2013 phys . b * 88 * 155116 anous t , hartman t , rovai a and sonner j 2016 jhep * 07 * 123 mezei m and stanford d 2016 arxiv:1608.05101 magan j m 2016 arxiv:1601.04663 + jansen a and magan j m 2016 arxiv:1604.03772 n. lashkari , j. lin , h. ooguri , b. stoica and m. van raamsdonk , arxiv:1605.01075 sarosi g and ugajin t 2016 jhep * 07 * 114 blanco d d , casini h , hung l y and myers r c 2013 jhep * 1308 * 060 cardy j and tonni e arxiv:1608.01283 caputa p _ in preperation _ bhattacharya j , nozaki m , takayanagi t and ugajin t 2013 phys . lett . * 110 * 091602 giovannetti v , montangero s and fazio r 2008 phys . * 101 * 180503 + pfeifer r n c , evenbly g and vidal g 2009 phys . a * 79 * 040301 + evenbly g , corboz p and vidal g 2010 phys . b * 82 * 132411 + evenbly g and vidal , g 2013 quantum criticality with the multi - scale entanglement renormalization ansatz . in : avella , a. , mancini , f. ( eds . ) strongly correlated systems : numerical methods . springer , heidelberg , arxiv:1109.5334 peschel i 2003 j. phys . a : math.gen . * 36 * l205 + vidal g , latorre j i , rico e and kitaev a 2003 phys . rev . lett . * 90 * 227902 + jin b - q and korepin v e 2004 j. stat . phys . * 116 * 79 + peschel i and eisler v 2009 j. phys . a : math . theor . * 42 * 504003

coupled chaotic systems are recently calling much attention due to the verification that they may be useful to the understanding of natural systems in a variety of fields as in ecology @xcite , in neuroscience @xcite , in economy @xcite , and in lasers @xcite . it has been verified that despite of the higher dimensionality of a coupled chaotic system , the coupling among the elements might make them to synchronize @xcite , reducing the dynamics of the system to a few degrees of freedom . in this work , we focus our attention in the phenomenon of phase synchronization ( ps ) , which describes the appearance of a phase synchronous behavior between two interacting chaotic systems @xcite , i.e. , given two chaotic systems , their phase difference remains bounded , despite of the fact that their amplitudes may be uncorrelated . this phenomenon is particularly interesting since it can arise from a very small coupling strength . its presence was reported in a variety of experimental systems . it was experimentally demonstrated in electronic circuits @xcite , and latter in electrochemical oscillators @xcite . it was found in plasma @xcite , in the chua s circuit @xcite , and there were also found evidences of phase synchronization in communication processes in the human brain @xcite . to detect ps in a real - time experiment , one has to measure the phase of the chaotic trajectory @xcite . however , the phase is not always an easily accessible information . to overcome this difficulty , it is important to understand fundamental properties of phase synchronous systems , that could be experimentally easily verified . for chaotic systems that are phase synchronized with a periodic forcing @xcite , it was reported that a stroboscopic map of the trajectory was a subset of this attractor and occupies only partially the region delimitated by a projection of the attractor . this property was used to detect in a real - time experiment phase synchronization between the chua s circuit and a sinusoidal forcing @xcite . this approach of detecting phase synchronization through the stroboscopic map can be extended for coupled chaotic oscillators , in a formal way . the stroboscopic map is generalized to the * conditional poincar map*. given two oscillators , at least one being chaotic , the conditional poincar map is constructed by collecting points in one oscillator at the moment at which some event happens in the other one . if the set of discrete points generated by this conditional map does not visit any arbitrary region of a especial projection of the chaotic attractor , we call this set a * p - set*. this property of the conditional poincar map is called non - transitivity @xcite , i.e. , an initial condition under the conditional poincar map does not visit everywhere in a subspace of the attractor . alike the stroboscopic maps of oscillators that are in phase synchrony with a forcing , the conditional poincar maps of coupled chaotic oscillators , in ps , also only partially occupy a projection of the attractor . in this work , we show how the conditional poincar map can be used to detect ps , without actually having to measure the phase . for phase - coherent oscillators , a special type of p - set , that we call ps - set ( phase synchronization set ) , conversely , its existence also implies ps . we illustrate our findings and ideas with numerical and experimental analyzes in the forced chua s circuit , and the coupled rsller oscillator @xcite . further , we extend these results to non - phase coherent attractors . finally , we also introduce a phase of a chaotic trajectory to be a quantity related to the amount of rotation of the tangent vector . this definition can be used to chaotic attractors , independently whether they have phase - coherent or non phase - coherent dynamics . this work is organized as follows : in sec . [ sec : phase ] , we define a way to measure the phase of a chaotic flow , and discuss the relation between the average returning time and the average angular frequency . in sec . [ sec : ps ] , we discuss the conditions for ps states and , in sec . [ sec : chua ] , we describe the phenomenon of ps in the forced chua s circuit . we introduce the notion of a conditional poincar map in sec . [ cpm ] and the p - sets ( as well as the ps - sets ) in sec . [ p - sets ] . in sec . [ sec : bsc ] , we show how ps can be found by the detection of these sets in the forced chua s circuit and in sec . [ sec : bsr ] , for the coupled rssler oscillator . further , in sec . [ extension ] , we discuss the extension of these ideas to a class of non - coherent oscillators . in sec . [ sec : cr ] , we make some remarks and the conclusions of this work . in appendix a , we formally introduce the conditional poincar map and the p - set , and in appendix b , we show that for coherent dynamics the ps - sets exist if , and only if , there is ps . in other words , ps implies ps - sets and vice - versa . the phase of a chaotic attractor in a projection @xmath0 ( a subspace ) is defined to be the amount of rotation of the tangent vector in this projection , and can be represented by an integral function of the type @xmath1 with @xmath2 being an infinitesimal displacement of the tangent vector of the flow , from time @xmath3 to time @xmath4 , and @xmath5 . note that in eq . ( [ fase ] ) , we are measuring the amount of rotation , per unit time , of a projection of the tangent vector of the flow , on the same subspace @xmath0 where the phase is defined . we call this subspace @xmath6 . the attractor @xmath7 , projected on the subspace @xmath6 is regarded as @xmath8 . the instantaneous angular frequency of the trajectory in @xmath8 , named @xmath9 is given by @xmath10 . so , from eq . ( [ fase ] ) , @xmath11 and , the average angular frequency @xmath12 is @xmath13 @xmath14 represents the average . equation ( [ natural_frequence_2 ] ) can be put into the form @xmath15 = @xmath16 . whenever a poincar section can be defined , the average period of the chaotic attractor on the subspace @xmath6 is calculated by @xmath17 where @xmath18 , and @xmath19 represents the time at which the trajectory in the subspace @xmath6 makes the @xmath20-th crossing with this poincar section . we introduce @xmath21 to be the average displacement of the phase for a typical period as @xmath22 with @xmath23 being the phase associated to the subspace @xmath6 at the moment that the @xmath24-th crossing between the trajectory and the poincar section happens . thus , we can put eq . ( [ natural_frequence_2 ] ) as @xmath25 for the forced chua s circuit , the subspace @xmath26 is defined by a suitable projection of the circuit variable . we have that @xmath27 . so , eq . ( [ natural_frequence_1 ] ) can be written as @xmath28 . for the coupled rssler oscillator , the quantities in eq . ( [ natural_frequence_1 ] ) can be calculated in two subspaces , one associated to the variables of one rssler , the subspace @xmath26 , and the subspace @xmath29 , associated to the variables of the other rssler system . as shown in @xcite , @xmath30 might slightly differs from @xmath31 , and thus , @xmath32 . ( [ natural_frequence_2 ] ) and ( [ natural_frequence_1 ] ) relate the average period , the average angular frequency , and the phase of a chaotic trajectory . this shows that the average period ( recurrence ) and the average angular frequency are intimately connected in phase coherent chaotic systems , and both these quantities can be calculated from the phase . having defined phase , ps exists whenever the following condition is satisfied @xmath33 the minimal bound for the phase difference , @xmath30 , in terms of the phase as defined by eq . ( [ ps ] ) , was theoretically estimated in ref . @xcite . equation ( [ ps ] ) means that the phase difference between the two coupled systems is always bounded , and @xmath34 is a rational constant @xcite . also , @xmath35 in this work , we will consider cases for @xmath34=1 . otherwise , a simple change of variables could eliminate this constant from eq . ( [ nc ] ) . for the forced chua s circuit @xmath36 , with @xmath37 , representing the angular frequency of the forcing . there is ps , if @xmath38=@xmath39 . therefore , @xmath40=@xmath41 . for the coupled rssler , if ps exists , @xmath42 = @xmath43 , @xmath44=@xmath45 , and @xmath46=@xmath47 , and therefore , we could have in the right term of eq . ( [ ps ] ) @xmath48 , instead of @xmath49 . the circuit is represented by : @xmath50 where @xmath51 , @xmath52 , and @xmath53 represent , respectively , the tension across two capacitors and the current through the inductor ( see @xcite for more details ) , @xmath39 and @xmath54 are the angular frequency and the amplitude of the forcing , respectively . the piecewise - linear function , @xmath55 , is given by : @xmath56\ ] ] where we have chosen the parameters @xmath57=0.1 , @xmath58=0.574 , @xmath59=1 , @xmath60=1/6 , @xmath61=-0.5 , and @xmath62=-0.8 , such that we obtain a rssler - type attractor , for @xmath54=0 . to calculate the phase of the chaotic trajectory , we first define the subspace @xmath26 to be given by the pair of variables @xmath63 , and then , we use eq . ( [ fase ] ) . in fig . [ ps_fig6 ] , we show the difference between the phase of the chaotic circuit [ as calculated by eq . ( [ fase ] ) ] and the phase of the forcing , @xmath64 . in ( a ) , the phase difference is bounded and the average period of the chaotic attractor , defined as the average recurrence time of trajectories that cross the section @xmath65 , is equal to @xmath66=3.57015 , which is equal to 1/f , since @xmath67=0.2801 . the average angular frequency can be calculated using eq . ( [ natural_frequence_1 ] ) , which gives us @xmath68=1.75992 . or , from eq . ( [ natural_frequence_2 ] ) , which gives us @xmath38=1.75992 . note that the average growing of the phase [ calculated by eq . ( [ fase ] ) ] for a typical average period is @xmath69 which is @xmath31 . in fig . [ ps_fig6](b ) , we have that @xmath70=3.57006 which is different from @xmath41 , since @xmath67=0.279 . so , in ( b ) there is no ps , and consequently inequality ( [ ps ] ) is not satisfied . in ref . @xcite , we detected @xmath71 experimentally in the forced chua s circuit , whenever stroboscopic maps could be constructed for a time interval equal to @xmath72 , such that this map , projected into the same subspace considered to calculate the phase , does not occupy the region occupied by the attractor projected on the same subspace . to understand this technique , we assume that @xmath73 , the phase of the chaotic trajectory , is the angle ( on the lift ) described by the vector position of this trajectory ( rssler - like attractor ) , and @xmath74=@xmath75 the phase of the forcing . if eq . ( [ ps ] ) is satisfied at any time , then it is satisfied at multiples of the period of the forcing @xmath76 . so , we get @xmath77 , which means that a stroboscopic map has to be concentrated in an angular section smaller than @xmath31 . the stroboscopic map , which is already a subset of the chaotic flow , projected into the same subspace considered to calculate the phase , does not occupy the whole region occupied by the attractor projected in this same subspace . another property of the stroboscopic map is that points in it are mapped into it by looking at the trajectory after a time interval given by @xmath78 , so , it is a subset that is recurrent to itself . using this technique , and for the same parameters as @xcite , we show in fig . [ ps_fig1](a ) the experimental synchronization region for the forced chua s circuit in the parameter space @xmath79 . the triangular shaped region represents parameters for which the stroboscopic map has the points concentrated in an angular section smaller than 2@xmath80 . the bump at the bottom right side of the ps region is due to non - synchronous states that present a long bounded phase difference . by a typical time interval within which the experiment is realized , which was of the order of 40,000 cycles , the system seemed to be phase synchronized , i.e. , localized stroboscopic maps were found . in fact , we have detected these maps even for observation times corresponding to 150,000 cycles , in the region of the bump . in a short , the bump region is an extended structure in the circuit parameter space , that presents intermittent behavior in the phase difference @xcite , but with a long laminar regime , even for parameters far away from the border between the ps and the non ps region . this intermittency differs from the usual one , observed in the transition to ps , in the fact that this latter happens very close to the border between the ps and the non ps region . the reason for this intermittency is due to the presence of a periodic window , close to the region of the bump . simulation is shown in fig . [ ps_fig1](b ) , where black points represent perturbing parameters for which eq . ( [ ps ] ) is satisfied , with @xmath81 defined in eq . ( [ fase ] ) . one sees that the @xmath71 region resembles a triangle . the triangular shaped region , denoted by the light gray dashed line , represents the region where the system is not phase synchronized , but the phase difference remains bounded for a long time interval , that might be longer than 100,000 cycles of the systems . so , we reproduce numerically the same atypical intermittency observed experimentally , i.e. , long laminar regime in the phase difference , for parameter regions away from the border between ps and non ps states . this happens associated with a periodic window , as the one shown in fig . [ ps_fig1](b ) . the shape of the synchronization region in fig . [ ps_fig1](b ) is equivalent to the region in the experiment , constructed by detecting the stroboscopic maps contained within a small angular section . this proposes an equivalence between the existence of a recurrent subset and the verification of eq . ( [ ps ] ) . inside the synchronization region , a stroboscopic map appears like in fig . [ ps_fig7](a ) , where the light gray points represent the attractor , and the dark filled circles , the map . outside of the @xmath71 region , there are parameter sets for which the stroboscopic maps do not occupy the region occupied by the attractor , at the projection in which the phase is calculated . as one sees in fig . [ ps_fig7](b ) [ for the parameters represented by the plus symbol in fig . [ ps_fig1](b ) ] there is a region of the projected attractor ( pointed by the arrow ) for which the stroboscopic map never visits . the difference between the stroboscopic map that appears while there is ps in ( a ) , and the stroboscopic map that appears while there is not ps in ( b ) , is that points of the stroboscopic map in ( a ) are all concentrated in an angular section of the attractor . as it will be further classified , the stroboscopic map in ( a ) is a ps - set and in ( b ) is a p - set . as a way to better characterize the ps phenomena in the chua s circuit , we calculate the lyapunov exponents . as usually expected , the transition to ps is associated to one of the exponents becoming smaller than zero . since there is already one exponent that is smaller than zero , ps induces the creation of a second stable direction in the chua s circuit . previous to the transition to phase synchronization , this exponent was zero . in fig . [ ps_fig8 ] , we show the exponent ( and the error bars ) associated to ps in black and the quantity @xmath82 ( in gray , for a fixed amplitude of @xmath54=0.0015 ) with respect to the frequency . in the region that the exponent becomes smaller than zero , @xmath38=@xmath39 , satisfying eq . ( [ nc ] ) . the finding of maps of the attractor that appear as localized structures implies ps . the conditional poincar map introduced in this chapter as a generalization of the stroboscopic map , is an efficient way of revealing the existence of such special mappings . the stroboscopic map technique defined for periodically driven chaotic systems was explained in the previous sections . to extent the idea of stroboscopic map in coupled chaotic oscillators we came up with the conditional poincar map , which is a map of the flow , constructed by observing it for specific times at which events occur in the subsystems @xmath8 . an event is considered to happen when the trajectory of one subsystem crosses a poincar section . so , given two oscillators @xmath83 and @xmath84 ( at least one being chaotic ) , with trajectories in the subspaces where the phase is defined , the conditional poincar map of @xmath83 is the trajectory position at the moment that a series of equal events happens , in @xmath84 . analogously , the conditional poincar map of @xmath84 is the trajectory position at the moment that a series of equal events happens , in @xmath83 . in the case of periodically forced chaotic systems , an event may be defined to happen whenever the forcing reaches a specific value , and the conditional poincar map is the usual stroboscopic map , because the time interval between two events is constant . in coupled chaotic systems , the time interval between two successive events is no longer constant . we define a time series of events @xmath85 , by the following rule : * @xmath86 represents the time at which the @xmath20-th crossing of the trajectory of @xmath84 occurs in a poincar plane . * @xmath87 represents the time at which the @xmath20-th crossing of the trajectory of @xmath83 occurs in a poincar plane . the discrete set of points observed at the times @xmath85 is called set @xmath88 . this set , projected at the subspaces @xmath6 ( where the phase is calculated ) , is named @xmath89 . the conditional poincar map is represented by @xmath90 . so , we say that @xmath89 is the set of points generated by @xmath90 . the next step is to define when @xmath89 can be regarded as either a p - set or a ps - set , this last set implying phase synchronization . the @xmath89 set is a p - set , if it does not completely fulfill the projection @xmath8 of the attractor . in other words , a discrete set @xmath89 is considered to be a p - set , if for balls of radius @xmath91 , centered in all points of the attractor projection @xmath8 , one does not find points of @xmath89 inside of all these balls . if @xmath89 completely fulfill @xmath8 , we say that these two sets are equivalent ( and we represent this by the symbol @xmath92 ) . more details , see appendix a. so , a p - set exists , if the conditional poincar map is not transitive on @xmath8 @xcite . that is , the flow , observed by the times for which the conditional poincar map is defined , does not visit arbitrary regions of @xmath8 . note that however , the attractor is chaotic , and therefore , the chaotic set is always transitive through the flow . so , given a set of initial conditions , its evolution through the flow eventually reaches arbitrary open subsets of the original chaotic attractor . we can classify three relevant types of sets generated by the conditional poincar map type - a : : @xmath89 is equivalent to @xmath8 ( @xmath93 ) . the conditional poincar map @xmath90 is transitive in @xmath8 . p - set : : @xmath89 is not equivalent to @xmath8 ( @xmath94 ) . the conditional poincar map is not transitive in @xmath8 . ps - set : : @xmath89 is p - set , with the additional condition that it is localized in the vicinity of the poincar section chosen to define the events . in the following , we comment each case : _ these sets appear whenever there is no ps_. if two non - identical coupled chaotic systems ( topologically similar ) are not phase synchronized , the chaotic trajectories do not make correlated events in both subspaces @xmath26 and @xmath29 . as a consequence while the trajectory is positioned at the specified poincar plane , at the subspace @xmath26 , the trajectory in the subspace @xmath29 is everywhere in this subspace , making the set @xmath89 to be equivalent to @xmath8 . an interesting illustration is the case of two uncoupled equal chaotic systems , but with different initial conditions . as we construct the conditional poincar maps , they will be a type - a set , since also the distance between the trajectories in the two oscillators are sensitive to the initial conditions , and will diverge exponentially . _ these sets constitutes a necessary , but not sufficient , condition to describe ps . _ in periodically forced chaotic systems , they might appear when there is not ps . as we already mentioned , the points in these sets are not localized in special spots of the attractor projection . as a consequence , the domain of the absolute difference between the time at which the same number of events happen in both oscillators has a broad character . _ ps - sets imply ps and vice - versa_. they exist , if and only if , there is phase synchronization , as shown in appendix b , and illustrated with the examples in sec . [ sec : chua ] , and throughout this work . another important point is that the ps - set provides a real - time detection that can be easily experimentally implemented ( sec . [ sec : chua ] ) , and easily constructed from a data set . ps - set implies ps because the difference between the time at which the @xmath24-th event happens in both oscillators is small , which means that the time difference @xmath95 is smaller than a finite constant value . as a consequence , the points in the conditional poincar map of one oscillator are confined around the poincar section chosen to define the events . therefore , the detection of a ps - set can be done by observing this characteristic of the conditional poincar map . for rssler - like oscillators , in which the trajectory spirals around an equilibrium point , the ps - set is confined within an angular section . having found the ps - set , we can study properties of these sets that give us the level of organization and coherence of the oscillators a ps - set , @xmath89 , is said to have length 1 , if the set is constructed by the time series of the same events . defining the event to be given by the crossing of the trajectory to a poincar plane , the corresponding time series of events @xmath96 is given by @xmath97 . for this ps - set , points in @xmath89 are mapped in @xmath89 , after one application of the conditional poincar map . a ps - set is said to have length 2 , if it is obtained by a time series of two different events . so , a ps - set can be constructed from more complex series of events . we construct a length-2 basic set using a time series of events @xmath98 given by @xmath99 @xcite . as an example , for the perturbed chua s circuit , the length-2 basic set is constructed by a stroboscopic map that collects points every half period of the forcing . a length-2 basic set is assumed to be composed by two other subsets , named minimal sets , the subsets @xmath100 and @xmath101 . they have the property that if a point @xmath102 is such that @xmath103 , this point , iterated by the conditional poincar map , goes to the minimal set @xmath101 , and if @xmath102 is such that @xmath104 , this point , iterated by the conditional poincar map , goes to the minimal set @xmath101 . thus , points in @xmath101 are mapped to itself after 2 applications of the conditional poincar map . the minimal set @xmath101 is said to be disjoint to the set @xmath105 , if they do not intersect , i.e. , @xmath106 . for some systems that present a strong phase - coherent state , as the ones here studied , in which the instantaneous trajectory velocity does not differ too much from the average velocity on typical orbits , it is possible to find length-2 basic set , with disjoint minimal sets , when ps is present . for a general case , we do not expect to find a length-2 ps - set with disjoint minimal sets . as an example , one can think of a spiking - firing oscillator , phase synchronized with a periodic forcing . due to the fact that the spiking - firing dynamics has a fast and a slow mode , the conditional poincar maps might overlap . here , we explain through a diagram the possible emerging sets from the conditional poincar map . @xmath107^{\mathcal{p}_j } \ar[ld]_{\mathcal{p}_j \circ t } & \\ \mathcal{d}_j \ar[rd ] & & \mathcal{x}_j \ar[ld]\\ & \mathcal{d}_j \stackrel{?}{\equiv } \mathcal{x}_j \ar[ld]_{\txt{yes } } \ar[rd]^{\txt{no } } & \\ type - a & & p - set \ar[d ] \\ & ps \ar@{<=>}[r ] & ps - set } \ ] ] starting from the chaotic attractor @xmath7 , the set @xmath88 is constructed from the conditional poincar map , represented by @xmath108 , we project @xmath88 and @xmath7 into the subspace @xmath6 , obtaining the sets @xmath89 and @xmath8 , respectively . we classify the set @xmath89 into type - a set or p - set , by checking whether the conditional poincar map is transitive in @xmath8 , i.e. , by verifying the equivalence between the sets @xmath8 and @xmath89 . then , if the p - set is localized in the vicinity of the poincar section where the events occur , the p - set is a ps - set , which means that ps is present . for applying our formalism to the periodically forced chua s circuit , the event times are @xmath109 , with @xmath110 representing the forcing period . the time series for the length-2 basic set is given by @xmath111 . everywhere inside the @xmath71 region , we find length-1 [ as an example , see fig . [ ps_fig7](a ) ] and length-2 [ as an example , see fig . [ ps_fig3](a ) ] * ps*-sets , this latter with disjoint minimal sets . in fig . [ ps_fig3](a ) , the application of the conditional poincar map in the minimal set @xmath112 leads to the minimal set @xmath113 . both sets form a length-2 ps - set . note that both @xmath112 and @xmath113 can be regarded as a length-1 ps - set . to our numerical precision , we have checked that there are not basic sets beneath the synchronization region tip . which means that * type - a * set is present for very small but finite amplitude forcing . outside the @xmath71 region , there is a * p - set * , i.e. , a non - transitive conditional poincar map on the chaotic attractor projection . so , @xmath114 . in this case , there is not ps . more examples of length - q ps - sets in the phase synchronous forced chua s circuit can be seen in @xcite . we can use the formalism of the conditional poincar map to study the appearance of ps in coupled chaotic systems , as the two coupled rssler oscillators given by : @xmath115 with @xmath116 , and @xmath117 . the index denotes systems 1 and 2 . the subspaces @xmath6 are defined by @xmath6=@xmath118 . in a coupled chaotic system , @xmath119 ( @xmath120 ) does not increase uniformly , but it is given by the time the trajectory crosses the poincar plane @xmath121=0 ( @xmath122=0 ) . for these times , and using the time series @xmath98 , we construct the minimal sets @xmath112 and @xmath113 ( @xmath123 and @xmath124 ) . the parameters are @xmath125=0.01 and @xmath126 = 0.001 . in fig . [ ps_fig3](b ) , we show a length-2 basic set with disjoint minimal sets . the application of the conditional poincar map in the minimal set @xmath123 leads to the minimal set @xmath124 , and vice - versa . these two sets form together a length-2 ps - set , but each one separated can be regarded as a length-1 ps - set . the characteristic of these ps - sets is that they appear as localized structures around the poincar section chosen to define the events . as one can see , the set @xmath123 is localized in the neighborhood of the line @xmath127 , where the poincar section is chosen . in fact , this length-2 ps - set with disjoint minimal sets ( as well as a length-1 ps - set ) is found everywhere in the ps region , as shown in fig . [ ps_fig4 ] . in it , filled squares represent parameters for which these special ps - sets are found , and empty circles parameters for which ps exists . a ps - set of length - q is detected using in eq . ( [ practice_basic_sets ] ) @xmath128 , from which we can check whether @xmath129 occupies ( * type - a * discrete set @xmath88 ) the whole space occupied by @xmath130 . the set @xmath131 in eq . ( [ practice_basic_sets ] ) is constructed assuming squares of size @xmath132 , in points of the set @xmath128 . in fig . [ ps_fig4 ] , we show a case for @xmath133=2 . in fig . [ ps_fig40](a ) , we show the attractor in the subspace @xmath26 , the subset @xmath130 in gray , and the discrete set @xmath134 in dark empty circles . in ( b ) , we show a magnification of the box in ( a ) . note that neighborhoods of arbitrary points in the trajectory of @xmath130 ( gray ) always contain a point of the discrete set @xmath134 ( dark empty circles ) . so , the set @xmath130 is equivalent to the set @xmath134 and , therefore , @xmath134 is not a ps - set , and therefore , it does not exist ps . differently from the chua s circuit , the rssler coupled system presents no p - sets for parameters outside of the ps region . the approach presented in this paper can be extended for non - coherent attractors . as noticed in ref . @xcite , attractors that present non - coherent phase motion in the phase space may present a coherent motion in the space of the velocities , i.e. @xmath135 , which is the case of the funnel attractor @xcite . in this case , the extension of our approach to the non - coherent phase motion is straightforward . instead of defining a conditional poincar map in the phase space , we analyze the dynamics in the velocity space , in which the phase is coherent , and therefore , all the ideas introduced herein can be applied in this new space . to some chaotic attractors , it might not be possible to define a poincar section to construct the conditional poincar map . however , one can still construct these conditional maps by defining different types of events . for example , in coupled neurons , this event can be chosen as being a beginning or the ending of the bursts , which can be well defined by the crossing of the trajectory with some given threshold . a chaotic set is always transitive through the flow . so , given a set of initial conditions , its evolution through the flow eventually reaches arbitrary open subsets of the original chaotic attractor . however , a stroboscopic map of the flow , whose generalization is here called conditional poincar map , might not possess the transitive property . that is , given a set of initial conditions , its evolution through the conditional poincar map * _ might _ * not reach arbitrary open subsets of the chaotic attractor . the introduction of the term `` conditional '' in the map nomenclature comes from the unconventional and non rigidly way we adapt the established definition of a stroboscopic poincar map . for coupled chaotic oscillators , this conditional map is constructed based on events , conveniently chosen at the same subspaces where the phase of a chaotic system is defined . in addition , the application of this map through the flow , which results in a discrete set , is inspected not in the whole phase space , but also in the same subspaces where the phase is defined . if phase synchronization exists the conditional map generates special discrete sets , named ps - sets . the contrary is also true . i.e. , a ps - set implies phase synchronization . this was illustrated in the periodically forced chua s circuit , in the coupled rssler oscillator , and in other more general ( topologically equivalent ) coupled chaotic systems . the ideas introduced here provide an efficient way of detecting phase synchronization , without having actually to measure the phase of the chaotic trajectory . indeed , this detection can be done in experiments in a real - time , as it was done here in the perturbed chua s circuit . it is worthy to say that the ps - sets are robust under small additive noise that could corrupt the data in an experiment . this is so , because a small additive noise does not interfere much with the time that the trajectory crosses the poincar section , but just deviates the timing of the crossing . then , the ps - sets still remains . this robustness against the noise is an important property in order to apply these ideas in experiments , as done in this work . we have also introduced the phase as a quantity that measures the velocity of rotation of a projection of the tangent vector along the trajectory . this definition can be used to arbitrary flows , independently whether or not they present a coherent or non - coherent phase dynamics . finally , our formalism of the conditional poincar map can be used in coupled maps ( or perturbed ) to detect synchronous behavior ( not * phase * synchronization ) between the systems ( or between the forcing ) , for the case where one does not find full synchronization between the maps . one particular example where that happens is in the periodically forced logistic equation @xcite or for a system of coupled logistic maps @xcite , where one can find a finite number of synchronous chaotic subsets , the basic sets . * acknowledgment * research partially financed by fapesp , alexander von humboldt foundation , and cnpq . next , we present some formalism using elements of ergodic theory in order to introduce the conditional poincar map and the @xmath89 sets . given a flow @xmath136 , we call @xmath7 the chaotic attractor . a subset of the attractor is its projection on the subspace @xmath6 , that we call @xmath8 . we assume @xmath137 . the notion of stroboscopic map of a flow can be generalized to any temporal translation on the trajectory . we define the temporal translation to be a transformation represented by @xmath138 , called conditional poincar map . the initial condition @xmath139 is iterated under the temporal transformation @xmath90 , to the point @xmath140 . applying the transformation @xmath90 in a typical trajectory , for the time sequence @xmath141 , @xmath142 , @xmath143 , give us the points @xmath144 , @xmath145 , and @xmath146 . so , @xmath145 is the point @xmath144 integrated by the flow for a time interval given by @xmath147 . applying the transformation @xmath90 , for an infinite series of @xmath141 , that is @xmath148 , gives us a discrete set that we call @xmath88 . the projection of @xmath88 on the subspace @xmath6 is named @xmath89 . now , we introduce the notion of transitivity . let us assume we have a chaotic set @xmath149 . let us choose two disjoint subsets @xmath150 and @xmath151 in @xmath149 . so , @xmath152 . there is a transformation @xmath153 that generates the set @xmath149 , with the property that @xmath154 and @xmath155 . so , clearly the transformation @xmath153 and its @xmath156-fold application ( @xmath157 ) can always place an arbitrary initial condition , belonging either in @xmath150 or @xmath151 , anywhere in the set @xmath149 . this property makes the transformation @xmath153 to be transitive in the set @xmath149 . however , the conditional poincar map might no longer be transitive . for example , in the case of the @xmath158-fold of the transformation @xmath108 , i.e. , @xmath159 . given an initial condition in the set @xmath150 , its iteration by @xmath159 will never reach the set @xmath151 . therefore , the conditional poincar map @xmath159 is not a transitive transformation in the set @xmath149 . in the case of flows , we have seen that the temporal transformation represented by @xmath90 , whenever ps is present , does not place arbitrary points of the attractor everywhere in the subsets @xmath8 of the chaotic attractor . therefore , if ps is present the transformation @xmath90 is not transitive to this subset of the attractor . the conditional poincar map @xmath90 is topologically transitive in a set @xmath149 @xcite if for any two open sets @xmath160 , @xmath161 to detect whether the conditional poincar map is transitive in some subset , we introduce the notion of equivalence . two sets @xmath149 and @xmath150 are ( not ) equivalent [ equivalence is represented by the symbol * @xmath92 * , and non equivalence by * @xmath162 * ] if they ( do not ) occupy the same neighboring space . in a more general way : two sets @xmath149 and @xmath150 are equivalent , @xmath163 , if @xmath164 , a set @xmath151 can be constructed by the union of open sets @xmath165 , open @xmath166 volumes centered at @xmath167 with length @xmath168 , such that @xmath169 @xmath170 @xmath171 , and @xmath172 if @xmath173 . the set @xmath88 is a p - set if @xmath174 . if @xmath89 can be decomposed into a collection @xmath175 of subsets of @xmath176 , with @xmath177 , such that a point in @xmath178 iterated by the conditional poincar map goes to @xmath179 , we refer to each minimal set @xmath180 as a _ recurrent decomposition_. in the particular case we have @xmath181 ( constant ) this minimal set is a _ periodic decomposition_. the number of sets @xmath133 is called _ length _ of the decomposition @xcite . if @xmath182 is transitive on @xmath8 , then @xmath89 @xmath183 . if @xmath182 is non transitive in @xmath8 , then a subset @xmath149 can be constructed , such that @xmath184 and @xmath185 . proofs of propositions 1 and 2 can be done by using eq . ( [ transitive ] ) and the definitions . definition 3 can be understood by the ps - set of length-1 and length-2 considered in this work . if the length-1 ps - set is constructed by the time series @xmath97 , the length-2 is obtained by the time series given by @xmath99 @xcite . for the particular case of a periodically forced system , @xmath186=@xmath41 , and so , the length-1 ps - set is the standard stroboscopic map , that collects point every period of the forcing . the length-2 ps - set is constructed by a stroboscopic map that collects points every half period of the forcing . for the length-2 ps - set , we have two minimal sets , the sets @xmath100 and @xmath101 , with the property that if @xmath187 , this point iterated by the conditional poincar map goes to @xmath101 and @xmath101 goes to @xmath100 , under the conditional poincar map . to check whether @xmath94 , we do the following , for @xmath188 it exists @xmath189 such that @xmath190 where @xmath131 is a open ball of radius @xmath91 centered at the point @xmath167 . @xmath91 is a small positive value . in this appendices , we show that a ps - set exists if , and only if , phase synchronization exists , which in other words ps - set implies ps and ps implies ps - sets . given a dynamical system @xmath191 , let @xmath192 be the flow and @xmath7 the attractor generated by it , we suppose that we have a chaotic dynamics now on . let @xmath193 be the poincar section in the subspace @xmath6 , and let @xmath194 be the poincar map associated to the section @xmath193 , such that given a point @xmath195 , thus @xmath196 = @xmath197 . from now on we use a rescaled time @xmath198 . for a slight abuse of notation we omit the symbol @xmath199 . to show the _ if _ in the preposition let us start by considering the time interval associated to the return of the point @xmath200 to the point @xmath201 is @xmath202 = @xmath203 , with @xmath142 being the times at which the subsystem @xmath8 crosses the poincar section @xmath193 , in the subspace @xmath6 . as already introduced , the average return time is given by @xmath204 , and the time is rescaled , such that @xmath205 our hypothesis is that the subsystem @xmath8 has a phase - coherent oscillation , so there is a number @xmath206 for which holds @xcite : @xmath207 the number @xmath208 measures the coherence in the phase oscillation , and is linked to the phase diffusion @xcite . this equation holds for all @xmath24 , so it implies that for a single oscillation is also true that @xmath209 . in the case of two systems that present ps , it holds @xcite : let us introduce a new variable that measures the difference between the time interval of two events in @xmath130 and @xmath213 . this new variable is @xmath214 , from eq . ( [ diferenca_temporal ] ) , it is true that @xmath215 . now , we analyze one typical oscillation . given the following initial conditions @xmath216 and @xmath217 , we evolve both until @xmath218 returns to @xmath219 . in other words , we evolve both initial conditions for a time @xmath220 . so , @xmath221 . analogously , @xmath222 . so , given an initial condition in @xmath219 ( subsystem @xmath130 ) evaluated by the time initial conditions return in the section @xmath225 ( of subsystem @xmath213 ) , it returns near the section @xmath219 , and vice - versa . for a general case , we have to show that an initial condition , on the section @xmath219 , evolved by the flow for the time @xmath226 still remains close to this section . in other words , we have to show that the approximation in eq . ( [ tiagoiii ] ) is valid for an arbitrary number of events @xmath24 in the subspace @xmath213 . now , noting that @xmath227 , from our hypotheses of phase coherent dynamics @xmath228 . the same arguments used to derive eq . ( [ tiagoiii ] ) for one oscillation , can be extended to an arbitrary number @xmath24 , so we proof the _ if _ of the proposition ( ps implies the existence of ps - sets ) . now , to show the _ only if _ ( ps - sets imply ps ) , let us say that there is a ps - set . as a consequence , eq . ( [ diferenca_temporal ] ) is valid . then using the definition of phase coherence @xcite , and noting that in ps the average return times in a given poincar section are the same , we see that eq . ( [ diferenca_temporal ] ) comes from the boundness of the phase , concluding the proposition . furthermore , let us suppose that the trajectory of the oscillators are perturbed by a small perturbation , which does not destroy the phase synchronous dynamics . the effect of the small perturbation in the time return of the trajectory to a poincar section is to deviate this time according @xmath229 , where @xmath230 represents the perturbation in system @xmath231 at the moment of the i - th event , with @xmath232 . under these hypotheses about the perturbation , we conclude the following result . this result shows that a ps - set can be constructed in ps states . moreover , for the coupled rssler - like systems , this result states that this set is confined in an angular region , which is a consequence of eq . ( [ diferenca_temporal ] ) . to see the relation between the constant @xmath233 and the size of the ps - set , we do the following . by the time normalization , average time interval between points in phase space are proportional to their distances . so , from eq . ( [ tiagoiii ] ) we write @xmath234 , with @xmath235 being the average half length of the ps - set . a rough calculation shows that in our experiment with the chua s circuit , we have that @xmath236 , and for the coupled rssler oscillators , we have that @xmath237 . this results completely agree with the theoretical approach done in @xcite . a. pikovsky , m. rosenblum , and j. kurths , _ synchronization : a universal concept in nonlinear sciences _ , cambridge university press , 2001 ; s. boccaletti , j. kurths , g. osipov , d. valladares , and c. zhou , phys . rep . * 366 * ( 2002 ) 1 . phase can be defined as a hilbert transformation of a trajectory component , as the angle of a trajectory point into a special projection of the attractor , as a function that grows @xmath31 every time the chaotic trajectory crosses some specific surface , or as the angle of the projected vector field into some subspace with respect to some rotation point . a chaotic set is always transitive through the flow . so , given a set of initial conditions , its evolution through the flow eventually reaches arbitrary open subsets of the original chaotic attractor . however , the conditional poincar map , might not possesses the transitive property . that is , given a set of initial conditions , its evolution through this map * _ might _ * not reaches arbitrary open subsets of the especial projections of the chaotic attractor , its dynamics stay confined to a subset of the attractor . @xmath34 is considered to be a rational . however , as shown in ref . @xcite , ps , as defined by the boundness of the phase difference was found in two chaotic systems for a finite but very large time interval , as @xmath34 approaches an irrational . therefore , although in this work we consider @xmath34 to be rational , we should make the remark that for the especial situation as the one presented in ref . @xcite , eq . ( [ ps ] ) can only be satisfied for a finite but large time if @xmath34 is considered to be irrational . the intermittency observed in the phase difference is characterized as an usual intermittency , by the alternance between a laminar regime and a burst regime . if the phase difference remains bounded in the interval @xmath238 $ ] , we say that we have a laminar regime . if the phase difference leaves this interval , we have a burst , also known as phase slip . as one approaches the border between the ps and the non ps region , the laminar regime in the phase difference becomes longer . the choice of these time intervals is not unique . it depends on what type of event one wants to identify in the system . a length-1 basic set is constructed by measuring time intervals between the occurrence of two events of the same type . @xmath239 length-2 basic set is constructed by measuring the time interval between the occurrence of an event @xmath239 and an event @xmath240 , and then , between the event @xmath240 and finally the event @xmath239 , and so , on .

in his classical paper on quantisation of magnetic poles dirac ( 1931 ) remarked that `` non - euclidean geometry and non - commutative algebra , which were at one time considered to be purely fictions of the mind and pastimes for logical thinkers , have now been found to be very necessary for the description of general facts of the physical world . it seems likely that this process of increasing abstraction will continue in the future and that advance in physics is to be associated with a continual modification and generalisation of the axioms at the base of the mathematics rather than with a logical development of any one mathematical scheme on a fixed foundation . '' in accordance with this principle , the present paper further extends the work of dirac by exploring the ` dequantisation ' of magnetic poles . diverse and numerous versions of the magnetic pole construction and the associated charge quantisation condition of dirac ( 1931 , 1948 ) have appeared in the literature , but the basic model can be most concisely and accurately formulated in terms of the hopf fibration ( wu & yang 1975 ; trautman 1977 ; yang 1977 ) . from a mathematical viewpoint , the rationale is as follows . the monopole potential is modelled as a connection @xmath0 on a nontrivial principal @xmath1-bundle @xmath2 over a subset of minkowski space . a charged matter field interacting with the monopole is accordingly modelled as a section of a vector bundle @xmath3 associated with @xmath2 via the representation @xmath4 of @xmath1 on a complex vector space . since @xmath0 then induces the connection @xmath5 on @xmath3 , the integer @xmath6 can be identified with the electric charge of the matter field ( see urbantke 2003 for a detailed account of the description of a magnetic monopole in the language of the hopf fibration ) . however , following the spirit of dirac , if we describe the monopole by a more general mathematical scheme , then the interaction of matter fields and magnetic poles _ with arbitrary real charges _ can also be modelled in a consistent manner . the present paper introduces a new sheaf - theoretic framework permitting an explicit construction of arbitrarily charged magnetic monopoles . this framework is likewise applied to generic @xmath1-bundles and also yields , as a by - product , the notion of a quasispin structure defined on arbitrary space - times . the results suggest that topological quantisation in general can be viewed from a more flexible perspective . although magnetic monopoles have not been observed experimentally , one important physical consequence of the present model is that their detection would not necessarily imply the quantisation of electric charge . likewise , an observed violation of charge quantisation would not necessarily imply the nonexistence of magnetic monopoles . furthermore , grand unified theories require the existence of magnetic monopoles , and , according to conventional field theory described in terms of manifolds this necessarily implies charge quantisation , hence , the detection of nonintegral electric charge would indicate that the present scheme , based on sheaves rather than manifolds , may be physically more realistic . also , a sheaf construction permits the global description of fermion fields on lorentzian manifolds possessing no conventional spin structure , as an alternative to the cumbersome machinery of khler fermions . the paper is organised as follows . in [ sec:2 ] we present a very brief sketch of sheaf theory for the benefit of readers less acquainted with the subject . this is intended to provide the bare minimum of information necessary for following the ensuing discussion ; for further details on sheaf theory , see , e.g. , bredon ( 1997 ) , kultze ( 1970 ) , or wells ( 2008 ) . in [ sec:3 ] we develop the basic mathematical machinery used in later sections . the key idea here is the construction of a principal @xmath7-sheaf bundle which generalises the conventional notion of a principal bundle . in [ sec:4 ] we prove that equivalence classes of principal @xmath8-bundles can , under certain hypotheses , be mapped injectively into equivalence classes of @xmath7-sheaf bundles . this implies that the dirac sheaf bundle constructed in [ sec:5 ] does indeed constitute a generalisation of the conventional dirac monopole . the spherically symmetrical connection and curvature of the dirac sheaf bundle are constructed in [ sec:6 ] , demonstrating that the magnetic charge of this model can assume arbitrary real values . the interaction of the generalised monopole with a charged matter field is considered in [ sec:7 ] . dequantisation of more general @xmath1 bundles is considered in [ sec:8 ] , with particular attention to gravitational and electromagnetic instantons . in [ sec:9 ] the basic machinery developed earlier is applied to spin structures and dirac operators . in [ sec:10 ] the paper concludes with a brief discussion of possible implications in diverse areas of physics . the concept of a sheaf over a manifold @xmath9 provides a way of interpolating local data and global data on x. we begin with the definition of a presheaf . a _ presheaf _ @xmath10 on @xmath9 is a functor assigning , to each open @xmath11 , a group @xmath12 , abelian or otherwise , such that for each @xmath13 the restriction map @xmath14 , @xmath15 , defines a homomorphism and such that @xmath16 for @xmath17 . an element @xmath18 is referred to as a _ section _ of @xmath12 over @xmath19 . the restriction of @xmath20 on @xmath13 is thus given by @xmath21 . the sections @xmath18 and @xmath22 are said to be _ equivalent _ at @xmath23 if there is a neighbourhood @xmath24 of @xmath25 such that @xmath26 . the equivalence class containing @xmath18 is called the _ germ _ of @xmath27 at @xmath25 , and the set of all such germs for any fixed @xmath25 is denoted by @xmath28 . the disjoint union @xmath29 of all the sets @xmath28 provides local information about the structure of @xmath10 . however , information concerning global structure has been lost , since we have discarded relations between the @xmath28 for varying @xmath25 . to retrieve some global structure , we introduce a topology in the following manner . for a fixed @xmath18 the set of all germs @xmath30 for @xmath31 is taken to be an open set in @xmath29 , and the topology of @xmath32 is defined as that generated by these open sets . the projection @xmath33 mapping @xmath28 into @xmath25 has the property that for any point @xmath34 with @xmath35 there is a neighbourhood @xmath36 for @xmath18 and @xmath37 such that the restriction @xmath38 is a homeomorphism onto a neighbourhood of @xmath39 . these ideas can be summarised as follows : a _ sheaf _ of groups on @xmath9 is a pair @xmath40 such that * @xmath29 is a topological space ( in general , not hausdorff ) ; * @xmath41 is a local homeomorphism of @xmath29 onto @xmath9 ; * each @xmath42 , @xmath43 , is a group called the _ stalk _ of @xmath29 at @xmath25 ; and * the group operations are continuous with respect to the relative topology on the subset @xmath44 of @xmath45 . [ def:1 ] for @xmath11 , a continuous map @xmath46 such that @xmath47 is called a _ section _ of @xmath32 over @xmath19 . the totality of such @xmath27 will be denoted @xmath48 . note that every element of the group @xmath12 specified by the presheaf functor naturally determines an element of @xmath48 , but the converse is only true locally . some of the key properties of a sheaf are as follows . if @xmath49 is an open covering of an open set @xmath50 , and if @xmath51 are such that @xmath52 . furthermore , if @xmath53 are such that latexmath:[$\sigma_\alpha|_{u_\alpha\cap u_\beta } = \sigma_\beta there exists an element @xmath18 with @xmath55 for all @xmath56 . the first property implies that if the restrictions of a pair of sections always agree , then the two sections are identical thus a section over @xmath19 is determined by the totality of its restrictions to subsets of @xmath19 . the second property , somewhat complementary to the first , implies that if pairs of sections always agree on their overlapping regions then global section can be constructed from the local data thus a section over @xmath19 may be assembled from consistent local sections on subsets of @xmath19 . a sheaf @xmath29 contains localised information concerning the topological space @xmath9 . global information about @xmath9 can then be extracted from @xmath29 by consideration of exact sequences , quotients , and so on . given two sheaves @xmath29 and @xmath57 over @xmath9 , a sheaf homomorphism @xmath58 is a continuous map such that the stalk map @xmath59 is a homomorphism of @xmath60 into @xmath61 for each @xmath43 . a sequence of sheaf homomorphisms of the form @xmath62 such that the corresponding sequence of stalk maps is exact for all @xmath43 is called a short exact sequence of sheaves . evidently , exactness is a local property . given a short exact sequence of sheaves , the induced sequence @xmath63 is exact at @xmath64 and @xmath65 , but in general not at @xmath66 . that is to say , the local exactness of a sequence does not imply exactness with respect to the global sections over @xmath9 . the measure of inexactness at @xmath66 can then be characterised by cohomology . recall that in the cohomology theories of @xmath9 one computes @xmath67 where @xmath8 is an abelian group . in sheaf cohomology the coefficients are not elements of a fixed group @xmath8 but are , rather , local sections of some sheaf @xmath29 over @xmath9 . more precisely , let @xmath68 be an open covering of @xmath9 . for any @xmath69 such that @xmath70 , we define the set of @xmath71-cochains by @xmath72 . for any @xmath73 define the coboundary operator @xmath74 by @xmath75 where @xmath76 and @xmath77 is the sheaf restriction map . these coboundary operators define a complex @xmath78 and the cohomology groups of this complex are then defined in the usual manner : @xmath79 passing to a direct limit over progressively finer coverings , we obtain the sheaf cohomology groups @xmath80 . the reader will notice that the definitions in this section run parallel , _ mutatis mutandis _ , to the conventional definitions in the theory of fibre bundles . the @xmath7-sheaf plays the role of a trivial and the @xmath7-sheaf bundle that of a generally nontrivial fibre bundle . let @xmath9 and @xmath81 be topological spaces and @xmath82 a sheaf of groups over @xmath9 . @xmath7-sheaf _ over @xmath9 is a triple @xmath83 such that * @xmath41 is a local homeomorphism of @xmath81 onto @xmath9 ; * for each @xmath84 , the stalk @xmath85 operates ( by left action ) upon @xmath86 ; and * if @xmath87 is equipped with the relative topology in @xmath88 , then the mapping @xmath89 defined by @xmath90 is continuous . [ def:2 ] for any subset @xmath91 , the obvious restriction maps define a @xmath92-sheaf @xmath93 over @xmath0 . in the sequel , when @xmath0 is clearly understood , we shall , for brevity , use the term @xmath7-sheaf in place of @xmath94-sheaf . we call @xmath29 a _ principal @xmath7-sheaf _ if @xmath95 and @xmath7 operates by left translation . given two @xmath7-sheaves @xmath96 and @xmath97 over @xmath9 , a continuous map @xmath98 will be called a _ @xmath7-sheaf map _ of @xmath29 into @xmath57 provided * @xmath99 ; and * for each @xmath84 , the induced map @xmath100 satisfies @xmath101 for all @xmath102 and @xmath103 . [ def:3 ] again , if @xmath104 , then a @xmath105-sheaf map will , for brevity , be called a @xmath7-sheaf map when the restriction is clearly understood . a _ @xmath7-sheaf isomorphism _ is a bijective @xmath7-sheaf map . a _ @xmath7-sheaf bundle _ @xmath106 over a topological space @xmath9 is defined by the following data : * an open covering @xmath107 of @xmath9 ; * for each @xmath108 , a @xmath7-sheaf @xmath109 over @xmath110 ; and * for each nonempty intersection @xmath111 , a @xmath7-sheaf isomorphism @xmath112 such that the cocycle condition @xmath113 is satisfied on @xmath114 . [ def:4 ] we call @xmath106 a _ principal @xmath7-sheaf bundle _ if each @xmath109 is a principal @xmath115-sheaf . two @xmath7-sheaf bundles @xmath106 and @xmath116 over @xmath9 , defined in terms of the respective open coverings @xmath117 and @xmath118 , are _ equivalent _ provided that the @xmath7-sheaf bundles @xmath119 and @xmath120 induced by the respective restriction maps on some common refinement @xmath121 of @xmath117 and @xmath122 are such that there exist @xmath7-sheaf isomorphisms @xmath123 for each @xmath108 satisfying @xmath124 for all @xmath125 . remark : of academic interest is the fact that , given a @xmath115-sheaf bundle @xmath106 , one can , from the presheaf defined by local sections of @xmath106 , define a @xmath7-sheaf @xmath126 which in turn uniquely determines the structure of @xmath106 up to equivalence ( kashiwara & schapira 2006 ) . however , this fact will not be required in the sequel . [ def:5 ] let @xmath9 be a smooth manifold , @xmath8 a lie group , and @xmath127 a closed central subgroup of @xmath8 . for each open subset @xmath19 of @xmath9 , let @xmath128 denote the totality of smooth maps @xmath129 , which forms a group under pointwise multiplication , i.e. @xmath130 . let @xmath131 denote the totality of constant maps @xmath132 , which forms a central subgroup of @xmath128 . assign to each open @xmath133 the quotient group @xmath134 with the obvious restriction maps , this defines a presheaf ; denote by @xmath135 the corresponding sheaf of groups over @xmath9 , regarded as a principal @xmath136-sheaf , and its restriction to a subset @xmath0 of @xmath9 by @xmath137 . a principal @xmath7-sheaf bundle over @xmath9 such that @xmath138 for some open covering @xmath139 of @xmath9 will , for brevity , be called a _ @xmath140-bundle_. _ example 1_. with the notation of definition [ def:4 ] , let @xmath141 be a principal @xmath8-bundle , and @xmath142 an open covering of @xmath9 by trivialising neighbourhoods . on each nonempty intersection @xmath143 the bundle transition function @xmath144 defines an element of @xmath145 and hence a continuous section @xmath146 of the sheaf @xmath147 . the germs of @xmath148 , acting by right multiplication , provide a @xmath7-sheaf map @xmath149 , which is obviously bijective , i.e. is a @xmath115-sheaf isomorphism . moreover , the cocycle condition ( [ eq:3.3 ] ) follows immediately from the corresponding cocycle condition for the bundle transition functions . thus , we obtain a @xmath140-bundle @xmath150 over @xmath9 . the following result shows that under a certain simple hypothesis the conventional principal @xmath8-bundles may be identified with a subset of the @xmath140-bundles over @xmath9 . throughout the sequel , we shall assume that @xmath9 is paracompact , i.e. that every open covering of @xmath9 has a locally finite subcovering . [ prop:1 ] for fixed @xmath127 , the correspondence @xmath151 induces an injective map of equivalence classes of principal @xmath8-bundles into equivalence classes of @xmath7-sheaf bundles over @xmath9 , provided that the ech cohomology group @xmath152 . we first prove three lemmata . [ lem:2.2 ] let @xmath8 be an arbitrary group . a mapping @xmath153 which commutes with all left translations is a right translation . proof . given that @xmath154 for all @xmath155 , let @xmath156 . then @xmath157 for all @xmath158 . @xmath159 recall that @xmath160 denotes the totality of continuous sections of a sheaf @xmath29 over a subset @xmath161 . [ lem:2.3 ] let @xmath19 be an open subset of the topological space @xmath9 , @xmath8 a group , @xmath29 a principal @xmath7-sheaf over @xmath9 and @xmath162 a @xmath7-sheaf isomorphism . then , for each point @xmath163 , there exists a neighbourhood @xmath164 of @xmath25 in @xmath19 and a @xmath165 such that @xmath166 for every section @xmath167 and all @xmath168 . proof . by lemma [ lem:2.2 ] and condition ii ) of definition [ def:3 ] , for each @xmath163 there exists a @xmath169 such that @xmath170 for all @xmath171 . now , fix @xmath172 , and choose any @xmath173 and any section @xmath174 with @xmath175 open in @xmath19 , @xmath176 and @xmath177 . since @xmath178 is continuous and stalk - preserving , @xmath179 ( @xmath180 ) defines a continuous section of @xmath29 , and by the continuity of the group operations @xmath181 @xmath182 is also a continuous section . for any other section @xmath183 , writing @xmath184 we have @xmath185 . hence , @xmath186 has the desired property . @xmath159 [ lem:2.4 ] let @xmath187 and @xmath188 be sheaves , @xmath189 an epimorphic sheaf map and @xmath190 . then , for any @xmath84 , there exists a neighbourhood @xmath164 of @xmath25 and a @xmath191 such that @xmath192 . this is an elementary fact of sheaf theory ( kultze 1970 , hilfsatz 3.4 ) . @xmath159 proof of proposition [ prop:1 ] . clearly , equivalent principal bundles give rise to equivalent sheaf bundles . conversely , let @xmath193 and @xmath194 be principal @xmath8-bundles defined in terms of trivialising open coverings @xmath117 and @xmath122 , respectively . choosing , if necessary , a common refinement , we may suppose that @xmath195 . then , with the foregoing notation , @xmath196 iff there exist @xmath7-sheaf isomorphisms @xmath197 , @xmath108 , such that @xmath198 on @xmath199 ( @xmath125 ) . the @xmath7-sheaves @xmath200 and @xmath201 are identical , both arising from the presheaf functor @xmath202 , @xmath203 open in @xmath204 . let @xmath205 denote the sheaf over @xmath110 defined by the presheaf functor @xmath206 . then , the presheaf epimorphism @xmath207 induces an epimorphic sheaf map @xmath208 , hence , by lemmata [ lem:2.3 ] and [ lem:2.4 ] , for each @xmath209 there exists a neighbourhood @xmath210 and a section @xmath211 such that @xmath212 = [ \sigma_{\alpha}][g_{\alpha}^{x } ] = [ \sigma_{\alpha}g_{\alpha}^{x}]$ ] for all @xmath213 , where the square brackets denote cosets in @xmath214 , regarded as continuous sections in @xmath215 . then , choosing the covering @xmath216 sufficiently fine , we may , by paracompactness , omit the superscript @xmath25 and simply write @xmath212= [ \sigma_{\alpha}g_{\alpha}]$ ] for all @xmath217 , with square brackets denoting cosets in @xmath218 , again regarded as elements of @xmath219 . hence , recalling the definition of the @xmath220 in example 1 , relation ( [ eq:4.1 ] ) implies that @xmath221 = [ \sigma_{\alpha}g_{\beta}^{-1 } g_{\alpha\beta } g_\alpha ] \label{eq:4.2}\end{aligned}\ ] ] on @xmath222 . since @xmath223 is arbitrary , this clearly means that @xmath224 for certain constant functions @xmath225 applying the cocycle condition for principal bundles to both members of ( [ eq:4.3 ] ) , we deduce that the constants @xmath226 also satisfy the cocycle condition , and therefore define an element of @xmath227 . hence , if the covering @xmath228 is chosen sufficiently fine , then , by hypothesis , there exists a 0-cochain @xmath229 such that @xmath230 . defining @xmath231 we find that ( [ eq:4.3 ] ) becomes @xmath232 which is the condition for equivalence of @xmath2 and @xmath233 . @xmath159 as used in the foregoing proof , lemmata [ lem:2.3 ] and [ lem:2.4 ] show that for a sufficiently fine covering @xmath234 the transition sheaf isomorphisms of a @xmath140-bundle are of the form @xmath235 for fixed @xmath236 and arbitrary @xmath237 . likewise , the @xmath7-sheaf isomorphisms @xmath238 in the definition of @xmath7-sheaf bundle equivalence are of the form @xmath239 for fixed @xmath240 and arbitrary @xmath241 . a standard theorem states that equivalence classes of principal @xmath8-bundles over @xmath9 correspond biuniquely to elements of the sheaf cohomology set @xmath242 . in view of ( [ eq:4.6 ] ) , the argument used in proving this theorem ( see , e.g. , lawson & michelsohn 1989 , appendix a ) can also be applied in a straightforward manner to yield the following result . the equivalence classes of @xmath140-bundles over @xmath9 correspond biuniquely to elements of the sheaf cohomology set @xmath243 . for @xmath244 , this correspondence reduces to the standard theorem cited above . [ cor:4.05 ] if @xmath8 is abelian , then proposition [ prop:1 ] follows more directly from standard results as follows . let @xmath245 denote the sheaf arising from the constant presheaf defined by @xmath246 . the short exact sequence @xmath247 induces a sequence of sheaf maps @xmath248 which is easily seen to be exact . this in turn induces a long exact cohomology sequence which includes , in particular , the segment @xmath249 for a paracompact @xmath8-manifold @xmath9 the sheaf cohomology group @xmath250 is known to be isomorphic to the ech cohomology group @xmath251 as well as the singular cohomology group @xmath252 with coefficients in @xmath127 ( spanier 1994 , chapter 6 ) . hence , if @xmath253 , then the exactness of the above sequence implies that @xmath254 is injective . denoting the totality of equivalence classes of principal @xmath8-bundles ( resp . @xmath140-bundles ) by @xmath255 ( resp . @xmath256 ) , we note that the diagram @xmath257 where the upper arrow represents the correspondence @xmath258 and the vertical arrows the biunique correspondence of corollary [ cor:4.05 ] , is commutative . the result follows . relations ( [ eq:4.6 ] ) and ( [ eq:4.7 ] ) also imply : for the special case @xmath259 ( the identity of @xmath8 ) , the @xmath140-bundles can be identified with conventional principal @xmath8-bundles , and @xmath7-sheaf equivalence is merely conventional bundle equivalence . [ cor:4.5 ] thus , for @xmath259 the present theory yields nothing new . however , for @xmath260 and @xmath261 we obtain a nontrivial extension of the conventional theory of principal bundles , as we shall see , in particular , from the example in the next section , wherein @xmath262 . note that the set of data @xmath263 ( see definition [ def:5 ] and ( [ eq:4.6 ] ) ) , subsequently referred to as a _ presentation _ of the @xmath140-bundle under consideration , plays a role analogous to that of a system of local trivialisations in the conventional theory of principal bundles . if @xmath118 is a refinement of @xmath264 , then @xmath265 induces , in the obvious manner , a presentation @xmath266 associated with @xmath118 . recalling the definition of @xmath7-sheaf equivalence ( see ( [ eq:3.4 ] ) ) , relations ( [ eq:4.6 ] ) and ( [ eq:4.7 ] ) imply that two presentations @xmath265 and @xmath266 associated with the same covering @xmath267 define equivalent @xmath140-bundles iff there exist elements @xmath268 such that the respective transition elements @xmath236 and @xmath269 satisfy @xmath270 for all @xmath271 and @xmath272 . if the associated open coverings @xmath117 and @xmath118 are different , then the condition for equivalence is given by ( [ eq:4.8 ] ) with respect to some common refinement @xmath273 of @xmath117 and @xmath118 . another immediate consequence of ( [ eq:4.6 ] ) is the fact that @xmath140-bundles , like conventional @xmath8-bundles , are functorial , i.e. a @xmath140-bundle @xmath274 over @xmath9 and a continuous map @xmath275 naturally induce a @xmath140-bundle @xmath276 over @xmath277 , since the transition sections @xmath236 on @xmath9 pull back to sections on @xmath277 which obviously satisfy the cocycle condition . moreover , one can prove that if @xmath9 and @xmath277 are compact hausdorff spaces and the maps @xmath278 and @xmath279 are homotopic , then @xmath276 and @xmath280 are equivalent ( cf . lawson & michelsohn 1990 , appendix a ) . let @xmath281 and @xmath282 . we represent @xmath283 by the unit sphere in @xmath284 , with spherical polar coordinates @xmath285 , ( @xmath286 , @xmath287 ) , and @xmath1 by @xmath288 . for the open covering @xmath289 of @xmath283 , we have @xmath290 . let @xmath7 be the sheaf of smooth @xmath1-valued functions over @xmath283 . let @xmath10 denote the presheaf over @xmath283 defined for each open @xmath203 by @xmath291 @xmath29 the associated @xmath7-sheaf , and @xmath292 @xmath293 . if @xmath294 is an arbitrary real number , define a @xmath7-sheaf isomorphism @xmath295 by @xmath296 for @xmath297 , where @xmath298 is the section of @xmath299 defined locally by @xmath300 . \label{eq:5.5}\end{aligned}\ ] ] here the square brackets indicate the equivalence class modulo constant sections @xmath301 . denote the @xmath140-bundle so defined by @xmath302 . if @xmath303 is an integer , then @xmath304 is just the principal @xmath115-sheaf bundle @xmath305 arising from the conventional @xmath1-bundle @xmath306 of charge @xmath307 ( in the appropriate units ) , in accordance with the construction of example 1 . in general , for an arbitrary abelian group @xmath127 , the ech cohomology groups are given by @xmath308 ( see , e.g. , spanier 1994 ) . in particular , we have @xmath309 . hence , by proposition [ prop:1 ] , the sheaf bundles @xmath304 are mutually inequivalent . that @xmath310 implies @xmath311 for arbitrary real @xmath294 and @xmath312 will be proved in [ sec:6 ] below . from the foregoing construction , the truth of the following proposition should be evident . [ prop:5.1 ] let @xmath313 be the transition functions for some trivialising atlas of a principal @xmath1-bundle @xmath2 over the base space @xmath9 . then , for any real number @xmath294 , the powers @xmath314 define the transition isomorphisms of a @xmath315-bundle @xmath316 over @xmath9 . if the ech cohomology group @xmath317 , then the principal @xmath1-bundles @xmath306 defined by the transition functions @xmath318 , if mutually inequivalent , correspond bi - uniquely with mutually inequivalent @xmath315-bundles @xmath319 . the final assertion is merely an application of proposition [ prop:1 ] . the lie algebra of @xmath8 will be denoted by @xmath320 and the algebra of smooth @xmath320-valued differential forms on a manifold @xmath321 by @xmath322 . a _ connection _ @xmath323 on a @xmath140-bundle @xmath29 given in terms of a presentation @xmath324 is specified by a family of @xmath320-valued 1-forms @xmath325 satisfying is a matrix group . ] @xmath326 for all @xmath327 and @xmath328 . [ def:6 ] both terms on the right side of ( [ eq:6.1 ] ) are unambiguously defined as follows . the value of @xmath236 at @xmath329 is an element of @xmath330 , represented by @xmath331 for some neighbourhood @xmath164 of @xmath25 , and @xmath332 is in turn represented by a section @xmath333 . the first term on the right side of ( [ eq:6.1 ] ) is then defined by @xmath334 , and the second by @xmath335 . any two possible choices @xmath313 differ only by a constant factor in @xmath127 , which affects neither term . two such connections @xmath323 and @xmath336 defined in terms of the respective presentations @xmath265 and @xmath266 are _ equivalent _ , provided that , for some common refinement @xmath267 of the two respectively associated open coverings @xmath117 and @xmath118 with ( cf . ( [ eq:4.8 ] ) ) @xmath337 the relation @xmath338 holds for all @xmath339 and @xmath340 . here the right side of ( [ eq:6.3 ] ) is interpreted in the same manner as that of ( [ eq:6.1 ] ) . by choosing a sufficiently fine covering , the above - mentioned neighbourhoods @xmath164 may , by paracompactness , be identified with the trivialising neighbourhoods @xmath110 , so that the representative sections @xmath341 and @xmath342 may be regarded as elements of @xmath343 and @xmath344 , respectively . the @xmath341 then satisfy the cocycle condition modulo locally constant sections in @xmath127 , which suffices to ensure the mutual consistency of the relations ( 6.1 ) as the indices @xmath56 and @xmath345 vary over @xmath346 . the _ curvature _ of such a connection is defined in the conventional manner , i.e. for each @xmath110 we have @xmath347 . \label{eq:6.4}\end{aligned}\ ] ] by the same calculation as that used for conventional principal bundles , one finds , using ( 6.1 ) , that @xmath348 for @xmath328 . furthermore , if @xmath349 and @xmath350 are related by ( [ eq:6.3 ] ) , then @xmath351 if @xmath193 is a principal @xmath8-bundle and @xmath127 a central closed subgroup of @xmath8 , then an ordinary connection @xmath0 on @xmath2 clearly gives rise to a connection @xmath323 on @xmath352 in the sense of definition [ def:6 ] , and the conventional local curvature forms then coincide with those of @xmath323 as defined in ( [ eq:6.4 ] ) . by virtue of the relations ( [ eq:6.5 ] ) and ( [ eq:6.6 ] ) , characteristic classes of @xmath140-bundles can be defined and shown to be independent of the choice of the presentation and the choice of the connection @xmath323 , as for conventional principal bundles . we now define a connection on the dirac sheaf bundle @xmath353 ( @xmath354 ) . using the notation of [ sec:5 ] , let @xmath355 denote the presentation @xmath356 , and let @xmath357 \rd\phi \in \lambda^1(u_\alpha , { \mathfrak u}(1 ) ) \label{eq:6.7}\end{aligned}\ ] ] be the corresponding local 1-forms . on @xmath358 we have @xmath359 in accordance with ( [ eq:6.1 ] ) . the curvature is given by @xmath360 and hence the first chen class is @xmath361 , where @xmath362 is the area form on the unit 2-sphere . this shows that the sheaf bundles @xmath302 for distinct @xmath294 are inequivalent , as claimed above . for simplicity of exposition and to emphasise the analogy with the conventional hopf fibration @xmath363 , the dirac sheaf bundle and the corresponding monopole connection have been constructed over @xmath364 as base space . however , with a view to physical applications , the same construction applies verbatim if the base space is the subset of minkowski space defined by @xmath365 . again , expressing @xmath366 by spherical polar coordinates , formulae ( 5.1 ) , ( 5.2 ) and ( 5.6 ) remain unchanged . thus , for arbitrary real @xmath294 , one obtains a @xmath315-bundle over @xmath367 with field strength ( [ eq:6.9 ] ) . remark : according to the conventional bundle picture , there are no singularity - free gauge potentials @xmath368 with the properties that ( i ) their curls are equal to the field , and that ( ii ) they are related by gauge transformations on the overlapping region @xmath358 , unless electromagnetic charges are quantised ( wu & yang 1975 , theorem 3 ) . the above result demonstrates that this conclusion is not valid in the present generalised model . let @xmath8 be a lie group , @xmath127 a closed subgroup of @xmath8 , and @xmath369 a representation of @xmath8 on a ( real or complex ) vector space @xmath370 . let @xmath9 be a smooth paracompact manifold , and for each open @xmath371 , let @xmath372 denote the totality of smooth maps @xmath373 . if @xmath374 and @xmath375 , define @xmath376 by @xmath377 . denote by @xmath378 the quotient space of @xmath379 under this action of @xmath127 . if @xmath380 , then the restriction of @xmath379 to @xmath381 obviously commutes with the action of @xmath127 , and thus one obtains a restriction map @xmath382 . if @xmath383 , then @xmath384 , hence the system @xmath385 defines a presheaf @xmath386 , and a corresponding sheaf @xmath387 are to be interpreted with groups replaced by vector spaces and homomorphisms by linear transformations . ] next , for @xmath84 , define a left action of the stalk @xmath388 of the sheaf @xmath389 ( see definition 5 ) upon @xmath390 as follows . for some neighbourhood @xmath19 of @xmath25 , the germ @xmath391 is represented by an element @xmath392 and @xmath393 in turn by a section @xmath394 , while @xmath395 is represented by an element @xmath395 of @xmath378 and @xmath395 in turn by a section @xmath396 . define @xmath397 as the element of @xmath390 represented by the section @xmath398 . one can readily check that this is independent of the choices of @xmath399 and @xmath400 . let @xmath401 be a @xmath140-bundle defined by a presentation @xmath402 . a _ particle field _ is a system of sections @xmath403 satisfying the condition @xmath404 for every @xmath271 and @xmath272 . the totality of such particle fields will be denoted by @xmath405 . [ def:7 ] inspection of the foregoing construction clearly shows that for @xmath406 this definition effectively reduces to that of a conventional particle field ( cf . corollary [ cor:4.5 ] ) , i.e. a section of the associated bundle @xmath407 defined by a principal @xmath8-bundle and a representation @xmath408 . also , note that if @xmath409 , then @xmath405 is _ not _ a vector space . nevertheless , one can define covariant derivatives on @xmath410 , as is shown by the ensuing examples 2 and 3 . let @xmath411 be the local potentials of a connection @xmath323 with respect to some presentation @xmath265 , as in definition 6 . consider a point @xmath412 , and choose local coordinates @xmath413 in a neighbourhood @xmath414 of @xmath25 so that @xmath415 with @xmath416 . let @xmath417 and choose a section @xmath418 representing @xmath419 in some neighbourhood @xmath420 of @xmath25 , with @xmath421 . then define @xmath422(x ) = ( ( \partial_\mu - a_{\alpha\mu})v_\alpha)_x , \label{eq:7.2}\end{aligned}\ ] ] where @xmath423 acts upon @xmath424 in accordance with the lie algebra representation induced by @xmath425 . this germ is clearly independent of the choice of @xmath400 , and thus one obtains a well - defined section of @xmath426 . moreover , for @xmath427 , @xmath428 that is , if @xmath419 and @xmath429 are related by ( [ eq:7.1 ] ) , then so are their covariant derivatives . to establish ( [ eq:7.3 ] ) note that both sides of the equation are well - defined local sections of @xmath430 . hence we may choose arbitrary representatives @xmath424 , @xmath431 , and @xmath313 of @xmath419 , @xmath432 , and @xmath236 , respectively , to perform the calculation . we choose representatives satisfying the relation @xmath433 , and by a routine computation , as for conventional minimal coupling , we verify that @xmath434 which implies ( [ eq:7.3 ] ) . choosing a sufficiently fine open covering , we may , by paracompactness , identify the aforesaid open sets @xmath420 with the @xmath110 , and hence assume that @xmath435 . accordingly , we let @xmath436 denote the element of @xmath437 determined by ( [ eq:7.3 ] ) . for each @xmath56 one can define a 1-form @xmath438 on @xmath110 assuming values in the sheaf @xmath439 , and ( [ eq:7.3 ] ) shows that these combine to provide a global @xmath439-valued form @xmath440 on @xmath9 . the quantisation of the gauge and particle fields described above will be investigated elsewhere . accordingly , we shall not attempt to define lagrangians in the present paper . rather , we shall merely postulate that the equations of motion derived in conventional gauge field theory are also valid in the present context . _ example 2_. _ generalised interaction of dirac monopole with charged scalar field_. consider the fundamental representation of @xmath1 , that is , @xmath441 for @xmath442 . we consider the dirac sheaf bundle @xmath302 and connection below , @xmath443 , where @xmath0 is the connection form for the conventional dirac monopole . ] @xmath444 defined over the subset @xmath445 of minkowski space , as described in 6 . let @xmath446 be a particle field associated with @xmath302 by the representation @xmath425 , as in definition [ def:7 ] . the classical equation of motion for a spin-0 charged particle interacting with an electromagnetic potential @xmath0 on minkowski space is @xmath447 where the charge factor @xmath448 has been absorbed into @xmath0 . if @xmath449 is replaced by @xmath450 , then , by virtue of relation ( [ eq:7.3 ] ) , both sides of the equation @xmath451 are well - defined elements of @xmath452 . remark : a solution of ( [ eq:7.7 ] ) might be physically described as a ` wave function ' determined only up to a locally constant phase factor . current dogma holds that a globally constant phase factor is undetectable , but the detectability of a locally constant phase factor using a physical measuring apparatus also seems problematical . hence , the model described in the present example appears to be physically plausible . the use of induced bundles to describe conventional interactions between gauge fields and particle fields initially defined in terms of different principal bundles ( see bleecker 1981 ) can be extended to @xmath140-bundles in a straightforward manner . if @xmath29 and @xmath57 are , respectively , @xmath140- and @xmath453-bundles defined over the same base space @xmath9 , then , passing to a common refinement if necessary , we may assume that @xmath29 and @xmath57 are given in terms of presentations @xmath265 and @xmath266 over the same system of trivialising neighbourhoods @xmath139 with transition isomorphisms determined by @xmath236 and @xmath269 , respectively . then , the pairs @xmath454 define a @xmath455-bundle @xmath456 over @xmath9 , with a presentation @xmath457 . furthermore , if @xmath458 and @xmath459 are the local 1-forms of connections @xmath323 and @xmath336 relative to the presentations @xmath265 and @xmath266 , respectively , then the local 1-forms @xmath460 determine a connection on @xmath461 relative to the presentation @xmath462 . now , let @xmath408 and @xmath463 be linear representations on a vector space @xmath370 such that @xmath464 for all @xmath465 . then @xmath466 defines a linear representation @xmath467 . the elements of @xmath468 represent particle fields which interact with potentials defined on @xmath29 as well as those defined on @xmath57 . _ example 3_. _ generalised interaction of dirac monopole and spinor field_. let @xmath425 denote the representation of @xmath1 on @xmath469 defined by scalar multiplication @xmath470 , and let @xmath471 , where @xmath472 and @xmath473 . then @xmath425 and @xmath474 obviously commute in the foregoing sense , hence @xmath475 defines a representation of @xmath476 on @xmath469 . let @xmath302 be the dirac @xmath315-bundle over @xmath477 , with connection forms @xmath478 ( @xmath479 ) relative to the presentation @xmath355 , as in ( [ eq:6.7 ] ) , extended from @xmath283 to @xmath477 in the obvious manner , as described in [ sec:6 ] . let @xmath57 be the trivial @xmath480-bundle over @xmath477 , or essentially @xmath481 , with trivial connection forms @xmath482 corresponding to the trivial presentation @xmath483 . thus , in this case the local 1-forms @xmath484 so the action of @xmath485 upon the local section @xmath486 is simply pointwise multiplication by the imaginary number @xmath487 . the conventional dirac equation in the presence of an electromagnetic potential , transcribed in the present context , becomes @xmath488 again , by virtue of ( [ eq:7.3 ] ) , both sides of ( [ eq:7.8 ] ) represent well - defined elements of @xmath489 . as indicated by the remarks in [ sec:1 ] , the mechanism of charge quantisation _ a la _ dirac is not an exclusive feature of the classical dirac monopole alone . the essential ingredient is a _ nontrivial _ @xmath1 bundle over some spacetime .. ] therefore , consideration of various other examples could be theoretically instructive as well as suggesting possible examples of applications to physical phenomena . the foregoing ansatz for construction of the dirac sheaf bundle can likewise be applied to cases where the underlying manifold @xmath9 is not a two - sphere . recall that given a principal bundle @xmath2 with abelian structure group @xmath8 and transition functions @xmath313 for some trivialising atlas @xmath117 , the powers @xmath490 for integral @xmath6 are the transition functions of a principal @xmath8-bundle @xmath306 with respect to the same atlas . moreover , if @xmath491 are the local 1-forms of a connection @xmath0 on @xmath2 , then one can easily check that @xmath492 provide the local 1-forms of a connection @xmath493 on @xmath306 . furthermore , if @xmath494 , then @xmath495 for the curvature ( field strength ) of @xmath493 . this means that if @xmath496 is a characteristic class of @xmath2 , then the corresponding class of @xmath306 is just @xmath497 . in particular , this applies to the chen classes . now , in accordance with the discussion in [ sec:6 ] , we can proceed similarly in the context of @xmath315-bundles , replacing the integer @xmath6 by the arbitrary real number @xmath294 , and deduce that if @xmath498 is the @xmath499th chen class of the circle bundle @xmath2 , then @xmath500 is the corresponding chen class of the @xmath315-bundle @xmath316 . to illustrate , consider the canonical connection on a circle bundle over the complex projective plane , which defines a gravitational and electromagnetic instanton . as local minima of the riemannian hilbert - maxwell action , such instantons provide significant contributions to the partition function in the joint path integral quantisation of the gravitational and electromagnetic fields , thus playing a role analogous to that of the usual instantons in the pure yang - mills theory ( cf . gibbons & hawking 1979 ; eguchi & hanson 1979 ) . _ example 4_. _ gravitational and electromagnetic instanton_. let @xmath501 and consider the canonical @xmath1 bundle over @xmath9 . the complex projective plane @xmath502 with coordinates @xmath503 satisfying @xmath504 is the quotient space of the five - sphere @xmath505 by the circle action @xmath506 . we regard @xmath505 as a subspace of @xmath507 : @xmath508 is defined by @xmath509 . parameterising @xmath505 by @xmath510 where @xmath511 $ ] , @xmath512 , formula ( 8.2 ) becomes @xmath513 as trivialising neighbourhoods , let @xmath514 . writing @xmath515 for the inhomogeneous coordinates on @xmath516 , a section @xmath517 over @xmath516 is given by @xmath518 similarly , we have @xmath519 for sections over @xmath520 and @xmath521 , respectively . by virtue of the relation @xmath522 the transition function of the @xmath1 bundle on , say , @xmath523 is given by @xmath524 . similarly , we have @xmath525 and @xmath526 . thus the generic transition function is simply @xmath527 . the canonical connection on the bundle is given by the hermitian inner product @xmath528 . in terms of the local coordinates on @xmath516 , given by @xmath529 and @xmath530 the local connection form is @xmath531 $ ] . a similar calculation shows that the local connection form on @xmath532 reads @xmath533 $ ] and @xmath534 $ ] . using the relation @xmath535 , we see at once that the local connection forms are related by the gauge transformation @xmath536 , where @xmath537 . similarly we have @xmath538 with @xmath539 and @xmath540 with @xmath541 . calculating the field strength @xmath542 on @xmath516 , say , we obtain @xmath543 . \label{eq:8.9}\end{aligned}\ ] ] this expression for @xmath81 in ( [ eq:8.9 ] ) agrees with the one obtained by trautman ( 1977 ) ; an alternative expression is given by gibbons & pope ( 1978 ) using different local coordinates . the field @xmath81 is self - dual with vanishing energy - momentum tensor , and thus , along with the fubini - study metric on @xmath544 , solves the einstein - maxwell equation with cosmological constant . at this point , one can merely speculate upon the possible incorporation of the corresponding @xmath315-bundles @xmath316 into the above - mentioned path integral formalism . _ example 5_. the field @xmath81 defined on @xmath545 induces solutions to the maxwell equation on analytic submanifolds of @xmath545 . trautman ( 1977 ) considered an example given by the veronese embedding ( cf . brody & hughston 2001 ) of @xmath546 in @xmath545 . for @xmath547 this is the embedding @xmath548 , which defines a conic @xmath549 in @xmath544 . a short calculation shows that , in terms of the spherical polar coordinates @xmath285 of @xmath550 , the local connection forms of the bundle on the two hemispheres @xmath532 are given by @xmath551 . comparison with ( [ eq:6.7 ] ) for @xmath552 shows that @xmath553 . thus , the electromagnetic field induced on @xmath283 corresponds to a magnetic pole of unit strength , which might appropriately be called the _ trautman monopole_. another elementary example is the solution to the maxwell equation arising from the segr embedding of @xmath554 in @xmath555 ; this defines a quadric @xmath556 in @xmath555 . in terms of the spherical polar coordinates @xmath557 of @xmath556 one finds at once that the local connection forms on @xmath556 are given by @xmath558 $ ] in the respective trivialising neighbourhoods @xmath532 . thus , we obtain a pair of disjoint dirac monopoles . the potential physical significance of the extensions to nonintegral charges @xmath294 would , of course , be similar to that of the simple dirac sheaf bundles described in [ sec:5 ] . remark : a nontrivial example of a new solution to the maxwell equation on a torus @xmath559 can be constructed by pulling the field @xmath81 back to @xmath559 via the elliptic curve @xmath560 in @xmath544 . here , @xmath561 is the complex coordinate of the torus and @xmath562 denotes the weierstra @xmath563-function . remark : since the equivalence classes of principal @xmath1-bundles over @xmath9 are indexed by @xmath564 ( see lawson & michelsohn 1990 , appendix a ) , any realisation of the usual hypothesis of charge quantisation through a topological mechanism is excluded in cosmological models over a contractible base , such as the schwarzschild universe ( see also trautman 1979 ) . this constitutes an apparent contradiction with certain grand unified theories which require the existence of monopoles ( t hooft 1974 ) . the author is not aware of any ansatz whereby this apparent contradiction could be resolved . consider the short exact sequence of lie groups @xmath565 with @xmath127 closed and central in @xmath8 . as described in example 1 , a principal @xmath8-bundle @xmath193 canonically determines a @xmath140-bundle @xmath566 over @xmath9 . in the special case where @xmath127 is discrete , a principal @xmath567-bundle @xmath568 canonically determines a @xmath140-bundle over @xmath9 as follows . let @xmath569 denote the transition functions of @xmath570 relative to a trivialising open covering @xmath571 of @xmath9 . since @xmath127 is discrete , the above epimorphism @xmath572 is a local homeomorphism , hence , for any @xmath328 , there exists a neighbourhood @xmath164 of @xmath25 , with @xmath573 , and a smooth mapping @xmath574 such that @xmath575 , and any two such @xmath576 and @xmath577 are related by @xmath578 for some constant @xmath579 . therefore , @xmath576 uniquely defines an element @xmath580 of @xmath214 ( see ( [ eq:3.5 ] ) ) and hence a continuous section @xmath581 of the sheaf @xmath582 . for @xmath583 we have @xmath584 hence , @xmath585 , and since @xmath127 is discrete , this product must be constant in some neighbourhood @xmath586 of @xmath25 . thus , @xmath587 in @xmath588 , and we replace @xmath164 by @xmath420 . passing to a suitable refinement of @xmath117 , we may , by paracompactness , identify the @xmath420 with the intersections @xmath589 , discard the superscripts @xmath25 from the @xmath590 , and thereby obtain , as in example 1 , transition sheaf isomorphisms @xmath591 where @xmath592 , which satisfy the cocycle condition by virtue of ( [ eq:9.2 ] ) . we denote the resulting @xmath140-bundle by @xmath593 . in particular , if @xmath2 is a principal @xmath8-bundle covering @xmath570 in the sense that the transition functions @xmath313 of @xmath2 with respect to some common trivialising open covering for @xmath2 and @xmath570 satisfy the relation @xmath594 then we may select @xmath595 in the foregoing construction , and thus @xmath593 coincides with the @xmath140-bundle @xmath150 described in example 1 . let @xmath596 denote the lie algebra of @xmath567 , and let @xmath0 be a conventional connection on @xmath570 . relative to a given system of local trivialisations over the open covering @xmath597 , with transition functions @xmath598 , @xmath0 is determined by a family of @xmath599-valued 1-forms @xmath600 satisfying @xmath601 for all @xmath328 . relative to another system of local trivialisations over @xmath117 , with transition functions @xmath602 , the same connection @xmath0 is represented by the 1-forms @xmath603 for a sufficiently fine covering , we can choose sections @xmath604 and @xmath605 such that @xmath606 and , as above , @xmath607 . these choices are unique modulo constant multiplicative factors in @xmath608 and @xmath609 , respectively . furthermore , the covering map @xmath449 induces a lie algebra isomophism @xmath610 . applying @xmath611 to ( [ eq:9.5 ] ) and ( [ eq:9.6 ] ) , we obtain @xmath612 and @xmath613 let @xmath236 and @xmath614 denote the sections of @xmath615 and @xmath616 determined by @xmath313 and @xmath617 , respectively , and write @xmath618 , @xmath619 . then ( [ eq:9.7 ] ) and ( [ eq:9.8 ] ) imply the relations @xmath620 and @xmath621 in accordance with the explanatory remarks following ( [ eq:6.1 ] ) and ( [ eq:6.3 ] ) . thus , the family of 1-forms @xmath622 , corresponding to the presentation @xmath623 , where , as usual , @xmath624 , determines a connection @xmath323 on the @xmath140-bundle @xmath593 in the sense of definition [ def:6 ] . of particular physical interest is the case where @xmath625 , @xmath626 ( the superscript @xmath627 denoting the component of the identity ) and @xmath449 is the canonical homomorphism with kernel @xmath628 . given a principal @xmath567-bundle @xmath570 over @xmath9 , a well - known theorem ( lawson & michelsohn 1990 , friedrich 2000 ) states that @xmath570 is covered by some principal @xmath8-bundle @xmath2 in the above sense iff the second stiefel class @xmath629 vanishes , and then the equivalence classes ( i.e. bundle equivalence compatible with the covering map onto @xmath570 ) of such coverings @xmath2 ( known as _ spin structures _ ) are parameterised by the elements of @xmath630 . inequivalent spin structures may ( or may not ) be inequivalent as abstract @xmath8-bundles over @xmath9 . nevertheless , this can entail no inconsistency with proposition [ prop:1 ] , since , if the hypothesis of that proposition is satisfied , then , by the above - cited theorem , there exists at most one spin structure @xmath2 over @xmath570 , and hence at most one @xmath150 , which , if existent , coincides with @xmath593 . however , the _ quasispin structure _ @xmath593 exists in any case , even if @xmath150 does not . specifically , let @xmath9 be a smooth lorentzian manifold and @xmath570 the orthonormal frame bundle of @xmath9 . the levi - civita connection @xmath631 on @xmath570 then determines , as above , a connection @xmath632 on @xmath593 , and the spin representation @xmath633 , where @xmath634 denotes the spin module , gives rise to a space @xmath635 of particle fields ( see definition [ def:7 ] ) . covariant differentiation in @xmath439 is then well - defined , just as was indicated , using local coordinates , in examples 2 and 3 . relative to a given trivialisation over @xmath110 in a sufficiently fine open covering , a quasispinor field @xmath636 is represented by a smooth section @xmath637 , determined up to a constant factor @xmath638 , that is , up to a sign @xmath639 . using the local 1-form @xmath640 relative to the given trivialisation , one obtains a conventional dirac operator @xmath641 over each @xmath110 , which , applied to @xmath642 , yields a section @xmath643 , likewise defined up to a sign . let @xmath644 denote the element of @xmath645 determined by @xmath642 . since the dirac operator behaves covariantly with respect to gauge transformations @xmath646 , the relations ( [ eq:7.1 ] ) , with @xmath647 , imply corresponding relations for the sections @xmath643 , which thus combine to yield an element @xmath648 . according to a well - known theorem of geroch ( 1968 ) , the orthonormal frame bundle of a noncompact lorentzian manifold @xmath9 possesses a spin structure iff @xmath9 is parallelisable . a specific example of a noncompact lorentzian manifold not admitting a spin structure is described , e.g. , in clarke ( 1971 ) . the present formalism permits the global description of fermionic fields on such spacetimes , and the local properties of quasispinor fields and quasi - dirac operators in this context are clearly identical with those of their conventional counterparts . the foregoing considerations are likewise applicable to the construction of quasispinor fields and quasi - dirac operators on oriented riemannian manifolds , with @xmath649 , @xmath650 and @xmath628 . in the conventional theory , the index of the dirac operator on a ` spinnable ' manifold @xmath9 is expressible , by the atiyah - singer index formula , in terms of the chen classes of @xmath9 ( lawson & michelsohn 1990 ; berline _ et al_. 1992 ) . even if @xmath9 is not spinnable , this expression is still well - defined . on the other hand , since @xmath439 is not a vector space , the index of the quasi - dirac operator @xmath651 can not be defined in the conventional manner . of course , one could , by fiat , simply define the index of the quasi - dirac operator as the value of the atiyah - singer expression . however , a more interesting approach would consist in providing a general geometrical definition of the index for differential operators on @xmath439 , and proving that its value for the quasi - dirac operator is given by the atiyah - singer formula . this appears to pose a not entirely trivial problem . in the pseudoriemannian case , if @xmath9 is not space and time orientable , then one deals with pin structures ( see chamblin and gibbons 1997 for a discussion of physical applications ) . since the homomorphism @xmath652 is also a double covering , the foregoing construction is directly applicable , and one thus obtains _ quasipin structures _ , even if true pin structures fail to exist . @xmath653-structures can also be generalised in a similar vein , but since in this case the central subgroup @xmath654 is not discrete , the situation is somewhat more complicated and will be dealt with elsewhere . mathematically sophisticated expositions of gauge theory tend to define connections and curvature _ ab initio _ on the total space of a principal bundle before pulling these quantities down to the base space via local trivialisations . some readers may , perhaps , inquire why we have not adopted this more elegant approach . the reason is that our sheaf bundles are not hausdorff spaces , hence not manifolds , so the tangent spaces are undefined . consequently , we must adopt the orthodox physical practice of working in terms of local sections . in the relevant literature one occasionally observes heuristic statements to the effect that topological solitons already display certain quantal effects at the classical level . since charge renormalisation and running coupling constants can only result from field quantisation in the conventional theory , one could , in a similar vein , assert that the generalised solitons described above already display quantum field theoretic effects at the classical level . the coupling constants can walk even before they begin to run . we have constructed a family of generalised dirac monopoles @xmath302 which include the conventional @xmath304 as special cases . the @xmath304 interact with conventional wave functions @xmath655 , i.e. sections of the associated vector bundles , and likewise , mathematical consistency in the description of interactions involving the @xmath302 requires the introduction of generalised wave functions @xmath656 which are sections of certain sheaves . every conventional @xmath655 defines a @xmath656 , but there also exist @xmath656 that do not arise from any @xmath655 . for example , consider normalised wave functions defined over the circle @xmath657 . in the conventional model , these include , e.g. , @xmath658 for integral @xmath6 , but not @xmath659 for arbitrary real @xmath294 . in our model , however , the @xmath660 ( and more generally , @xmath661 , where @xmath662 is an ordinary normalised complex - valued function on @xmath663 ) represent well - defined sections of the sheaf @xmath664 , where @xmath282 . thus , rather than demanding knowledge of a global function on @xmath9 , determined only up to a globally constant phase factor , we merely demand , for any point @xmath84 , the knowledge of a function on some neighbourhood @xmath19 of @xmath25 , and determined only up to a constant phase factor on @xmath19 . the probabilistic interpretation of such a generalised wave function remains the same as in the conventional theory . moreover , differentiation with respect to @xmath25 and multiplication by ordinary functions @xmath665 are well - defined , so we can apply hamiltonian operators such as @xmath666 . magnetic monopoles , of either integral charge or otherwise , have not yet been encountered in experimental studies ( see , e.g. , milton 2006 for the current status of the experimental limits ) . the discovery of either nonintegral magnetic monopoles or nonintegral electric charges would tend to indicate that the above - described model is not merely more general mathematically but also more realistic physically than the conventional principal bundle model . such a discovery might therefore cast doubt upon the long - standing assumption that the venues for physical phenomena are necessarily manifolds , and suggest that sheaves , which provide a more local description , constitute the appropriate arenas for physical models . the sheaf - theoretic dequantisation of solitons and instantons for nonabelian gauge groups constitutes a challenging topic for research , and might have implications for grand unified theories . the above treatment of the dirac sheaf bundle can be adapted to cases where the dirac monopole is embedded within a nonabelian principal bundle , such as the t hooft monopole . the validity of the model constructed above would have certain obvious physical consequences . firstly , the detection of a magnetic monopole would not necessarily imply the quantisation of electromagnetic charges . conversely , an observed violation of charge quantisation would not necessarily imply the nonexistence of magnetic monopoles . a definite conclusion concerning the actual physical quantisation of electric charges remains elusive . for example , explosions of electron bubbles in liquid helium have been observed in the laboratory , suggesting that fractionally charged particles can possibly exist in isolation ( konstantinov & maris 2003 ) , although the interpretation of such experiments is controversial ( jackiw _ et al_. 2001 ; bender _ et al_. 2005 ) . other physical observations apparently consistent with nonintegral charges include the fractional quantum hall effect ( laughlin 1983 ) and geometric phase measurements in anisotropic spin systems ( bruno 2004 ) . the possible variability of the fine structure constant over the cosmological timescale ( bekenstein 1982 ) and the postulated fractional charges of the quarks in the deconfinement phase could also be relevant to this issue . a sheaf - theoretic formulation of the underlying quantum theory along the foregoing lines might conceivably clarify some of these diverse phenomena . the author hereby expresses his gratitude to c. m. bender , e. j. brody , g. w. gibbons , t. w. b. kibble , b. muratori , and r. p. w. thomas for stimulating and informative discussions in connection with the foregoing material .

the stability and structure of nuclei at the upper end of the nuclear chart is a hot topic in contemporary nuclear physics . it is a challenging task to answer which heavy nuclides may exist and what properties they may have . therefore strong efforts have been made in the experimental developments as well as in the theoretical description of super - heavy elements ( she s ) . in the last decades a huge progress in the synthesis of new elements has been achieved in world leading laboratories like the gsi , darmstadt @xcite , jinr , dubna @xcite , and riken , tokyo @xcite . the first she s with @xmath2 and @xmath3 were produced in cold fusion reactions . in these experiments , involving neutron rich projectiles and spherical targets ( @xmath4pb or @xmath5bi ) , weakly excited compound nuclei were produced which cooled down by the emission of only one or two neutrons . further experimental progress in the synthesis of the heaviest elements was achieved by hot fusion reactions in which targets of deformed actinide were bombarded with the doubly - magic nucleus @xmath6ca . the compound nucleus created in this way was more excited and three or more neutrons were emitted . these reactions succeeded in the synthesis of new elements up to @xmath7 and @xmath8 @xcite . the first observation of the element 117 was possible lately @xcite using a radioactive @xmath9bk target . new possibilities for the synthesis of new isotopes will be opened in heavy ion collisions with radioactive ion beams @xcite . nowadays , many other laboratories are involved in the exploration of she s . thus the experimental groups from berkeley @xcite , ganil @xcite , livermore @xcite , jyvskyl @xcite , and oak ridge @xcite are working in this direction . they would bring in the nearest future further information on the stability and properties of the she s and an independent verification of the existing data . in a parallel way to the experimental efforts , the properties of the she s have been also investigated in various nuclear models . a proper description of trans - fermium nuclei is a great challenge for any theoretical model . usually , the parameters of the theories on atomic nuclei are adjusted to the stable isotopes and then extrapolated to the region of heavier systems . therefore many tries in different theoretical approaches are performed to foresee the stability and the structure of the heaviest nuclei . a detailed review of the theoretical analysis of she s can be found in ref . the first theoretical investigations on the stability of heavy nuclei were made in the 60 s . it was noticed that shell effects could stabilize nuclei heavier than those known at that time @xcite . calculations made in the macroscopic - microscopic model with strutinsky shell correction predicted the island of stability " . large values of shell energies were obtained at @xmath10 for prolate deformed nuclei and @xmath11 for spherical ones @xcite . in the last decades many calculations have been made providing more and more precise predictions . the fission barriers and the ground state properties were calculated using macroscopic - microscopic methods with large range of deformation parameters and nuclear shapes , including reflection and axial symmetry breaking @xcite . self - consistent methods also provided many results on fission barriers and half - lives . important results have been obtained in the relativistic mean - field ( rmf ) @xcite , the hartree - fock ( hf ) approach with skyrme forces @xcite and the hartree fock - bogoliubov ( hfb ) theory with gogny forces @xcite . the first calculations of fission barriers were performed in the axial and reflection symmetric regime but later all relevant deformations were considered in the minimization of the energy . it is well known that the liquid drop model does not predict a fission barrier in the heaviest nuclei and that the stability of the trans - fermium nuclei is achieved by the shell effects . the self - consistent quantum mechanical methods ( rmf , hf and hfb ) as microscopic theories are the perfect tools for the analysis of she s . moreover , in the self - consistent calculations all possible shapes of a nucleus are considered in the minimization process . in contrast , the most of the macroscopic - microscopic models are restricted to some pre - defined classes of deformations and only optimal shapes " @xcite allow to obtain any configuration of a nucleus . therefore they are very suitable to describe large deformations of nuclei around the scission point . a degree of freedom which plays an important role along the fission path are the pairing correlations @xcite . since in the hfb theory the particle - hole and the particle - particle matrix elements are treated on the same footing , the proper consideration of pairing along the whole fission path is guaranteed . this method has been successfully applied in many aspects of low energy nuclear physics , in particular in the description of fission barriers of heavy nuclei @xcite . another theoretical quest was to discover a semi - empirical formula describing @xmath1 emission half - lives @xcite . these investigations are very important as @xmath1 radioactivity is the dominant decay channel in many she s . the purpose of this article is to perform a systematic study of she s with respect to their stability and ground state properties in the framework of the hfb theory with the density dependent finite range gogny force and the d1s parametrization . in our analysis we include the region of the well known fermium ( fm , @xmath12 ) and nobelium ( no , @xmath13 ) elements to compare our predictions with the available experimental data . we show results for the heavier even - even nuclei rutherfordium ( rf , @xmath14 ) , seaborgium ( sg , @xmath15 ) , hassium ( hs , @xmath16 ) , darmstadtium ( ds , @xmath17 ) and copernicium ( cn , @xmath18 , which was named two years ago @xcite ) . heavier elements with @xmath19 , without given name so far , are also considered . we limit our study to @xmath20 isotopes . these nuclei are nowadays in the main stream of interest of experiments with she s . a large amount of information on nuclear structure and stability can be obtained from properties of the ground states of the she s . consequently , we start our investigation with the description of the ground state characteristics . deformations , pairing energies and two - nucleon separation energies are analyzed and collated with the single - particle energy level scheme . the ground state energies can be used to evaluate the @xmath21 values which are necessary to calculate the probability of @xmath1 emission - one of the dominant decay modes in she . a competing process to @xmath1 decay is the spontaneous fission . to analyze this mode we determine the fission barriers for all mentioned she s as a function of the quadrupole moment @xmath22 . the calculations were performed in an axial basis , although we are aware of the non - axial effects on the height of the barrier and we discuss them in a few selected cases . the impact of the octupole deformation on the potential energy along the fission path is crucial in the determination of fission barriers of she s . this is taken into account by allowing non - reflexion symmetrical shapes . the calculations were performed in a large deformed harmonic oscillator basis paying special attention to the proper optimization of the oscillator lengths and to the convergence of the calculations with the size of the basis . next , using the wkb approximation , we calculate the fission half - lives . finally , the comparison of the half - lives for @xmath1 decay and spontaneous fission allows to predict the stability of the heaviest nuclei . this article is organized as follows . in sec . ii we briefly describe the constrained hartree - fock - bogoliubov calculations . the description of the ground - state properties of she s is shown in sec . iii , fission barriers in sec . half - lives of she s are discussed in sec . v. finally , sec . vi contains a summary and some concluding remarks . in our research we will apply the self consistent hartree - fock - bogoliubov theory with the finite range density dependent gogny force . in the numerical applications we use the d1s @xcite parameterization of the gogny interaction . the d1s parameters were adjusted @xcite to give a better surface energy term ( crucial for a proper description of the fission phenomenon ) . the choice of the gogny force with the d1s parameterization is based on the fact that whenever this interaction has been used to describe low energy nuclear structure phenomena an , at least , reasonable agreement with experiment has been obtained . this degree of agreement has been obtained both for calculations at the mean field level and beyond @xcite . in the microscopic hfb calculations we have used the computer code presented in ref . @xcite , see also @xcite where special attention was paid to an accurate computation of the matrix elements of the gogny interaction for very large basis like the one used in this paper . the self - consistent equations have been solved by expanding the quasiparticle creation and annihilation operators on finite bases of axially symmetric deformed harmonic oscillator ( ho ) eigenfunctions . the size of the bases used depends upon two parameters , @xmath23 and @xmath24 , which are related to the allowed range of the ho quantum numbers through the relation @xmath25 along the perpendicular direction we take @xmath23 shells , ( i.e. @xmath26 shells depending on the value of @xmath27 . in the present study we have used @xmath28 , a value which is suited for the elongated shapes along the @xmath29 direction typical of the fission process , and @xmath30 . other parameters characterizing the ho bases are the oscillator lengths @xmath31 and @xmath32 . these two quantities have been determined , for each calculated wave function , as to minimize the hfb energy . in order to study the different paths to fission , in our calculations we have used as constraints the axial quadrupole ( @xmath22 ) and octupole ( @xmath33 ) moments , with @xmath34 . higher multipolarities are adjusted in the self - consistent process to minimize the energy . to study the impact of triaxiality effects we have also carried out calculations for a few nuclei where the axial symmetry requirement was released but the left - right symmetry was imposed . in the calculations the coulomb exchange energy has been treated in the slater approximation @xcite . additionally , the coulomb and the spin - orbit contributions to the pairing field have been neglected . finally , the two body kinetic energy correction ( 2b - kec ) is not included in the variation process because , for heavy nuclei , it remains almost constant for most of the physical configurations . as this term was included in the fitting of the force , we have to include its contribution at the end of the calculation in order to obtain reasonable binding energies . see also @xcite for a detailed discussion of the relevance of neglected or approximated terms in these calculations . we have also subtracted from the hfb energy the rotational energy corrections ( rec ) stemming from the restoration of the rotational symmetry . this correction has a considerable influence on the energy landscape ( and therefore on the height of the fission barriers ) as it is somehow proportional to the degree of symmetry breaking and therefore proportional to the quadrupole moment . a full calculation of the rec would imply an angular momentum projection @xcite which is only feasible for light nuclei . in order to estimate the rec we have followed the usual recipe @xcite of subtracting to the hfb energy the quantity @xmath35 , where @xmath36 is the fluctuation associated with the angular momentum operators in the hfb wave function and @xmath37 is the yoccoz moment of inertia @xcite . this moment of inertia has been computed using the cranking " approximation in which the full linear response matrix appearing in its expression is replaced by the zero order approximation . the effect of the cranking approximation " in the yoccoz moment of inertia was analyzed with the gogny interaction for heavy nuclei in @xcite by comparing it with the one extracted from an angular momentum projected calculation ( see also @xcite for a comparison in light nuclei ) . the conclusion is that the exact rec is a factor 0.7 smaller than the one computed with the cranking " approximation to the yoccoz moment of inertia for strongly deformed configurations ( a similar behavior has been observed for the thouless - valatin moment of inertia in @xcite ) . we have taken this phenomenological factor into account in our calculation of the rec . the evaluation of the spontaneous fission half - life is carried out in the standard wkb framework where @xmath38 is given ( in seconds ) by @xcite @xmath39 in this expression @xmath40 is the action along the @xmath22 constrained path which is given by @xmath41 @xmath42 corresponds to the @xmath22 value of the ground state and @xmath43 the @xmath22 value where the potential energy equals the one of the ground state . for the collective quadrupole inertia @xmath44 we have used the adiabatic time dependent hfb ( atdhfb ) expression computed again in the cranking " approximation and given by @xcite @xmath45 with @xmath46 here @xmath47 is the two - particle - zero - hole component of the quadrupole operator @xmath48 in the quasiparticle representation @xcite and @xmath49 are the quasiparticle energies obtained in the solution of the hfb equation . in the expression for the action the collective potential @xmath50 is given by the hfb energy ( with the 2b - kec and rec corrections ) minus the zero point energy ( zpe ) correction @xmath51 associated with the quadrupole motion . this zpe correction is given by @xmath52 where @xmath53 finally , in the expression for the action an additional parameter @xmath54 is introduced . this parameter can be taken as the hfb energy of the ( metastable ) ground state . however , it is argued that in a quantal treatment of the problem the ground state energy is given by the hfb energy plus the zero point energy associated to the collective motion . to account for this fact , the usual recipe @xcite is to add an estimation of the zero point energy to the hfb energy in order to obtain @xmath54 . in our calculations we have taken a zero point energy of 0.5 mev for all the isotopes considered . in some she s around @xmath8 a weakly oblate deformed ground state can be found . the fission path in these nuclei obviously does not go along axial symmetric shapes through the spherical configuration . such a nucleus rather takes triaxial shapes to reach the prolate saddle point with the similar absolute value of the quadrupole moment . as the energies on the triaxial part of the fission barrier are very small in comparison with the saddle point @xcite they will contribute insignificantly to the action integral . therefore we will neglect them in the calculation of half - lives . to calculate the @xmath1-decay half - lives we use the phenomenological formula of viola and seaborg @xcite @xmath55 = ( az+b)(q_{\alpha})^{-1/2 } + ( cz + d ) -7.5\ ] ] with @xmath56 the atomic number of the parent nucleus . the @xmath57-factor of the decay , @xmath58 , is obtained from the calculated ground state binding energies with the experimental value @xmath59 mev @xcite @xmath60 the constants in eq . ( [ eq : ta ] ) are : @xmath61 and @xmath62 ( taken from @xcite ) . in fig . [ fig : all_nuclide ] we present an overview of all the nuclei covered in the present work . the chosen region ranges from the heaviest trans - actinides , well known from numerous experiments , up to beyond the neutron magic shell number @xmath63 predicted by many theoretical models . the upper limit is provided by the vanishing of the two - proton separation energy . the latter approximates the proton drip line which is determined by the one - proton separation energy with some correction due to the influence of the centrifugal barrier @xcite . the region of neutron rich isotopes beyond the @xmath64-stability line ( indicated in fig . [ fig : all_nuclide ] by black squares ) has been omitted as it is out of reach for the current experimental methods and can not be produced in heavy ion collisions . as it can be seen in fig . [ fig : all_nuclide ] the experimentally known she s are located in the center of the investigated region . since we consider the ground state properties as specially relevant for the understanding of the underlying physics we present these properties in the first subsection while the following subsection is entirely devoted to the study of the fission barriers . the theoretical approach discussed in the previous section has been applied to perform a systematic study of the properties of 160 super - heavy nuclei in the region @xmath65 , @xmath66 . earlier calculations in this region have been performed in the microscopic - macroscopic approach @xcite , in the hf plus bcs with the skyrme force in the p - h channel and a monopole pairing force in the p - p channel @xcite and in the full hfb approach by @xcite with the gogny force . the calculations by berger et al . are rather similar to ours in the basic aspects . however , in our work the configuration space is larger , more appropriate for fission , and we allow more general hfb wave functions ( with more simultaneously broken symmetries ) . furthermore our study is rather detailed and systematic . to gain insight into the relevant physics of the she s we display in fig . [ fig : spe ] the single particles energies ( s.p.e . ) for the nucleus @xmath67hs in reflection symmetric configurations as a function of the quadrupole deformation @xmath68 and the quadrupole moment @xmath22 . we have chosen this nucleus as a representative of the whole region for the following reasons . since it is in the center of the area of the calculated nuclei its s.p.e . spectrum is characteristic for all she s , furthermore its proton and neutron pairing energies are small and the extraction of the s.p.e . from the hfb calculations is more reliable . the s.p.e . are obtained as usual : after solving the hfb equations the one - body hf hamiltonian is diagonalized in that basis . for the neutron system we find large spherical gaps at neutron numbers 164 , 184 and 228 ; prolate gaps at 162 and oblate gaps at 172 and 178 . in nuclei with neutron number close to 160 the spherical minimum at 164 is overwhelmed by the prolate one at 162 . for protons there are several gaps . the more relevant are the spherical ones for @xmath56 values 92 , 114 and 126 , the spherical - oblate one at 120 and the prolate ones at 104 and 108 . as we will see below many nuclear properties can be understood just by looking at these s.p.e.s . let s now discuss some ground state properties starting with the deformation parameters @xmath69 . since most of the studies on these parameters have been done in the macroscopic - microscopic models we will use a definition of the @xmath69 s in the spirit of these models . the deformation parameters @xmath69 entering in @xmath70 $ ] are determined in such a way that the multipole moments , ( @xmath71 ) , calculated with @xmath72 and the hfb self consistent ones do coincide . the deformation parameters @xmath68 of the calculated she s are given in the third column of table [ tab1 ] and they are visualized in fig . [ fig : beta]a . the quadrupole deformation parameter @xmath68 barely depends on the proton number z and decreases slowly with increasing neutron number . for all z values the nuclei with @xmath73 have a prolate deformation with @xmath74 . in the neutron number region @xmath75 , the potential energy surfaces ( pes ) @xcite of these nuclei present coexisting prolate and oblate minima . the quoted values for those nuclei are the ones corresponding to the deeper minimum . the prolate minimum is deeper in lighter nuclei with @xmath76 and @xmath77 , while the oblate one is deeper in the heavier ones . with increasing neutron number the depth of the prolate well becomes smaller compared to the oblate one with the oblate minimum becoming the lowest one around n=174 ( see also fig . [ fig : fb106 - 110 ] ) . at the same time the absolute value of the quadrupole deformation parameter of the ground state decreases from @xmath78 for @xmath79 to @xmath80 for @xmath81 . the barrier between the two wells diminishes and finally disappears at @xmath63 , where the nuclei become spherical . we also observe in this region a weak z dependency . for isotopes with @xmath82 small prolate deformation can be observed in the ground state . as compared with other calculations @xcite of she s our @xmath68 are somewhat larger but this may be connected to the slightly different methods of computation of the deformation parameters @xmath69 . in the presence of coexisting minima , one must be aware that when the restriction to axially symmetric shapes is released one of the two minima may not be a true minimum , but rather a saddle point @xcite . a clear example is the case of the heavier ds isotopes . in fig . [ fig : fb106 - 110 ] one observes two minima ( prolate and oblate ) around @xmath83 b , but looking at triaxial calculations one finds that the prolate minimum is a saddle point . the gross behavior of the deformation parameter can be qualitatively understood by looking at fig . [ fig : spe ] . here we observe that shell effects favor strong prolate deformations for the nuclei with @xmath84 and @xmath85 , i.e. , deformation driving ( down - sloping ) levels are being populated and spherical - driving ( up - sloping ) depopulated . in the region @xmath86 and @xmath87 neutron shell effects favor smaller prolate deformations while in the proton system small prolate as well as spherical shapes are favored . around @xmath79 the oblate shapes are favored while larger n values prefer smaller oblaticity . this effect is reinforced by the soft oblate proton gap at @xmath88 . the spherical shape is the privileged one by neutron shell structure at @xmath63 . this coincides with the subsequent proton spherical gaps at @xmath89 , @xmath88 , and @xmath90 which cover most of the @xmath63 isotones considered here . in columns 4 to 7 of table [ tab1 ] as well as in fig . [ fig : beta]b , c , d we present the deformation parameters of higher multipolarities . most of the nuclei are reflection symmetric in the ground states and only for nuclei with @xmath82 we obtain octupole deformed ground states independent of the proton number . mller et al . @xcite also found non - zero @xmath91 deformations for the she s in this region . the odd multipolarities @xmath92 and @xmath93 , not shown here , are different from zero only for those nuclei whose ground state is octupole deformed , however the numerical values are very small ( e.g. @xmath94 values are less than 0.009 ) . concerning the even deformations higher than two , see fig . [ fig : beta]c , d , they are different from zero only for nuclei with @xmath95 , i.e. , for all nuclei with the exception of the few isotopes around @xmath96 . as we can observe there is a smaller dependence with @xmath56 than with @xmath97 . nuclei with @xmath98 have small negative hexadecapole deformations . for increasing @xmath97 , @xmath99 increases up to @xmath100 where it reaches the largest absolute value . large negative hexadecapole moment together with large positive quadrupole moment produces barrel - like shapes in the ground state of nuclei in this region . from here on @xmath99 decreases rather smoothly up to @xmath8 where it sharply decreases to very small deformations and zero values at the largest neutron number analyzed . this sharp decrease is associated to the prolate oblate shape transition that takes place at this neutron numbers . for @xmath97-values @xmath101 , i.e. , the prolate part , there is a clear tendency with @xmath56 for a given @xmath97 : with increasing proton number @xmath99 gets larger . for the oblate - spherical part , there is almost no dependence with @xmath56 . the behavior of the deformation parameter @xmath102 is quite different from @xmath103 . for @xmath104 we obtain negative values decreasing in absolute value as @xmath97 increases . around @xmath105 , @xmath102 becomes almost zero and for @xmath106 we obtain increasing values of @xmath102 as @xmath97 increases . for @xmath107 we obtain rather small @xmath102 deformations . our results for @xmath103 and @xmath102 are rather similar to those of mller et al @xcite and sobiczewski et al @xcite . contrary to the @xmath103 and @xmath102 parameters , @xmath108 is positive in the region @xmath109 . the nuclei with @xmath110 and 166 have either zero or very small @xmath108 values . the nuclei with @xmath111 have mostly negative @xmath108 values . with the exception of the @xmath112 isotopes that have positive deformations , all the nuclei with @xmath113 have @xmath108 deformations almost zero . we have to notice that the maximal absolute values of the deformation parameters decrease with increasing multipolarity indicating a decreasing relevance . this means that higher multipolarities can be omitted in the minimization of the energy of the ground state in non - self - consistent approximations . ll ddddd ddddd dd + + & & & & & & & & & & & & & + + + + 100 & 150 & 0.329 & 0.000 & 0.003 & -0.028 & 0.005 & -4.28 & -7.31 & 7.35 & 13.20 & 8.86 & 4.04 & 13.17 + 100 & 152 & 0.331 & 0.000 & -0.009 & -0.029 & 0.007 & -3.85 & -7.69 & 7.12 & 12.32 & 9.93 & 4.97 & 14.32 + 100 & 154 & 0.332 & 0.000 & -0.021 & -0.028 & 0.010 & -5.87 & -8.29 & 7.02 & 11.35 & 10.87 & 5.39 & 12.82 + 100 & 156 & 0.329 & 0.000 & -0.032 & -0.024 & 0.011 & -6.88 & -9.26 & 6.66 & 10.77 & 11.74 & 6.98 & 7.07 + 100 & 158 & 0.321 & 0.000 & -0.044 & -0.018 & 0.011 & -7.01 & -10.66 & 6.21 & 10.35 & 12.57 & 9.15 & -0.06 + 100 & 160 & 0.310 & 0.000 & -0.056 & -0.011 & 0.009 & -5.80 & -12.25 & 5.72 & 10.01 & 13.37 & 11.81 & 0.37 + 100 & 162 & 0.300 & 0.000 & -0.069 & -0.004 & 0.008 & -0.70 & -13.67 & 5.43 & 9.50 & 14.16 & 13.55 & 0.16 + 100 & 164 & 0.285 & 0.000 & -0.071 & 0.003 & 0.005 & -6.73 & -14.36 & 5.73 & 8.41 & 14.72 & 11.75 & -3.02 + 100 & 166 & 0.267 & 0.000 & -0.074 & 0.009 & 0.001 & -9.02 & -15.28 & 5.56 & 8.02 & 15.25 & 12.75 & -6.38 + + 102 & 150 & 0.331 & 0.000 & -0.006 & -0.027 & 0.005 & -4.47 & -6.59 & 8.12 & 14.29 & 6.98 & 1.97 & 7.63 + 102 & 152 & 0.334 & 0.000 & -0.017 & -0.028 & 0.007 & -3.43 & -6.51 & 7.93 & 13.39 & 8.05 & 2.64 & 10.02 + 102 & 154 & 0.335 & 0.000 & -0.029 & -0.028 & 0.010 & -5.55 & -6.77 & 7.91 & 12.34 & 9.04 & 2.71 & 8.23 + 102 & 156 & 0.332 & 0.000 & -0.040 & -0.024 & 0.012 & -6.62 & -7.45 & 7.57 & 11.69 & 9.96 & 3.99 & 0.76 + 102 & 158 & 0.325 & 0.000 & -0.050 & -0.018 & 0.011 & -6.88 & -8.61 & 7.09 & 11.25 & 10.86 & 5.94 & 0.57 + 102 & 160 & 0.315 & 0.000 & -0.060 & -0.011 & 0.009 & -5.60 & -9.93 & 6.55 & 10.89 & 11.74 & 8.38 & 1.65 + 102 & 162 & 0.305 & 0.000 & -0.073 & -0.004 & 0.009 & -0.33 & -11.43 & 6.19 & 10.37 & 12.61 & 10.19 & 2.23 + 102 & 164 & 0.293 & 0.000 & -0.075 & 0.003 & 0.005 & -6.53 & -12.43 & 6.72 & 8.97 & 13.17 & 7.58 & -1.43 + 102 & 166 & 0.276 & 0.000 & -0.079 & 0.010 & 0.001 & -8.91 & -13.85 & 6.54 & 8.59 & 13.74 & 8.43 & -4.70 + + 104 & 150 & 0.331 & 0.000 & -0.016 & -0.025 & 0.005 & -4.85 & -4.76 & 8.91 & 15.35 & 5.10 & 0.11 & -0.41 + 104 & 152 & 0.333 & 0.000 & -0.026 & -0.026 & 0.007 & -3.35 & -4.37 & 8.78 & 14.42 & 6.13 & 0.52 & 1.85 + 104 & 154 & 0.334 & 0.000 & -0.037 & -0.025 & 0.010 & -5.63 & -4.53 & 8.85 & 13.32 & 7.11 & 0.30 & 1.77 + 104 & 156 & 0.332 & 0.000 & -0.047 & -0.021 & 0.011 & -6.78 & -5.03 & 8.52 & 12.67 & 8.09 & 1.36 & 1.90 + 104 & 158 & 0.325 & 0.000 & -0.057 & -0.016 & 0.010 & -6.92 & -5.99 & 7.98 & 12.23 & 9.07 & 3.24 & 2.03 + 104 & 160 & 0.317 & 0.000 & -0.067 & -0.010 & 0.009 & -5.39 & -7.17 & 7.40 & 11.83 & 10.01 & 5.47 & 3.86 + 104 & 162 & 0.309 & 0.000 & -0.078 & -0.004 & 0.009 & -0.06 & -8.76 & 6.99 & 11.30 & 10.94 & 7.22 & 5.36 + 104 & 164 & 0.299 & 0.000 & -0.081 & 0.004 & 0.005 & -6.28 & -9.99 & 7.75 & 9.61 & 11.58 & 4.09 & 1.11 + 104 & 166 & 0.285 & 0.000 & -0.085 & 0.011 & 0.001 & -8.87 & -11.54 & 7.54 & 9.18 & 12.17 & 4.91 & -2.64 + + 106 & 150 & 0.329 & 0.000 & -0.024 & -0.021 & 0.004 & -5.50 & -6.23 & 10.04 & 16.27 & 2.91 & -2.43 & -3.01 + 106 & 152 & 0.331 & 0.000 & -0.035 & -0.022 & 0.006 & -3.81 & -5.49 & 9.98 & 15.41 & 3.90 & -2.28 & -0.11 + 106 & 154 & 0.331 & 0.000 & -0.046 & -0.020 & 0.009 & -6.10 & -5.02 & 10.03 & 14.37 & 4.95 & -2.41 & 0.70 + 106 & 156 & 0.328 & 0.000 & -0.057 & -0.017 & 0.010 & -7.21 & -5.03 & 9.61 & 13.74 & 6.02 & -1.26 & 1.20 + 106 & 158 & 0.324 & 0.000 & -0.066 & -0.012 & 0.010 & -7.08 & -5.22 & 8.97 & 13.31 & 7.10 & 0.63 & 2.68 + 106 & 160 & 0.317 & 0.000 & -0.075 & -0.007 & 0.009 & -5.25 & -5.66 & 8.31 & 12.89 & 8.16 & 2.82 & 5.75 + 106 & 162 & 0.310 & 0.000 & -0.084 & -0.002 & 0.009 & -0.01 & -6.39 & 7.81 & 12.33 & 9.19 & 4.65 & 9.30 + 106 & 164 & 0.301 & 0.000 & -0.088 & 0.006 & 0.004 & -5.87 & -7.37 & 8.77 & 10.34 & 9.92 & 1.27 & 3.72 + 106 & 166 & 0.288 & 0.000 & -0.092 & 0.013 & 0.001 & -8.28 & -8.84 & 8.46 & 9.92 & 10.66 & 2.30 & -0.11 + 106 & 168 & 0.272 & 0.000 & -0.095 & 0.019 & -0.002 & -8.81 & -10.72 & 8.03 & 9.61 & 11.24 & 3.82 & -4.13 + 106 & 170 & 0.246 & 0.000 & -0.092 & 0.022 & -0.004 & -8.29 & -13.34 & 7.69 & 9.37 & 11.82 & 5.12 & -5.79 + 106 & 172 & 0.196 & 0.000 & -0.071 & 0.016 & -0.002 & -6.31 & -17.22 & 7.18 & 9.30 & 12.40 & 7.23 & -5.47 + + 108 & 150 & 0.324 & 0.000 & -0.029 & -0.017 & 0.004 & -6.11 & -7.61 & 11.11 & 17.08 & 0.92 & -4.45 & -5.43 + 108 & 152 & 0.327 & 0.000 & -0.041 & -0.018 & 0.006 & -4.46 & -6.01 & 11.06 & 16.32 & 1.83 & -4.34 & -2.77 + 108 & 154 & 0.327 & 0.000 & -0.054 & -0.016 & 0.008 & -6.76 & -4.49 & 11.10 & 15.37 & 2.83 & -4.43 & -0.95 + 108 & 156 & 0.325 & 0.000 & -0.065 & -0.012 & 0.009 & -7.82 & -3.47 & 10.66 & 14.81 & 3.90 & -3.38 & 0.35 + 108 & 158 & 0.321 & 0.000 & -0.075 & -0.007 & 0.009 & -7.40 & -3.05 & 10.01 & 14.39 & 4.98 & -1.71 & 2.64 + 108 & 160 & 0.316 & 0.000 & -0.083 & -0.003 & 0.009 & -5.33 & -2.89 & 9.36 & 13.96 & 6.05 & 0.14 & 6.49 + 108 & 162 & 0.310 & 0.000 & -0.092 & 0.001 & 0.008 & -0.02 & -3.04 & 8.84 & 13.41 & 7.13 & 1.76 & 12.83 + 108 & 164 & 0.300 & 0.000 & -0.096 & 0.009 & 0.004 & -5.29 & -4.26 & 10.02 & 11.15 & 7.94 & -1.74 & 5.89 + 108 & 166 & 0.290 & 0.000 & -0.100 & 0.016 & 0.000 & -7.58 & -5.88 & 9.65 & 10.71 & 8.73 & -0.71 & 2.66 + 108 & 168 & 0.276 & 0.000 & -0.103 & 0.022 & -0.003 & -8.16 & -8.10 & 9.22 & 10.35 & 9.47 & 0.56 & -1.41 + 108 & 170 & 0.250 & 0.000 & -0.098 & 0.025 & -0.004 & -7.84 & -11.26 & 8.68 & 10.15 & 10.25 & 2.29 & -3.84 + 108 & 172 & 0.195 & 0.000 & -0.071 & 0.016 & -0.002 & -6.07 & -15.49 & 8.08 & 9.97 & 10.92 & 4.40 & -3.54 + 108 & 174 & 0.170 & 0.000 & -0.070 & 0.018 & -0.003 & -3.95 & -17.17 & 7.32 & 10.06 & 11.56 & 7.45 & -4.43 + 108 & 176 & -0.162 & 0.000 & -0.022 & 0.005 & 0.002 & -4.53 & -19.62 & 6.72 & 10.02 & 11.99 & 10.21 & 6.25 + 108 & 178 & -0.146 & 0.000 & -0.031 & 0.008 & 0.006 & -0.01 & -20.36 & 6.49 & 9.82 & 12.67 & 11.37 & 14.21 + 108 & 180 & -0.090 & 0.000 & -0.024 & -0.002 & 0.001 & -3.12 & -22.30 & 6.69 & 8.94 & 13.23 & 10.36 & 5.76 + 108 & 182 & -0.060 & 0.000 & -0.030 & -0.008 & -0.001 & 0.00 & -23.23 & 6.44 & 8.63 & 13.83 & 11.63 & 14.10 + 108 & 184 & -0.037 & 0.000 & 0.000 & 0.000 & 0.000 & 0.00 & -23.95 & 7.06 & 7.41 & 14.28 & 8.61 & 8.40 + 108 & 186 & 0.034 & -0.062 & 0.005 & 0.000 & 0.000 & -5.76 & -23.57 & 7.00 & 7.02 & 14.50 & 8.88 & 0.68 + + 110 & 154 & 0.319 & 0.000 & -0.052 & -0.012 & 0.006 & -7.43 & -8.77 & 12.31 & 16.15 & 0.62 & -6.48 & -4.27 + 110 & 156 & 0.314 & 0.000 & -0.065 & -0.007 & 0.007 & -8.68 & -7.91 & 11.97 & 15.71 & 1.52 & -5.78 & -2.38 + 110 & 158 & 0.310 & 0.000 & -0.076 & -0.002 & 0.007 & -8.12 & -7.25 & 11.41 & 15.37 & 2.50 & -4.55 & 0.14 + 110 & 160 & 0.306 & 0.000 & -0.087 & 0.002 & 0.006 & -5.66 & -6.75 & 10.79 & 15.01 & 3.55 & -3.09 & 3.94 + 110 & 162 & 0.303 & 0.000 & -0.096 & 0.006 & 0.007 & -0.03 & -6.39 & 10.32 & 14.43 & 4.57 & -1.90 & 9.02 + 110 & 164 & 0.293 & 0.000 & -0.101 & 0.014 & 0.003 & -4.93 & -6.71 & 11.41 & 12.32 & 5.74 & -4.55 & 5.87 + 110 & 166 & 0.283 & 0.000 & -0.106 & 0.021 & -0.001 & -6.63 & -7.26 & 10.77 & 11.79 & 6.82 & -3.04 & 3.63 + 110 & 168 & 0.271 & 0.000 & -0.110 & 0.027 & -0.004 & -6.90 & -8.28 & 10.14 & 11.34 & 7.81 & -1.42 & 1.94 + 110 & 170 & 0.247 & 0.000 & -0.103 & 0.027 & -0.005 & -7.25 & -10.38 & 9.63 & 10.86 & 8.52 & 0.01 & -1.98 + 110 & 172 & 0.197 & 0.000 & -0.076 & 0.017 & -0.002 & -5.86 & -13.21 & 8.99 & 10.79 & 9.34 & 1.98 & -0.76 + 110 & 174 & 0.175 & 0.000 & -0.075 & 0.021 & -0.004 & -3.28 & -14.84 & 8.24 & 10.72 & 10.00 & 4.57 & -1.87 + 110 & 176 & -0.160 & 0.000 & -0.023 & 0.004 & 0.002 & -4.45 & -18.70 & 7.72 & 10.58 & 10.56 & 6.58 & -0.48 + 110 & 178 & -0.145 & 0.000 & -0.032 & 0.008 & 0.006 & 0.00 & -19.31 & 7.24 & 10.50 & 11.24 & 8.63 & 4.61 + 110 & 180 & -0.090 & 0.000 & -0.024 & -0.002 & 0.001 & -3.08 & -21.19 & 7.53 & 9.53 & 11.83 & 7.37 & 5.67 + 110 & 182 & -0.060 & 0.000 & -0.030 & -0.008 & -0.001 & -0.01 & -22.18 & 7.24 & 9.23 & 12.43 & 8.63 & 13.90 + 110 & 184 & 0.000 & -0.002 & 0.000 & 0.000 & 0.000 & 0.00 & -23.43 & 8.03 & 7.84 & 12.86 & 5.36 & 6.25 + 110 & 186 & 0.035 & -0.035 & 0.004 & 0.000 & 0.000 & -6.59 & -23.04 & 8.29 & 7.15 & 12.99 & 4.39 & -10.25 + + 112 & 158 & 0.297 & 0.000 & -0.076 & 0.002 & 0.004 & -8.73 & -9.36 & 12.07 & 16.32 & 0.86 & -5.43 & -2.42 + 112 & 160 & 0.295 & 0.000 & -0.088 & 0.006 & 0.005 & -6.01 & -8.62 & 11.50 & 15.94 & 1.79 & -4.18 & 1.58 + 112 & 162 & 0.294 & 0.000 & -0.098 & 0.009 & 0.005 & -0.04 & -7.98 & 11.17 & 15.34 & 2.70 & -3.41 & 6.17 + 112 & 164 & 0.283 & 0.000 & -0.103 & 0.017 & 0.001 & -4.70 & -7.77 & 12.10 & 13.50 & 3.88 & -5.50 & 4.78 + 112 & 166 & 0.273 & 0.000 & -0.109 & 0.025 & -0.002 & -6.09 & -7.78 & 11.52 & 12.90 & 4.99 & -4.23 & 3.56 + 112 & 168 & 0.262 & 0.000 & -0.113 & 0.031 & -0.005 & -6.12 & -8.13 & 10.94 & 12.37 & 6.02 & -2.86 & 2.87 + 112 & 170 & 0.253 & 0.000 & -0.114 & 0.035 & -0.007 & -6.00 & -8.62 & 10.32 & 11.96 & 7.12 & -1.27 & -0.27 + 112 & 172 & 0.211 & 0.000 & -0.087 & 0.024 & -0.004 & -5.83 & -10.26 & 9.46 & 11.72 & 8.05 & 1.19 & 2.36 + 112 & 174 & 0.191 & 0.000 & -0.086 & 0.026 & -0.005 & -1.90 & -11.45 & 8.76 & 11.49 & 8.82 & 3.46 & 3.11 + 112 & 176 & -0.157 & 0.000 & -0.023 & 0.004 & 0.002 & -4.34 & -17.21 & 8.60 & 10.88 & 9.12 & 4.01 & 8.14 + 112 & 178 & -0.144 & 0.000 & -0.032 & 0.008 & 0.006 & -0.01 & -17.75 & 8.00 & 11.18 & 9.80 & 6.25 & 16.60 + 112 & 180 & -0.088 & 0.000 & -0.023 & -0.002 & 0.001 & -3.07 & -19.53 & 8.36 & 10.14 & 10.41 & 4.88 & 7.05 + 112 & 182 & -0.060 & 0.000 & -0.030 & -0.008 & -0.001 & 0.00 & -20.57 & 8.02 & 9.87 & 11.05 & 6.17 & 15.09 + 112 & 184 & 0.000 & -0.001 & 0.000 & 0.000 & 0.000 & 0.00 & -21.93 & 8.82 & 8.43 & 11.64 & 3.25 & 7.29 + 112 & 186 & 0.052 & -0.007 & 0.005 & 0.001 & 0.001 & -7.35 & -20.94 & 9.80 & 6.86 & 11.35 & 0.18 & -8.67 + 112 & 188 & 0.015 & -0.072 & 0.004 & -0.001 & 0.000 & -11.05 & -22.03 & 8.68 & 8.27 & 12.86 & 3.73 & -8.10 + 112 & 190 & 0.028 & -0.123 & 0.011 & -0.005 & 0.000 & -10.70 & -21.64 & 9.01 & 6.43 & 13.26 & 2.62 & -14.23 + + 114 & 160 & 0.284 & 0.000 & -0.089 & 0.010 & 0.003 & -6.36 & -8.58 & 12.08 & 16.84 & 0.28 & -4.90 & -0.80 + 114 & 162 & 0.285 & 0.000 & -0.100 & 0.012 & 0.004 & -0.10 & -7.72 & 11.87 & 16.15 & 1.09 & -4.44 & 3.72 + 114 & 164 & 0.272 & 0.000 & -0.105 & 0.020 & 0.000 & -4.62 & -6.79 & 12.53 & 14.68 & 2.27 & -5.85 & 3.62 + 114 & 166 & 0.262 & 0.000 & -0.110 & 0.027 & -0.003 & -5.88 & -6.35 & 12.05 & 13.98 & 3.35 & -4.84 & 3.27 + 114 & 168 & 0.251 & 0.000 & -0.113 & 0.032 & -0.005 & -6.00 & -6.29 & 11.56 & 13.39 & 4.37 & -3.74 & 3.65 + 114 & 170 & 0.231 & 0.000 & -0.106 & 0.031 & -0.006 & -6.70 & -6.67 & 10.97 & 12.96 & 5.37 & -2.33 & 1.60 + 114 & 172 & 0.210 & 0.000 & -0.094 & 0.028 & -0.005 & -5.48 & -7.17 & 10.28 & 12.65 & 6.30 & -0.53 & 5.00 + 114 & 174 & 0.190 & 0.000 & -0.091 & 0.030 & -0.006 & -1.13 & -8.09 & 9.69 & 12.31 & 7.12 & 1.17 & 6.38 + 114 & 176 & -0.153 & 0.000 & -0.022 & 0.005 & 0.002 & -4.32 & -15.07 & 9.74 & 11.44 & 7.68 & 1.02 & 10.60 + 114 & 178 & -0.141 & 0.000 & -0.031 & 0.008 & 0.006 & -0.01 & -15.58 & 8.76 & 11.86 & 8.36 & 4.18 & 18.54 + 114 & 180 & -0.085 & 0.000 & -0.022 & -0.002 & 0.000 & -3.10 & -17.31 & 9.13 & 10.81 & 9.03 & 2.92 & 9.45 + 114 & 182 & -0.059 & 0.000 & -0.029 & -0.007 & -0.001 & -0.01 & -18.41 & 8.78 & 10.49 & 9.65 & 4.11 & 17.10 + 114 & 184 & 0.000 & 0.000 & 0.000 & 0.000 & 0.000 & 0.00 & -19.84 & 9.68 & 8.97 & 10.19 & 1.20 & 8.59 + 114 & 186 & 0.052 & -0.005 & 0.004 & 0.001 & 0.001 & -7.66 & -18.85 & 10.60 & 7.51 & 10.84 & -1.39 & -6.76 + 114 & 188 & 0.032 & -0.064 & 0.006 & 0.000 & 0.000 & -10.96 & -20.07 & 8.81 & 8.65 & 11.22 & 4.00 & -7.58 + 114 & 190 & -0.019 & -0.072 & 0.003 & -0.001 & 0.000 & -16.36 & -19.89 & 9.91 & 7.17 & 11.96 & 0.52 & -12.52 + + 116 & 164 & 0.262 & 0.000 & -0.106 & 0.023 & 0.000 & -4.71 & -3.91 & 13.12 & 15.74 & 0.50 & -6.50 & 1.17 + 116 & 166 & 0.252 & 0.000 & -0.112 & 0.030 & -0.004 & -5.81 & -2.26 & 12.76 & 15.04 & 1.56 & -5.78 & 3.84 + 116 & 168 & 0.242 & 0.000 & -0.115 & 0.035 & -0.006 & -5.83 & -1.49 & 12.32 & 14.42 & 2.59 & -4.87 & 5.14 + 116 & 170 & 0.231 & 0.000 & -0.114 & 0.037 & -0.007 & -6.02 & -1.32 & 11.74 & 13.97 & 3.60 & -3.59 & 2.78 + 116 & 172 & 0.189 & 0.000 & -0.089 & 0.027 & -0.005 & -5.08 & -4.87 & 11.20 & 13.50 & 4.45 & -2.31 & 6.02 + 116 & 174 & 0.188 & 0.000 & -0.095 & 0.033 & -0.007 & -0.44 & -4.55 & 10.67 & 13.18 & 5.32 & -0.96 & 8.01 + 116 & 176 & -0.147 & 0.000 & -0.021 & 0.005 & 0.002 & -4.30 & -12.14 & 10.66 & 12.32 & 6.20 & -0.93 & 13.72 + 116 & 178 & -0.138 & 0.000 & -0.029 & 0.008 & 0.006 & -0.01 & -12.67 & 9.55 & 12.55 & 6.89 & 2.26 & 21.11 + 116 & 180 & -0.081 & 0.000 & -0.020 & -0.002 & 0.000 & -3.17 & -14.50 & 9.90 & 11.51 & 7.59 & 1.20 & 12.61 + 116 & 182 & -0.057 & 0.000 & -0.028 & -0.007 & -0.001 & -0.01 & -15.68 & 9.57 & 11.14 & 8.24 & 2.20 & 19.87 + 116 & 184 & 0.000 & 0.000 & 0.000 & 0.000 & 0.000 & 0.00 & -17.19 & 10.51 & 9.55 & 8.82 & -0.53 & 11.02 + 116 & 186 & 0.051 & -0.008 & 0.003 & 0.001 & 0.001 & -7.91 & -16.20 & 11.31 & 8.17 & 9.48 & -2.58 & -4.28 + 116 & 188 & 0.033 & -0.039 & 0.004 & 0.000 & 0.000 & -12.40 & -17.04 & 9.89 & 8.93 & 9.76 & 1.23 & -6.10 + 116 & 190 & 0.015 & -0.078 & 0.004 & -0.002 & 0.000 & -15.32 & -18.02 & 10.38 & 8.16 & 10.75 & -0.17 & -9.85 + + 118 & 170 & 0.208 & 0.000 & -0.095 & 0.029 & -0.005 & -7.20 & -2.98 & 13.37 & 15.09 & 0.96 & -6.47 & 1.39 + 118 & 172 & 0.171 & 0.000 & -0.081 & 0.024 & -0.004 & -4.96 & -3.14 & 12.57 & 14.77 & 2.23 & -4.86 & 6.27 + 118 & 174 & 0.168 & 0.000 & -0.088 & 0.030 & -0.006 & -1.39 & -2.40 & 11.92 & 14.15 & 3.20 & -3.44 & 9.44 + 118 & 176 & -0.143 & 0.000 & -0.020 & 0.006 & 0.002 & -4.32 & -8.21 & 11.33 & 13.77 & 4.65 & -2.04 & 17.87 + 118 & 178 & -0.135 & 0.000 & -0.027 & 0.009 & 0.005 & -0.01 & -8.89 & 10.42 & 13.23 & 5.33 & 0.34 & 24.12 + 118 & 180 & -0.075 & 0.000 & -0.020 & -0.002 & 0.000 & -3.30 & -11.26 & 10.70 & 12.27 & 6.09 & -0.42 & 16.37 + 118 & 182 & -0.056 & 0.000 & -0.028 & -0.007 & -0.001 & -0.01 & -12.42 & 10.40 & 11.81 & 6.76 & 0.40 & 22.89 + 118 & 184 & 0.000 & 0.000 & 0.000 & 0.000 & 0.000 & 0.00 & -14.04 & 11.37 & 10.17 & 7.38 & -2.14 & 13.92 + 118 & 186 & 0.051 & -0.008 & 0.001 & 0.001 & 0.000 & -8.12 & -13.06 & 12.15 & 8.77 & 7.98 & -3.95 & -2.03 + 118 & 188 & 0.049 & -0.054 & 0.006 & 0.001 & 0.000 & -11.50 & -13.75 & 10.73 & 9.59 & 8.64 & -0.50 & -4.10 + 118 & 190 & 0.016 & -0.054 & 0.003 & -0.001 & 0.000 & -16.98 & -14.30 & 10.80 & 8.86 & 9.34 & -0.69 & -7.35 + + 120 & 172 & -0.176 & 0.000 & -0.002 & 0.011 & 0.000 & -3.70 & -2.47 & 12.63 & 15.84 & 0.90 & -4.46 & 12.15 + 120 & 174 & -0.154 & 0.000 & -0.010 & 0.008 & 0.001 & -5.41 & -1.89 & 12.26 & 15.14 & 1.89 & -3.65 & 15.82 + 120 & 176 & -0.138 & 0.000 & -0.018 & 0.007 & 0.003 & -4.35 & -3.44 & 11.73 & 14.68 & 2.80 & -2.43 & 19.57 + 120 & 178 & -0.130 & 0.000 & -0.025 & 0.009 & 0.005 & -0.01 & -4.81 & 11.53 & 13.97 & 3.54 & -1.95 & 24.62 + 120 & 180 & -0.066 & 0.000 & -0.019 & -0.003 & 0.000 & -3.56 & -7.97 & 11.60 & 13.16 & 4.43 & -2.12 & 19.80 + 120 & 182 & -0.055 & 0.000 & -0.029 & -0.008 & -0.001 & -0.01 & -8.83 & 11.29 & 12.58 & 5.20 & -1.36 & 25.60 + 120 & 184 & 0.000 & 0.000 & 0.000 & 0.000 & 0.000 & 0.00 & -10.57 & 12.30 & 10.80 & 5.83 & -3.74 & 16.29 + 120 & 186 & 0.050 & -0.012 & 0.001 & 0.001 & 0.000 & -8.16 & -9.69 & 13.05 & 9.42 & 6.48 & -5.33 & 0.27 + 120 & 188 & 0.048 & -0.059 & 0.005 & 0.001 & 0.000 & -11.37 & -10.34 & 11.57 & 10.25 & 7.14 & -2.05 & -1.88 + 120 & 190 & 0.016 & -0.052 & 0.002 & 0.000 & 0.000 & -17.16 & -10.62 & 11.61 & 9.55 & 7.83 & -2.15 & -5.01 + + 122 & 176 & -0.109 & 0.000 & -0.020 & 0.002 & 0.002 & -5.35 & -4.68 & 13.02 & 15.62 & 0.60 & -4.75 & 17.54 + 122 & 178 & -0.109 & 0.000 & -0.024 & 0.004 & 0.003 & -1.11 & -4.56 & 12.93 & 14.77 & 1.40 & -4.56 & 19.76 + 122 & 180 & -0.062 & 0.000 & -0.023 & -0.004 & 0.000 & -3.62 & -4.98 & 12.50 & 14.40 & 2.64 & -3.64 & 21.94 + 122 & 182 & -0.055 & 0.000 & -0.031 & -0.008 & -0.001 & 0.00 & -5.24 & 12.27 & 13.39 & 3.45 & -3.13 & 28.05 + 122 & 184 & 0.000 & 0.000 & 0.000 & 0.000 & 0.000 & 0.00 & -7.19 & 13.33 & 11.52 & 4.17 & -5.38 & 18.02 + 122 & 186 & 0.049 & -0.036 & 0.000 & 0.001 & 0.000 & -7.63 & -6.64 & 13.52 & 10.61 & 5.36 & -5.76 & 3.69 + 122 & 188 & 0.031 & -0.059 & 0.002 & 0.001 & 0.000 & -11.75 & -7.17 & 12.10 & 10.84 & 5.95 & -2.74 & 1.10 + 122 & 190 & 0.015 & -0.068 & 0.002 & 0.000 & 0.000 & -16.17 & -7.55 & 12.47 & 9.88 & 6.28 & -3.58 & -3.44 + + 124 & 180 & -0.065 & 0.000 & -0.028 & -0.006 & 0.000 & -3.44 & -1.51 & 13.59 & 15.21 & 0.31 & -5.40 & 20.40 + 124 & 182 & -0.058 & 0.000 & -0.033 & -0.010 & -0.001 & 0.00 & -1.15 & 13.76 & 14.23 & 1.15 & -5.73 & 29.94 + 124 & 184 & 0.000 & -0.001 & 0.000 & 0.000 & 0.000 & 0.00 & -4.30 & 14.62 & 12.53 & 2.16 & -7.32 & 17.90 + 124 & 186 & 0.032 & -0.050 & 0.001 & 0.002 & 0.000 & -6.81 & -4.27 & 13.95 & 12.19 & 3.74 & -6.09 & 9.63 + 124 & 188 & 0.015 & -0.068 & 0.002 & 0.002 & 0.000 & -11.72 & -4.48 & 13.55 & 11.01 & 3.91 & -5.32 & 2.93 + 124 & 190 & -0.002 & -0.078 & 0.002 & 0.001 & 0.000 & -16.02 & -4.73 & 14.21 & 10.18 & 4.21 & -6.58 & -2.94 + + 126 & 186 & 0.015 & -0.073 & 0.001 & 0.003 & 0.000 & -6.38 & -1.22 & 15.04 & 13.53 & 1.07 & -7.59 & 9.82 + 126 & 188 & -0.002 & -0.086 & 0.002 & 0.004 & 0.000 & -11.40 & -1.03 & 16.13 & 11.10 & 1.16 & -9.37 & 1.03 + 126 & 190 & 0.011 & -0.121 & 0.003 & 0.004 & 0.000 & -13.12 & -0.48 & 16.05 & 11.09 & 2.07 & -9.25 & -4.06 + + ( color online ) ground state proton and neutron pairing energies for super - heavy nuclei . ] in the hfb approach @xcite , the pairing energy is given by @xmath114 with @xmath115 the pairing field and @xmath116 the pairing tensor . the neutron pairing energies of each nucleus are given in the eighth column of table [ tab1 ] and in the upper panel of fig . [ fig : pair ] . the general behavior of these results can be easily understood by looking again at fig . [ fig : spe ] . we know that small pairing energies correspond to situations of low level density of the neutron s.p.e in particular we expect zero neutron pairing energies at @xmath117 and 184 . notice that the fact that for @xmath105 the pairing energies are zero is consistent with the prolate deformation of these nuclei and for @xmath112 with the oblate one . for @xmath81 and 184 we have spherical nuclei . appropriate neutron shell gaps can be found in fig . [ fig : spe ] for the mentioned cases . for the same reason large pairing energies are associated with high level density , for example , we find large pairing energies for @xmath118 and 190 . we observe that the general pattern , qualitatively , does not depend on the proton number . the proton pairing energies are given in the ninth column of table [ tab1 ] and the lower panel of fig . [ fig : pair ] . in this case the general pattern looks more complicated . for the same proton number some nuclei are prolate , some oblate and some spherical and therefore only part of them fit to the proton shell gaps in fig . [ fig : spe ] . again the small pairing energies found at @xmath56 values 104 , 108 , 116 , 120 and 126 have to do with the low proton level density found in fig . [ fig : spe ] for those proton number . concerning the behavior for the different isotopes , for example for @xmath119 , from @xmath120 up to @xmath121 we have proton pairing energies around @xmath122 mev and from @xmath8 on around @xmath123 mev . the reason for this change is that in the first interval the nuclei are prolate and in the second either oblate or spherical . similar arguments apply for other cases . it is interesting to notice that with the exception of the few nuclei with @xmath124 none of the nuclei studied has absolute values of the neutron pairing energies larger than 9 mev . on the other hand , many of the analyzed nuclei have proton pairing energies much larger than 10 mev . ( color online ) @xmath21 values , theoretical values marked by dots are compared with experimental data @xcite . the different lines correspond to the indicated isotope ] in fig . [ fig : qalpha ] the @xmath58 values are plotted as a function of the neutron number [ see eq . ( [ eq : qalpha ] ) ] . the exact numerical values are also given in column 10 of table [ tab1 ] . the patterns displayed by the different isotopes are easily understood just looking at eq . ( [ eq : qalpha ] ) : minima appear when the mother nucleus are more bound than the average and maxima correspond to more bound daughter nuclei . thus we observe minima corresponding to the neutron numbers for which energy gaps appear in the single particle energies of fig . [ fig : spe ] , namely 162 , 178 and 182 - 184 . for the @xmath125 values @xmath126 and @xmath127 there is no structure because the energy gaps that one finds for these neutron numbers are not large enough to provide energy minima at those deformations , see column three of table [ tab1 ] . experimental values @xcite for some isotopes of the nuclei fm , no , rf , sg , hs , ds , 114 116 and 120 are also displayed in the figure . the agreement between theory and experiment is very satisfactory , in most of the cases we obtain a quantitative agreement and for the others at least the tendency is the right one . concerning the proton dependence we observe a similar situation , we find large energy spacing for the lines whose z numbers correspond to energy gaps in the single particle diagram of fig . [ fig : spe ] , for example , hs , 116 , 120 , etc . ( color online ) top panel : two neutron separation energies as a function of the neutron number . lower panel : two proton separation energies as a function of the proton number . the lines represent isotonic chains , the one at the lower left corner corresponds to n=150 , and the one closer to the upper right corner to a=190 . experimental data are taken from ref . ] in fig . [ fig : s2n ] , upper panel , we present the 2-neutron separation energy as a function of the neutron number ( see also column 11 of table [ tab1 ] ) . as expected we obtain a decreasing behavior of @xmath128 with increasing neutron number since we get closer to the neutron drip line . the smooth decline of @xmath128 is only disturbed at the neutron numbers corresponding to the single particle shell gaps 162 , 184 , etc . the two - proton separation energies are shown , in the lower panel of fig . [ fig : s2n ] , as a function of the proton number ( see also column 12 of table [ tab1 ] ) . the general behavior of decreasing @xmath129 for a given isotonic chain with growing z illustrates the fact that we are getting closer to the proton drip line . the fact that the @xmath129 presents less structure than the @xmath128 is obviously related with the fact that , in the region of interest , the shell gaps in fig . [ fig : spe ] are smaller for protons than for neutrons . [ fig : s2n ] also includes the available experimental values for some fm and no isotopes . we observe that in the case of @xmath128 the theoretical values are a around 1 mev smaller as compared to experiment while in the case of @xmath129 they are around 1 mev larger . in both cases , however , the trend is correctly described . this is the well known binding energy drift that takes place with the d1s parametrization of the gogny force for most isotopic chains . to correct this drift chappert et al @xcite have recently done a new parametrization of the gogny force , the d1n , which reduces considerably the drift but otherwise keeps the quality of the d1s parametrization . in table [ tab : d1snm ] we have included calculations with the d1n parametrization and we observe a considerable improvement in the agreement . we also include in the table the results of calculations with the d1 m parametrization of the gogny force obtained by including beyond mean field effects in the fit @xcite , which also do not present the mentioned drift . lldddd & @xmath97 & d1s & d1n & d1 m & exp + & 150 & 13.20 & 13.86 & 14.00 & 13.98 + & 152 & 12.32 & 13.01 & 13.14 & 13.40 + & 154 & 11.35 & 12.13 & 12.23 & 12.06 + & 156 & 10.77 & 11.64 & 11.73 & 11.56 + & 150 & 8.86 & 7.80 & 7.62 & 7.74 + & 152 & 9.93 & 8.79 & 8.59 & 8.93 + & 154 & 10.87 & 9.67 & 9.47 & 9.71 + & 156 & 11.74 & 10.50 & 10.31 & 10.43 + & 150 & 7.35 & 7.66 & 7.70 & 7.56 + & 152 & 7.12 & 7.49 & 7.54 & 7.15 + & 154 & 7.02 & 7.38 & 7.48 & 7.31 + & 156 & 6.66 & 6.99 & 7.10 & 7.03 + the d1n and d1 m parametrizations also improve slightly the agreement of @xmath21 with the experiment ( see table [ tab : d1snm ] ) , but this effect is much less pronounced than in the case of the separation energies as the influence of protons and neutrons cancel each other . the large size basis used in these calculations and the large number of nuclei studied prevents to make a systematic study with totally unrestricted symmetry - breaking wave functions . to clarify the effects of these restrictions we show in fig . [ 8panel3 ] , as an example , the behavior of several relevant quantities along the fission path for the nuclei @xmath130hs ( left panel ) , @xmath131ds ( middle panel ) and @xmath132cn ( right panel ) in different approximations . the panels ( d ) , in the middle row of fig . [ 8panel3 ] , display three different fission paths for each nucleus corresponding to the following constraints on the wave functions : 1.- the axially symmetric ( @xmath133 ) and reflexion symmetric ( @xmath134 ) fission path ( continuous line ) , which we shall call in the following as - rs path . 2.- the axially symmetric and non - reflexion symmetric ( @xmath135 ) path ( long - dashed lines ) , in the following as - nrs path . 3.- the non - axially symmetric ( @xmath136 ) and reflexion symmetric path ( short - dashed lines ) , in the following nas - rs path . in the ground state these nuclei are well quadrupole deformed , @xmath137 b , see panels ( d ) , and in the way to fission they have to tunnel through a barrier of several mev height . the barriers represent the potential energy needed to deform the nucleus . it is related , therefore , with the single particle levels around the fermi surface available at the corresponding deformation along the path . for the three nuclei the paths in the as - rs case present two hump barriers where the height of the second barrier decreases with the mass number while the first one remains more or less constant for @xmath130hs , @xmath131ds and increases for @xmath132cn . the width of the barrier in this approach is also similar for the three nuclei . the origin of the two humps can be easily understood looking at the single particle energies of fig . [ fig : spe ] . if we follow the neutron fermi level to the prolate side we find two regions with a clearly developed low level density . the first one at @xmath138 b with the shell gaps n=162 and z=108 corresponds to the ground state minimum . on the way from this point to larger deformations we find a larger level density region which corresponds to the energy increase of the first barrier . behind that we arrive at the second region at @xmath139 b , which corresponds to the super deformed minimum . at even larger deformations one finds a high level density region that provides the energy rise of the second barrier . finally above @xmath140 b , we observe intruder states of a high lying @xmath141i@xmath142 orbital which could be an indication of the scission point . incidentally , since the neutron shell gap is at n=162 and the proton one at z=112 the super - deformed minimum is deeper in @xmath132cn than in the lighter nuclei displayed in fig . [ 8panel3 ] . this behavior changes remarkably in the as - nrs approach . one can follow in panels ( f ) the portion of the paths where a lower solution with @xmath143 is found . as one can see in panels ( f ) the octupole degree of freedom plays an important role for quadrupole deformations starting around @xmath144 b , i.e. , close to the super deformed minima . as a matter of fact in this channel the second hump of the barriers diminishes strongly and the paths in this region look like the continuation of the first hump . the as - nrs path reaches up to large deformations , @xmath145 b for @xmath130hs , @xmath146 b for @xmath131ds and decreases to @xmath147 b for @xmath132cn . we can understand the onset of octupole deformation looking again at fig . [ fig : spe ] . here we observe that the @xmath148 and @xmath149 levels of the @xmath141k@xmath150 shell cross the fermi surface around 26 b , while the @xmath151 level crosses around 35 b. interestingly , at these @xmath22 values the @xmath151 level of the @xmath141h@xmath152 orbit also approaches the fermi surface . in fig . [ fig : spe ] and at zero deformations the @xmath141k@xmath150 subshell lies at about 0.4 mev and the @xmath141h@xmath152 at about @xmath153 mev . the k@xmath150 and h@xmath152 subshells interact strongly through the @xmath154 , i.e. , the octupole interaction . that means if we allow for reflexion symmetry breaking we can increase the quadrupole deformation at a lower energy cost @xcite . we can observe in fig . [ 8panel3 ] that around these values the as - nrs fission paths get lower in energy than the as - rs ones . in the ( g ) panels we can follow the behavior of the hexadecapole moment along the fission path . in the ground state it is close to zero in both approximations and from this point on it grows linearly with @xmath22 in the as - nrs approach . in the as - rs approach , however , first it increases linearly up to the scission point where a kink is observed followed by a linear increase . in the nas - rs path we can observe the effect of the triaxial shapes along the fission paths . in panels ( e ) we can see the two portions of the trajectory where triaxial solutions are found . the first one , close to the ground state , spans a smaller interval of @xmath22 values than the second one and does not have a large impact on the energy . the second one , as in the as - nrs case , is relevant around the second hump causing a significant lowering of its height , i.e. , in this case we still have to deal with two humped barriers . the width of the barrier , at variance with the as - nrs case , is more or less like the as - rs one . though the shape of the barriers is very relevant to calculate lifetimes one has to consider , however , that other quantities entering in the corresponding formula , eqs . ( [ tsf ] ) and ( [ action ] ) , do play an important role . a relevant parameter is the collective quadrupole inertia @xmath44 , since mass parameters are strongly influenced by pairing correlations and these itself by the single particle level density which we expect to vary along the different fission paths . in panels ( a ) and ( b ) of fig . [ 8panel3 ] we display the neutron and protons pairing energies , respectively . we indeed observe big differences in both of them along the portions of the paths where the symmetry breaking takes place . in particular we observe that the as - rs solutions provide always the largest pairing energies . as we can observe in panels ( c ) , where the @xmath44 s are plotted in the different approaches , in the relevant parts the as - rs provides the smaller masses followed by the as - nrs ones . this implies , that not necessarily the smallest fission barriers provide the shortest lifetimes @xcite . since the action @xmath40 in eq . ( [ action ] ) can be seen as a line integral of the function @xmath155 , the area of the surface delimited by this line and the x - axis provides the value of the action @xmath40 . in panel ( h ) we display @xmath156 in the three approaches . for the nucleus @xmath130hs we find that though the nas - rs fission path has a smaller fission barrier than the as - rs , the actions @xmath40 for both approaches are very close . the actual values of @xmath40 are 26.69 ( as - rs ) and 26.49 ( nas - rs ) , while in the as - nrs case we obtain a much larger value , namely 41.66 . in the nucleus @xmath131ds , the fission paths alone would predict that the as - nrs and the nas - rs approaches would provide much shorter lifetimes than the as - rs one . however , in panel ( h ) one finds that the three areas look rather similar . actually the precise numbers 25.88 ( as - rs ) , 26.97 ( as - nrs ) and 26.49 ( nas - rs ) show this to be the case . lastly , for @xmath132cn , the prediction of the fission paths is more or less in accordance with the one of panel ( h ) and the actual numbers 23.34 ( as - rs ) , 18.32 ( as - nrs ) and 22.52 ( nas - rs ) corroborate that . we can conclude that the restriction to axially symmetric paths is , in general , a good approximation , though as we we will see later one can find some exceptions . the low and short nrs barrier in nuclei with @xmath157 makes the most probable fission through octupole deformed shapes . in these nuclei we expect to find asymmetric mass distribution of fission fragments @xcite . in the one dimensional fission paths plotted in fig . [ 8panel3 ] we find crossings between the two paths giving the impression that one could switch from one path to the other without further problem . however , if we look in a higher dimensional plot one can see that this is not the case . to illustrate this point we have drawn in fig . [ q2-q3 ] potential energy contour lines versus the quadrupole , @xmath22 , and the octupole moment , @xmath33 , for the nuclei @xmath130hs and @xmath132cn . in this figure we can follow the as - rs and the as - nrs paths of fig . [ 8panel3 ] for the respective nucleus . the as - rs path corresponds to @xmath158 and goes along the x - axis and the as - nrs one goes along the bullets . it is interesting to see how the self - consistent path goes along a valley in both nuclei . we can also see that no alternative paths are present . in the @xmath130hs case we find that , at @xmath159 b where both paths seem to cross in fig . [ 8panel3](d ) , in reality both paths are separated by a 4 to 5 mev high barrier . in the following to perform a systematic description of the fission barriers of the 160 she s , we restrict ourselves to the axial approximation in which we have performed two kinds of calculations , namely the reflexion symmetric , as - rs , and the non - reflexion symmetric , as - nrs . all the fission barriers are presented in the series of figures [ fig : fb100 - 104]-[fig : fb124 - 126 ] . in fig . [ fig : fb100 - 104 ] we present the fission barriers for the isotopes of the elements fm , no and rf for quadrupole values from @xmath123 b up to @xmath160 b ( continuous line for as - rs and dashed for as - nrs ) . we first discuss the as - rs results . in panel ( a ) we present the fm results for neutron number 150 up to 166 . all isotopes present a well prolate deformed minimum around 15 b. in addition in the lighter isotopes a shallow superdeformed ( sd ) minimum appears around 50 b , at @xmath161 we find a very flat minimum and for the heavier isotopes no sd minimum is found . the common characteristic of these nuclei is the presence of a big broad barrier . for @xmath162 the barrier is centered at @xmath163 b , has a height of about 12 mev and a width of 18 b. with increasing neutron number the center of the barrier shifts to larger deformations and the height diminishes . for @xmath164 the center is around @xmath165 b and the height is about 8 mev . in the heavier isotopes we find some structure in the first barrier , namely the development of a shoulder around @xmath166 b with increasing neutron number . the presence of a sd minimum in the lighter isotopes drives the existence of a second barrier . since the minima are rather shallow the second barriers are broad but not high . these properties will contribute in general to a tendency of shorter lifetimes with increasing neutron number , though the particular behavior must be analyzed case by case . the fission paths for the no isotopes are shown in panel ( b ) for the same neutron numbers as the fm case . the structure of the first barriers look roughly like the ones in fm . an important point is the fact that the sd minimum is somewhat lower in energy than in the fm case and its role is therefore much less relevant . the results for the rf isotopes are displayed in panel ( c ) . the tendency observed in the no isotopes is reinforced , the sd minimum gets even deeper and the second barrier disappears for most isotopes . this fact will provide in general shorter lifetimes for rf than for no and for no shorter than for fm . these are general tendencies but since the lifetimes are very sensitive to small energy differences along the fission path , to make quantitative predictions calculations involving also collective inertia have to be performed . let us now describe the as - nrs results . for all isotopes of the three elements , the fission barriers are much larger along the non - reflexion symmetric paths than along the reflexion symmetric path . the lifetimes are therefore considerably longer in the as - nrs path than in the as - rs one . the fission in the as - nrs mode is completely impossible . an exhaustive discussion of the fission barriers of the nuclei @xmath167fm , @xmath168fm , @xmath169fm , @xmath169no , and @xmath170rf can be found in ref . @xcite . in these nuclei the elongated fission " mode can be observed . this mode is connected to octupole deformed fission paths which start at @xmath171 b. we have not observed the elongated fission " in heavier nuclei therefore we will not discuss it here . in fig . [ fig : fb106 - 110 ] we present the paths for the nuclei of sg , hs and ds elements . in panel ( a ) we display the results from @xmath162 up to @xmath121 for the sg isotopes . these nuclei have a trend somewhat different from the preceding ones : the sd minimum does not play a relevant role in the fission process since it is always deeper than the ground state and as a matter of fact if the sd minimum were not that shallow it would be the ground state for the lightest sg isotopes . furthermore the barriers get flatter , most of them being lower than 10 mev . apart from this feature the situation for the isotopes @xmath172 is similar to the nuclei fm , no and rf , i.e. , the ground states are prolate deformed ( @xmath173 b ) with large negative deformation energies and high barriers at larger @xmath22 values . for @xmath174 the situation changes very fast , and the following properties get reinforced as the neutron number increases : the ground states get less deformed , the shoulders around @xmath175 b get deeper and become real minima and as a consequence the original barrier becomes a two humped one . the height of the spherical maximum decreases and a coexisting oblate minimum develops at @xmath176 b. in particular for @xmath79 the oblate and prolate minima are degenerated . we therefore expect a strong reduction in the fission lifetime of heavier isotopes as compared to the lighter ones . in panel ( b ) the @xmath162 up to @xmath177 hs isotopes are shown . in the lighter isotopes the fission barriers are flatter and the sd minima deeper than in the corresponding sg isotones . the effect of the sd minima on the fission process is much smaller . furthermore the second fission barrier develops much earlier . since the single particle energies of fig . [ fig : spe ] are for @xmath67hs we can confront the general discussion of sect . [ sect : iva ] with the shape of the fission path in this nucleus . aside from these facts the hs isotopes from @xmath178 up to @xmath179 behave to a large extend like the corresponding isotones of the sg isotopes . for @xmath8 the oblate minimum becomes the ground state and the two humps of the fission barrier are very similar in size and height ( about 5 mev ) . with increasing number of neutrons , the prolate minimum shifts towards @xmath180 and the spherical minimum becomes the ground state at the @xmath63 shell closure . as a consequence the two humps separate from each other , the outer one shifting to larger @xmath22 values and the inner one to smaller ones . at the same time the inner barrier gets bigger and the outer smaller . from this behavior one expects the lifetimes for fission to be smaller around neutron number 170 - 174 . the nucleus @xmath181hs is octupole deformed in its ground state , see fig . [ fig : beta ] . in the self - consistent @xmath22-constrained calculations this nucleus remains weakly octupole deformed up to @xmath182 b where it turns reflexion symmetric ( see short dashed part of the fission path ) . since the fission fragments at the scission point are characteristic of a reflexion symmetric fission we still denote this mode as rs fission . the same situation is found with the @xmath82 isotones for the heavier she . the paths along @xmath183 close to the ground state are plotted with short dash lines in the corresponding figures . the ds results for the isotopes @xmath184 up to @xmath185 are displayed in panel ( c ) . the main characteristics of these fission paths are the following : for the lighter isotopes the fission barriers are flatter than for the corresponding isotones in hs and sg . for the light and medium mass isotopes , for a given isotone number , we find an increase of the first barrier moving from sg to hs and from this to ds . the opposite effect is observed for the second barrier , in particular for the very heavy isotopes this barrier disappears at the highest neutron number studied . concerning the as - nrs results of fig . [ fig : fb106 - 110 ] and for the sg isotopes , we find that though the as - nrs fission barriers are smaller than the as - rs ones for medium @xmath22 values , for larger ones they are much higher and as a result the non - reflexion symmetric fission paths are not favored as compared with the reflexion symmetric ones . for the hs isotopes the same behavior as the sg isotopes is observed : for lighter isotopes up - bending tails of the fission paths make the as - nrs paths very unfavored , but for medium - heavy n values the tails bend down and though the path is longer the second barrier is smaller for the as - nrs path than for the as - rs one , i.e. , around @xmath186 and above the as - nrs path may compete with the as - rs one ( cf . [ q2-q3 ] ) . for the heaviest hs isotopes in the as - rs approach the second barrier decreases considerably and the as - nrs barrier becomes longer than the other one . for the ds isotopes the as - nrs path is even more favorable because the down - bending tendency gets reinforced and we have some as - nrs paths which are clearly favored , for example for the nuclei with @xmath187 . for larger n values the vanishing of the second barrier in the as - rs case again favors this approach . notice that in general octupole effects set in for larger @xmath22 values as the neutron number increases . in fig . [ fig : fb112 - 116 ] we present the fission paths for the isotopes of the elements cn and @xmath89 and 116 . as before we first discuss the as - rs paths . in panel ( a ) , where the isotopes @xmath188 of cn are shown , we observe that compared with the corresponding isotones discussed before , the second hump of the barriers is lower and that the slopes of the tails of the fission paths are more pronounced . these facts point to shorter lifetimes of the cn isotopes as compared with lighter isotones . in panel ( b ) of the figure the @xmath89 isotopes are shown . the main difference with respect to the cn isotopes is the increase of the first hump and the decrease of the second one . in the isotopes with one hump barriers these are higher and a bit broader than for the corresponding isotones in cn . altogether , it seems that , in general , the lifetimes of the @xmath89 isotopes will be somewhat longer than the one at the corresponding @xmath119 isotones . in panel ( c ) we display the @xmath189 isotopes . here the same trend as in the previous nuclei is observed : a reinforcement of the tendency to increase the first hump of the double humped barriers and in the case of only one hump an increase of this . concerning the as - nrs results for cn and the @xmath89 and 116 elements , we find that the onset of octupolarity is energetically favored after the level crossing of the higher shells " and with the exception of the lightest isotopes the barriers are much smaller in the as - nrs path . this is due to the disappearance of the second hump of the barrier , i.e. , in the as - nrs path we have only one - humped barriers . we expect therefore a shortening of the fission lifetimes along these paths . in fig . [ fig : fb118 - 122 ] the results for the @xmath190 and 122 isotopes is shown . the same tendency as before is observed in the as - rs calculations , larger first barriers as @xmath56 increases . the role of the octupole degree of freedom is also relevant and all as - nrs fission paths do have smaller fission barriers . the same comments do also apply to the @xmath191 and 126 isotopes in fig . [ fig : fb124 - 126 ] . ( color online ) the dominant decay modes of she s . logarithm of the shortest half - lives ( in s ) is also indicated . ] one of the key issues in the theoretical description is the prediction of the decay modes and the half - lives of the she s . the agreement between the experimental data and the theoretical predictions states not only the quality of both of them but it may be also a criterion for the identification of a particular isotope synthesized in the corresponding fusion reaction . moreover , since the contemporary experimental techniques do not allow the detection of nuclei with half - lives shorter than @xmath192s one must estimate which isotopes can survive long enough to be detected and the way in which they disintegrates . therefore , the half - lives of the two main competing processes , spontaneous fission and @xmath1 emission , should be evaluated . the shortest half - life determines the dominant decay channel and the total half - life . if the branching ratio between two modes is equal to 50 % the logarithm of the total half - life would be smaller than the shorter partial half - life by not more than 0.3 . differences in half - lives between two modes of one order of magnitude lead to a logarithm of the total half - life only 0.05 smaller than the logarithm of the half - life of the fastest decay . the half - lives for @xmath1 decay and spontaneous fission calculated in the hfb theory are collected in the last two columns of table [ tab1 ] . for an easier analysis these data are also presented in fig . [ fig : t1/2 ] where the isotopic chain of each element is shown in a separate panel . in some nuclei two fission half - lives can be calculated along paths leading to distinct fragment mass asymmetry . both solutions are depicted in fig . [ fig : t1/2 ] and discussed below , however , in table i only the shorter half - life of the dominant mode is given . the available experimental data are also plotted in fig . [ fig : t1/2 ] . a first look into the panels of fig . [ fig : t1/2 ] reveals similar tendencies for the different isotopic chains . it is easy to distinguish the intervals of neutron number where common features are characteristic for many elements despite of differences in the absolute values . therefore we will discuss our results collected in groups of similar neutron number starting from the lighter ones . as we have seen in the discussion of the precedent section nrs effects influence the fission paths , and thereby the fission half - lives , in three @xmath22 regions : 1.- for small @xmath22 values , @xmath193 b , they affect nuclei with octupole deformed ground states , i.e. nuclei with @xmath194 . 2.- starting at medium ( @xmath195 b ) and extending up to large @xmath22 values , these effects are present in all nuclei . 3.- for larger @xmath22 values , @xmath196 b , they appear in the light isotopes of the elements fm , no , rf and sg . according to these effects we are using three symbols in fig . [ fig : t1/2 ] for the fission half - lives depending on the paths used in the calculations : as - rs ( triangle - up ) includes the genuine as - rs ones plus those with octupole effects close to the ground state . the reason to include the latter under a `` as - rs '' denomination is that at @xmath197 b the nucleus takes as - rs shape and the fission takes place exactly in the same way as in the pure as - rs case with symmetric fragment mass distribution . as - nrs ( triangle - down ) denotes the real " nrs paths leading to fission with fragments of different masses , which correspond to the long dashed lines in the fission barrier plots . as - rs / nrs ( star ) labels the paths where the first part is as - rs and only for @xmath196 b one follows the nas branch . the first characteristic region covers the lighter isotopes with @xmath198 , clearly delimited by the @xmath105 deformed `` shell closure '' . in this range of @xmath97 the half - lives for @xmath1 emission increase monotonically with the neutron number in all elements from fm to @xmath89 . it can be also noticed that an increase of the proton number by two units leads to the decrease of the @xmath1 half - lives by two or three orders of magnitude . as we will also see for the other regions , this tendency is a direct consequence of the calculated values of @xmath21 shown in fig . [ fig : qalpha ] and is consistent with well - known properties of @xmath1 decay in heavy nuclei . the three lightest fm and no isotopes are a specific group of nuclei in which the hfb calculations predict relatively long fission half - lives . the small fission probability is the consequence of the shape of the fission barrier in these nuclei which extends to large quadrupole deformation . in the as - rs path the second barrier extends up to 120 b providing very long half - lives . a somewhat shorter half - life is obtained in the as - nrs approach where the barrier is constructed from two humps : the first a rs and the second a nrs one starting at @xmath199 b ( fig . [ fig : fb100 - 104 ] ) . in the slightly heavier isotopes the second nrs barrier also can be noticed in the pes ( fig . [ fig : fb106 - 110 ] ) , but in contrast to the group of lighter fm and no nuclei the second , super deformed minimum has an energy below the ground state . thus the second hump does not affect the barrier tunneling but it governs the fragment mass asymmetry . in these nuclei the reduction of the width of the barrier leads to a decrease of the half - lives by a few orders of magnitude what can be seen in the fm and no isotopes with @xmath200 as well as the rf and the sg isotopes . all nuclei with the second nrs hump leading to the so - called elongated fission " @xcite ( i.e. fm and no with @xmath201 as well as rf and sg with @xmath202 ) are marked in fig . [ fig : t1/2 ] by the blue stars . in the heavier isotopes of the @xmath198 interval , where only the single rs barrier remains , one observes a rise of the fission half - lives with the neutron number . the slope of this trend changes from an almost flat dependence in fm to a very steep one in sg , ds , cn , and @xmath89 . this trend is caused by a broadening of the barrier that eventually becomes a two humped one ( figs . [ fig : fb100 - 104 ] , [ fig : fb106 - 110 ] , [ fig : fb112 - 116 ] ) . the local maximum found in all elements in the partial half - lives at @xmath105 with respect to @xmath1 and fission decays indicates the special character of this neutron number in the chart of nuclides , see also fig . [ fig : spe ] . these isotones are the most stable in the close vicinity , hence , in many papers @xmath105 is called deformed magic number " . this name stresses the significant difference from the classical magic nuclei which are spherical in their ground states . many experimental data coming from the cold fusion " reactions are available in the region of isotopes with @xmath203 @xcite . the agreement of the @xmath1 decay half - lives with the theoretical predictions is noticeable although some discrepancies are observed . most calculated fission half - lives overestimate the experimental data . the agreement would be better if triaxial effects were taken into account in the saddle point of the first barrier . the consideration of the @xmath204-deformation in the calculations of nuclei in the fm region reduces the barrier heights by around 2 mev decreasing thereby the theoretical fission half - lives by around 2 orders of magnitude @xcite . at variance with the discussion of sect . [ sect : iva ] in these nuclei this reduction is not fully compensated by an increase of the inertia parameters . therefore discrepancies between the theory and the experiment diminishes considerably . it should be pointed out that the fastest decay is properly predicted in all cases . furthermore , when the experimental data for @xmath1 emission and fission provide comparable decay probability for both processes the theoretical predictions are also similar for both partial half - lives . almost all nuclei in this region do have half - lives long enough to be considered as of experimental interest . the next region covers nuclei with neutron number from @xmath110 to @xmath112 . at @xmath110 a kink of 2 - 4 orders of magnitude in the @xmath1 decay half - lives can be observed corresponding to a local maximum in @xmath21 , see fig . [ fig : qalpha ] . the increase of the neutron number for a given isotope causes a linear growth in the @xmath1 half - life following the tendency already observed in the lighter nuclei . the half - life rise up to the neutron number @xmath112 where the locally longest living isotopes are found , see also fig . [ fig : qalpha ] . they reach values from @xmath206 s in @xmath207 to @xmath208 s in hs . again , the half - lives calculated for heavier isotones are smaller and , consequently , the @xmath1 emission becomes the most probable decay channel in the proton rich nuclei . the as - rs fission half - lives for nuclei with @xmath209 behave completely different than those for @xmath198 . after passing the maximum at @xmath105 they decrease up to the local minimum at @xmath186 . the drop is very steep for the lighter elements ( rf , sg , hs ) and gentler for the heavier ones ( ds , cn ) , whereas for the isotopes of @xmath89 almost no change is observed . this behavior can be explained by the decreasing potential energy along almost the whole barrier , associated with a diminishing of the saddle point energy and the clear development of the second minimum ( figs . [ fig : fb100 - 104 ] , [ fig : fb106 - 110 ] , [ fig : fb112 - 116 ] ) . the inverse trend of increasing the potential energy at small deformations is noticed in heavier elements , starting from cn ( fig . [ fig : fb112 - 116 ] ) . the first rs hump of the barrier grows up with increasing mass of a nucleus . in the heavy isotopes with @xmath210 it is higher than the second rs barrier . the half - lives calculated for @xmath211 along the nrs path diminish more rapidly with the increase on @xmath97 than in the rs mode . this is induced by the changes that take place in the nrs barrier , namely the fast decrease of its height and , more important , the narrowing of its width , see figs . [ fig : fb106 - 110 ] , [ fig : fb112 - 116 ] . in nuclei with @xmath212 the nrs fission barrier allows to avoid the second hump of the rs barrier making the asymmetric fission the most probable mode . very short fission half - lives , even below @xmath213 ms , can be found in the hs , ds , cn and @xmath89 isotopes , where the nrs fission is the dominant decay mode being even faster than the @xmath1 emission . in @xmath89 the nrs fission has half - lives comparable to the @xmath1 decay . the available experimental data around @xmath79 obtained from the hot fusion " experiments @xcite , fit perfectly to the theoretical predictions . in cn two spontaneous fission half - lives correspond to the prediction of the nrs fission mode . in @xmath214 the observed 50% branching ratio is very well reproduced . in the isotope with two more neutrons the detected @xmath1 decay is predicted with only slightly longer half - life than in the dominant fission channel . finally , in @xmath189 and @xmath207 the dominant @xmath1 radioactivity is properly predicted by the theoretical analysis with a good estimation of the half - lives . while approaching @xmath112 the fission half - lives increase with a slope which grows with the proton number . this effect is governed by two trends observed in the evolution of the pes s . the first one is the aforementioned growth of the first hump of the barrier which is the highest one in almost all nuclei with @xmath215 . the other factor is the lowering of the energy of the oblate minimum which becomes the ground state in the nuclei around @xmath112 ( cf . discussion of ground - state deformations in sec . [ ssgsp ] ) . the shift of the ground state from the prolate to the oblate minimum , see also fig . [ fig : spe ] , gives an additional increase of the barrier height up to over 1 mev . the energy difference between prolate and oblate minima has its largest value in the @xmath112 isotones what , added to the large first barrier , produces extremely long half - lives . they exceed ( in the nrs mode ) @xmath216 s in cn and @xmath89 and @xmath217 s in @xmath88 . in the neutron rich hs isotopes the nrs barriers extend up to relatively large values of the quadrupole moment . therefore their transition probabilities are smaller than in the rs mode . this fact together with the long @xmath1 decay half - lifes implies that in this region one can find isotopes with very long half - lives . the rs fission is the dominant decay mode . however these very neutron rich hs isotopes are extremely difficult to synthesize using contemporary experimental techniques . the saddle point of the first hump of the barrier takes its maximal energy value at @xmath81 . we can observe in these isotones another very high maximum of the fission half - lives in all elements . in contrast , the @xmath219 isotones decay a few orders of magnitude faster than the neighboring nuclei with @xmath112 and @xmath81 . the influence of the oblate minimum and the high first barrier is not strong enough to enlarge the fission half - lives here . in two isotones with @xmath219 , namely cn and @xmath89 , the nrs fission is the dominant process with half - lives shorter than for the @xmath1 decay . the energy difference between the oblate and the prolate minima as well as between the absolute values of their quadrupole moments shrink in @xmath219 and @xmath81 continuously . finally , one founds that all @xmath63 isotones do have a spherical ground state . this indicates a magic number at @xmath63 . nevertheless , this feature does not have a big impact on the fission half - lives and the region of the most stable nuclei is slightly shifted towards the neutron deficient isotopes . at @xmath112 and @xmath81 the @xmath1 decay half - lives also reach their maxima although they are less pronounced than for the fission half - lives . they correspond to the minima of @xmath21 that can be observed for these neutron numbers , see fig . [ fig : qalpha ] . for the @xmath220 elements the @xmath1 emission is the fastest decay process for isotopes lighter than @xmath63 . most of them live long enough to allow the synthesis of these nuclei . the extremely long fission half - lives for the @xmath112 and @xmath81 isotopes are larger than the @xmath1 decay half - lives in cn and @xmath89 whereas for @xmath219 nrs fission is the dominant mode . the isotopes with neutron number larger than the magic @xmath63 are characterized by a rapid decrease of the fission half - lives with increasing neutron number . this is a consequence of the decrease of energy along the whole energy barrier . the second minimum goes down below the ground state and the height of the first barrier reduces substantially when heavier nuclei are considered . these strong trends can not be balanced by the few mev decrease of the ground state energy due to the octupole deformation . consequently , for the elements from ds to @xmath88 , fission becomes the dominant decay mode with half - lives below @xmath192s . the @xmath1 decay half - lives do not vary strongly along the isotopic chain in this region . the fluctuations are associated with the changes of the ground state deformations of the parent and the daughter nuclei . in this region the decrease of the @xmath1 decay half - lives with the proton number is also visible . most isotopes in this region can not be synthesized due to the very short fission ( in the proton deficient nuclei ) or @xmath1 emission ( in the proton rich isotones ) half - lives . the experimental limit of @xmath221 ms is exceeded only for a few nuclei . to conclude this section we would like to present in fig . [ fig : t1/2_main ] the shortest half - lives of each isotope in the form of the chart of she s . from this figure it is easy to distinguish the regions where each decay mode plays the most important role . the predominant decay mode , specially in the proton rich region , is the @xmath1 emission . roughly speaking , the spontaneous fission in the rs mode is dominant in the proton deficient nuclei with @xmath222 for @xmath223 , @xmath224 for @xmath157 , and @xmath225 for @xmath226 . in the central part of the diagram the region with the fastest decay in the nrs fission channel is defined by @xmath227 and @xmath228 . two regions of long living nuclei can be found also in fig . [ fig : t1/2_main ] . the first one includes nuclei around @xmath229sg@xmath230 where half - lives reach @xmath231 s. the other island of stability " is associated with the anomalous long fission half - lives at @xmath112 and @xmath81 . at these neutron numbers ( and also at @xmath63 ) long living isotopes of cn and @xmath89 elements can be found . the half - lives of these isotopes are also longer than @xmath232 s. very long half - lives characterize also the hs isotopes decaying through the rs fission . in this work the hatree - fock - bogoliubov theory with large basis size and the density dependent gogny force as interaction have been used to study the most relevant properties of 160 heavy and super - heavy elements as well as their predominant decay modes . in order to keep the calculations as general as possible we allow for wave functions with different symmetries , namely axially symmetric and reflexion symmetric , axially symmetric and non reflexion symmetric and lastly triaxial wave functions . after a thorough analysis along different fission paths and considering that our calculations are extremely cpu time demanding due to the large basis to study 160 nuclei we perform most calculations in the axially symmetric approaches . the ground state deformations @xmath233 and @xmath102 as well as @xmath21 factors , pairing properties and the two neutron and two proton separation energies are thoroughly discussed . the single particle energies are used as a guide for the interpretation of these properties . the fission paths for three representative nuclei , namely @xmath130hs , @xmath131ds and @xmath132cn , are analyzed in detail . properties like mass parameters , pairing features and the variation of the action among others are calculated along the fission path with wave functions of the three types mentioned above . we find that though the shape of the fission barrier has a large impact on the fission half - lives , the mass parameter also plays an important role . in such a way that wave functions with larger barriers and smaller mass parameter may tunnel easier than alternative ones with smaller barriers and larger mass . since in general axially symmetric wave functions do have smaller masses than the triaxial ones , the restriction to axial symmetry is a good option to perform a systematic study of the half - lives of she . in the second part of the paper a thorough study of the shapes of the barriers in the as - rs and as - nrs is performed . a reasonable explanation of heights and shapes as well as of octupole effects is obtained on the basis of the single particle energy levels . the two dimensional ( @xmath234 ) energy contour plots for the nuclei @xmath130hs and @xmath132cn allow to disentangle the different fission paths and its possible interconnections . in the third part of the paper the half - lives of all studied nuclei are calculated for the different decay modes , namely @xmath1-decay and along the different fission paths . we find clear tendencies with the neutron number easily explainable on the basis of the behavior of @xmath21 factors and barrier shapes , respectively . in particular , we find that the @xmath1 emission is the predominant decay mode , specially in the proton rich region . concerning the spontaneous fission we obtain that the rs mode is dominant for the proton deficient nuclei in medium mass she s with @xmath222 for @xmath223 , @xmath224 for @xmath157 , and in the region of the heaviest she s with @xmath225 for @xmath226 . the fastest decay in the nrs fission channel takes place for @xmath227 and @xmath228 . the long living nuclei can be found two regions . the first one is in the vicinity of @xmath229sg@xmath230 where half - lives around @xmath231 s are obtained . the anomalous long fission half - lives for proton deficient isotones with @xmath112 , @xmath81 and @xmath63 create the second region . at these neutron numbers several isotopes of the elements cn and @xmath89 are found with half - lives longer than @xmath232 s. very long half - lives characterize also the hs isotopes decaying through the rs fission . the nuclei beyond @xmath88 and @xmath63 have half - lives too short to be detected within contemporary experimental limit of @xmath192s . in conclusion we have presented a systematic study of she within the self - consistent hfb with very general wave functions and large configuration space . our calculations provide an overall interpretation of the systematics and global properties of these elements and its decay modes . in general we find a reasonable good agreement with the experimental data . the authors would like to thank prof . krzysztof pomorski for a careful reading of the manuscript . contribution of luis m. robledo to this paper is gratefully acknowledged . work partially supported by grant no . dec-2011/01/b / st2/03667 from the national science centre ( poland ) and by the spanish ministerio de educacin y ciencia under contracts fpa2009 - 13377-c02 - 01 and fpa2011 - 29854-c04 - 04 and the spanish consolider - ingenio 2010 programme cpan ( csd2007 - 00042 ) .